Clang Project

clang_source_code/lib/CodeGen/CGBuiltin.cpp
1//===---- CGBuiltin.cpp - Emit LLVM Code for builtins ---------------------===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This contains code to emit Builtin calls as LLVM code.
10//
11//===----------------------------------------------------------------------===//
12
13#include "CGCXXABI.h"
14#include "CGObjCRuntime.h"
15#include "CGOpenCLRuntime.h"
16#include "CGRecordLayout.h"
17#include "CodeGenFunction.h"
18#include "CodeGenModule.h"
19#include "ConstantEmitter.h"
20#include "TargetInfo.h"
21#include "clang/AST/ASTContext.h"
22#include "clang/AST/Decl.h"
23#include "clang/AST/OSLog.h"
24#include "clang/Basic/TargetBuiltins.h"
25#include "clang/Basic/TargetInfo.h"
26#include "clang/CodeGen/CGFunctionInfo.h"
27#include "llvm/ADT/SmallPtrSet.h"
28#include "llvm/ADT/StringExtras.h"
29#include "llvm/IR/DataLayout.h"
30#include "llvm/IR/InlineAsm.h"
31#include "llvm/IR/Intrinsics.h"
32#include "llvm/IR/MDBuilder.h"
33#include "llvm/Support/ConvertUTF.h"
34#include "llvm/Support/ScopedPrinter.h"
35#include "llvm/Support/TargetParser.h"
36#include <sstream>
37
38using namespace clang;
39using namespace CodeGen;
40using namespace llvm;
41
42static
43int64_t clamp(int64_t Value, int64_t Low, int64_t High) {
44  return std::min(High, std::max(Low, Value));
45}
46
47/// getBuiltinLibFunction - Given a builtin id for a function like
48/// "__builtin_fabsf", return a Function* for "fabsf".
49llvm::Constant *CodeGenModule::getBuiltinLibFunction(const FunctionDecl *FD,
50                                                     unsigned BuiltinID) {
51  assert(Context.BuiltinInfo.isLibFunction(BuiltinID));
52
53  // Get the name, skip over the __builtin_ prefix (if necessary).
54  StringRef Name;
55  GlobalDecl D(FD);
56
57  // If the builtin has been declared explicitly with an assembler label,
58  // use the mangled name. This differs from the plain label on platforms
59  // that prefix labels.
60  if (FD->hasAttr<AsmLabelAttr>())
61    Name = getMangledName(D);
62  else
63    Name = Context.BuiltinInfo.getName(BuiltinID) + 10;
64
65  llvm::FunctionType *Ty =
66    cast<llvm::FunctionType>(getTypes().ConvertType(FD->getType()));
67
68  return GetOrCreateLLVMFunction(Name, Ty, D, /*ForVTable=*/false);
69}
70
71/// Emit the conversions required to turn the given value into an
72/// integer of the given size.
73static Value *EmitToInt(CodeGenFunction &CGF, llvm::Value *V,
74                        QualType T, llvm::IntegerType *IntType) {
75  V = CGF.EmitToMemory(V, T);
76
77  if (V->getType()->isPointerTy())
78    return CGF.Builder.CreatePtrToInt(V, IntType);
79
80  getType() == IntType", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 80, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(V->getType() == IntType);
81  return V;
82}
83
84static Value *EmitFromInt(CodeGenFunction &CGF, llvm::Value *V,
85                          QualType T, llvm::Type *ResultType) {
86  V = CGF.EmitFromMemory(V, T);
87
88  if (ResultType->isPointerTy())
89    return CGF.Builder.CreateIntToPtr(V, ResultType);
90
91  getType() == ResultType", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 91, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(V->getType() == ResultType);
92  return V;
93}
94
95/// Utility to insert an atomic instruction based on Intrinsic::ID
96/// and the expression node.
97static Value *MakeBinaryAtomicValue(
98    CodeGenFunction &CGF, llvm::AtomicRMWInst::BinOp Kind, const CallExpr *E,
99    AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) {
100  QualType T = E->getType();
101  getArg(0)->getType()->isPointerType()", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 101, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(E->getArg(0)->getType()->isPointerType());
102  getArg(0)->getType()->getPointeeType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 103, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(CGF.getContext().hasSameUnqualifiedType(T,
103getArg(0)->getType()->getPointeeType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 103, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">                                  E->getArg(0)->getType()->getPointeeType()));
104  getArg(1)->getType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 104, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()));
105
106  llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0));
107  unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace();
108
109  llvm::IntegerType *IntType =
110    llvm::IntegerType::get(CGF.getLLVMContext(),
111                           CGF.getContext().getTypeSize(T));
112  llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace);
113
114  llvm::Value *Args[2];
115  Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType);
116  Args[1] = CGF.EmitScalarExpr(E->getArg(1));
117  llvm::Type *ValueType = Args[1]->getType();
118  Args[1] = EmitToInt(CGF, Args[1], T, IntType);
119
120  llvm::Value *Result = CGF.Builder.CreateAtomicRMW(
121      Kind, Args[0], Args[1], Ordering);
122  return EmitFromInt(CGF, Result, T, ValueType);
123}
124
125static Value *EmitNontemporalStore(CodeGenFunction &CGF, const CallExpr *E) {
126  Value *Val = CGF.EmitScalarExpr(E->getArg(0));
127  Value *Address = CGF.EmitScalarExpr(E->getArg(1));
128
129  // Convert the type of the pointer to a pointer to the stored type.
130  Val = CGF.EmitToMemory(Val, E->getArg(0)->getType());
131  Value *BC = CGF.Builder.CreateBitCast(
132      Address, llvm::PointerType::getUnqual(Val->getType()), "cast");
133  LValue LV = CGF.MakeNaturalAlignAddrLValue(BC, E->getArg(0)->getType());
134  LV.setNontemporal(true);
135  CGF.EmitStoreOfScalar(Val, LV, false);
136  return nullptr;
137}
138
139static Value *EmitNontemporalLoad(CodeGenFunction &CGF, const CallExpr *E) {
140  Value *Address = CGF.EmitScalarExpr(E->getArg(0));
141
142  LValue LV = CGF.MakeNaturalAlignAddrLValue(Address, E->getType());
143  LV.setNontemporal(true);
144  return CGF.EmitLoadOfScalar(LV, E->getExprLoc());
145}
146
147static RValue EmitBinaryAtomic(CodeGenFunction &CGF,
148                               llvm::AtomicRMWInst::BinOp Kind,
149                               const CallExpr *E) {
150  return RValue::get(MakeBinaryAtomicValue(CGF, Kind, E));
151}
152
153/// Utility to insert an atomic instruction based Intrinsic::ID and
154/// the expression node, where the return value is the result of the
155/// operation.
156static RValue EmitBinaryAtomicPost(CodeGenFunction &CGF,
157                                   llvm::AtomicRMWInst::BinOp Kind,
158                                   const CallExpr *E,
159                                   Instruction::BinaryOps Op,
160                                   bool Invert = false) {
161  QualType T = E->getType();
162  getArg(0)->getType()->isPointerType()", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 162, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(E->getArg(0)->getType()->isPointerType());
163  getArg(0)->getType()->getPointeeType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 164, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(CGF.getContext().hasSameUnqualifiedType(T,
164getArg(0)->getType()->getPointeeType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 164, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">                                  E->getArg(0)->getType()->getPointeeType()));
165  getArg(1)->getType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 165, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(CGF.getContext().hasSameUnqualifiedType(T, E->getArg(1)->getType()));
166
167  llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0));
168  unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace();
169
170  llvm::IntegerType *IntType =
171    llvm::IntegerType::get(CGF.getLLVMContext(),
172                           CGF.getContext().getTypeSize(T));
173  llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace);
174
175  llvm::Value *Args[2];
176  Args[1] = CGF.EmitScalarExpr(E->getArg(1));
177  llvm::Type *ValueType = Args[1]->getType();
178  Args[1] = EmitToInt(CGF, Args[1], T, IntType);
179  Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType);
180
181  llvm::Value *Result = CGF.Builder.CreateAtomicRMW(
182      Kind, Args[0], Args[1], llvm::AtomicOrdering::SequentiallyConsistent);
183  Result = CGF.Builder.CreateBinOp(Op, Result, Args[1]);
184  if (Invert)
185    Result = CGF.Builder.CreateBinOp(llvm::Instruction::Xor, Result,
186                                     llvm::ConstantInt::get(IntType, -1));
187  Result = EmitFromInt(CGF, Result, T, ValueType);
188  return RValue::get(Result);
189}
190
191/// Utility to insert an atomic cmpxchg instruction.
192///
193/// @param CGF The current codegen function.
194/// @param E   Builtin call expression to convert to cmpxchg.
195///            arg0 - address to operate on
196///            arg1 - value to compare with
197///            arg2 - new value
198/// @param ReturnBool Specifies whether to return success flag of
199///                   cmpxchg result or the old value.
200///
201/// @returns result of cmpxchg, according to ReturnBool
202///
203/// Note: In order to lower Microsoft's _InterlockedCompareExchange* intrinsics
204/// invoke the function EmitAtomicCmpXchgForMSIntrin.
205static Value *MakeAtomicCmpXchgValue(CodeGenFunction &CGF, const CallExpr *E,
206                                     bool ReturnBool) {
207  QualType T = ReturnBool ? E->getArg(1)->getType() : E->getType();
208  llvm::Value *DestPtr = CGF.EmitScalarExpr(E->getArg(0));
209  unsigned AddrSpace = DestPtr->getType()->getPointerAddressSpace();
210
211  llvm::IntegerType *IntType = llvm::IntegerType::get(
212      CGF.getLLVMContext(), CGF.getContext().getTypeSize(T));
213  llvm::Type *IntPtrType = IntType->getPointerTo(AddrSpace);
214
215  Value *Args[3];
216  Args[0] = CGF.Builder.CreateBitCast(DestPtr, IntPtrType);
217  Args[1] = CGF.EmitScalarExpr(E->getArg(1));
218  llvm::Type *ValueType = Args[1]->getType();
219  Args[1] = EmitToInt(CGF, Args[1], T, IntType);
220  Args[2] = EmitToInt(CGF, CGF.EmitScalarExpr(E->getArg(2)), T, IntType);
221
222  Value *Pair = CGF.Builder.CreateAtomicCmpXchg(
223      Args[0], Args[1], Args[2], llvm::AtomicOrdering::SequentiallyConsistent,
224      llvm::AtomicOrdering::SequentiallyConsistent);
225  if (ReturnBool)
226    // Extract boolean success flag and zext it to int.
227    return CGF.Builder.CreateZExt(CGF.Builder.CreateExtractValue(Pair, 1),
228                                  CGF.ConvertType(E->getType()));
229  else
230    // Extract old value and emit it using the same type as compare value.
231    return EmitFromInt(CGF, CGF.Builder.CreateExtractValue(Pair, 0), T,
232                       ValueType);
233}
234
235/// This function should be invoked to emit atomic cmpxchg for Microsoft's
236/// _InterlockedCompareExchange* intrinsics which have the following signature:
237/// T _InterlockedCompareExchange(T volatile *Destination,
238///                               T Exchange,
239///                               T Comparand);
240///
241/// Whereas the llvm 'cmpxchg' instruction has the following syntax:
242/// cmpxchg *Destination, Comparand, Exchange.
243/// So we need to swap Comparand and Exchange when invoking
244/// CreateAtomicCmpXchg. That is the reason we could not use the above utility
245/// function MakeAtomicCmpXchgValue since it expects the arguments to be
246/// already swapped.
247
248static
249Value *EmitAtomicCmpXchgForMSIntrin(CodeGenFunction &CGF, const CallExpr *E,
250    AtomicOrdering SuccessOrdering = AtomicOrdering::SequentiallyConsistent) {
251  getArg(0)->getType()->isPointerType()", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 251, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(E->getArg(0)->getType()->isPointerType());
252  getType(), E->getArg(0)->getType()->getPointeeType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 253, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(CGF.getContext().hasSameUnqualifiedType(
253getType(), E->getArg(0)->getType()->getPointeeType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 253, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">      E->getType(), E->getArg(0)->getType()->getPointeeType()));
254  getType(), E->getArg(1)->getType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 255, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),
255getType(), E->getArg(1)->getType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 255, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">                                                 E->getArg(1)->getType()));
256  getType(), E->getArg(2)->getType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 257, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(CGF.getContext().hasSameUnqualifiedType(E->getType(),
257getType(), E->getArg(2)->getType())", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 257, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">                                                 E->getArg(2)->getType()));
258
259  auto *Destination = CGF.EmitScalarExpr(E->getArg(0));
260  auto *Comparand = CGF.EmitScalarExpr(E->getArg(2));
261  auto *Exchange = CGF.EmitScalarExpr(E->getArg(1));
262
263  // For Release ordering, the failure ordering should be Monotonic.
264  auto FailureOrdering = SuccessOrdering == AtomicOrdering::Release ?
265                         AtomicOrdering::Monotonic :
266                         SuccessOrdering;
267
268  auto *Result = CGF.Builder.CreateAtomicCmpXchg(
269                   Destination, Comparand, Exchange,
270                   SuccessOrdering, FailureOrdering);
271  Result->setVolatile(true);
272  return CGF.Builder.CreateExtractValue(Result, 0);
273}
274
275static Value *EmitAtomicIncrementValue(CodeGenFunction &CGF, const CallExpr *E,
276    AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) {
277  getArg(0)->getType()->isPointerType()", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 277, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(E->getArg(0)->getType()->isPointerType());
278
279  auto *IntTy = CGF.ConvertType(E->getType());
280  auto *Result = CGF.Builder.CreateAtomicRMW(
281                   AtomicRMWInst::Add,
282                   CGF.EmitScalarExpr(E->getArg(0)),
283                   ConstantInt::get(IntTy, 1),
284                   Ordering);
285  return CGF.Builder.CreateAdd(Result, ConstantInt::get(IntTy, 1));
286}
287
288static Value *EmitAtomicDecrementValue(CodeGenFunction &CGF, const CallExpr *E,
289    AtomicOrdering Ordering = AtomicOrdering::SequentiallyConsistent) {
290  getArg(0)->getType()->isPointerType()", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 290, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(E->getArg(0)->getType()->isPointerType());
291
292  auto *IntTy = CGF.ConvertType(E->getType());
293  auto *Result = CGF.Builder.CreateAtomicRMW(
294                   AtomicRMWInst::Sub,
295                   CGF.EmitScalarExpr(E->getArg(0)),
296                   ConstantInt::get(IntTy, 1),
297                   Ordering);
298  return CGF.Builder.CreateSub(Result, ConstantInt::get(IntTy, 1));
299}
300
301// Build a plain volatile load.
302static Value *EmitISOVolatileLoad(CodeGenFunction &CGF, const CallExpr *E) {
303  Value *Ptr = CGF.EmitScalarExpr(E->getArg(0));
304  QualType ElTy = E->getArg(0)->getType()->getPointeeType();
305  CharUnits LoadSize = CGF.getContext().getTypeSizeInChars(ElTy);
306  llvm::Type *ITy =
307      llvm::IntegerType::get(CGF.getLLVMContext(), LoadSize.getQuantity() * 8);
308  Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo());
309  llvm::LoadInst *Load = CGF.Builder.CreateAlignedLoad(Ptr, LoadSize);
310  Load->setVolatile(true);
311  return Load;
312}
313
314// Build a plain volatile store.
315static Value *EmitISOVolatileStore(CodeGenFunction &CGF, const CallExpr *E) {
316  Value *Ptr = CGF.EmitScalarExpr(E->getArg(0));
317  Value *Value = CGF.EmitScalarExpr(E->getArg(1));
318  QualType ElTy = E->getArg(0)->getType()->getPointeeType();
319  CharUnits StoreSize = CGF.getContext().getTypeSizeInChars(ElTy);
320  llvm::Type *ITy =
321      llvm::IntegerType::get(CGF.getLLVMContext(), StoreSize.getQuantity() * 8);
322  Ptr = CGF.Builder.CreateBitCast(Ptr, ITy->getPointerTo());
323  llvm::StoreInst *Store =
324      CGF.Builder.CreateAlignedStore(Value, Ptr, StoreSize);
325  Store->setVolatile(true);
326  return Store;
327}
328
329// Emit a simple mangled intrinsic that has 1 argument and a return type
330// matching the argument type.
331static Value *emitUnaryBuiltin(CodeGenFunction &CGF,
332                               const CallExpr *E,
333                               unsigned IntrinsicID) {
334  llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
335
336  Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
337  return CGF.Builder.CreateCall(F, Src0);
338}
339
340// Emit an intrinsic that has 2 operands of the same type as its result.
341static Value *emitBinaryBuiltin(CodeGenFunction &CGF,
342                                const CallExpr *E,
343                                unsigned IntrinsicID) {
344  llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
345  llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
346
347  Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
348  return CGF.Builder.CreateCall(F, { Src0, Src1 });
349}
350
351// Emit an intrinsic that has 3 operands of the same type as its result.
352static Value *emitTernaryBuiltin(CodeGenFunction &CGF,
353                                 const CallExpr *E,
354                                 unsigned IntrinsicID) {
355  llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
356  llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
357  llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2));
358
359  Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
360  return CGF.Builder.CreateCall(F, { Src0, Src1, Src2 });
361}
362
363// Emit an intrinsic that has 1 float or double operand, and 1 integer.
364static Value *emitFPIntBuiltin(CodeGenFunction &CGF,
365                               const CallExpr *E,
366                               unsigned IntrinsicID) {
367  llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
368  llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
369
370  Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
371  return CGF.Builder.CreateCall(F, {Src0, Src1});
372}
373
374/// EmitFAbs - Emit a call to @llvm.fabs().
375static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) {
376  Function *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType());
377  llvm::CallInst *Call = CGF.Builder.CreateCall(F, V);
378  Call->setDoesNotAccessMemory();
379  return Call;
380}
381
382/// Emit the computation of the sign bit for a floating point value. Returns
383/// the i1 sign bit value.
384static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) {
385  LLVMContext &C = CGF.CGM.getLLVMContext();
386
387  llvm::Type *Ty = V->getType();
388  int Width = Ty->getPrimitiveSizeInBits();
389  llvm::Type *IntTy = llvm::IntegerType::get(C, Width);
390  V = CGF.Builder.CreateBitCast(V, IntTy);
391  if (Ty->isPPC_FP128Ty()) {
392    // We want the sign bit of the higher-order double. The bitcast we just
393    // did works as if the double-double was stored to memory and then
394    // read as an i128. The "store" will put the higher-order double in the
395    // lower address in both little- and big-Endian modes, but the "load"
396    // will treat those bits as a different part of the i128: the low bits in
397    // little-Endian, the high bits in big-Endian. Therefore, on big-Endian
398    // we need to shift the high bits down to the low before truncating.
399    Width >>= 1;
400    if (CGF.getTarget().isBigEndian()) {
401      Value *ShiftCst = llvm::ConstantInt::get(IntTy, Width);
402      V = CGF.Builder.CreateLShr(V, ShiftCst);
403    }
404    // We are truncating value in order to extract the higher-order
405    // double, which we will be using to extract the sign from.
406    IntTy = llvm::IntegerType::get(C, Width);
407    V = CGF.Builder.CreateTrunc(V, IntTy);
408  }
409  Value *Zero = llvm::Constant::getNullValue(IntTy);
410  return CGF.Builder.CreateICmpSLT(V, Zero);
411}
412
413static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *FD,
414                              const CallExpr *E, llvm::Constant *calleeValue) {
415  CGCallee callee = CGCallee::forDirect(calleeValue, GlobalDecl(FD));
416  return CGF.EmitCall(E->getCallee()->getType(), callee, E, ReturnValueSlot());
417}
418
419/// Emit a call to llvm.{sadd,uadd,ssub,usub,smul,umul}.with.overflow.*
420/// depending on IntrinsicID.
421///
422/// \arg CGF The current codegen function.
423/// \arg IntrinsicID The ID for the Intrinsic we wish to generate.
424/// \arg X The first argument to the llvm.*.with.overflow.*.
425/// \arg Y The second argument to the llvm.*.with.overflow.*.
426/// \arg Carry The carry returned by the llvm.*.with.overflow.*.
427/// \returns The result (i.e. sum/product) returned by the intrinsic.
428static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF,
429                                          const llvm::Intrinsic::ID IntrinsicID,
430                                          llvm::Value *X, llvm::Value *Y,
431                                          llvm::Value *&Carry) {
432  // Make sure we have integers of the same width.
433   (0) . __assert_fail ("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 435, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(X->getType() == Y->getType() &&
434 (0) . __assert_fail ("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 435, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">         "Arguments must be the same type. (Did you forget to make sure both "
435 (0) . __assert_fail ("X->getType() == Y->getType() && \"Arguments must be the same type. (Did you forget to make sure both \" \"arguments have the same integer width?)\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 435, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">         "arguments have the same integer width?)");
436
437  Function *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType());
438  llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y});
439  Carry = CGF.Builder.CreateExtractValue(Tmp, 1);
440  return CGF.Builder.CreateExtractValue(Tmp, 0);
441}
442
443static Value *emitRangedBuiltin(CodeGenFunction &CGF,
444                                unsigned IntrinsicID,
445                                int low, int high) {
446    llvm::MDBuilder MDHelper(CGF.getLLVMContext());
447    llvm::MDNode *RNode = MDHelper.createRange(APInt(32, low), APInt(32, high));
448    Function *F = CGF.CGM.getIntrinsic(IntrinsicID, {});
449    llvm::Instruction *Call = CGF.Builder.CreateCall(F);
450    Call->setMetadata(llvm::LLVMContext::MD_range, RNode);
451    return Call;
452}
453
454namespace {
455  struct WidthAndSignedness {
456    unsigned Width;
457    bool Signed;
458  };
459}
460
461static WidthAndSignedness
462getIntegerWidthAndSignedness(const clang::ASTContext &context,
463                             const clang::QualType Type) {
464   (0) . __assert_fail ("Type->isIntegerType() && \"Given type is not an integer.\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 464, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Type->isIntegerType() && "Given type is not an integer.");
465  unsigned Width = Type->isBooleanType() ? 1 : context.getTypeInfo(Type).Width;
466  bool Signed = Type->isSignedIntegerType();
467  return {Width, Signed};
468}
469
470// Given one or more integer types, this function produces an integer type that
471// encompasses them: any value in one of the given types could be expressed in
472// the encompassing type.
473static struct WidthAndSignedness
474EncompassingIntegerType(ArrayRef<struct WidthAndSignedness> Types) {
475   (0) . __assert_fail ("Types.size() > 0 && \"Empty list of types.\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 475, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Types.size() > 0 && "Empty list of types.");
476
477  // If any of the given types is signed, we must return a signed type.
478  bool Signed = false;
479  for (const auto &Type : Types) {
480    Signed |= Type.Signed;
481  }
482
483  // The encompassing type must have a width greater than or equal to the width
484  // of the specified types.  Additionally, if the encompassing type is signed,
485  // its width must be strictly greater than the width of any unsigned types
486  // given.
487  unsigned Width = 0;
488  for (const auto &Type : Types) {
489    unsigned MinWidth = Type.Width + (Signed && !Type.Signed);
490    if (Width < MinWidth) {
491      Width = MinWidth;
492    }
493  }
494
495  return {Width, Signed};
496}
497
498Value *CodeGenFunction::EmitVAStartEnd(Value *ArgValue, bool IsStart) {
499  llvm::Type *DestType = Int8PtrTy;
500  if (ArgValue->getType() != DestType)
501    ArgValue =
502        Builder.CreateBitCast(ArgValue, DestType, ArgValue->getName().data());
503
504  Intrinsic::ID inst = IsStart ? Intrinsic::vastart : Intrinsic::vaend;
505  return Builder.CreateCall(CGM.getIntrinsic(inst), ArgValue);
506}
507
508/// Checks if using the result of __builtin_object_size(p, @p From) in place of
509/// __builtin_object_size(p, @p To) is correct
510static bool areBOSTypesCompatible(int From, int To) {
511  // Note: Our __builtin_object_size implementation currently treats Type=0 and
512  // Type=2 identically. Encoding this implementation detail here may make
513  // improving __builtin_object_size difficult in the future, so it's omitted.
514  return From == To || (From == 0 && To == 1) || (From == 3 && To == 2);
515}
516
517static llvm::Value *
518getDefaultBuiltinObjectSizeResult(unsigned Type, llvm::IntegerType *ResType) {
519  return ConstantInt::get(ResType, (Type & 2) ? 0 : -1, /*isSigned=*/true);
520}
521
522llvm::Value *
523CodeGenFunction::evaluateOrEmitBuiltinObjectSize(const Expr *E, unsigned Type,
524                                                 llvm::IntegerType *ResType,
525                                                 llvm::Value *EmittedE,
526                                                 bool IsDynamic) {
527  uint64_t ObjectSize;
528  if (!E->tryEvaluateObjectSize(ObjectSize, getContext(), Type))
529    return emitBuiltinObjectSize(E, Type, ResType, EmittedE, IsDynamic);
530  return ConstantInt::get(ResType, ObjectSize, /*isSigned=*/true);
531}
532
533/// Returns a Value corresponding to the size of the given expression.
534/// This Value may be either of the following:
535///   - A llvm::Argument (if E is a param with the pass_object_size attribute on
536///     it)
537///   - A call to the @llvm.objectsize intrinsic
538///
539/// EmittedE is the result of emitting `E` as a scalar expr. If it's non-null
540/// and we wouldn't otherwise try to reference a pass_object_size parameter,
541/// we'll call @llvm.objectsize on EmittedE, rather than emitting E.
542llvm::Value *
543CodeGenFunction::emitBuiltinObjectSize(const Expr *E, unsigned Type,
544                                       llvm::IntegerType *ResType,
545                                       llvm::Value *EmittedE, bool IsDynamic) {
546  // We need to reference an argument if the pointer is a parameter with the
547  // pass_object_size attribute.
548  if (auto *D = dyn_cast<DeclRefExpr>(E->IgnoreParenImpCasts())) {
549    auto *Param = dyn_cast<ParmVarDecl>(D->getDecl());
550    auto *PS = D->getDecl()->getAttr<PassObjectSizeAttr>();
551    if (Param != nullptr && PS != nullptr &&
552        areBOSTypesCompatible(PS->getType(), Type)) {
553      auto Iter = SizeArguments.find(Param);
554      assert(Iter != SizeArguments.end());
555
556      const ImplicitParamDecl *D = Iter->second;
557      auto DIter = LocalDeclMap.find(D);
558      assert(DIter != LocalDeclMap.end());
559
560      return EmitLoadOfScalar(DIter->second, /*volatile=*/false,
561                              getContext().getSizeType(), E->getBeginLoc());
562    }
563  }
564
565  // LLVM can't handle Type=3 appropriately, and __builtin_object_size shouldn't
566  // evaluate E for side-effects. In either case, we shouldn't lower to
567  // @llvm.objectsize.
568  if (Type == 3 || (!EmittedE && E->HasSideEffects(getContext())))
569    return getDefaultBuiltinObjectSizeResult(Type, ResType);
570
571  Value *Ptr = EmittedE ? EmittedE : EmitScalarExpr(E);
572   (0) . __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 573, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Ptr->getType()->isPointerTy() &&
573 (0) . __assert_fail ("Ptr->getType()->isPointerTy() && \"Non-pointer passed to __builtin_object_size?\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 573, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">         "Non-pointer passed to __builtin_object_size?");
574
575  Function *F =
576      CGM.getIntrinsic(Intrinsic::objectsize, {ResType, Ptr->getType()});
577
578  // LLVM only supports 0 and 2, make sure that we pass along that as a boolean.
579  Value *Min = Builder.getInt1((Type & 2) != 0);
580  // For GCC compatibility, __builtin_object_size treat NULL as unknown size.
581  Value *NullIsUnknown = Builder.getTrue();
582  Value *Dynamic = Builder.getInt1(IsDynamic);
583  return Builder.CreateCall(F, {Ptr, Min, NullIsUnknown, Dynamic});
584}
585
586namespace {
587/// A struct to generically describe a bit test intrinsic.
588struct BitTest {
589  enum ActionKind : uint8_t { TestOnly, Complement, Reset, Set };
590  enum InterlockingKind : uint8_t {
591    Unlocked,
592    Sequential,
593    Acquire,
594    Release,
595    NoFence
596  };
597
598  ActionKind Action;
599  InterlockingKind Interlocking;
600  bool Is64Bit;
601
602  static BitTest decodeBitTestBuiltin(unsigned BuiltinID);
603};
604} // namespace
605
606BitTest BitTest::decodeBitTestBuiltin(unsigned BuiltinID) {
607  switch (BuiltinID) {
608    // Main portable variants.
609  case Builtin::BI_bittest:
610    return {TestOnly, Unlocked, false};
611  case Builtin::BI_bittestandcomplement:
612    return {Complement, Unlocked, false};
613  case Builtin::BI_bittestandreset:
614    return {Reset, Unlocked, false};
615  case Builtin::BI_bittestandset:
616    return {Set, Unlocked, false};
617  case Builtin::BI_interlockedbittestandreset:
618    return {Reset, Sequential, false};
619  case Builtin::BI_interlockedbittestandset:
620    return {Set, Sequential, false};
621
622    // X86-specific 64-bit variants.
623  case Builtin::BI_bittest64:
624    return {TestOnly, Unlocked, true};
625  case Builtin::BI_bittestandcomplement64:
626    return {Complement, Unlocked, true};
627  case Builtin::BI_bittestandreset64:
628    return {Reset, Unlocked, true};
629  case Builtin::BI_bittestandset64:
630    return {Set, Unlocked, true};
631  case Builtin::BI_interlockedbittestandreset64:
632    return {Reset, Sequential, true};
633  case Builtin::BI_interlockedbittestandset64:
634    return {Set, Sequential, true};
635
636    // ARM/AArch64-specific ordering variants.
637  case Builtin::BI_interlockedbittestandset_acq:
638    return {Set, Acquire, false};
639  case Builtin::BI_interlockedbittestandset_rel:
640    return {Set, Release, false};
641  case Builtin::BI_interlockedbittestandset_nf:
642    return {Set, NoFence, false};
643  case Builtin::BI_interlockedbittestandreset_acq:
644    return {Reset, Acquire, false};
645  case Builtin::BI_interlockedbittestandreset_rel:
646    return {Reset, Release, false};
647  case Builtin::BI_interlockedbittestandreset_nf:
648    return {Reset, NoFence, false};
649  }
650  llvm_unreachable("expected only bittest intrinsics");
651}
652
653static char bitActionToX86BTCode(BitTest::ActionKind A) {
654  switch (A) {
655  case BitTest::TestOnly:   return '\0';
656  case BitTest::Complement: return 'c';
657  case BitTest::Reset:      return 'r';
658  case BitTest::Set:        return 's';
659  }
660  llvm_unreachable("invalid action");
661}
662
663static llvm::Value *EmitX86BitTestIntrinsic(CodeGenFunction &CGF,
664                                            BitTest BT,
665                                            const CallExpr *E, Value *BitBase,
666                                            Value *BitPos) {
667  char Action = bitActionToX86BTCode(BT.Action);
668  char SizeSuffix = BT.Is64Bit ? 'q' : 'l';
669
670  // Build the assembly.
671  SmallString<64> Asm;
672  raw_svector_ostream AsmOS(Asm);
673  if (BT.Interlocking != BitTest::Unlocked)
674    AsmOS << "lock ";
675  AsmOS << "bt";
676  if (Action)
677    AsmOS << Action;
678  AsmOS << SizeSuffix << " $2, ($1)\n\tsetc ${0:b}";
679
680  // Build the constraints. FIXME: We should support immediates when possible.
681  std::string Constraints = "=r,r,r,~{cc},~{flags},~{fpsr}";
682  llvm::IntegerType *IntType = llvm::IntegerType::get(
683      CGF.getLLVMContext(),
684      CGF.getContext().getTypeSize(E->getArg(1)->getType()));
685  llvm::Type *IntPtrType = IntType->getPointerTo();
686  llvm::FunctionType *FTy =
687      llvm::FunctionType::get(CGF.Int8Ty, {IntPtrType, IntType}, false);
688
689  llvm::InlineAsm *IA =
690      llvm::InlineAsm::get(FTy, Asm, Constraints, /*SideEffects=*/true);
691  return CGF.Builder.CreateCall(IA, {BitBase, BitPos});
692}
693
694static llvm::AtomicOrdering
695getBitTestAtomicOrdering(BitTest::InterlockingKind I) {
696  switch (I) {
697  case BitTest::Unlocked:   return llvm::AtomicOrdering::NotAtomic;
698  case BitTest::Sequential: return llvm::AtomicOrdering::SequentiallyConsistent;
699  case BitTest::Acquire:    return llvm::AtomicOrdering::Acquire;
700  case BitTest::Release:    return llvm::AtomicOrdering::Release;
701  case BitTest::NoFence:    return llvm::AtomicOrdering::Monotonic;
702  }
703  llvm_unreachable("invalid interlocking");
704}
705
706/// Emit a _bittest* intrinsic. These intrinsics take a pointer to an array of
707/// bits and a bit position and read and optionally modify the bit at that
708/// position. The position index can be arbitrarily large, i.e. it can be larger
709/// than 31 or 63, so we need an indexed load in the general case.
710static llvm::Value *EmitBitTestIntrinsic(CodeGenFunction &CGF,
711                                         unsigned BuiltinID,
712                                         const CallExpr *E) {
713  Value *BitBase = CGF.EmitScalarExpr(E->getArg(0));
714  Value *BitPos = CGF.EmitScalarExpr(E->getArg(1));
715
716  BitTest BT = BitTest::decodeBitTestBuiltin(BuiltinID);
717
718  // X86 has special BT, BTC, BTR, and BTS instructions that handle the array
719  // indexing operation internally. Use them if possible.
720  llvm::Triple::ArchType Arch = CGF.getTarget().getTriple().getArch();
721  if (Arch == llvm::Triple::x86 || Arch == llvm::Triple::x86_64)
722    return EmitX86BitTestIntrinsic(CGF, BT, E, BitBase, BitPos);
723
724  // Otherwise, use generic code to load one byte and test the bit. Use all but
725  // the bottom three bits as the array index, and the bottom three bits to form
726  // a mask.
727  // Bit = BitBaseI8[BitPos >> 3] & (1 << (BitPos & 0x7)) != 0;
728  Value *ByteIndex = CGF.Builder.CreateAShr(
729      BitPos, llvm::ConstantInt::get(BitPos->getType(), 3), "bittest.byteidx");
730  Value *BitBaseI8 = CGF.Builder.CreatePointerCast(BitBase, CGF.Int8PtrTy);
731  Address ByteAddr(CGF.Builder.CreateInBoundsGEP(CGF.Int8Ty, BitBaseI8,
732                                                 ByteIndex, "bittest.byteaddr"),
733                   CharUnits::One());
734  Value *PosLow =
735      CGF.Builder.CreateAnd(CGF.Builder.CreateTrunc(BitPos, CGF.Int8Ty),
736                            llvm::ConstantInt::get(CGF.Int8Ty, 0x7));
737
738  // The updating instructions will need a mask.
739  Value *Mask = nullptr;
740  if (BT.Action != BitTest::TestOnly) {
741    Mask = CGF.Builder.CreateShl(llvm::ConstantInt::get(CGF.Int8Ty, 1), PosLow,
742                                 "bittest.mask");
743  }
744
745  // Check the action and ordering of the interlocked intrinsics.
746  llvm::AtomicOrdering Ordering = getBitTestAtomicOrdering(BT.Interlocking);
747
748  Value *OldByte = nullptr;
749  if (Ordering != llvm::AtomicOrdering::NotAtomic) {
750    // Emit a combined atomicrmw load/store operation for the interlocked
751    // intrinsics.
752    llvm::AtomicRMWInst::BinOp RMWOp = llvm::AtomicRMWInst::Or;
753    if (BT.Action == BitTest::Reset) {
754      Mask = CGF.Builder.CreateNot(Mask);
755      RMWOp = llvm::AtomicRMWInst::And;
756    }
757    OldByte = CGF.Builder.CreateAtomicRMW(RMWOp, ByteAddr.getPointer(), Mask,
758                                          Ordering);
759  } else {
760    // Emit a plain load for the non-interlocked intrinsics.
761    OldByte = CGF.Builder.CreateLoad(ByteAddr, "bittest.byte");
762    Value *NewByte = nullptr;
763    switch (BT.Action) {
764    case BitTest::TestOnly:
765      // Don't store anything.
766      break;
767    case BitTest::Complement:
768      NewByte = CGF.Builder.CreateXor(OldByte, Mask);
769      break;
770    case BitTest::Reset:
771      NewByte = CGF.Builder.CreateAnd(OldByte, CGF.Builder.CreateNot(Mask));
772      break;
773    case BitTest::Set:
774      NewByte = CGF.Builder.CreateOr(OldByte, Mask);
775      break;
776    }
777    if (NewByte)
778      CGF.Builder.CreateStore(NewByte, ByteAddr);
779  }
780
781  // However we loaded the old byte, either by plain load or atomicrmw, shift
782  // the bit into the low position and mask it to 0 or 1.
783  Value *ShiftedByte = CGF.Builder.CreateLShr(OldByte, PosLow, "bittest.shr");
784  return CGF.Builder.CreateAnd(
785      ShiftedByte, llvm::ConstantInt::get(CGF.Int8Ty, 1), "bittest.res");
786}
787
788namespace {
789enum class MSVCSetJmpKind {
790  _setjmpex,
791  _setjmp3,
792  _setjmp
793};
794}
795
796/// MSVC handles setjmp a bit differently on different platforms. On every
797/// architecture except 32-bit x86, the frame address is passed. On x86, extra
798/// parameters can be passed as variadic arguments, but we always pass none.
799static RValue EmitMSVCRTSetJmp(CodeGenFunction &CGF, MSVCSetJmpKind SJKind,
800                               const CallExpr *E) {
801  llvm::Value *Arg1 = nullptr;
802  llvm::Type *Arg1Ty = nullptr;
803  StringRef Name;
804  bool IsVarArg = false;
805  if (SJKind == MSVCSetJmpKind::_setjmp3) {
806    Name = "_setjmp3";
807    Arg1Ty = CGF.Int32Ty;
808    Arg1 = llvm::ConstantInt::get(CGF.IntTy, 0);
809    IsVarArg = true;
810  } else {
811    Name = SJKind == MSVCSetJmpKind::_setjmp ? "_setjmp" : "_setjmpex";
812    Arg1Ty = CGF.Int8PtrTy;
813    if (CGF.getTarget().getTriple().getArch() == llvm::Triple::aarch64) {
814      Arg1 = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(Intrinsic::sponentry));
815    } else
816      Arg1 = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(Intrinsic::frameaddress),
817                                    llvm::ConstantInt::get(CGF.Int32Ty, 0));
818  }
819
820  // Mark the call site and declaration with ReturnsTwice.
821  llvm::Type *ArgTypes[2] = {CGF.Int8PtrTy, Arg1Ty};
822  llvm::AttributeList ReturnsTwiceAttr = llvm::AttributeList::get(
823      CGF.getLLVMContext(), llvm::AttributeList::FunctionIndex,
824      llvm::Attribute::ReturnsTwice);
825  llvm::FunctionCallee SetJmpFn = CGF.CGM.CreateRuntimeFunction(
826      llvm::FunctionType::get(CGF.IntTy, ArgTypes, IsVarArg), Name,
827      ReturnsTwiceAttr, /*Local=*/true);
828
829  llvm::Value *Buf = CGF.Builder.CreateBitOrPointerCast(
830      CGF.EmitScalarExpr(E->getArg(0)), CGF.Int8PtrTy);
831  llvm::Value *Args[] = {Buf, Arg1};
832  llvm::CallBase *CB = CGF.EmitRuntimeCallOrInvoke(SetJmpFn, Args);
833  CB->setAttributes(ReturnsTwiceAttr);
834  return RValue::get(CB);
835}
836
837// Many of MSVC builtins are on x64, ARM and AArch64; to avoid repeating code,
838// we handle them here.
839enum class CodeGenFunction::MSVCIntrin {
840  _BitScanForward,
841  _BitScanReverse,
842  _InterlockedAnd,
843  _InterlockedDecrement,
844  _InterlockedExchange,
845  _InterlockedExchangeAdd,
846  _InterlockedExchangeSub,
847  _InterlockedIncrement,
848  _InterlockedOr,
849  _InterlockedXor,
850  _InterlockedExchangeAdd_acq,
851  _InterlockedExchangeAdd_rel,
852  _InterlockedExchangeAdd_nf,
853  _InterlockedExchange_acq,
854  _InterlockedExchange_rel,
855  _InterlockedExchange_nf,
856  _InterlockedCompareExchange_acq,
857  _InterlockedCompareExchange_rel,
858  _InterlockedCompareExchange_nf,
859  _InterlockedOr_acq,
860  _InterlockedOr_rel,
861  _InterlockedOr_nf,
862  _InterlockedXor_acq,
863  _InterlockedXor_rel,
864  _InterlockedXor_nf,
865  _InterlockedAnd_acq,
866  _InterlockedAnd_rel,
867  _InterlockedAnd_nf,
868  _InterlockedIncrement_acq,
869  _InterlockedIncrement_rel,
870  _InterlockedIncrement_nf,
871  _InterlockedDecrement_acq,
872  _InterlockedDecrement_rel,
873  _InterlockedDecrement_nf,
874  __fastfail,
875};
876
877Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID,
878                                            const CallExpr *E) {
879  switch (BuiltinID) {
880  case MSVCIntrin::_BitScanForward:
881  case MSVCIntrin::_BitScanReverse: {
882    Value *ArgValue = EmitScalarExpr(E->getArg(1));
883
884    llvm::Type *ArgType = ArgValue->getType();
885    llvm::Type *IndexType =
886      EmitScalarExpr(E->getArg(0))->getType()->getPointerElementType();
887    llvm::Type *ResultType = ConvertType(E->getType());
888
889    Value *ArgZero = llvm::Constant::getNullValue(ArgType);
890    Value *ResZero = llvm::Constant::getNullValue(ResultType);
891    Value *ResOne = llvm::ConstantInt::get(ResultType, 1);
892
893    BasicBlock *Begin = Builder.GetInsertBlock();
894    BasicBlock *End = createBasicBlock("bitscan_end", this->CurFn);
895    Builder.SetInsertPoint(End);
896    PHINode *Result = Builder.CreatePHI(ResultType, 2, "bitscan_result");
897
898    Builder.SetInsertPoint(Begin);
899    Value *IsZero = Builder.CreateICmpEQ(ArgValue, ArgZero);
900    BasicBlock *NotZero = createBasicBlock("bitscan_not_zero", this->CurFn);
901    Builder.CreateCondBr(IsZero, End, NotZero);
902    Result->addIncoming(ResZero, Begin);
903
904    Builder.SetInsertPoint(NotZero);
905    Address IndexAddress = EmitPointerWithAlignment(E->getArg(0));
906
907    if (BuiltinID == MSVCIntrin::_BitScanForward) {
908      Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
909      Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()});
910      ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false);
911      Builder.CreateStore(ZeroCount, IndexAddress, false);
912    } else {
913      unsigned ArgWidth = cast<llvm::IntegerType>(ArgType)->getBitWidth();
914      Value *ArgTypeLastIndex = llvm::ConstantInt::get(IndexType, ArgWidth - 1);
915
916      Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
917      Value *ZeroCount = Builder.CreateCall(F, {ArgValue, Builder.getTrue()});
918      ZeroCount = Builder.CreateIntCast(ZeroCount, IndexType, false);
919      Value *Index = Builder.CreateNSWSub(ArgTypeLastIndex, ZeroCount);
920      Builder.CreateStore(Index, IndexAddress, false);
921    }
922    Builder.CreateBr(End);
923    Result->addIncoming(ResOne, NotZero);
924
925    Builder.SetInsertPoint(End);
926    return Result;
927  }
928  case MSVCIntrin::_InterlockedAnd:
929    return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E);
930  case MSVCIntrin::_InterlockedExchange:
931    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E);
932  case MSVCIntrin::_InterlockedExchangeAdd:
933    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E);
934  case MSVCIntrin::_InterlockedExchangeSub:
935    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E);
936  case MSVCIntrin::_InterlockedOr:
937    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E);
938  case MSVCIntrin::_InterlockedXor:
939    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E);
940  case MSVCIntrin::_InterlockedExchangeAdd_acq:
941    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E,
942                                 AtomicOrdering::Acquire);
943  case MSVCIntrin::_InterlockedExchangeAdd_rel:
944    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E,
945                                 AtomicOrdering::Release);
946  case MSVCIntrin::_InterlockedExchangeAdd_nf:
947    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E,
948                                 AtomicOrdering::Monotonic);
949  case MSVCIntrin::_InterlockedExchange_acq:
950    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E,
951                                 AtomicOrdering::Acquire);
952  case MSVCIntrin::_InterlockedExchange_rel:
953    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E,
954                                 AtomicOrdering::Release);
955  case MSVCIntrin::_InterlockedExchange_nf:
956    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xchg, E,
957                                 AtomicOrdering::Monotonic);
958  case MSVCIntrin::_InterlockedCompareExchange_acq:
959    return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Acquire);
960  case MSVCIntrin::_InterlockedCompareExchange_rel:
961    return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Release);
962  case MSVCIntrin::_InterlockedCompareExchange_nf:
963    return EmitAtomicCmpXchgForMSIntrin(*this, E, AtomicOrdering::Monotonic);
964  case MSVCIntrin::_InterlockedOr_acq:
965    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E,
966                                 AtomicOrdering::Acquire);
967  case MSVCIntrin::_InterlockedOr_rel:
968    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E,
969                                 AtomicOrdering::Release);
970  case MSVCIntrin::_InterlockedOr_nf:
971    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E,
972                                 AtomicOrdering::Monotonic);
973  case MSVCIntrin::_InterlockedXor_acq:
974    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E,
975                                 AtomicOrdering::Acquire);
976  case MSVCIntrin::_InterlockedXor_rel:
977    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E,
978                                 AtomicOrdering::Release);
979  case MSVCIntrin::_InterlockedXor_nf:
980    return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E,
981                                 AtomicOrdering::Monotonic);
982  case MSVCIntrin::_InterlockedAnd_acq:
983    return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E,
984                                 AtomicOrdering::Acquire);
985  case MSVCIntrin::_InterlockedAnd_rel:
986    return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E,
987                                 AtomicOrdering::Release);
988  case MSVCIntrin::_InterlockedAnd_nf:
989    return MakeBinaryAtomicValue(*this, AtomicRMWInst::And, E,
990                                 AtomicOrdering::Monotonic);
991  case MSVCIntrin::_InterlockedIncrement_acq:
992    return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Acquire);
993  case MSVCIntrin::_InterlockedIncrement_rel:
994    return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Release);
995  case MSVCIntrin::_InterlockedIncrement_nf:
996    return EmitAtomicIncrementValue(*this, E, AtomicOrdering::Monotonic);
997  case MSVCIntrin::_InterlockedDecrement_acq:
998    return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Acquire);
999  case MSVCIntrin::_InterlockedDecrement_rel:
1000    return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Release);
1001  case MSVCIntrin::_InterlockedDecrement_nf:
1002    return EmitAtomicDecrementValue(*this, E, AtomicOrdering::Monotonic);
1003
1004  case MSVCIntrin::_InterlockedDecrement:
1005    return EmitAtomicDecrementValue(*this, E);
1006  case MSVCIntrin::_InterlockedIncrement:
1007    return EmitAtomicIncrementValue(*this, E);
1008
1009  case MSVCIntrin::__fastfail: {
1010    // Request immediate process termination from the kernel. The instruction
1011    // sequences to do this are documented on MSDN:
1012    // https://msdn.microsoft.com/en-us/library/dn774154.aspx
1013    llvm::Triple::ArchType ISA = getTarget().getTriple().getArch();
1014    StringRef Asm, Constraints;
1015    switch (ISA) {
1016    default:
1017      ErrorUnsupported(E, "__fastfail call for this architecture");
1018      break;
1019    case llvm::Triple::x86:
1020    case llvm::Triple::x86_64:
1021      Asm = "int $$0x29";
1022      Constraints = "{cx}";
1023      break;
1024    case llvm::Triple::thumb:
1025      Asm = "udf #251";
1026      Constraints = "{r0}";
1027      break;
1028    case llvm::Triple::aarch64:
1029      Asm = "brk #0xF003";
1030      Constraints = "{w0}";
1031    }
1032    llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, {Int32Ty}, false);
1033    llvm::InlineAsm *IA =
1034        llvm::InlineAsm::get(FTy, Asm, Constraints, /*SideEffects=*/true);
1035    llvm::AttributeList NoReturnAttr = llvm::AttributeList::get(
1036        getLLVMContext(), llvm::AttributeList::FunctionIndex,
1037        llvm::Attribute::NoReturn);
1038    llvm::CallInst *CI = Builder.CreateCall(IA, EmitScalarExpr(E->getArg(0)));
1039    CI->setAttributes(NoReturnAttr);
1040    return CI;
1041  }
1042  }
1043  llvm_unreachable("Incorrect MSVC intrinsic!");
1044}
1045
1046namespace {
1047// ARC cleanup for __builtin_os_log_format
1048struct CallObjCArcUse final : EHScopeStack::Cleanup {
1049  CallObjCArcUse(llvm::Value *object) : object(object) {}
1050  llvm::Value *object;
1051
1052  void Emit(CodeGenFunction &CGF, Flags flags) override {
1053    CGF.EmitARCIntrinsicUse(object);
1054  }
1055};
1056}
1057
1058Value *CodeGenFunction::EmitCheckedArgForBuiltin(const Expr *E,
1059                                                 BuiltinCheckKind Kind) {
1060   (0) . __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1061, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert((Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero)
1061 (0) . __assert_fail ("(Kind == BCK_CLZPassedZero || Kind == BCK_CTZPassedZero) && \"Unsupported builtin check kind\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1061, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">          && "Unsupported builtin check kind");
1062
1063  Value *ArgValue = EmitScalarExpr(E);
1064  if (!SanOpts.has(SanitizerKind::Builtin) || !getTarget().isCLZForZeroUndef())
1065    return ArgValue;
1066
1067  SanitizerScope SanScope(this);
1068  Value *Cond = Builder.CreateICmpNE(
1069      ArgValue, llvm::Constant::getNullValue(ArgValue->getType()));
1070  EmitCheck(std::make_pair(Cond, SanitizerKind::Builtin),
1071            SanitizerHandler::InvalidBuiltin,
1072            {EmitCheckSourceLocation(E->getExprLoc()),
1073             llvm::ConstantInt::get(Builder.getInt8Ty(), Kind)},
1074            None);
1075  return ArgValue;
1076}
1077
1078/// Get the argument type for arguments to os_log_helper.
1079static CanQualType getOSLogArgType(ASTContext &C, int Size) {
1080  QualType UnsignedTy = C.getIntTypeForBitwidth(Size * 8, /*Signed=*/false);
1081  return C.getCanonicalType(UnsignedTy);
1082}
1083
1084llvm::Function *CodeGenFunction::generateBuiltinOSLogHelperFunction(
1085    const analyze_os_log::OSLogBufferLayout &Layout,
1086    CharUnits BufferAlignment) {
1087  ASTContext &Ctx = getContext();
1088
1089  llvm::SmallString<64> Name;
1090  {
1091    raw_svector_ostream OS(Name);
1092    OS << "__os_log_helper";
1093    OS << "_" << BufferAlignment.getQuantity();
1094    OS << "_" << int(Layout.getSummaryByte());
1095    OS << "_" << int(Layout.getNumArgsByte());
1096    for (const auto &Item : Layout.Items)
1097      OS << "_" << int(Item.getSizeByte()) << "_"
1098         << int(Item.getDescriptorByte());
1099  }
1100
1101  if (llvm::Function *F = CGM.getModule().getFunction(Name))
1102    return F;
1103
1104  llvm::SmallVector<QualType, 4> ArgTys;
1105  llvm::SmallVector<ImplicitParamDecl, 4> Params;
1106  Params.emplace_back(Ctx, nullptr, SourceLocation(), &Ctx.Idents.get("buffer"),
1107                      Ctx.VoidPtrTy, ImplicitParamDecl::Other);
1108  ArgTys.emplace_back(Ctx.VoidPtrTy);
1109
1110  for (unsigned int I = 0, E = Layout.Items.size(); I < E; ++I) {
1111    char Size = Layout.Items[I].getSizeByte();
1112    if (!Size)
1113      continue;
1114
1115    QualType ArgTy = getOSLogArgType(Ctx, Size);
1116    Params.emplace_back(
1117        Ctx, nullptr, SourceLocation(),
1118        &Ctx.Idents.get(std::string("arg") + llvm::to_string(I)), ArgTy,
1119        ImplicitParamDecl::Other);
1120    ArgTys.emplace_back(ArgTy);
1121  }
1122
1123  FunctionArgList Args;
1124  for (auto &P : Params)
1125    Args.push_back(&P);
1126
1127  QualType ReturnTy = Ctx.VoidTy;
1128  QualType FuncionTy = Ctx.getFunctionType(ReturnTy, ArgTys, {});
1129
1130  // The helper function has linkonce_odr linkage to enable the linker to merge
1131  // identical functions. To ensure the merging always happens, 'noinline' is
1132  // attached to the function when compiling with -Oz.
1133  const CGFunctionInfo &FI =
1134      CGM.getTypes().arrangeBuiltinFunctionDeclaration(ReturnTy, Args);
1135  llvm::FunctionType *FuncTy = CGM.getTypes().GetFunctionType(FI);
1136  llvm::Function *Fn = llvm::Function::Create(
1137      FuncTy, llvm::GlobalValue::LinkOnceODRLinkage, Name, &CGM.getModule());
1138  Fn->setVisibility(llvm::GlobalValue::HiddenVisibility);
1139  CGM.SetLLVMFunctionAttributes(GlobalDecl(), FI, Fn);
1140  CGM.SetLLVMFunctionAttributesForDefinition(nullptr, Fn);
1141
1142  // Attach 'noinline' at -Oz.
1143  if (CGM.getCodeGenOpts().OptimizeSize == 2)
1144    Fn->addFnAttr(llvm::Attribute::NoInline);
1145
1146  auto NL = ApplyDebugLocation::CreateEmpty(*this);
1147  IdentifierInfo *II = &Ctx.Idents.get(Name);
1148  FunctionDecl *FD = FunctionDecl::Create(
1149      Ctx, Ctx.getTranslationUnitDecl(), SourceLocation(), SourceLocation(), II,
1150      FuncionTy, nullptr, SC_PrivateExtern, false, false);
1151
1152  StartFunction(FD, ReturnTy, Fn, FI, Args);
1153
1154  // Create a scope with an artificial location for the body of this function.
1155  auto AL = ApplyDebugLocation::CreateArtificial(*this);
1156
1157  CharUnits Offset;
1158  Address BufAddr(Builder.CreateLoad(GetAddrOfLocalVar(&Params[0]), "buf"),
1159                  BufferAlignment);
1160  Builder.CreateStore(Builder.getInt8(Layout.getSummaryByte()),
1161                      Builder.CreateConstByteGEP(BufAddr, Offset++, "summary"));
1162  Builder.CreateStore(Builder.getInt8(Layout.getNumArgsByte()),
1163                      Builder.CreateConstByteGEP(BufAddr, Offset++, "numArgs"));
1164
1165  unsigned I = 1;
1166  for (const auto &Item : Layout.Items) {
1167    Builder.CreateStore(
1168        Builder.getInt8(Item.getDescriptorByte()),
1169        Builder.CreateConstByteGEP(BufAddr, Offset++, "argDescriptor"));
1170    Builder.CreateStore(
1171        Builder.getInt8(Item.getSizeByte()),
1172        Builder.CreateConstByteGEP(BufAddr, Offset++, "argSize"));
1173
1174    CharUnits Size = Item.size();
1175    if (!Size.getQuantity())
1176      continue;
1177
1178    Address Arg = GetAddrOfLocalVar(&Params[I]);
1179    Address Addr = Builder.CreateConstByteGEP(BufAddr, Offset, "argData");
1180    Addr = Builder.CreateBitCast(Addr, Arg.getPointer()->getType(),
1181                                 "argDataCast");
1182    Builder.CreateStore(Builder.CreateLoad(Arg), Addr);
1183    Offset += Size;
1184    ++I;
1185  }
1186
1187  FinishFunction();
1188
1189  return Fn;
1190}
1191
1192RValue CodeGenFunction::emitBuiltinOSLogFormat(const CallExpr &E) {
1193   (0) . __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1194, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(E.getNumArgs() >= 2 &&
1194 (0) . __assert_fail ("E.getNumArgs() >= 2 && \"__builtin_os_log_format takes at least 2 arguments\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1194, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">         "__builtin_os_log_format takes at least 2 arguments");
1195  ASTContext &Ctx = getContext();
1196  analyze_os_log::OSLogBufferLayout Layout;
1197  analyze_os_log::computeOSLogBufferLayout(Ctx, &E, Layout);
1198  Address BufAddr = EmitPointerWithAlignment(E.getArg(0));
1199  llvm::SmallVector<llvm::Value *, 4> RetainableOperands;
1200
1201  // Ignore argument 1, the format string. It is not currently used.
1202  CallArgList Args;
1203  Args.add(RValue::get(BufAddr.getPointer()), Ctx.VoidPtrTy);
1204
1205  for (const auto &Item : Layout.Items) {
1206    int Size = Item.getSizeByte();
1207    if (!Size)
1208      continue;
1209
1210    llvm::Value *ArgVal;
1211
1212    if (Item.getKind() == analyze_os_log::OSLogBufferItem::MaskKind) {
1213      uint64_t Val = 0;
1214      for (unsigned I = 0, E = Item.getMaskType().size(); I < E; ++I)
1215        Val |= ((uint64_t)Item.getMaskType()[I]) << I * 8;
1216      ArgVal = llvm::Constant::getIntegerValue(Int64Ty, llvm::APInt(64, Val));
1217    } else if (const Expr *TheExpr = Item.getExpr()) {
1218      ArgVal = EmitScalarExpr(TheExpr, /*Ignore*/ false);
1219
1220      // Check if this is a retainable type.
1221      if (TheExpr->getType()->isObjCRetainableType()) {
1222         (0) . __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1223, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(getEvaluationKind(TheExpr->getType()) == TEK_Scalar &&
1223 (0) . __assert_fail ("getEvaluationKind(TheExpr->getType()) == TEK_Scalar && \"Only scalar can be a ObjC retainable type\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1223, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">               "Only scalar can be a ObjC retainable type");
1224        // Check if the object is constant, if not, save it in
1225        // RetainableOperands.
1226        if (!isa<Constant>(ArgVal))
1227          RetainableOperands.push_back(ArgVal);
1228      }
1229    } else {
1230      ArgVal = Builder.getInt32(Item.getConstValue().getQuantity());
1231    }
1232
1233    unsigned ArgValSize =
1234        CGM.getDataLayout().getTypeSizeInBits(ArgVal->getType());
1235    llvm::IntegerType *IntTy = llvm::Type::getIntNTy(getLLVMContext(),
1236                                                     ArgValSize);
1237    ArgVal = Builder.CreateBitOrPointerCast(ArgVal, IntTy);
1238    CanQualType ArgTy = getOSLogArgType(Ctx, Size);
1239    // If ArgVal has type x86_fp80, zero-extend ArgVal.
1240    ArgVal = Builder.CreateZExtOrBitCast(ArgVal, ConvertType(ArgTy));
1241    Args.add(RValue::get(ArgVal), ArgTy);
1242  }
1243
1244  const CGFunctionInfo &FI =
1245      CGM.getTypes().arrangeBuiltinFunctionCall(Ctx.VoidTy, Args);
1246  llvm::Function *F = CodeGenFunction(CGM).generateBuiltinOSLogHelperFunction(
1247      Layout, BufAddr.getAlignment());
1248  EmitCall(FI, CGCallee::forDirect(F), ReturnValueSlot(), Args);
1249
1250  // Push a clang.arc.use cleanup for each object in RetainableOperands. The
1251  // cleanup will cause the use to appear after the final log call, keeping
1252  // the object valid while it’s held in the log buffer.  Note that if there’s
1253  // a release cleanup on the object, it will already be active; since
1254  // cleanups are emitted in reverse order, the use will occur before the
1255  // object is released.
1256  if (!RetainableOperands.empty() && getLangOpts().ObjCAutoRefCount &&
1257      CGM.getCodeGenOpts().OptimizationLevel != 0)
1258    for (llvm::Value *Object : RetainableOperands)
1259      pushFullExprCleanup<CallObjCArcUse>(getARCCleanupKind(), Object);
1260
1261  return RValue::get(BufAddr.getPointer());
1262}
1263
1264/// Determine if a binop is a checked mixed-sign multiply we can specialize.
1265static bool isSpecialMixedSignMultiply(unsigned BuiltinID,
1266                                       WidthAndSignedness Op1Info,
1267                                       WidthAndSignedness Op2Info,
1268                                       WidthAndSignedness ResultInfo) {
1269  return BuiltinID == Builtin::BI__builtin_mul_overflow &&
1270         std::max(Op1Info.Width, Op2Info.Width) >= ResultInfo.Width &&
1271         Op1Info.Signed != Op2Info.Signed;
1272}
1273
1274/// Emit a checked mixed-sign multiply. This is a cheaper specialization of
1275/// the generic checked-binop irgen.
1276static RValue
1277EmitCheckedMixedSignMultiply(CodeGenFunction &CGF, const clang::Expr *Op1,
1278                             WidthAndSignedness Op1Info, const clang::Expr *Op2,
1279                             WidthAndSignedness Op2Info,
1280                             const clang::Expr *ResultArg, QualType ResultQTy,
1281                             WidthAndSignedness ResultInfo) {
1282   (0) . __assert_fail ("isSpecialMixedSignMultiply(Builtin..BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1284, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(isSpecialMixedSignMultiply(Builtin::BI__builtin_mul_overflow, Op1Info,
1283 (0) . __assert_fail ("isSpecialMixedSignMultiply(Builtin..BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1284, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">                                    Op2Info, ResultInfo) &&
1284 (0) . __assert_fail ("isSpecialMixedSignMultiply(Builtin..BI__builtin_mul_overflow, Op1Info, Op2Info, ResultInfo) && \"Not a mixed-sign multipliction we can specialize\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1284, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">         "Not a mixed-sign multipliction we can specialize");
1285
1286  // Emit the signed and unsigned operands.
1287  const clang::Expr *SignedOp = Op1Info.Signed ? Op1 : Op2;
1288  const clang::Expr *UnsignedOp = Op1Info.Signed ? Op2 : Op1;
1289  llvm::Value *Signed = CGF.EmitScalarExpr(SignedOp);
1290  llvm::Value *Unsigned = CGF.EmitScalarExpr(UnsignedOp);
1291  unsigned SignedOpWidth = Op1Info.Signed ? Op1Info.Width : Op2Info.Width;
1292  unsigned UnsignedOpWidth = Op1Info.Signed ? Op2Info.Width : Op1Info.Width;
1293
1294  // One of the operands may be smaller than the other. If so, [s|z]ext it.
1295  if (SignedOpWidth < UnsignedOpWidth)
1296    Signed = CGF.Builder.CreateSExt(Signed, Unsigned->getType(), "op.sext");
1297  if (UnsignedOpWidth < SignedOpWidth)
1298    Unsigned = CGF.Builder.CreateZExt(Unsigned, Signed->getType(), "op.zext");
1299
1300  llvm::Type *OpTy = Signed->getType();
1301  llvm::Value *Zero = llvm::Constant::getNullValue(OpTy);
1302  Address ResultPtr = CGF.EmitPointerWithAlignment(ResultArg);
1303  llvm::Type *ResTy = ResultPtr.getElementType();
1304  unsigned OpWidth = std::max(Op1Info.Width, Op2Info.Width);
1305
1306  // Take the absolute value of the signed operand.
1307  llvm::Value *IsNegative = CGF.Builder.CreateICmpSLT(Signed, Zero);
1308  llvm::Value *AbsOfNegative = CGF.Builder.CreateSub(Zero, Signed);
1309  llvm::Value *AbsSigned =
1310      CGF.Builder.CreateSelect(IsNegative, AbsOfNegative, Signed);
1311
1312  // Perform a checked unsigned multiplication.
1313  llvm::Value *UnsignedOverflow;
1314  llvm::Value *UnsignedResult =
1315      EmitOverflowIntrinsic(CGF, llvm::Intrinsic::umul_with_overflow, AbsSigned,
1316                            Unsigned, UnsignedOverflow);
1317
1318  llvm::Value *Overflow, *Result;
1319  if (ResultInfo.Signed) {
1320    // Signed overflow occurs if the result is greater than INT_MAX or lesser
1321    // than INT_MIN, i.e when |Result| > (INT_MAX + IsNegative).
1322    auto IntMax =
1323        llvm::APInt::getSignedMaxValue(ResultInfo.Width).zextOrSelf(OpWidth);
1324    llvm::Value *MaxResult =
1325        CGF.Builder.CreateAdd(llvm::ConstantInt::get(OpTy, IntMax),
1326                              CGF.Builder.CreateZExt(IsNegative, OpTy));
1327    llvm::Value *SignedOverflow =
1328        CGF.Builder.CreateICmpUGT(UnsignedResult, MaxResult);
1329    Overflow = CGF.Builder.CreateOr(UnsignedOverflow, SignedOverflow);
1330
1331    // Prepare the signed result (possibly by negating it).
1332    llvm::Value *NegativeResult = CGF.Builder.CreateNeg(UnsignedResult);
1333    llvm::Value *SignedResult =
1334        CGF.Builder.CreateSelect(IsNegative, NegativeResult, UnsignedResult);
1335    Result = CGF.Builder.CreateTrunc(SignedResult, ResTy);
1336  } else {
1337    // Unsigned overflow occurs if the result is < 0 or greater than UINT_MAX.
1338    llvm::Value *Underflow = CGF.Builder.CreateAnd(
1339        IsNegative, CGF.Builder.CreateIsNotNull(UnsignedResult));
1340    Overflow = CGF.Builder.CreateOr(UnsignedOverflow, Underflow);
1341    if (ResultInfo.Width < OpWidth) {
1342      auto IntMax =
1343          llvm::APInt::getMaxValue(ResultInfo.Width).zext(OpWidth);
1344      llvm::Value *TruncOverflow = CGF.Builder.CreateICmpUGT(
1345          UnsignedResult, llvm::ConstantInt::get(OpTy, IntMax));
1346      Overflow = CGF.Builder.CreateOr(Overflow, TruncOverflow);
1347    }
1348
1349    // Negate the product if it would be negative in infinite precision.
1350    Result = CGF.Builder.CreateSelect(
1351        IsNegative, CGF.Builder.CreateNeg(UnsignedResult), UnsignedResult);
1352
1353    Result = CGF.Builder.CreateTrunc(Result, ResTy);
1354  }
1355   (0) . __assert_fail ("Overflow && Result && \"Missing overflow or result\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1355, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Overflow && Result && "Missing overflow or result");
1356
1357  bool isVolatile =
1358      ResultArg->getType()->getPointeeType().isVolatileQualified();
1359  CGF.Builder.CreateStore(CGF.EmitToMemory(Result, ResultQTy), ResultPtr,
1360                          isVolatile);
1361  return RValue::get(Overflow);
1362}
1363
1364static llvm::Value *dumpRecord(CodeGenFunction &CGF, QualType RType,
1365                               Value *&RecordPtr, CharUnits Align,
1366                               llvm::FunctionCallee Func, int Lvl) {
1367  const auto *RT = RType->getAs<RecordType>();
1368  ASTContext &Context = CGF.getContext();
1369  RecordDecl *RD = RT->getDecl()->getDefinition();
1370  ASTContext &Ctx = RD->getASTContext();
1371  const ASTRecordLayout &RL = Ctx.getASTRecordLayout(RD);
1372  std::string Pad = std::string(Lvl * 4, ' ');
1373
1374  Value *GString =
1375      CGF.Builder.CreateGlobalStringPtr(RType.getAsString() + " {\n");
1376  Value *Res = CGF.Builder.CreateCall(Func, {GString});
1377
1378  static llvm::DenseMap<QualType, const char *> Types;
1379  if (Types.empty()) {
1380    Types[Context.CharTy] = "%c";
1381    Types[Context.BoolTy] = "%d";
1382    Types[Context.SignedCharTy] = "%hhd";
1383    Types[Context.UnsignedCharTy] = "%hhu";
1384    Types[Context.IntTy] = "%d";
1385    Types[Context.UnsignedIntTy] = "%u";
1386    Types[Context.LongTy] = "%ld";
1387    Types[Context.UnsignedLongTy] = "%lu";
1388    Types[Context.LongLongTy] = "%lld";
1389    Types[Context.UnsignedLongLongTy] = "%llu";
1390    Types[Context.ShortTy] = "%hd";
1391    Types[Context.UnsignedShortTy] = "%hu";
1392    Types[Context.VoidPtrTy] = "%p";
1393    Types[Context.FloatTy] = "%f";
1394    Types[Context.DoubleTy] = "%f";
1395    Types[Context.LongDoubleTy] = "%Lf";
1396    Types[Context.getPointerType(Context.CharTy)] = "%s";
1397    Types[Context.getPointerType(Context.getConstType(Context.CharTy))] = "%s";
1398  }
1399
1400  for (const auto *FD : RD->fields()) {
1401    uint64_t Off = RL.getFieldOffset(FD->getFieldIndex());
1402    Off = Ctx.toCharUnitsFromBits(Off).getQuantity();
1403
1404    Value *FieldPtr = RecordPtr;
1405    if (RD->isUnion())
1406      FieldPtr = CGF.Builder.CreatePointerCast(
1407          FieldPtr, CGF.ConvertType(Context.getPointerType(FD->getType())));
1408    else
1409      FieldPtr = CGF.Builder.CreateStructGEP(CGF.ConvertType(RType), FieldPtr,
1410                                             FD->getFieldIndex());
1411
1412    GString = CGF.Builder.CreateGlobalStringPtr(
1413        llvm::Twine(Pad)
1414            .concat(FD->getType().getAsString())
1415            .concat(llvm::Twine(' '))
1416            .concat(FD->getNameAsString())
1417            .concat(" : ")
1418            .str());
1419    Value *TmpRes = CGF.Builder.CreateCall(Func, {GString});
1420    Res = CGF.Builder.CreateAdd(Res, TmpRes);
1421
1422    QualType CanonicalType =
1423        FD->getType().getUnqualifiedType().getCanonicalType();
1424
1425    // We check whether we are in a recursive type
1426    if (CanonicalType->isRecordType()) {
1427      Value *TmpRes =
1428          dumpRecord(CGF, CanonicalType, FieldPtr, Align, Func, Lvl + 1);
1429      Res = CGF.Builder.CreateAdd(TmpRes, Res);
1430      continue;
1431    }
1432
1433    // We try to determine the best format to print the current field
1434    llvm::Twine Format = Types.find(CanonicalType) == Types.end()
1435                             ? Types[Context.VoidPtrTy]
1436                             : Types[CanonicalType];
1437
1438    Address FieldAddress = Address(FieldPtr, Align);
1439    FieldPtr = CGF.Builder.CreateLoad(FieldAddress);
1440
1441    // FIXME Need to handle bitfield here
1442    GString = CGF.Builder.CreateGlobalStringPtr(
1443        Format.concat(llvm::Twine('\n')).str());
1444    TmpRes = CGF.Builder.CreateCall(Func, {GString, FieldPtr});
1445    Res = CGF.Builder.CreateAdd(Res, TmpRes);
1446  }
1447
1448  GString = CGF.Builder.CreateGlobalStringPtr(Pad + "}\n");
1449  Value *TmpRes = CGF.Builder.CreateCall(Func, {GString});
1450  Res = CGF.Builder.CreateAdd(Res, TmpRes);
1451  return Res;
1452}
1453
1454static bool
1455TypeRequiresBuiltinLaunderImp(const ASTContext &Ctx, QualType Ty,
1456                              llvm::SmallPtrSetImpl<const Decl *> &Seen) {
1457  if (const auto *Arr = Ctx.getAsArrayType(Ty))
1458    Ty = Ctx.getBaseElementType(Arr);
1459
1460  const auto *Record = Ty->getAsCXXRecordDecl();
1461  if (!Record)
1462    return false;
1463
1464  // We've already checked this type, or are in the process of checking it.
1465  if (!Seen.insert(Record).second)
1466    return false;
1467
1468   (0) . __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1469, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Record->hasDefinition() &&
1469 (0) . __assert_fail ("Record->hasDefinition() && \"Incomplete types should already be diagnosed\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 1469, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">         "Incomplete types should already be diagnosed");
1470
1471  if (Record->isDynamicClass())
1472    return true;
1473
1474  for (FieldDecl *F : Record->fields()) {
1475    if (TypeRequiresBuiltinLaunderImp(Ctx, F->getType(), Seen))
1476      return true;
1477  }
1478  return false;
1479}
1480
1481/// Determine if the specified type requires laundering by checking if it is a
1482/// dynamic class type or contains a subobject which is a dynamic class type.
1483static bool TypeRequiresBuiltinLaunder(CodeGenModule &CGM, QualType Ty) {
1484  if (!CGM.getCodeGenOpts().StrictVTablePointers)
1485    return false;
1486  llvm::SmallPtrSet<const Decl *, 16> Seen;
1487  return TypeRequiresBuiltinLaunderImp(CGM.getContext(), Ty, Seen);
1488}
1489
1490RValue CodeGenFunction::emitRotate(const CallExpr *E, bool IsRotateRight) {
1491  llvm::Value *Src = EmitScalarExpr(E->getArg(0));
1492  llvm::Value *ShiftAmt = EmitScalarExpr(E->getArg(1));
1493
1494  // The builtin's shift arg may have a different type than the source arg and
1495  // result, but the LLVM intrinsic uses the same type for all values.
1496  llvm::Type *Ty = Src->getType();
1497  ShiftAmt = Builder.CreateIntCast(ShiftAmt, Ty, false);
1498
1499  // Rotate is a special case of LLVM funnel shift - 1st 2 args are the same.
1500  unsigned IID = IsRotateRight ? Intrinsic::fshr : Intrinsic::fshl;
1501  Function *F = CGM.getIntrinsic(IID, Ty);
1502  return RValue::get(Builder.CreateCall(F, { Src, Src, ShiftAmt }));
1503}
1504
1505RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
1506                                        const CallExpr *E,
1507                                        ReturnValueSlot ReturnValue) {
1508  const FunctionDecl *FD = GD.getDecl()->getAsFunction();
1509  // See if we can constant fold this builtin.  If so, don't emit it at all.
1510  Expr::EvalResult Result;
1511  if (E->EvaluateAsRValue(Result, CGM.getContext()) &&
1512      !Result.hasSideEffects()) {
1513    if (Result.Val.isInt())
1514      return RValue::get(llvm::ConstantInt::get(getLLVMContext(),
1515                                                Result.Val.getInt()));
1516    if (Result.Val.isFloat())
1517      return RValue::get(llvm::ConstantFP::get(getLLVMContext(),
1518                                               Result.Val.getFloat()));
1519  }
1520
1521  // There are LLVM math intrinsics/instructions corresponding to math library
1522  // functions except the LLVM op will never set errno while the math library
1523  // might. Also, math builtins have the same semantics as their math library
1524  // twins. Thus, we can transform math library and builtin calls to their
1525  // LLVM counterparts if the call is marked 'const' (known to never set errno).
1526  if (FD->hasAttr<ConstAttr>()) {
1527    switch (BuiltinID) {
1528    case Builtin::BIceil:
1529    case Builtin::BIceilf:
1530    case Builtin::BIceill:
1531    case Builtin::BI__builtin_ceil:
1532    case Builtin::BI__builtin_ceilf:
1533    case Builtin::BI__builtin_ceill:
1534      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::ceil));
1535
1536    case Builtin::BIcopysign:
1537    case Builtin::BIcopysignf:
1538    case Builtin::BIcopysignl:
1539    case Builtin::BI__builtin_copysign:
1540    case Builtin::BI__builtin_copysignf:
1541    case Builtin::BI__builtin_copysignl:
1542    case Builtin::BI__builtin_copysignf128:
1543      return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::copysign));
1544
1545    case Builtin::BIcos:
1546    case Builtin::BIcosf:
1547    case Builtin::BIcosl:
1548    case Builtin::BI__builtin_cos:
1549    case Builtin::BI__builtin_cosf:
1550    case Builtin::BI__builtin_cosl:
1551      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::cos));
1552
1553    case Builtin::BIexp:
1554    case Builtin::BIexpf:
1555    case Builtin::BIexpl:
1556    case Builtin::BI__builtin_exp:
1557    case Builtin::BI__builtin_expf:
1558    case Builtin::BI__builtin_expl:
1559      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::exp));
1560
1561    case Builtin::BIexp2:
1562    case Builtin::BIexp2f:
1563    case Builtin::BIexp2l:
1564    case Builtin::BI__builtin_exp2:
1565    case Builtin::BI__builtin_exp2f:
1566    case Builtin::BI__builtin_exp2l:
1567      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::exp2));
1568
1569    case Builtin::BIfabs:
1570    case Builtin::BIfabsf:
1571    case Builtin::BIfabsl:
1572    case Builtin::BI__builtin_fabs:
1573    case Builtin::BI__builtin_fabsf:
1574    case Builtin::BI__builtin_fabsl:
1575    case Builtin::BI__builtin_fabsf128:
1576      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::fabs));
1577
1578    case Builtin::BIfloor:
1579    case Builtin::BIfloorf:
1580    case Builtin::BIfloorl:
1581    case Builtin::BI__builtin_floor:
1582    case Builtin::BI__builtin_floorf:
1583    case Builtin::BI__builtin_floorl:
1584      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::floor));
1585
1586    case Builtin::BIfma:
1587    case Builtin::BIfmaf:
1588    case Builtin::BIfmal:
1589    case Builtin::BI__builtin_fma:
1590    case Builtin::BI__builtin_fmaf:
1591    case Builtin::BI__builtin_fmal:
1592      return RValue::get(emitTernaryBuiltin(*this, E, Intrinsic::fma));
1593
1594    case Builtin::BIfmax:
1595    case Builtin::BIfmaxf:
1596    case Builtin::BIfmaxl:
1597    case Builtin::BI__builtin_fmax:
1598    case Builtin::BI__builtin_fmaxf:
1599    case Builtin::BI__builtin_fmaxl:
1600      return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::maxnum));
1601
1602    case Builtin::BIfmin:
1603    case Builtin::BIfminf:
1604    case Builtin::BIfminl:
1605    case Builtin::BI__builtin_fmin:
1606    case Builtin::BI__builtin_fminf:
1607    case Builtin::BI__builtin_fminl:
1608      return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::minnum));
1609
1610    // fmod() is a special-case. It maps to the frem instruction rather than an
1611    // LLVM intrinsic.
1612    case Builtin::BIfmod:
1613    case Builtin::BIfmodf:
1614    case Builtin::BIfmodl:
1615    case Builtin::BI__builtin_fmod:
1616    case Builtin::BI__builtin_fmodf:
1617    case Builtin::BI__builtin_fmodl: {
1618      Value *Arg1 = EmitScalarExpr(E->getArg(0));
1619      Value *Arg2 = EmitScalarExpr(E->getArg(1));
1620      return RValue::get(Builder.CreateFRem(Arg1, Arg2, "fmod"));
1621    }
1622
1623    case Builtin::BIlog:
1624    case Builtin::BIlogf:
1625    case Builtin::BIlogl:
1626    case Builtin::BI__builtin_log:
1627    case Builtin::BI__builtin_logf:
1628    case Builtin::BI__builtin_logl:
1629      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log));
1630
1631    case Builtin::BIlog10:
1632    case Builtin::BIlog10f:
1633    case Builtin::BIlog10l:
1634    case Builtin::BI__builtin_log10:
1635    case Builtin::BI__builtin_log10f:
1636    case Builtin::BI__builtin_log10l:
1637      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log10));
1638
1639    case Builtin::BIlog2:
1640    case Builtin::BIlog2f:
1641    case Builtin::BIlog2l:
1642    case Builtin::BI__builtin_log2:
1643    case Builtin::BI__builtin_log2f:
1644    case Builtin::BI__builtin_log2l:
1645      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::log2));
1646
1647    case Builtin::BInearbyint:
1648    case Builtin::BInearbyintf:
1649    case Builtin::BInearbyintl:
1650    case Builtin::BI__builtin_nearbyint:
1651    case Builtin::BI__builtin_nearbyintf:
1652    case Builtin::BI__builtin_nearbyintl:
1653      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::nearbyint));
1654
1655    case Builtin::BIpow:
1656    case Builtin::BIpowf:
1657    case Builtin::BIpowl:
1658    case Builtin::BI__builtin_pow:
1659    case Builtin::BI__builtin_powf:
1660    case Builtin::BI__builtin_powl:
1661      return RValue::get(emitBinaryBuiltin(*this, E, Intrinsic::pow));
1662
1663    case Builtin::BIrint:
1664    case Builtin::BIrintf:
1665    case Builtin::BIrintl:
1666    case Builtin::BI__builtin_rint:
1667    case Builtin::BI__builtin_rintf:
1668    case Builtin::BI__builtin_rintl:
1669      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::rint));
1670
1671    case Builtin::BIround:
1672    case Builtin::BIroundf:
1673    case Builtin::BIroundl:
1674    case Builtin::BI__builtin_round:
1675    case Builtin::BI__builtin_roundf:
1676    case Builtin::BI__builtin_roundl:
1677      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::round));
1678
1679    case Builtin::BIsin:
1680    case Builtin::BIsinf:
1681    case Builtin::BIsinl:
1682    case Builtin::BI__builtin_sin:
1683    case Builtin::BI__builtin_sinf:
1684    case Builtin::BI__builtin_sinl:
1685      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::sin));
1686
1687    case Builtin::BIsqrt:
1688    case Builtin::BIsqrtf:
1689    case Builtin::BIsqrtl:
1690    case Builtin::BI__builtin_sqrt:
1691    case Builtin::BI__builtin_sqrtf:
1692    case Builtin::BI__builtin_sqrtl:
1693      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::sqrt));
1694
1695    case Builtin::BItrunc:
1696    case Builtin::BItruncf:
1697    case Builtin::BItruncl:
1698    case Builtin::BI__builtin_trunc:
1699    case Builtin::BI__builtin_truncf:
1700    case Builtin::BI__builtin_truncl:
1701      return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::trunc));
1702
1703    default:
1704      break;
1705    }
1706  }
1707
1708  switch (BuiltinID) {
1709  default: break;
1710  case Builtin::BI__builtin___CFStringMakeConstantString:
1711  case Builtin::BI__builtin___NSStringMakeConstantString:
1712    return RValue::get(ConstantEmitter(*this).emitAbstract(E, E->getType()));
1713  case Builtin::BI__builtin_stdarg_start:
1714  case Builtin::BI__builtin_va_start:
1715  case Builtin::BI__va_start:
1716  case Builtin::BI__builtin_va_end:
1717    return RValue::get(
1718        EmitVAStartEnd(BuiltinID == Builtin::BI__va_start
1719                           ? EmitScalarExpr(E->getArg(0))
1720                           : EmitVAListRef(E->getArg(0)).getPointer(),
1721                       BuiltinID != Builtin::BI__builtin_va_end));
1722  case Builtin::BI__builtin_va_copy: {
1723    Value *DstPtr = EmitVAListRef(E->getArg(0)).getPointer();
1724    Value *SrcPtr = EmitVAListRef(E->getArg(1)).getPointer();
1725
1726    llvm::Type *Type = Int8PtrTy;
1727
1728    DstPtr = Builder.CreateBitCast(DstPtr, Type);
1729    SrcPtr = Builder.CreateBitCast(SrcPtr, Type);
1730    return RValue::get(Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy),
1731                                          {DstPtr, SrcPtr}));
1732  }
1733  case Builtin::BI__builtin_abs:
1734  case Builtin::BI__builtin_labs:
1735  case Builtin::BI__builtin_llabs: {
1736    // X < 0 ? -X : X
1737    // The negation has 'nsw' because abs of INT_MIN is undefined.
1738    Value *ArgValue = EmitScalarExpr(E->getArg(0));
1739    Value *NegOp = Builder.CreateNSWNeg(ArgValue, "neg");
1740    Constant *Zero = llvm::Constant::getNullValue(ArgValue->getType());
1741    Value *CmpResult = Builder.CreateICmpSLT(ArgValue, Zero, "abscond");
1742    Value *Result = Builder.CreateSelect(CmpResult, NegOp, ArgValue, "abs");
1743    return RValue::get(Result);
1744  }
1745  case Builtin::BI__builtin_conj:
1746  case Builtin::BI__builtin_conjf:
1747  case Builtin::BI__builtin_conjl: {
1748    ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
1749    Value *Real = ComplexVal.first;
1750    Value *Imag = ComplexVal.second;
1751    Value *Zero =
1752      Imag->getType()->isFPOrFPVectorTy()
1753        ? llvm::ConstantFP::getZeroValueForNegation(Imag->getType())
1754        : llvm::Constant::getNullValue(Imag->getType());
1755
1756    Imag = Builder.CreateFSub(Zero, Imag, "sub");
1757    return RValue::getComplex(std::make_pair(Real, Imag));
1758  }
1759  case Builtin::BI__builtin_creal:
1760  case Builtin::BI__builtin_crealf:
1761  case Builtin::BI__builtin_creall:
1762  case Builtin::BIcreal:
1763  case Builtin::BIcrealf:
1764  case Builtin::BIcreall: {
1765    ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
1766    return RValue::get(ComplexVal.first);
1767  }
1768
1769  case Builtin::BI__builtin_dump_struct: {
1770    llvm::Type *LLVMIntTy = getTypes().ConvertType(getContext().IntTy);
1771    llvm::FunctionType *LLVMFuncType = llvm::FunctionType::get(
1772        LLVMIntTy, {llvm::Type::getInt8PtrTy(getLLVMContext())}, true);
1773
1774    Value *Func = EmitScalarExpr(E->getArg(1)->IgnoreImpCasts());
1775    CharUnits Arg0Align = EmitPointerWithAlignment(E->getArg(0)).getAlignment();
1776
1777    const Expr *Arg0 = E->getArg(0)->IgnoreImpCasts();
1778    QualType Arg0Type = Arg0->getType()->getPointeeType();
1779
1780    Value *RecordPtr = EmitScalarExpr(Arg0);
1781    Value *Res = dumpRecord(*this, Arg0Type, RecordPtr, Arg0Align,
1782                            {LLVMFuncType, Func}, 0);
1783    return RValue::get(Res);
1784  }
1785
1786  case Builtin::BI__builtin_cimag:
1787  case Builtin::BI__builtin_cimagf:
1788  case Builtin::BI__builtin_cimagl:
1789  case Builtin::BIcimag:
1790  case Builtin::BIcimagf:
1791  case Builtin::BIcimagl: {
1792    ComplexPairTy ComplexVal = EmitComplexExpr(E->getArg(0));
1793    return RValue::get(ComplexVal.second);
1794  }
1795
1796  case Builtin::BI__builtin_clrsb:
1797  case Builtin::BI__builtin_clrsbl:
1798  case Builtin::BI__builtin_clrsbll: {
1799    // clrsb(x) -> clz(x < 0 ? ~x : x) - 1 or
1800    Value *ArgValue = EmitScalarExpr(E->getArg(0));
1801
1802    llvm::Type *ArgType = ArgValue->getType();
1803    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
1804
1805    llvm::Type *ResultType = ConvertType(E->getType());
1806    Value *Zero = llvm::Constant::getNullValue(ArgType);
1807    Value *IsNeg = Builder.CreateICmpSLT(ArgValue, Zero, "isneg");
1808    Value *Inverse = Builder.CreateNot(ArgValue, "not");
1809    Value *Tmp = Builder.CreateSelect(IsNeg, Inverse, ArgValue);
1810    Value *Ctlz = Builder.CreateCall(F, {Tmp, Builder.getFalse()});
1811    Value *Result = Builder.CreateSub(Ctlz, llvm::ConstantInt::get(ArgType, 1));
1812    Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1813                                   "cast");
1814    return RValue::get(Result);
1815  }
1816  case Builtin::BI__builtin_ctzs:
1817  case Builtin::BI__builtin_ctz:
1818  case Builtin::BI__builtin_ctzl:
1819  case Builtin::BI__builtin_ctzll: {
1820    Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CTZPassedZero);
1821
1822    llvm::Type *ArgType = ArgValue->getType();
1823    Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
1824
1825    llvm::Type *ResultType = ConvertType(E->getType());
1826    Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef());
1827    Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef});
1828    if (Result->getType() != ResultType)
1829      Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1830                                     "cast");
1831    return RValue::get(Result);
1832  }
1833  case Builtin::BI__builtin_clzs:
1834  case Builtin::BI__builtin_clz:
1835  case Builtin::BI__builtin_clzl:
1836  case Builtin::BI__builtin_clzll: {
1837    Value *ArgValue = EmitCheckedArgForBuiltin(E->getArg(0), BCK_CLZPassedZero);
1838
1839    llvm::Type *ArgType = ArgValue->getType();
1840    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
1841
1842    llvm::Type *ResultType = ConvertType(E->getType());
1843    Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef());
1844    Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef});
1845    if (Result->getType() != ResultType)
1846      Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1847                                     "cast");
1848    return RValue::get(Result);
1849  }
1850  case Builtin::BI__builtin_ffs:
1851  case Builtin::BI__builtin_ffsl:
1852  case Builtin::BI__builtin_ffsll: {
1853    // ffs(x) -> x ? cttz(x) + 1 : 0
1854    Value *ArgValue = EmitScalarExpr(E->getArg(0));
1855
1856    llvm::Type *ArgType = ArgValue->getType();
1857    Function *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
1858
1859    llvm::Type *ResultType = ConvertType(E->getType());
1860    Value *Tmp =
1861        Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}),
1862                          llvm::ConstantInt::get(ArgType, 1));
1863    Value *Zero = llvm::Constant::getNullValue(ArgType);
1864    Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero");
1865    Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs");
1866    if (Result->getType() != ResultType)
1867      Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1868                                     "cast");
1869    return RValue::get(Result);
1870  }
1871  case Builtin::BI__builtin_parity:
1872  case Builtin::BI__builtin_parityl:
1873  case Builtin::BI__builtin_parityll: {
1874    // parity(x) -> ctpop(x) & 1
1875    Value *ArgValue = EmitScalarExpr(E->getArg(0));
1876
1877    llvm::Type *ArgType = ArgValue->getType();
1878    Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType);
1879
1880    llvm::Type *ResultType = ConvertType(E->getType());
1881    Value *Tmp = Builder.CreateCall(F, ArgValue);
1882    Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1));
1883    if (Result->getType() != ResultType)
1884      Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1885                                     "cast");
1886    return RValue::get(Result);
1887  }
1888  case Builtin::BI__lzcnt16:
1889  case Builtin::BI__lzcnt:
1890  case Builtin::BI__lzcnt64: {
1891    Value *ArgValue = EmitScalarExpr(E->getArg(0));
1892
1893    llvm::Type *ArgType = ArgValue->getType();
1894    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
1895
1896    llvm::Type *ResultType = ConvertType(E->getType());
1897    Value *Result = Builder.CreateCall(F, {ArgValue, Builder.getFalse()});
1898    if (Result->getType() != ResultType)
1899      Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1900                                     "cast");
1901    return RValue::get(Result);
1902  }
1903  case Builtin::BI__popcnt16:
1904  case Builtin::BI__popcnt:
1905  case Builtin::BI__popcnt64:
1906  case Builtin::BI__builtin_popcount:
1907  case Builtin::BI__builtin_popcountl:
1908  case Builtin::BI__builtin_popcountll: {
1909    Value *ArgValue = EmitScalarExpr(E->getArg(0));
1910
1911    llvm::Type *ArgType = ArgValue->getType();
1912    Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType);
1913
1914    llvm::Type *ResultType = ConvertType(E->getType());
1915    Value *Result = Builder.CreateCall(F, ArgValue);
1916    if (Result->getType() != ResultType)
1917      Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
1918                                     "cast");
1919    return RValue::get(Result);
1920  }
1921  case Builtin::BI__builtin_unpredictable: {
1922    // Always return the argument of __builtin_unpredictable. LLVM does not
1923    // handle this builtin. Metadata for this builtin should be added directly
1924    // to instructions such as branches or switches that use it.
1925    return RValue::get(EmitScalarExpr(E->getArg(0)));
1926  }
1927  case Builtin::BI__builtin_expect: {
1928    Value *ArgValue = EmitScalarExpr(E->getArg(0));
1929    llvm::Type *ArgType = ArgValue->getType();
1930
1931    Value *ExpectedValue = EmitScalarExpr(E->getArg(1));
1932    // Don't generate llvm.expect on -O0 as the backend won't use it for
1933    // anything.
1934    // Note, we still IRGen ExpectedValue because it could have side-effects.
1935    if (CGM.getCodeGenOpts().OptimizationLevel == 0)
1936      return RValue::get(ArgValue);
1937
1938    Function *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType);
1939    Value *Result =
1940        Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval");
1941    return RValue::get(Result);
1942  }
1943  case Builtin::BI__builtin_assume_aligned: {
1944    const Expr *Ptr = E->getArg(0);
1945    Value *PtrValue = EmitScalarExpr(Ptr);
1946    Value *OffsetValue =
1947      (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) : nullptr;
1948
1949    Value *AlignmentValue = EmitScalarExpr(E->getArg(1));
1950    ConstantInt *AlignmentCI = cast<ConstantInt>(AlignmentValue);
1951    unsigned Alignment = (unsigned)AlignmentCI->getZExtValue();
1952
1953    EmitAlignmentAssumption(PtrValue, Ptr,
1954                            /*The expr loc is sufficient.*/ SourceLocation(),
1955                            Alignment, OffsetValue);
1956    return RValue::get(PtrValue);
1957  }
1958  case Builtin::BI__assume:
1959  case Builtin::BI__builtin_assume: {
1960    if (E->getArg(0)->HasSideEffects(getContext()))
1961      return RValue::get(nullptr);
1962
1963    Value *ArgValue = EmitScalarExpr(E->getArg(0));
1964    Function *FnAssume = CGM.getIntrinsic(Intrinsic::assume);
1965    return RValue::get(Builder.CreateCall(FnAssume, ArgValue));
1966  }
1967  case Builtin::BI__builtin_bswap16:
1968  case Builtin::BI__builtin_bswap32:
1969  case Builtin::BI__builtin_bswap64: {
1970    return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bswap));
1971  }
1972  case Builtin::BI__builtin_bitreverse8:
1973  case Builtin::BI__builtin_bitreverse16:
1974  case Builtin::BI__builtin_bitreverse32:
1975  case Builtin::BI__builtin_bitreverse64: {
1976    return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::bitreverse));
1977  }
1978  case Builtin::BI__builtin_rotateleft8:
1979  case Builtin::BI__builtin_rotateleft16:
1980  case Builtin::BI__builtin_rotateleft32:
1981  case Builtin::BI__builtin_rotateleft64:
1982  case Builtin::BI_rotl8: // Microsoft variants of rotate left
1983  case Builtin::BI_rotl16:
1984  case Builtin::BI_rotl:
1985  case Builtin::BI_lrotl:
1986  case Builtin::BI_rotl64:
1987    return emitRotate(E, false);
1988
1989  case Builtin::BI__builtin_rotateright8:
1990  case Builtin::BI__builtin_rotateright16:
1991  case Builtin::BI__builtin_rotateright32:
1992  case Builtin::BI__builtin_rotateright64:
1993  case Builtin::BI_rotr8: // Microsoft variants of rotate right
1994  case Builtin::BI_rotr16:
1995  case Builtin::BI_rotr:
1996  case Builtin::BI_lrotr:
1997  case Builtin::BI_rotr64:
1998    return emitRotate(E, true);
1999
2000  case Builtin::BI__builtin_constant_p: {
2001    llvm::Type *ResultType = ConvertType(E->getType());
2002    if (CGM.getCodeGenOpts().OptimizationLevel == 0)
2003      // At -O0, we don't perform inlining, so we don't need to delay the
2004      // processing.
2005      return RValue::get(ConstantInt::get(ResultType, 0));
2006
2007    const Expr *Arg = E->getArg(0);
2008    QualType ArgType = Arg->getType();
2009    if (!hasScalarEvaluationKind(ArgType) || ArgType->isFunctionType())
2010      // We can only reason about scalar types.
2011      return RValue::get(ConstantInt::get(ResultType, 0));
2012
2013    Value *ArgValue = EmitScalarExpr(Arg);
2014    if (ArgType->isObjCObjectPointerType()) {
2015      // Convert Objective-C objects to id because we cannot distinguish between
2016      // LLVM types for Obj-C classes as they are opaque.
2017      ArgType = CGM.getContext().getObjCIdType();
2018      ArgValue = Builder.CreateBitCast(ArgValue, ConvertType(ArgType));
2019    }
2020    Function *F =
2021        CGM.getIntrinsic(Intrinsic::is_constant, ConvertType(ArgType));
2022    Value *Result = Builder.CreateCall(F, ArgValue);
2023    if (Result->getType() != ResultType)
2024      Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/false);
2025    return RValue::get(Result);
2026  }
2027  case Builtin::BI__builtin_dynamic_object_size:
2028  case Builtin::BI__builtin_object_size: {
2029    unsigned Type =
2030        E->getArg(1)->EvaluateKnownConstInt(getContext()).getZExtValue();
2031    auto *ResType = cast<llvm::IntegerType>(ConvertType(E->getType()));
2032
2033    // We pass this builtin onto the optimizer so that it can figure out the
2034    // object size in more complex cases.
2035    bool IsDynamic = BuiltinID == Builtin::BI__builtin_dynamic_object_size;
2036    return RValue::get(emitBuiltinObjectSize(E->getArg(0), Type, ResType,
2037                                             /*EmittedE=*/nullptr, IsDynamic));
2038  }
2039  case Builtin::BI__builtin_prefetch: {
2040    Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0));
2041    // FIXME: Technically these constants should of type 'int', yes?
2042    RW = (E->getNumArgs() > 1) ? EmitScalarExpr(E->getArg(1)) :
2043      llvm::ConstantInt::get(Int32Ty, 0);
2044    Locality = (E->getNumArgs() > 2) ? EmitScalarExpr(E->getArg(2)) :
2045      llvm::ConstantInt::get(Int32Ty, 3);
2046    Value *Data = llvm::ConstantInt::get(Int32Ty, 1);
2047    Function *F = CGM.getIntrinsic(Intrinsic::prefetch);
2048    return RValue::get(Builder.CreateCall(F, {Address, RW, Locality, Data}));
2049  }
2050  case Builtin::BI__builtin_readcyclecounter: {
2051    Function *F = CGM.getIntrinsic(Intrinsic::readcyclecounter);
2052    return RValue::get(Builder.CreateCall(F));
2053  }
2054  case Builtin::BI__builtin___clear_cache: {
2055    Value *Begin = EmitScalarExpr(E->getArg(0));
2056    Value *End = EmitScalarExpr(E->getArg(1));
2057    Function *F = CGM.getIntrinsic(Intrinsic::clear_cache);
2058    return RValue::get(Builder.CreateCall(F, {Begin, End}));
2059  }
2060  case Builtin::BI__builtin_trap:
2061    return RValue::get(EmitTrapCall(Intrinsic::trap));
2062  case Builtin::BI__debugbreak:
2063    return RValue::get(EmitTrapCall(Intrinsic::debugtrap));
2064  case Builtin::BI__builtin_unreachable: {
2065    EmitUnreachable(E->getExprLoc());
2066
2067    // We do need to preserve an insertion point.
2068    EmitBlock(createBasicBlock("unreachable.cont"));
2069
2070    return RValue::get(nullptr);
2071  }
2072
2073  case Builtin::BI__builtin_powi:
2074  case Builtin::BI__builtin_powif:
2075  case Builtin::BI__builtin_powil: {
2076    Value *Base = EmitScalarExpr(E->getArg(0));
2077    Value *Exponent = EmitScalarExpr(E->getArg(1));
2078    llvm::Type *ArgType = Base->getType();
2079    Function *F = CGM.getIntrinsic(Intrinsic::powi, ArgType);
2080    return RValue::get(Builder.CreateCall(F, {Base, Exponent}));
2081  }
2082
2083  case Builtin::BI__builtin_isgreater:
2084  case Builtin::BI__builtin_isgreaterequal:
2085  case Builtin::BI__builtin_isless:
2086  case Builtin::BI__builtin_islessequal:
2087  case Builtin::BI__builtin_islessgreater:
2088  case Builtin::BI__builtin_isunordered: {
2089    // Ordered comparisons: we know the arguments to these are matching scalar
2090    // floating point values.
2091    Value *LHS = EmitScalarExpr(E->getArg(0));
2092    Value *RHS = EmitScalarExpr(E->getArg(1));
2093
2094    switch (BuiltinID) {
2095    default: llvm_unreachable("Unknown ordered comparison");
2096    case Builtin::BI__builtin_isgreater:
2097      LHS = Builder.CreateFCmpOGT(LHS, RHS, "cmp");
2098      break;
2099    case Builtin::BI__builtin_isgreaterequal:
2100      LHS = Builder.CreateFCmpOGE(LHS, RHS, "cmp");
2101      break;
2102    case Builtin::BI__builtin_isless:
2103      LHS = Builder.CreateFCmpOLT(LHS, RHS, "cmp");
2104      break;
2105    case Builtin::BI__builtin_islessequal:
2106      LHS = Builder.CreateFCmpOLE(LHS, RHS, "cmp");
2107      break;
2108    case Builtin::BI__builtin_islessgreater:
2109      LHS = Builder.CreateFCmpONE(LHS, RHS, "cmp");
2110      break;
2111    case Builtin::BI__builtin_isunordered:
2112      LHS = Builder.CreateFCmpUNO(LHS, RHS, "cmp");
2113      break;
2114    }
2115    // ZExt bool to int type.
2116    return RValue::get(Builder.CreateZExt(LHS, ConvertType(E->getType())));
2117  }
2118  case Builtin::BI__builtin_isnan: {
2119    Value *V = EmitScalarExpr(E->getArg(0));
2120    V = Builder.CreateFCmpUNO(V, V, "cmp");
2121    return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType())));
2122  }
2123
2124  case Builtin::BIfinite:
2125  case Builtin::BI__finite:
2126  case Builtin::BIfinitef:
2127  case Builtin::BI__finitef:
2128  case Builtin::BIfinitel:
2129  case Builtin::BI__finitel:
2130  case Builtin::BI__builtin_isinf:
2131  case Builtin::BI__builtin_isfinite: {
2132    // isinf(x)    --> fabs(x) == infinity
2133    // isfinite(x) --> fabs(x) != infinity
2134    // x != NaN via the ordered compare in either case.
2135    Value *V = EmitScalarExpr(E->getArg(0));
2136    Value *Fabs = EmitFAbs(*this, V);
2137    Constant *Infinity = ConstantFP::getInfinity(V->getType());
2138    CmpInst::Predicate Pred = (BuiltinID == Builtin::BI__builtin_isinf)
2139                                  ? CmpInst::FCMP_OEQ
2140                                  : CmpInst::FCMP_ONE;
2141    Value *FCmp = Builder.CreateFCmp(Pred, Fabs, Infinity, "cmpinf");
2142    return RValue::get(Builder.CreateZExt(FCmp, ConvertType(E->getType())));
2143  }
2144
2145  case Builtin::BI__builtin_isinf_sign: {
2146    // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0
2147    Value *Arg = EmitScalarExpr(E->getArg(0));
2148    Value *AbsArg = EmitFAbs(*this, Arg);
2149    Value *IsInf = Builder.CreateFCmpOEQ(
2150        AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf");
2151    Value *IsNeg = EmitSignBit(*this, Arg);
2152
2153    llvm::Type *IntTy = ConvertType(E->getType());
2154    Value *Zero = Constant::getNullValue(IntTy);
2155    Value *One = ConstantInt::get(IntTy, 1);
2156    Value *NegativeOne = ConstantInt::get(IntTy, -1);
2157    Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One);
2158    Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero);
2159    return RValue::get(Result);
2160  }
2161
2162  case Builtin::BI__builtin_isnormal: {
2163    // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min
2164    Value *V = EmitScalarExpr(E->getArg(0));
2165    Value *Eq = Builder.CreateFCmpOEQ(V, V, "iseq");
2166
2167    Value *Abs = EmitFAbs(*this, V);
2168    Value *IsLessThanInf =
2169      Builder.CreateFCmpULT(Abs, ConstantFP::getInfinity(V->getType()),"isinf");
2170    APFloat Smallest = APFloat::getSmallestNormalized(
2171                   getContext().getFloatTypeSemantics(E->getArg(0)->getType()));
2172    Value *IsNormal =
2173      Builder.CreateFCmpUGE(Abs, ConstantFP::get(V->getContext(), Smallest),
2174                            "isnormal");
2175    V = Builder.CreateAnd(Eq, IsLessThanInf, "and");
2176    V = Builder.CreateAnd(V, IsNormal, "and");
2177    return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType())));
2178  }
2179
2180  case Builtin::BI__builtin_flt_rounds: {
2181    Function *F = CGM.getIntrinsic(Intrinsic::flt_rounds);
2182
2183    llvm::Type *ResultType = ConvertType(E->getType());
2184    Value *Result = Builder.CreateCall(F);
2185    if (Result->getType() != ResultType)
2186      Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
2187                                     "cast");
2188    return RValue::get(Result);
2189  }
2190
2191  case Builtin::BI__builtin_fpclassify: {
2192    Value *V = EmitScalarExpr(E->getArg(5));
2193    llvm::Type *Ty = ConvertType(E->getArg(5)->getType());
2194
2195    // Create Result
2196    BasicBlock *Begin = Builder.GetInsertBlock();
2197    BasicBlock *End = createBasicBlock("fpclassify_end", this->CurFn);
2198    Builder.SetInsertPoint(End);
2199    PHINode *Result =
2200      Builder.CreatePHI(ConvertType(E->getArg(0)->getType()), 4,
2201                        "fpclassify_result");
2202
2203    // if (V==0) return FP_ZERO
2204    Builder.SetInsertPoint(Begin);
2205    Value *IsZero = Builder.CreateFCmpOEQ(V, Constant::getNullValue(Ty),
2206                                          "iszero");
2207    Value *ZeroLiteral = EmitScalarExpr(E->getArg(4));
2208    BasicBlock *NotZero = createBasicBlock("fpclassify_not_zero", this->CurFn);
2209    Builder.CreateCondBr(IsZero, End, NotZero);
2210    Result->addIncoming(ZeroLiteral, Begin);
2211
2212    // if (V != V) return FP_NAN
2213    Builder.SetInsertPoint(NotZero);
2214    Value *IsNan = Builder.CreateFCmpUNO(V, V, "cmp");
2215    Value *NanLiteral = EmitScalarExpr(E->getArg(0));
2216    BasicBlock *NotNan = createBasicBlock("fpclassify_not_nan", this->CurFn);
2217    Builder.CreateCondBr(IsNan, End, NotNan);
2218    Result->addIncoming(NanLiteral, NotZero);
2219
2220    // if (fabs(V) == infinity) return FP_INFINITY
2221    Builder.SetInsertPoint(NotNan);
2222    Value *VAbs = EmitFAbs(*this, V);
2223    Value *IsInf =
2224      Builder.CreateFCmpOEQ(VAbs, ConstantFP::getInfinity(V->getType()),
2225                            "isinf");
2226    Value *InfLiteral = EmitScalarExpr(E->getArg(1));
2227    BasicBlock *NotInf = createBasicBlock("fpclassify_not_inf", this->CurFn);
2228    Builder.CreateCondBr(IsInf, End, NotInf);
2229    Result->addIncoming(InfLiteral, NotNan);
2230
2231    // if (fabs(V) >= MIN_NORMAL) return FP_NORMAL else FP_SUBNORMAL
2232    Builder.SetInsertPoint(NotInf);
2233    APFloat Smallest = APFloat::getSmallestNormalized(
2234        getContext().getFloatTypeSemantics(E->getArg(5)->getType()));
2235    Value *IsNormal =
2236      Builder.CreateFCmpUGE(VAbs, ConstantFP::get(V->getContext(), Smallest),
2237                            "isnormal");
2238    Value *NormalResult =
2239      Builder.CreateSelect(IsNormal, EmitScalarExpr(E->getArg(2)),
2240                           EmitScalarExpr(E->getArg(3)));
2241    Builder.CreateBr(End);
2242    Result->addIncoming(NormalResult, NotInf);
2243
2244    // return Result
2245    Builder.SetInsertPoint(End);
2246    return RValue::get(Result);
2247  }
2248
2249  case Builtin::BIalloca:
2250  case Builtin::BI_alloca:
2251  case Builtin::BI__builtin_alloca: {
2252    Value *Size = EmitScalarExpr(E->getArg(0));
2253    const TargetInfo &TI = getContext().getTargetInfo();
2254    // The alignment of the alloca should correspond to __BIGGEST_ALIGNMENT__.
2255    unsigned SuitableAlignmentInBytes =
2256        CGM.getContext()
2257            .toCharUnitsFromBits(TI.getSuitableAlign())
2258            .getQuantity();
2259    AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size);
2260    AI->setAlignment(SuitableAlignmentInBytes);
2261    return RValue::get(AI);
2262  }
2263
2264  case Builtin::BI__builtin_alloca_with_align: {
2265    Value *Size = EmitScalarExpr(E->getArg(0));
2266    Value *AlignmentInBitsValue = EmitScalarExpr(E->getArg(1));
2267    auto *AlignmentInBitsCI = cast<ConstantInt>(AlignmentInBitsValue);
2268    unsigned AlignmentInBits = AlignmentInBitsCI->getZExtValue();
2269    unsigned AlignmentInBytes =
2270        CGM.getContext().toCharUnitsFromBits(AlignmentInBits).getQuantity();
2271    AllocaInst *AI = Builder.CreateAlloca(Builder.getInt8Ty(), Size);
2272    AI->setAlignment(AlignmentInBytes);
2273    return RValue::get(AI);
2274  }
2275
2276  case Builtin::BIbzero:
2277  case Builtin::BI__builtin_bzero: {
2278    Address Dest = EmitPointerWithAlignment(E->getArg(0));
2279    Value *SizeVal = EmitScalarExpr(E->getArg(1));
2280    EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
2281                        E->getArg(0)->getExprLoc(), FD, 0);
2282    Builder.CreateMemSet(Dest, Builder.getInt8(0), SizeVal, false);
2283    return RValue::get(nullptr);
2284  }
2285  case Builtin::BImemcpy:
2286  case Builtin::BI__builtin_memcpy: {
2287    Address Dest = EmitPointerWithAlignment(E->getArg(0));
2288    Address Src = EmitPointerWithAlignment(E->getArg(1));
2289    Value *SizeVal = EmitScalarExpr(E->getArg(2));
2290    EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
2291                        E->getArg(0)->getExprLoc(), FD, 0);
2292    EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(),
2293                        E->getArg(1)->getExprLoc(), FD, 1);
2294    Builder.CreateMemCpy(Dest, Src, SizeVal, false);
2295    return RValue::get(Dest.getPointer());
2296  }
2297
2298  case Builtin::BI__builtin_char_memchr:
2299    BuiltinID = Builtin::BI__builtin_memchr;
2300    break;
2301
2302  case Builtin::BI__builtin___memcpy_chk: {
2303    // fold __builtin_memcpy_chk(x, y, cst1, cst2) to memcpy iff cst1<=cst2.
2304    Expr::EvalResult SizeResult, DstSizeResult;
2305    if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) ||
2306        !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext()))
2307      break;
2308    llvm::APSInt Size = SizeResult.Val.getInt();
2309    llvm::APSInt DstSize = DstSizeResult.Val.getInt();
2310    if (Size.ugt(DstSize))
2311      break;
2312    Address Dest = EmitPointerWithAlignment(E->getArg(0));
2313    Address Src = EmitPointerWithAlignment(E->getArg(1));
2314    Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size);
2315    Builder.CreateMemCpy(Dest, Src, SizeVal, false);
2316    return RValue::get(Dest.getPointer());
2317  }
2318
2319  case Builtin::BI__builtin_objc_memmove_collectable: {
2320    Address DestAddr = EmitPointerWithAlignment(E->getArg(0));
2321    Address SrcAddr = EmitPointerWithAlignment(E->getArg(1));
2322    Value *SizeVal = EmitScalarExpr(E->getArg(2));
2323    CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this,
2324                                                  DestAddr, SrcAddr, SizeVal);
2325    return RValue::get(DestAddr.getPointer());
2326  }
2327
2328  case Builtin::BI__builtin___memmove_chk: {
2329    // fold __builtin_memmove_chk(x, y, cst1, cst2) to memmove iff cst1<=cst2.
2330    Expr::EvalResult SizeResult, DstSizeResult;
2331    if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) ||
2332        !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext()))
2333      break;
2334    llvm::APSInt Size = SizeResult.Val.getInt();
2335    llvm::APSInt DstSize = DstSizeResult.Val.getInt();
2336    if (Size.ugt(DstSize))
2337      break;
2338    Address Dest = EmitPointerWithAlignment(E->getArg(0));
2339    Address Src = EmitPointerWithAlignment(E->getArg(1));
2340    Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size);
2341    Builder.CreateMemMove(Dest, Src, SizeVal, false);
2342    return RValue::get(Dest.getPointer());
2343  }
2344
2345  case Builtin::BImemmove:
2346  case Builtin::BI__builtin_memmove: {
2347    Address Dest = EmitPointerWithAlignment(E->getArg(0));
2348    Address Src = EmitPointerWithAlignment(E->getArg(1));
2349    Value *SizeVal = EmitScalarExpr(E->getArg(2));
2350    EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
2351                        E->getArg(0)->getExprLoc(), FD, 0);
2352    EmitNonNullArgCheck(RValue::get(Src.getPointer()), E->getArg(1)->getType(),
2353                        E->getArg(1)->getExprLoc(), FD, 1);
2354    Builder.CreateMemMove(Dest, Src, SizeVal, false);
2355    return RValue::get(Dest.getPointer());
2356  }
2357  case Builtin::BImemset:
2358  case Builtin::BI__builtin_memset: {
2359    Address Dest = EmitPointerWithAlignment(E->getArg(0));
2360    Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)),
2361                                         Builder.getInt8Ty());
2362    Value *SizeVal = EmitScalarExpr(E->getArg(2));
2363    EmitNonNullArgCheck(RValue::get(Dest.getPointer()), E->getArg(0)->getType(),
2364                        E->getArg(0)->getExprLoc(), FD, 0);
2365    Builder.CreateMemSet(Dest, ByteVal, SizeVal, false);
2366    return RValue::get(Dest.getPointer());
2367  }
2368  case Builtin::BI__builtin___memset_chk: {
2369    // fold __builtin_memset_chk(x, y, cst1, cst2) to memset iff cst1<=cst2.
2370    Expr::EvalResult SizeResult, DstSizeResult;
2371    if (!E->getArg(2)->EvaluateAsInt(SizeResult, CGM.getContext()) ||
2372        !E->getArg(3)->EvaluateAsInt(DstSizeResult, CGM.getContext()))
2373      break;
2374    llvm::APSInt Size = SizeResult.Val.getInt();
2375    llvm::APSInt DstSize = DstSizeResult.Val.getInt();
2376    if (Size.ugt(DstSize))
2377      break;
2378    Address Dest = EmitPointerWithAlignment(E->getArg(0));
2379    Value *ByteVal = Builder.CreateTrunc(EmitScalarExpr(E->getArg(1)),
2380                                         Builder.getInt8Ty());
2381    Value *SizeVal = llvm::ConstantInt::get(Builder.getContext(), Size);
2382    Builder.CreateMemSet(Dest, ByteVal, SizeVal, false);
2383    return RValue::get(Dest.getPointer());
2384  }
2385  case Builtin::BI__builtin_wmemcmp: {
2386    // The MSVC runtime library does not provide a definition of wmemcmp, so we
2387    // need an inline implementation.
2388    if (!getTarget().getTriple().isOSMSVCRT())
2389      break;
2390
2391    llvm::Type *WCharTy = ConvertType(getContext().WCharTy);
2392
2393    Value *Dst = EmitScalarExpr(E->getArg(0));
2394    Value *Src = EmitScalarExpr(E->getArg(1));
2395    Value *Size = EmitScalarExpr(E->getArg(2));
2396
2397    BasicBlock *Entry = Builder.GetInsertBlock();
2398    BasicBlock *CmpGT = createBasicBlock("wmemcmp.gt");
2399    BasicBlock *CmpLT = createBasicBlock("wmemcmp.lt");
2400    BasicBlock *Next = createBasicBlock("wmemcmp.next");
2401    BasicBlock *Exit = createBasicBlock("wmemcmp.exit");
2402    Value *SizeEq0 = Builder.CreateICmpEQ(Size, ConstantInt::get(SizeTy, 0));
2403    Builder.CreateCondBr(SizeEq0, Exit, CmpGT);
2404
2405    EmitBlock(CmpGT);
2406    PHINode *DstPhi = Builder.CreatePHI(Dst->getType(), 2);
2407    DstPhi->addIncoming(Dst, Entry);
2408    PHINode *SrcPhi = Builder.CreatePHI(Src->getType(), 2);
2409    SrcPhi->addIncoming(Src, Entry);
2410    PHINode *SizePhi = Builder.CreatePHI(SizeTy, 2);
2411    SizePhi->addIncoming(Size, Entry);
2412    CharUnits WCharAlign =
2413        getContext().getTypeAlignInChars(getContext().WCharTy);
2414    Value *DstCh = Builder.CreateAlignedLoad(WCharTy, DstPhi, WCharAlign);
2415    Value *SrcCh = Builder.CreateAlignedLoad(WCharTy, SrcPhi, WCharAlign);
2416    Value *DstGtSrc = Builder.CreateICmpUGT(DstCh, SrcCh);
2417    Builder.CreateCondBr(DstGtSrc, Exit, CmpLT);
2418
2419    EmitBlock(CmpLT);
2420    Value *DstLtSrc = Builder.CreateICmpULT(DstCh, SrcCh);
2421    Builder.CreateCondBr(DstLtSrc, Exit, Next);
2422
2423    EmitBlock(Next);
2424    Value *NextDst = Builder.CreateConstInBoundsGEP1_32(WCharTy, DstPhi, 1);
2425    Value *NextSrc = Builder.CreateConstInBoundsGEP1_32(WCharTy, SrcPhi, 1);
2426    Value *NextSize = Builder.CreateSub(SizePhi, ConstantInt::get(SizeTy, 1));
2427    Value *NextSizeEq0 =
2428        Builder.CreateICmpEQ(NextSize, ConstantInt::get(SizeTy, 0));
2429    Builder.CreateCondBr(NextSizeEq0, Exit, CmpGT);
2430    DstPhi->addIncoming(NextDst, Next);
2431    SrcPhi->addIncoming(NextSrc, Next);
2432    SizePhi->addIncoming(NextSize, Next);
2433
2434    EmitBlock(Exit);
2435    PHINode *Ret = Builder.CreatePHI(IntTy, 4);
2436    Ret->addIncoming(ConstantInt::get(IntTy, 0), Entry);
2437    Ret->addIncoming(ConstantInt::get(IntTy, 1), CmpGT);
2438    Ret->addIncoming(ConstantInt::get(IntTy, -1), CmpLT);
2439    Ret->addIncoming(ConstantInt::get(IntTy, 0), Next);
2440    return RValue::get(Ret);
2441  }
2442  case Builtin::BI__builtin_dwarf_cfa: {
2443    // The offset in bytes from the first argument to the CFA.
2444    //
2445    // Why on earth is this in the frontend?  Is there any reason at
2446    // all that the backend can't reasonably determine this while
2447    // lowering llvm.eh.dwarf.cfa()?
2448    //
2449    // TODO: If there's a satisfactory reason, add a target hook for
2450    // this instead of hard-coding 0, which is correct for most targets.
2451    int32_t Offset = 0;
2452
2453    Function *F = CGM.getIntrinsic(Intrinsic::eh_dwarf_cfa);
2454    return RValue::get(Builder.CreateCall(F,
2455                                      llvm::ConstantInt::get(Int32Ty, Offset)));
2456  }
2457  case Builtin::BI__builtin_return_address: {
2458    Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0),
2459                                                   getContext().UnsignedIntTy);
2460    Function *F = CGM.getIntrinsic(Intrinsic::returnaddress);
2461    return RValue::get(Builder.CreateCall(F, Depth));
2462  }
2463  case Builtin::BI_ReturnAddress: {
2464    Function *F = CGM.getIntrinsic(Intrinsic::returnaddress);
2465    return RValue::get(Builder.CreateCall(F, Builder.getInt32(0)));
2466  }
2467  case Builtin::BI__builtin_frame_address: {
2468    Value *Depth = ConstantEmitter(*this).emitAbstract(E->getArg(0),
2469                                                   getContext().UnsignedIntTy);
2470    Function *F = CGM.getIntrinsic(Intrinsic::frameaddress);
2471    return RValue::get(Builder.CreateCall(F, Depth));
2472  }
2473  case Builtin::BI__builtin_extract_return_addr: {
2474    Value *Address = EmitScalarExpr(E->getArg(0));
2475    Value *Result = getTargetHooks().decodeReturnAddress(*this, Address);
2476    return RValue::get(Result);
2477  }
2478  case Builtin::BI__builtin_frob_return_addr: {
2479    Value *Address = EmitScalarExpr(E->getArg(0));
2480    Value *Result = getTargetHooks().encodeReturnAddress(*this, Address);
2481    return RValue::get(Result);
2482  }
2483  case Builtin::BI__builtin_dwarf_sp_column: {
2484    llvm::IntegerType *Ty
2485      = cast<llvm::IntegerType>(ConvertType(E->getType()));
2486    int Column = getTargetHooks().getDwarfEHStackPointer(CGM);
2487    if (Column == -1) {
2488      CGM.ErrorUnsupported(E, "__builtin_dwarf_sp_column");
2489      return RValue::get(llvm::UndefValue::get(Ty));
2490    }
2491    return RValue::get(llvm::ConstantInt::get(Ty, Column, true));
2492  }
2493  case Builtin::BI__builtin_init_dwarf_reg_size_table: {
2494    Value *Address = EmitScalarExpr(E->getArg(0));
2495    if (getTargetHooks().initDwarfEHRegSizeTable(*this, Address))
2496      CGM.ErrorUnsupported(E, "__builtin_init_dwarf_reg_size_table");
2497    return RValue::get(llvm::UndefValue::get(ConvertType(E->getType())));
2498  }
2499  case Builtin::BI__builtin_eh_return: {
2500    Value *Int = EmitScalarExpr(E->getArg(0));
2501    Value *Ptr = EmitScalarExpr(E->getArg(1));
2502
2503    llvm::IntegerType *IntTy = cast<llvm::IntegerType>(Int->getType());
2504     (0) . __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 2505, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert((IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) &&
2505 (0) . __assert_fail ("(IntTy->getBitWidth() == 32 || IntTy->getBitWidth() == 64) && \"LLVM's __builtin_eh_return only supports 32- and 64-bit variants\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 2505, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">           "LLVM's __builtin_eh_return only supports 32- and 64-bit variants");
2506    Function *F =
2507        CGM.getIntrinsic(IntTy->getBitWidth() == 32 ? Intrinsic::eh_return_i32
2508                                                    : Intrinsic::eh_return_i64);
2509    Builder.CreateCall(F, {Int, Ptr});
2510    Builder.CreateUnreachable();
2511
2512    // We do need to preserve an insertion point.
2513    EmitBlock(createBasicBlock("builtin_eh_return.cont"));
2514
2515    return RValue::get(nullptr);
2516  }
2517  case Builtin::BI__builtin_unwind_init: {
2518    Function *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init);
2519    return RValue::get(Builder.CreateCall(F));
2520  }
2521  case Builtin::BI__builtin_extend_pointer: {
2522    // Extends a pointer to the size of an _Unwind_Word, which is
2523    // uint64_t on all platforms.  Generally this gets poked into a
2524    // register and eventually used as an address, so if the
2525    // addressing registers are wider than pointers and the platform
2526    // doesn't implicitly ignore high-order bits when doing
2527    // addressing, we need to make sure we zext / sext based on
2528    // the platform's expectations.
2529    //
2530    // See: http://gcc.gnu.org/ml/gcc-bugs/2002-02/msg00237.html
2531
2532    // Cast the pointer to intptr_t.
2533    Value *Ptr = EmitScalarExpr(E->getArg(0));
2534    Value *Result = Builder.CreatePtrToInt(Ptr, IntPtrTy, "extend.cast");
2535
2536    // If that's 64 bits, we're done.
2537    if (IntPtrTy->getBitWidth() == 64)
2538      return RValue::get(Result);
2539
2540    // Otherwise, ask the codegen data what to do.
2541    if (getTargetHooks().extendPointerWithSExt())
2542      return RValue::get(Builder.CreateSExt(Result, Int64Ty, "extend.sext"));
2543    else
2544      return RValue::get(Builder.CreateZExt(Result, Int64Ty, "extend.zext"));
2545  }
2546  case Builtin::BI__builtin_setjmp: {
2547    // Buffer is a void**.
2548    Address Buf = EmitPointerWithAlignment(E->getArg(0));
2549
2550    // Store the frame pointer to the setjmp buffer.
2551    Value *FrameAddr =
2552      Builder.CreateCall(CGM.getIntrinsic(Intrinsic::frameaddress),
2553                         ConstantInt::get(Int32Ty, 0));
2554    Builder.CreateStore(FrameAddr, Buf);
2555
2556    // Store the stack pointer to the setjmp buffer.
2557    Value *StackAddr =
2558        Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave));
2559    Address StackSaveSlot = Builder.CreateConstInBoundsGEP(Buf, 2);
2560    Builder.CreateStore(StackAddr, StackSaveSlot);
2561
2562    // Call LLVM's EH setjmp, which is lightweight.
2563    Function *F = CGM.getIntrinsic(Intrinsic::eh_sjlj_setjmp);
2564    Buf = Builder.CreateBitCast(Buf, Int8PtrTy);
2565    return RValue::get(Builder.CreateCall(F, Buf.getPointer()));
2566  }
2567  case Builtin::BI__builtin_longjmp: {
2568    Value *Buf = EmitScalarExpr(E->getArg(0));
2569    Buf = Builder.CreateBitCast(Buf, Int8PtrTy);
2570
2571    // Call LLVM's EH longjmp, which is lightweight.
2572    Builder.CreateCall(CGM.getIntrinsic(Intrinsic::eh_sjlj_longjmp), Buf);
2573
2574    // longjmp doesn't return; mark this as unreachable.
2575    Builder.CreateUnreachable();
2576
2577    // We do need to preserve an insertion point.
2578    EmitBlock(createBasicBlock("longjmp.cont"));
2579
2580    return RValue::get(nullptr);
2581  }
2582  case Builtin::BI__builtin_launder: {
2583    const Expr *Arg = E->getArg(0);
2584    QualType ArgTy = Arg->getType()->getPointeeType();
2585    Value *Ptr = EmitScalarExpr(Arg);
2586    if (TypeRequiresBuiltinLaunder(CGM, ArgTy))
2587      Ptr = Builder.CreateLaunderInvariantGroup(Ptr);
2588
2589    return RValue::get(Ptr);
2590  }
2591  case Builtin::BI__sync_fetch_and_add:
2592  case Builtin::BI__sync_fetch_and_sub:
2593  case Builtin::BI__sync_fetch_and_or:
2594  case Builtin::BI__sync_fetch_and_and:
2595  case Builtin::BI__sync_fetch_and_xor:
2596  case Builtin::BI__sync_fetch_and_nand:
2597  case Builtin::BI__sync_add_and_fetch:
2598  case Builtin::BI__sync_sub_and_fetch:
2599  case Builtin::BI__sync_and_and_fetch:
2600  case Builtin::BI__sync_or_and_fetch:
2601  case Builtin::BI__sync_xor_and_fetch:
2602  case Builtin::BI__sync_nand_and_fetch:
2603  case Builtin::BI__sync_val_compare_and_swap:
2604  case Builtin::BI__sync_bool_compare_and_swap:
2605  case Builtin::BI__sync_lock_test_and_set:
2606  case Builtin::BI__sync_lock_release:
2607  case Builtin::BI__sync_swap:
2608    llvm_unreachable("Shouldn't make it through sema");
2609  case Builtin::BI__sync_fetch_and_add_1:
2610  case Builtin::BI__sync_fetch_and_add_2:
2611  case Builtin::BI__sync_fetch_and_add_4:
2612  case Builtin::BI__sync_fetch_and_add_8:
2613  case Builtin::BI__sync_fetch_and_add_16:
2614    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Add, E);
2615  case Builtin::BI__sync_fetch_and_sub_1:
2616  case Builtin::BI__sync_fetch_and_sub_2:
2617  case Builtin::BI__sync_fetch_and_sub_4:
2618  case Builtin::BI__sync_fetch_and_sub_8:
2619  case Builtin::BI__sync_fetch_and_sub_16:
2620    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Sub, E);
2621  case Builtin::BI__sync_fetch_and_or_1:
2622  case Builtin::BI__sync_fetch_and_or_2:
2623  case Builtin::BI__sync_fetch_and_or_4:
2624  case Builtin::BI__sync_fetch_and_or_8:
2625  case Builtin::BI__sync_fetch_and_or_16:
2626    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Or, E);
2627  case Builtin::BI__sync_fetch_and_and_1:
2628  case Builtin::BI__sync_fetch_and_and_2:
2629  case Builtin::BI__sync_fetch_and_and_4:
2630  case Builtin::BI__sync_fetch_and_and_8:
2631  case Builtin::BI__sync_fetch_and_and_16:
2632    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::And, E);
2633  case Builtin::BI__sync_fetch_and_xor_1:
2634  case Builtin::BI__sync_fetch_and_xor_2:
2635  case Builtin::BI__sync_fetch_and_xor_4:
2636  case Builtin::BI__sync_fetch_and_xor_8:
2637  case Builtin::BI__sync_fetch_and_xor_16:
2638    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xor, E);
2639  case Builtin::BI__sync_fetch_and_nand_1:
2640  case Builtin::BI__sync_fetch_and_nand_2:
2641  case Builtin::BI__sync_fetch_and_nand_4:
2642  case Builtin::BI__sync_fetch_and_nand_8:
2643  case Builtin::BI__sync_fetch_and_nand_16:
2644    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Nand, E);
2645
2646  // Clang extensions: not overloaded yet.
2647  case Builtin::BI__sync_fetch_and_min:
2648    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Min, E);
2649  case Builtin::BI__sync_fetch_and_max:
2650    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Max, E);
2651  case Builtin::BI__sync_fetch_and_umin:
2652    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMin, E);
2653  case Builtin::BI__sync_fetch_and_umax:
2654    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::UMax, E);
2655
2656  case Builtin::BI__sync_add_and_fetch_1:
2657  case Builtin::BI__sync_add_and_fetch_2:
2658  case Builtin::BI__sync_add_and_fetch_4:
2659  case Builtin::BI__sync_add_and_fetch_8:
2660  case Builtin::BI__sync_add_and_fetch_16:
2661    return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Add, E,
2662                                llvm::Instruction::Add);
2663  case Builtin::BI__sync_sub_and_fetch_1:
2664  case Builtin::BI__sync_sub_and_fetch_2:
2665  case Builtin::BI__sync_sub_and_fetch_4:
2666  case Builtin::BI__sync_sub_and_fetch_8:
2667  case Builtin::BI__sync_sub_and_fetch_16:
2668    return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Sub, E,
2669                                llvm::Instruction::Sub);
2670  case Builtin::BI__sync_and_and_fetch_1:
2671  case Builtin::BI__sync_and_and_fetch_2:
2672  case Builtin::BI__sync_and_and_fetch_4:
2673  case Builtin::BI__sync_and_and_fetch_8:
2674  case Builtin::BI__sync_and_and_fetch_16:
2675    return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::And, E,
2676                                llvm::Instruction::And);
2677  case Builtin::BI__sync_or_and_fetch_1:
2678  case Builtin::BI__sync_or_and_fetch_2:
2679  case Builtin::BI__sync_or_and_fetch_4:
2680  case Builtin::BI__sync_or_and_fetch_8:
2681  case Builtin::BI__sync_or_and_fetch_16:
2682    return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Or, E,
2683                                llvm::Instruction::Or);
2684  case Builtin::BI__sync_xor_and_fetch_1:
2685  case Builtin::BI__sync_xor_and_fetch_2:
2686  case Builtin::BI__sync_xor_and_fetch_4:
2687  case Builtin::BI__sync_xor_and_fetch_8:
2688  case Builtin::BI__sync_xor_and_fetch_16:
2689    return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Xor, E,
2690                                llvm::Instruction::Xor);
2691  case Builtin::BI__sync_nand_and_fetch_1:
2692  case Builtin::BI__sync_nand_and_fetch_2:
2693  case Builtin::BI__sync_nand_and_fetch_4:
2694  case Builtin::BI__sync_nand_and_fetch_8:
2695  case Builtin::BI__sync_nand_and_fetch_16:
2696    return EmitBinaryAtomicPost(*this, llvm::AtomicRMWInst::Nand, E,
2697                                llvm::Instruction::And, true);
2698
2699  case Builtin::BI__sync_val_compare_and_swap_1:
2700  case Builtin::BI__sync_val_compare_and_swap_2:
2701  case Builtin::BI__sync_val_compare_and_swap_4:
2702  case Builtin::BI__sync_val_compare_and_swap_8:
2703  case Builtin::BI__sync_val_compare_and_swap_16:
2704    return RValue::get(MakeAtomicCmpXchgValue(*this, E, false));
2705
2706  case Builtin::BI__sync_bool_compare_and_swap_1:
2707  case Builtin::BI__sync_bool_compare_and_swap_2:
2708  case Builtin::BI__sync_bool_compare_and_swap_4:
2709  case Builtin::BI__sync_bool_compare_and_swap_8:
2710  case Builtin::BI__sync_bool_compare_and_swap_16:
2711    return RValue::get(MakeAtomicCmpXchgValue(*this, E, true));
2712
2713  case Builtin::BI__sync_swap_1:
2714  case Builtin::BI__sync_swap_2:
2715  case Builtin::BI__sync_swap_4:
2716  case Builtin::BI__sync_swap_8:
2717  case Builtin::BI__sync_swap_16:
2718    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E);
2719
2720  case Builtin::BI__sync_lock_test_and_set_1:
2721  case Builtin::BI__sync_lock_test_and_set_2:
2722  case Builtin::BI__sync_lock_test_and_set_4:
2723  case Builtin::BI__sync_lock_test_and_set_8:
2724  case Builtin::BI__sync_lock_test_and_set_16:
2725    return EmitBinaryAtomic(*this, llvm::AtomicRMWInst::Xchg, E);
2726
2727  case Builtin::BI__sync_lock_release_1:
2728  case Builtin::BI__sync_lock_release_2:
2729  case Builtin::BI__sync_lock_release_4:
2730  case Builtin::BI__sync_lock_release_8:
2731  case Builtin::BI__sync_lock_release_16: {
2732    Value *Ptr = EmitScalarExpr(E->getArg(0));
2733    QualType ElTy = E->getArg(0)->getType()->getPointeeType();
2734    CharUnits StoreSize = getContext().getTypeSizeInChars(ElTy);
2735    llvm::Type *ITy = llvm::IntegerType::get(getLLVMContext(),
2736                                             StoreSize.getQuantity() * 8);
2737    Ptr = Builder.CreateBitCast(Ptr, ITy->getPointerTo());
2738    llvm::StoreInst *Store =
2739      Builder.CreateAlignedStore(llvm::Constant::getNullValue(ITy), Ptr,
2740                                 StoreSize);
2741    Store->setAtomic(llvm::AtomicOrdering::Release);
2742    return RValue::get(nullptr);
2743  }
2744
2745  case Builtin::BI__sync_synchronize: {
2746    // We assume this is supposed to correspond to a C++0x-style
2747    // sequentially-consistent fence (i.e. this is only usable for
2748    // synchronization, not device I/O or anything like that). This intrinsic
2749    // is really badly designed in the sense that in theory, there isn't
2750    // any way to safely use it... but in practice, it mostly works
2751    // to use it with non-atomic loads and stores to get acquire/release
2752    // semantics.
2753    Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent);
2754    return RValue::get(nullptr);
2755  }
2756
2757  case Builtin::BI__builtin_nontemporal_load:
2758    return RValue::get(EmitNontemporalLoad(*this, E));
2759  case Builtin::BI__builtin_nontemporal_store:
2760    return RValue::get(EmitNontemporalStore(*this, E));
2761  case Builtin::BI__c11_atomic_is_lock_free:
2762  case Builtin::BI__atomic_is_lock_free: {
2763    // Call "bool __atomic_is_lock_free(size_t size, void *ptr)". For the
2764    // __c11 builtin, ptr is 0 (indicating a properly-aligned object), since
2765    // _Atomic(T) is always properly-aligned.
2766    const char *LibCallName = "__atomic_is_lock_free";
2767    CallArgList Args;
2768    Args.add(RValue::get(EmitScalarExpr(E->getArg(0))),
2769             getContext().getSizeType());
2770    if (BuiltinID == Builtin::BI__atomic_is_lock_free)
2771      Args.add(RValue::get(EmitScalarExpr(E->getArg(1))),
2772               getContext().VoidPtrTy);
2773    else
2774      Args.add(RValue::get(llvm::Constant::getNullValue(VoidPtrTy)),
2775               getContext().VoidPtrTy);
2776    const CGFunctionInfo &FuncInfo =
2777        CGM.getTypes().arrangeBuiltinFunctionCall(E->getType(), Args);
2778    llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FuncInfo);
2779    llvm::FunctionCallee Func = CGM.CreateRuntimeFunction(FTy, LibCallName);
2780    return EmitCall(FuncInfo, CGCallee::forDirect(Func),
2781                    ReturnValueSlot(), Args);
2782  }
2783
2784  case Builtin::BI__atomic_test_and_set: {
2785    // Look at the argument type to determine whether this is a volatile
2786    // operation. The parameter type is always volatile.
2787    QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType();
2788    bool Volatile =
2789        PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified();
2790
2791    Value *Ptr = EmitScalarExpr(E->getArg(0));
2792    unsigned AddrSpace = Ptr->getType()->getPointerAddressSpace();
2793    Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace));
2794    Value *NewVal = Builder.getInt8(1);
2795    Value *Order = EmitScalarExpr(E->getArg(1));
2796    if (isa<llvm::ConstantInt>(Order)) {
2797      int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
2798      AtomicRMWInst *Result = nullptr;
2799      switch (ord) {
2800      case 0:  // memory_order_relaxed
2801      default: // invalid order
2802        Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2803                                         llvm::AtomicOrdering::Monotonic);
2804        break;
2805      case 1: // memory_order_consume
2806      case 2: // memory_order_acquire
2807        Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2808                                         llvm::AtomicOrdering::Acquire);
2809        break;
2810      case 3: // memory_order_release
2811        Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2812                                         llvm::AtomicOrdering::Release);
2813        break;
2814      case 4: // memory_order_acq_rel
2815
2816        Result = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2817                                         llvm::AtomicOrdering::AcquireRelease);
2818        break;
2819      case 5: // memory_order_seq_cst
2820        Result = Builder.CreateAtomicRMW(
2821            llvm::AtomicRMWInst::Xchg, Ptr, NewVal,
2822            llvm::AtomicOrdering::SequentiallyConsistent);
2823        break;
2824      }
2825      Result->setVolatile(Volatile);
2826      return RValue::get(Builder.CreateIsNotNull(Result, "tobool"));
2827    }
2828
2829    llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn);
2830
2831    llvm::BasicBlock *BBs[5] = {
2832      createBasicBlock("monotonic", CurFn),
2833      createBasicBlock("acquire", CurFn),
2834      createBasicBlock("release", CurFn),
2835      createBasicBlock("acqrel", CurFn),
2836      createBasicBlock("seqcst", CurFn)
2837    };
2838    llvm::AtomicOrdering Orders[5] = {
2839        llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Acquire,
2840        llvm::AtomicOrdering::Release, llvm::AtomicOrdering::AcquireRelease,
2841        llvm::AtomicOrdering::SequentiallyConsistent};
2842
2843    Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false);
2844    llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]);
2845
2846    Builder.SetInsertPoint(ContBB);
2847    PHINode *Result = Builder.CreatePHI(Int8Ty, 5, "was_set");
2848
2849    for (unsigned i = 0; i < 5; ++i) {
2850      Builder.SetInsertPoint(BBs[i]);
2851      AtomicRMWInst *RMW = Builder.CreateAtomicRMW(llvm::AtomicRMWInst::Xchg,
2852                                                   Ptr, NewVal, Orders[i]);
2853      RMW->setVolatile(Volatile);
2854      Result->addIncoming(RMW, BBs[i]);
2855      Builder.CreateBr(ContBB);
2856    }
2857
2858    SI->addCase(Builder.getInt32(0), BBs[0]);
2859    SI->addCase(Builder.getInt32(1), BBs[1]);
2860    SI->addCase(Builder.getInt32(2), BBs[1]);
2861    SI->addCase(Builder.getInt32(3), BBs[2]);
2862    SI->addCase(Builder.getInt32(4), BBs[3]);
2863    SI->addCase(Builder.getInt32(5), BBs[4]);
2864
2865    Builder.SetInsertPoint(ContBB);
2866    return RValue::get(Builder.CreateIsNotNull(Result, "tobool"));
2867  }
2868
2869  case Builtin::BI__atomic_clear: {
2870    QualType PtrTy = E->getArg(0)->IgnoreImpCasts()->getType();
2871    bool Volatile =
2872        PtrTy->castAs<PointerType>()->getPointeeType().isVolatileQualified();
2873
2874    Address Ptr = EmitPointerWithAlignment(E->getArg(0));
2875    unsigned AddrSpace = Ptr.getPointer()->getType()->getPointerAddressSpace();
2876    Ptr = Builder.CreateBitCast(Ptr, Int8Ty->getPointerTo(AddrSpace));
2877    Value *NewVal = Builder.getInt8(0);
2878    Value *Order = EmitScalarExpr(E->getArg(1));
2879    if (isa<llvm::ConstantInt>(Order)) {
2880      int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
2881      StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile);
2882      switch (ord) {
2883      case 0:  // memory_order_relaxed
2884      default: // invalid order
2885        Store->setOrdering(llvm::AtomicOrdering::Monotonic);
2886        break;
2887      case 3:  // memory_order_release
2888        Store->setOrdering(llvm::AtomicOrdering::Release);
2889        break;
2890      case 5:  // memory_order_seq_cst
2891        Store->setOrdering(llvm::AtomicOrdering::SequentiallyConsistent);
2892        break;
2893      }
2894      return RValue::get(nullptr);
2895    }
2896
2897    llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn);
2898
2899    llvm::BasicBlock *BBs[3] = {
2900      createBasicBlock("monotonic", CurFn),
2901      createBasicBlock("release", CurFn),
2902      createBasicBlock("seqcst", CurFn)
2903    };
2904    llvm::AtomicOrdering Orders[3] = {
2905        llvm::AtomicOrdering::Monotonic, llvm::AtomicOrdering::Release,
2906        llvm::AtomicOrdering::SequentiallyConsistent};
2907
2908    Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false);
2909    llvm::SwitchInst *SI = Builder.CreateSwitch(Order, BBs[0]);
2910
2911    for (unsigned i = 0; i < 3; ++i) {
2912      Builder.SetInsertPoint(BBs[i]);
2913      StoreInst *Store = Builder.CreateStore(NewVal, Ptr, Volatile);
2914      Store->setOrdering(Orders[i]);
2915      Builder.CreateBr(ContBB);
2916    }
2917
2918    SI->addCase(Builder.getInt32(0), BBs[0]);
2919    SI->addCase(Builder.getInt32(3), BBs[1]);
2920    SI->addCase(Builder.getInt32(5), BBs[2]);
2921
2922    Builder.SetInsertPoint(ContBB);
2923    return RValue::get(nullptr);
2924  }
2925
2926  case Builtin::BI__atomic_thread_fence:
2927  case Builtin::BI__atomic_signal_fence:
2928  case Builtin::BI__c11_atomic_thread_fence:
2929  case Builtin::BI__c11_atomic_signal_fence: {
2930    llvm::SyncScope::ID SSID;
2931    if (BuiltinID == Builtin::BI__atomic_signal_fence ||
2932        BuiltinID == Builtin::BI__c11_atomic_signal_fence)
2933      SSID = llvm::SyncScope::SingleThread;
2934    else
2935      SSID = llvm::SyncScope::System;
2936    Value *Order = EmitScalarExpr(E->getArg(0));
2937    if (isa<llvm::ConstantInt>(Order)) {
2938      int ord = cast<llvm::ConstantInt>(Order)->getZExtValue();
2939      switch (ord) {
2940      case 0:  // memory_order_relaxed
2941      default: // invalid order
2942        break;
2943      case 1:  // memory_order_consume
2944      case 2:  // memory_order_acquire
2945        Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID);
2946        break;
2947      case 3:  // memory_order_release
2948        Builder.CreateFence(llvm::AtomicOrdering::Release, SSID);
2949        break;
2950      case 4:  // memory_order_acq_rel
2951        Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID);
2952        break;
2953      case 5:  // memory_order_seq_cst
2954        Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID);
2955        break;
2956      }
2957      return RValue::get(nullptr);
2958    }
2959
2960    llvm::BasicBlock *AcquireBB, *ReleaseBB, *AcqRelBB, *SeqCstBB;
2961    AcquireBB = createBasicBlock("acquire", CurFn);
2962    ReleaseBB = createBasicBlock("release", CurFn);
2963    AcqRelBB = createBasicBlock("acqrel", CurFn);
2964    SeqCstBB = createBasicBlock("seqcst", CurFn);
2965    llvm::BasicBlock *ContBB = createBasicBlock("atomic.continue", CurFn);
2966
2967    Order = Builder.CreateIntCast(Order, Builder.getInt32Ty(), false);
2968    llvm::SwitchInst *SI = Builder.CreateSwitch(Order, ContBB);
2969
2970    Builder.SetInsertPoint(AcquireBB);
2971    Builder.CreateFence(llvm::AtomicOrdering::Acquire, SSID);
2972    Builder.CreateBr(ContBB);
2973    SI->addCase(Builder.getInt32(1), AcquireBB);
2974    SI->addCase(Builder.getInt32(2), AcquireBB);
2975
2976    Builder.SetInsertPoint(ReleaseBB);
2977    Builder.CreateFence(llvm::AtomicOrdering::Release, SSID);
2978    Builder.CreateBr(ContBB);
2979    SI->addCase(Builder.getInt32(3), ReleaseBB);
2980
2981    Builder.SetInsertPoint(AcqRelBB);
2982    Builder.CreateFence(llvm::AtomicOrdering::AcquireRelease, SSID);
2983    Builder.CreateBr(ContBB);
2984    SI->addCase(Builder.getInt32(4), AcqRelBB);
2985
2986    Builder.SetInsertPoint(SeqCstBB);
2987    Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent, SSID);
2988    Builder.CreateBr(ContBB);
2989    SI->addCase(Builder.getInt32(5), SeqCstBB);
2990
2991    Builder.SetInsertPoint(ContBB);
2992    return RValue::get(nullptr);
2993  }
2994
2995  case Builtin::BI__builtin_signbit:
2996  case Builtin::BI__builtin_signbitf:
2997  case Builtin::BI__builtin_signbitl: {
2998    return RValue::get(
2999        Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))),
3000                           ConvertType(E->getType())));
3001  }
3002  case Builtin::BI__annotation: {
3003    // Re-encode each wide string to UTF8 and make an MDString.
3004    SmallVector<Metadata *, 1> Strings;
3005    for (const Expr *Arg : E->arguments()) {
3006      const auto *Str = cast<StringLiteral>(Arg->IgnoreParenCasts());
3007      getCharByteWidth() == 2", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 3007, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Str->getCharByteWidth() == 2);
3008      StringRef WideBytes = Str->getBytes();
3009      std::string StrUtf8;
3010      if (!convertUTF16ToUTF8String(
3011              makeArrayRef(WideBytes.data(), WideBytes.size()), StrUtf8)) {
3012        CGM.ErrorUnsupported(E, "non-UTF16 __annotation argument");
3013        continue;
3014      }
3015      Strings.push_back(llvm::MDString::get(getLLVMContext(), StrUtf8));
3016    }
3017
3018    // Build and MDTuple of MDStrings and emit the intrinsic call.
3019    llvm::Function *F =
3020        CGM.getIntrinsic(llvm::Intrinsic::codeview_annotation, {});
3021    MDTuple *StrTuple = MDTuple::get(getLLVMContext(), Strings);
3022    Builder.CreateCall(F, MetadataAsValue::get(getLLVMContext(), StrTuple));
3023    return RValue::getIgnored();
3024  }
3025  case Builtin::BI__builtin_annotation: {
3026    llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0));
3027    llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::annotation,
3028                                      AnnVal->getType());
3029
3030    // Get the annotation string, go through casts. Sema requires this to be a
3031    // non-wide string literal, potentially casted, so the cast<> is safe.
3032    const Expr *AnnotationStrExpr = E->getArg(1)->IgnoreParenCasts();
3033    StringRef Str = cast<StringLiteral>(AnnotationStrExpr)->getString();
3034    return RValue::get(EmitAnnotationCall(F, AnnVal, Str, E->getExprLoc()));
3035  }
3036  case Builtin::BI__builtin_addcb:
3037  case Builtin::BI__builtin_addcs:
3038  case Builtin::BI__builtin_addc:
3039  case Builtin::BI__builtin_addcl:
3040  case Builtin::BI__builtin_addcll:
3041  case Builtin::BI__builtin_subcb:
3042  case Builtin::BI__builtin_subcs:
3043  case Builtin::BI__builtin_subc:
3044  case Builtin::BI__builtin_subcl:
3045  case Builtin::BI__builtin_subcll: {
3046
3047    // We translate all of these builtins from expressions of the form:
3048    //   int x = ..., y = ..., carryin = ..., carryout, result;
3049    //   result = __builtin_addc(x, y, carryin, &carryout);
3050    //
3051    // to LLVM IR of the form:
3052    //
3053    //   %tmp1 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %x, i32 %y)
3054    //   %tmpsum1 = extractvalue {i32, i1} %tmp1, 0
3055    //   %carry1 = extractvalue {i32, i1} %tmp1, 1
3056    //   %tmp2 = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %tmpsum1,
3057    //                                                       i32 %carryin)
3058    //   %result = extractvalue {i32, i1} %tmp2, 0
3059    //   %carry2 = extractvalue {i32, i1} %tmp2, 1
3060    //   %tmp3 = or i1 %carry1, %carry2
3061    //   %tmp4 = zext i1 %tmp3 to i32
3062    //   store i32 %tmp4, i32* %carryout
3063
3064    // Scalarize our inputs.
3065    llvm::Value *X = EmitScalarExpr(E->getArg(0));
3066    llvm::Value *Y = EmitScalarExpr(E->getArg(1));
3067    llvm::Value *Carryin = EmitScalarExpr(E->getArg(2));
3068    Address CarryOutPtr = EmitPointerWithAlignment(E->getArg(3));
3069
3070    // Decide if we are lowering to a uadd.with.overflow or usub.with.overflow.
3071    llvm::Intrinsic::ID IntrinsicId;
3072    switch (BuiltinID) {
3073    default: llvm_unreachable("Unknown multiprecision builtin id.");
3074    case Builtin::BI__builtin_addcb:
3075    case Builtin::BI__builtin_addcs:
3076    case Builtin::BI__builtin_addc:
3077    case Builtin::BI__builtin_addcl:
3078    case Builtin::BI__builtin_addcll:
3079      IntrinsicId = llvm::Intrinsic::uadd_with_overflow;
3080      break;
3081    case Builtin::BI__builtin_subcb:
3082    case Builtin::BI__builtin_subcs:
3083    case Builtin::BI__builtin_subc:
3084    case Builtin::BI__builtin_subcl:
3085    case Builtin::BI__builtin_subcll:
3086      IntrinsicId = llvm::Intrinsic::usub_with_overflow;
3087      break;
3088    }
3089
3090    // Construct our resulting LLVM IR expression.
3091    llvm::Value *Carry1;
3092    llvm::Value *Sum1 = EmitOverflowIntrinsic(*this, IntrinsicId,
3093                                              X, Y, Carry1);
3094    llvm::Value *Carry2;
3095    llvm::Value *Sum2 = EmitOverflowIntrinsic(*this, IntrinsicId,
3096                                              Sum1, Carryin, Carry2);
3097    llvm::Value *CarryOut = Builder.CreateZExt(Builder.CreateOr(Carry1, Carry2),
3098                                               X->getType());
3099    Builder.CreateStore(CarryOut, CarryOutPtr);
3100    return RValue::get(Sum2);
3101  }
3102
3103  case Builtin::BI__builtin_add_overflow:
3104  case Builtin::BI__builtin_sub_overflow:
3105  case Builtin::BI__builtin_mul_overflow: {
3106    const clang::Expr *LeftArg = E->getArg(0);
3107    const clang::Expr *RightArg = E->getArg(1);
3108    const clang::Expr *ResultArg = E->getArg(2);
3109
3110    clang::QualType ResultQTy =
3111        ResultArg->getType()->castAs<PointerType>()->getPointeeType();
3112
3113    WidthAndSignedness LeftInfo =
3114        getIntegerWidthAndSignedness(CGM.getContext(), LeftArg->getType());
3115    WidthAndSignedness RightInfo =
3116        getIntegerWidthAndSignedness(CGM.getContext(), RightArg->getType());
3117    WidthAndSignedness ResultInfo =
3118        getIntegerWidthAndSignedness(CGM.getContext(), ResultQTy);
3119
3120    // Handle mixed-sign multiplication as a special case, because adding
3121    // runtime or backend support for our generic irgen would be too expensive.
3122    if (isSpecialMixedSignMultiply(BuiltinID, LeftInfo, RightInfo, ResultInfo))
3123      return EmitCheckedMixedSignMultiply(*this, LeftArg, LeftInfo, RightArg,
3124                                          RightInfo, ResultArg, ResultQTy,
3125                                          ResultInfo);
3126
3127    WidthAndSignedness EncompassingInfo =
3128        EncompassingIntegerType({LeftInfo, RightInfo, ResultInfo});
3129
3130    llvm::Type *EncompassingLLVMTy =
3131        llvm::IntegerType::get(CGM.getLLVMContext(), EncompassingInfo.Width);
3132
3133    llvm::Type *ResultLLVMTy = CGM.getTypes().ConvertType(ResultQTy);
3134
3135    llvm::Intrinsic::ID IntrinsicId;
3136    switch (BuiltinID) {
3137    default:
3138      llvm_unreachable("Unknown overflow builtin id.");
3139    case Builtin::BI__builtin_add_overflow:
3140      IntrinsicId = EncompassingInfo.Signed
3141                        ? llvm::Intrinsic::sadd_with_overflow
3142                        : llvm::Intrinsic::uadd_with_overflow;
3143      break;
3144    case Builtin::BI__builtin_sub_overflow:
3145      IntrinsicId = EncompassingInfo.Signed
3146                        ? llvm::Intrinsic::ssub_with_overflow
3147                        : llvm::Intrinsic::usub_with_overflow;
3148      break;
3149    case Builtin::BI__builtin_mul_overflow:
3150      IntrinsicId = EncompassingInfo.Signed
3151                        ? llvm::Intrinsic::smul_with_overflow
3152                        : llvm::Intrinsic::umul_with_overflow;
3153      break;
3154    }
3155
3156    llvm::Value *Left = EmitScalarExpr(LeftArg);
3157    llvm::Value *Right = EmitScalarExpr(RightArg);
3158    Address ResultPtr = EmitPointerWithAlignment(ResultArg);
3159
3160    // Extend each operand to the encompassing type.
3161    Left = Builder.CreateIntCast(Left, EncompassingLLVMTy, LeftInfo.Signed);
3162    Right = Builder.CreateIntCast(Right, EncompassingLLVMTy, RightInfo.Signed);
3163
3164    // Perform the operation on the extended values.
3165    llvm::Value *Overflow, *Result;
3166    Result = EmitOverflowIntrinsic(*this, IntrinsicId, Left, Right, Overflow);
3167
3168    if (EncompassingInfo.Width > ResultInfo.Width) {
3169      // The encompassing type is wider than the result type, so we need to
3170      // truncate it.
3171      llvm::Value *ResultTrunc = Builder.CreateTrunc(Result, ResultLLVMTy);
3172
3173      // To see if the truncation caused an overflow, we will extend
3174      // the result and then compare it to the original result.
3175      llvm::Value *ResultTruncExt = Builder.CreateIntCast(
3176          ResultTrunc, EncompassingLLVMTy, ResultInfo.Signed);
3177      llvm::Value *TruncationOverflow =
3178          Builder.CreateICmpNE(Result, ResultTruncExt);
3179
3180      Overflow = Builder.CreateOr(Overflow, TruncationOverflow);
3181      Result = ResultTrunc;
3182    }
3183
3184    // Finally, store the result using the pointer.
3185    bool isVolatile =
3186      ResultArg->getType()->getPointeeType().isVolatileQualified();
3187    Builder.CreateStore(EmitToMemory(Result, ResultQTy), ResultPtr, isVolatile);
3188
3189    return RValue::get(Overflow);
3190  }
3191
3192  case Builtin::BI__builtin_uadd_overflow:
3193  case Builtin::BI__builtin_uaddl_overflow:
3194  case Builtin::BI__builtin_uaddll_overflow:
3195  case Builtin::BI__builtin_usub_overflow:
3196  case Builtin::BI__builtin_usubl_overflow:
3197  case Builtin::BI__builtin_usubll_overflow:
3198  case Builtin::BI__builtin_umul_overflow:
3199  case Builtin::BI__builtin_umull_overflow:
3200  case Builtin::BI__builtin_umulll_overflow:
3201  case Builtin::BI__builtin_sadd_overflow:
3202  case Builtin::BI__builtin_saddl_overflow:
3203  case Builtin::BI__builtin_saddll_overflow:
3204  case Builtin::BI__builtin_ssub_overflow:
3205  case Builtin::BI__builtin_ssubl_overflow:
3206  case Builtin::BI__builtin_ssubll_overflow:
3207  case Builtin::BI__builtin_smul_overflow:
3208  case Builtin::BI__builtin_smull_overflow:
3209  case Builtin::BI__builtin_smulll_overflow: {
3210
3211    // We translate all of these builtins directly to the relevant llvm IR node.
3212
3213    // Scalarize our inputs.
3214    llvm::Value *X = EmitScalarExpr(E->getArg(0));
3215    llvm::Value *Y = EmitScalarExpr(E->getArg(1));
3216    Address SumOutPtr = EmitPointerWithAlignment(E->getArg(2));
3217
3218    // Decide which of the overflow intrinsics we are lowering to:
3219    llvm::Intrinsic::ID IntrinsicId;
3220    switch (BuiltinID) {
3221    default: llvm_unreachable("Unknown overflow builtin id.");
3222    case Builtin::BI__builtin_uadd_overflow:
3223    case Builtin::BI__builtin_uaddl_overflow:
3224    case Builtin::BI__builtin_uaddll_overflow:
3225      IntrinsicId = llvm::Intrinsic::uadd_with_overflow;
3226      break;
3227    case Builtin::BI__builtin_usub_overflow:
3228    case Builtin::BI__builtin_usubl_overflow:
3229    case Builtin::BI__builtin_usubll_overflow:
3230      IntrinsicId = llvm::Intrinsic::usub_with_overflow;
3231      break;
3232    case Builtin::BI__builtin_umul_overflow:
3233    case Builtin::BI__builtin_umull_overflow:
3234    case Builtin::BI__builtin_umulll_overflow:
3235      IntrinsicId = llvm::Intrinsic::umul_with_overflow;
3236      break;
3237    case Builtin::BI__builtin_sadd_overflow:
3238    case Builtin::BI__builtin_saddl_overflow:
3239    case Builtin::BI__builtin_saddll_overflow:
3240      IntrinsicId = llvm::Intrinsic::sadd_with_overflow;
3241      break;
3242    case Builtin::BI__builtin_ssub_overflow:
3243    case Builtin::BI__builtin_ssubl_overflow:
3244    case Builtin::BI__builtin_ssubll_overflow:
3245      IntrinsicId = llvm::Intrinsic::ssub_with_overflow;
3246      break;
3247    case Builtin::BI__builtin_smul_overflow:
3248    case Builtin::BI__builtin_smull_overflow:
3249    case Builtin::BI__builtin_smulll_overflow:
3250      IntrinsicId = llvm::Intrinsic::smul_with_overflow;
3251      break;
3252    }
3253
3254
3255    llvm::Value *Carry;
3256    llvm::Value *Sum = EmitOverflowIntrinsic(*this, IntrinsicId, X, Y, Carry);
3257    Builder.CreateStore(Sum, SumOutPtr);
3258
3259    return RValue::get(Carry);
3260  }
3261  case Builtin::BI__builtin_addressof:
3262    return RValue::get(EmitLValue(E->getArg(0)).getPointer());
3263  case Builtin::BI__builtin_operator_new:
3264    return EmitBuiltinNewDeleteCall(
3265        E->getCallee()->getType()->castAs<FunctionProtoType>(), E, false);
3266  case Builtin::BI__builtin_operator_delete:
3267    return EmitBuiltinNewDeleteCall(
3268        E->getCallee()->getType()->castAs<FunctionProtoType>(), E, true);
3269
3270  case Builtin::BI__noop:
3271    // __noop always evaluates to an integer literal zero.
3272    return RValue::get(ConstantInt::get(IntTy, 0));
3273  case Builtin::BI__builtin_call_with_static_chain: {
3274    const CallExpr *Call = cast<CallExpr>(E->getArg(0));
3275    const Expr *Chain = E->getArg(1);
3276    return EmitCall(Call->getCallee()->getType(),
3277                    EmitCallee(Call->getCallee()), Call, ReturnValue,
3278                    EmitScalarExpr(Chain));
3279  }
3280  case Builtin::BI_InterlockedExchange8:
3281  case Builtin::BI_InterlockedExchange16:
3282  case Builtin::BI_InterlockedExchange:
3283  case Builtin::BI_InterlockedExchangePointer:
3284    return RValue::get(
3285        EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E));
3286  case Builtin::BI_InterlockedCompareExchangePointer:
3287  case Builtin::BI_InterlockedCompareExchangePointer_nf: {
3288    llvm::Type *RTy;
3289    llvm::IntegerType *IntType =
3290      IntegerType::get(getLLVMContext(),
3291                       getContext().getTypeSize(E->getType()));
3292    llvm::Type *IntPtrType = IntType->getPointerTo();
3293
3294    llvm::Value *Destination =
3295      Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), IntPtrType);
3296
3297    llvm::Value *Exchange = EmitScalarExpr(E->getArg(1));
3298    RTy = Exchange->getType();
3299    Exchange = Builder.CreatePtrToInt(Exchange, IntType);
3300
3301    llvm::Value *Comparand =
3302      Builder.CreatePtrToInt(EmitScalarExpr(E->getArg(2)), IntType);
3303
3304    auto Ordering =
3305      BuiltinID == Builtin::BI_InterlockedCompareExchangePointer_nf ?
3306      AtomicOrdering::Monotonic : AtomicOrdering::SequentiallyConsistent;
3307
3308    auto Result = Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange,
3309                                              Ordering, Ordering);
3310    Result->setVolatile(true);
3311
3312    return RValue::get(Builder.CreateIntToPtr(Builder.CreateExtractValue(Result,
3313                                                                         0),
3314                                              RTy));
3315  }
3316  case Builtin::BI_InterlockedCompareExchange8:
3317  case Builtin::BI_InterlockedCompareExchange16:
3318  case Builtin::BI_InterlockedCompareExchange:
3319  case Builtin::BI_InterlockedCompareExchange64:
3320    return RValue::get(EmitAtomicCmpXchgForMSIntrin(*this, E));
3321  case Builtin::BI_InterlockedIncrement16:
3322  case Builtin::BI_InterlockedIncrement:
3323    return RValue::get(
3324        EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E));
3325  case Builtin::BI_InterlockedDecrement16:
3326  case Builtin::BI_InterlockedDecrement:
3327    return RValue::get(
3328        EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E));
3329  case Builtin::BI_InterlockedAnd8:
3330  case Builtin::BI_InterlockedAnd16:
3331  case Builtin::BI_InterlockedAnd:
3332    return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E));
3333  case Builtin::BI_InterlockedExchangeAdd8:
3334  case Builtin::BI_InterlockedExchangeAdd16:
3335  case Builtin::BI_InterlockedExchangeAdd:
3336    return RValue::get(
3337        EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E));
3338  case Builtin::BI_InterlockedExchangeSub8:
3339  case Builtin::BI_InterlockedExchangeSub16:
3340  case Builtin::BI_InterlockedExchangeSub:
3341    return RValue::get(
3342        EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E));
3343  case Builtin::BI_InterlockedOr8:
3344  case Builtin::BI_InterlockedOr16:
3345  case Builtin::BI_InterlockedOr:
3346    return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E));
3347  case Builtin::BI_InterlockedXor8:
3348  case Builtin::BI_InterlockedXor16:
3349  case Builtin::BI_InterlockedXor:
3350    return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E));
3351
3352  case Builtin::BI_bittest64:
3353  case Builtin::BI_bittest:
3354  case Builtin::BI_bittestandcomplement64:
3355  case Builtin::BI_bittestandcomplement:
3356  case Builtin::BI_bittestandreset64:
3357  case Builtin::BI_bittestandreset:
3358  case Builtin::BI_bittestandset64:
3359  case Builtin::BI_bittestandset:
3360  case Builtin::BI_interlockedbittestandreset:
3361  case Builtin::BI_interlockedbittestandreset64:
3362  case Builtin::BI_interlockedbittestandset64:
3363  case Builtin::BI_interlockedbittestandset:
3364  case Builtin::BI_interlockedbittestandset_acq:
3365  case Builtin::BI_interlockedbittestandset_rel:
3366  case Builtin::BI_interlockedbittestandset_nf:
3367  case Builtin::BI_interlockedbittestandreset_acq:
3368  case Builtin::BI_interlockedbittestandreset_rel:
3369  case Builtin::BI_interlockedbittestandreset_nf:
3370    return RValue::get(EmitBitTestIntrinsic(*this, BuiltinID, E));
3371
3372    // These builtins exist to emit regular volatile loads and stores not
3373    // affected by the -fms-volatile setting.
3374  case Builtin::BI__iso_volatile_load8:
3375  case Builtin::BI__iso_volatile_load16:
3376  case Builtin::BI__iso_volatile_load32:
3377  case Builtin::BI__iso_volatile_load64:
3378    return RValue::get(EmitISOVolatileLoad(*this, E));
3379  case Builtin::BI__iso_volatile_store8:
3380  case Builtin::BI__iso_volatile_store16:
3381  case Builtin::BI__iso_volatile_store32:
3382  case Builtin::BI__iso_volatile_store64:
3383    return RValue::get(EmitISOVolatileStore(*this, E));
3384
3385  case Builtin::BI__exception_code:
3386  case Builtin::BI_exception_code:
3387    return RValue::get(EmitSEHExceptionCode());
3388  case Builtin::BI__exception_info:
3389  case Builtin::BI_exception_info:
3390    return RValue::get(EmitSEHExceptionInfo());
3391  case Builtin::BI__abnormal_termination:
3392  case Builtin::BI_abnormal_termination:
3393    return RValue::get(EmitSEHAbnormalTermination());
3394  case Builtin::BI_setjmpex:
3395    if (getTarget().getTriple().isOSMSVCRT())
3396      return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E);
3397    break;
3398  case Builtin::BI_setjmp:
3399    if (getTarget().getTriple().isOSMSVCRT()) {
3400      if (getTarget().getTriple().getArch() == llvm::Triple::x86)
3401        return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp3, E);
3402      else if (getTarget().getTriple().getArch() == llvm::Triple::aarch64)
3403        return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmpex, E);
3404      return EmitMSVCRTSetJmp(*this, MSVCSetJmpKind::_setjmp, E);
3405    }
3406    break;
3407
3408  case Builtin::BI__GetExceptionInfo: {
3409    if (llvm::GlobalVariable *GV =
3410            CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType()))
3411      return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy));
3412    break;
3413  }
3414
3415  case Builtin::BI__fastfail:
3416    return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E));
3417
3418  case Builtin::BI__builtin_coro_size: {
3419    auto & Context = getContext();
3420    auto SizeTy = Context.getSizeType();
3421    auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy));
3422    Function *F = CGM.getIntrinsic(Intrinsic::coro_size, T);
3423    return RValue::get(Builder.CreateCall(F));
3424  }
3425
3426  case Builtin::BI__builtin_coro_id:
3427    return EmitCoroutineIntrinsic(E, Intrinsic::coro_id);
3428  case Builtin::BI__builtin_coro_promise:
3429    return EmitCoroutineIntrinsic(E, Intrinsic::coro_promise);
3430  case Builtin::BI__builtin_coro_resume:
3431    return EmitCoroutineIntrinsic(E, Intrinsic::coro_resume);
3432  case Builtin::BI__builtin_coro_frame:
3433    return EmitCoroutineIntrinsic(E, Intrinsic::coro_frame);
3434  case Builtin::BI__builtin_coro_noop:
3435    return EmitCoroutineIntrinsic(E, Intrinsic::coro_noop);
3436  case Builtin::BI__builtin_coro_free:
3437    return EmitCoroutineIntrinsic(E, Intrinsic::coro_free);
3438  case Builtin::BI__builtin_coro_destroy:
3439    return EmitCoroutineIntrinsic(E, Intrinsic::coro_destroy);
3440  case Builtin::BI__builtin_coro_done:
3441    return EmitCoroutineIntrinsic(E, Intrinsic::coro_done);
3442  case Builtin::BI__builtin_coro_alloc:
3443    return EmitCoroutineIntrinsic(E, Intrinsic::coro_alloc);
3444  case Builtin::BI__builtin_coro_begin:
3445    return EmitCoroutineIntrinsic(E, Intrinsic::coro_begin);
3446  case Builtin::BI__builtin_coro_end:
3447    return EmitCoroutineIntrinsic(E, Intrinsic::coro_end);
3448  case Builtin::BI__builtin_coro_suspend:
3449    return EmitCoroutineIntrinsic(E, Intrinsic::coro_suspend);
3450  case Builtin::BI__builtin_coro_param:
3451    return EmitCoroutineIntrinsic(E, Intrinsic::coro_param);
3452
3453  // OpenCL v2.0 s6.13.16.2, Built-in pipe read and write functions
3454  case Builtin::BIread_pipe:
3455  case Builtin::BIwrite_pipe: {
3456    Value *Arg0 = EmitScalarExpr(E->getArg(0)),
3457          *Arg1 = EmitScalarExpr(E->getArg(1));
3458    CGOpenCLRuntime OpenCLRT(CGM);
3459    Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
3460    Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
3461
3462    // Type of the generic packet parameter.
3463    unsigned GenericAS =
3464        getContext().getTargetAddressSpace(LangAS::opencl_generic);
3465    llvm::Type *I8PTy = llvm::PointerType::get(
3466        llvm::Type::getInt8Ty(getLLVMContext()), GenericAS);
3467
3468    // Testing which overloaded version we should generate the call for.
3469    if (2U == E->getNumArgs()) {
3470      const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_2"
3471                                                             : "__write_pipe_2";
3472      // Creating a generic function type to be able to call with any builtin or
3473      // user defined type.
3474      llvm::Type *ArgTys[] = {Arg0->getType(), I8PTy, Int32Ty, Int32Ty};
3475      llvm::FunctionType *FTy = llvm::FunctionType::get(
3476          Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3477      Value *BCast = Builder.CreatePointerCast(Arg1, I8PTy);
3478      return RValue::get(
3479          Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3480                             {Arg0, BCast, PacketSize, PacketAlign}));
3481    } else {
3482       (0) . __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 3483, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(4 == E->getNumArgs() &&
3483 (0) . __assert_fail ("4 == E->getNumArgs() && \"Illegal number of parameters to pipe function\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 3483, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">             "Illegal number of parameters to pipe function");
3484      const char *Name = (BuiltinID == Builtin::BIread_pipe) ? "__read_pipe_4"
3485                                                             : "__write_pipe_4";
3486
3487      llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, I8PTy,
3488                              Int32Ty, Int32Ty};
3489      Value *Arg2 = EmitScalarExpr(E->getArg(2)),
3490            *Arg3 = EmitScalarExpr(E->getArg(3));
3491      llvm::FunctionType *FTy = llvm::FunctionType::get(
3492          Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3493      Value *BCast = Builder.CreatePointerCast(Arg3, I8PTy);
3494      // We know the third argument is an integer type, but we may need to cast
3495      // it to i32.
3496      if (Arg2->getType() != Int32Ty)
3497        Arg2 = Builder.CreateZExtOrTrunc(Arg2, Int32Ty);
3498      return RValue::get(Builder.CreateCall(
3499          CGM.CreateRuntimeFunction(FTy, Name),
3500          {Arg0, Arg1, Arg2, BCast, PacketSize, PacketAlign}));
3501    }
3502  }
3503  // OpenCL v2.0 s6.13.16 ,s9.17.3.5 - Built-in pipe reserve read and write
3504  // functions
3505  case Builtin::BIreserve_read_pipe:
3506  case Builtin::BIreserve_write_pipe:
3507  case Builtin::BIwork_group_reserve_read_pipe:
3508  case Builtin::BIwork_group_reserve_write_pipe:
3509  case Builtin::BIsub_group_reserve_read_pipe:
3510  case Builtin::BIsub_group_reserve_write_pipe: {
3511    // Composing the mangled name for the function.
3512    const char *Name;
3513    if (BuiltinID == Builtin::BIreserve_read_pipe)
3514      Name = "__reserve_read_pipe";
3515    else if (BuiltinID == Builtin::BIreserve_write_pipe)
3516      Name = "__reserve_write_pipe";
3517    else if (BuiltinID == Builtin::BIwork_group_reserve_read_pipe)
3518      Name = "__work_group_reserve_read_pipe";
3519    else if (BuiltinID == Builtin::BIwork_group_reserve_write_pipe)
3520      Name = "__work_group_reserve_write_pipe";
3521    else if (BuiltinID == Builtin::BIsub_group_reserve_read_pipe)
3522      Name = "__sub_group_reserve_read_pipe";
3523    else
3524      Name = "__sub_group_reserve_write_pipe";
3525
3526    Value *Arg0 = EmitScalarExpr(E->getArg(0)),
3527          *Arg1 = EmitScalarExpr(E->getArg(1));
3528    llvm::Type *ReservedIDTy = ConvertType(getContext().OCLReserveIDTy);
3529    CGOpenCLRuntime OpenCLRT(CGM);
3530    Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
3531    Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
3532
3533    // Building the generic function prototype.
3534    llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty, Int32Ty};
3535    llvm::FunctionType *FTy = llvm::FunctionType::get(
3536        ReservedIDTy, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3537    // We know the second argument is an integer type, but we may need to cast
3538    // it to i32.
3539    if (Arg1->getType() != Int32Ty)
3540      Arg1 = Builder.CreateZExtOrTrunc(Arg1, Int32Ty);
3541    return RValue::get(
3542        Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3543                           {Arg0, Arg1, PacketSize, PacketAlign}));
3544  }
3545  // OpenCL v2.0 s6.13.16, s9.17.3.5 - Built-in pipe commit read and write
3546  // functions
3547  case Builtin::BIcommit_read_pipe:
3548  case Builtin::BIcommit_write_pipe:
3549  case Builtin::BIwork_group_commit_read_pipe:
3550  case Builtin::BIwork_group_commit_write_pipe:
3551  case Builtin::BIsub_group_commit_read_pipe:
3552  case Builtin::BIsub_group_commit_write_pipe: {
3553    const char *Name;
3554    if (BuiltinID == Builtin::BIcommit_read_pipe)
3555      Name = "__commit_read_pipe";
3556    else if (BuiltinID == Builtin::BIcommit_write_pipe)
3557      Name = "__commit_write_pipe";
3558    else if (BuiltinID == Builtin::BIwork_group_commit_read_pipe)
3559      Name = "__work_group_commit_read_pipe";
3560    else if (BuiltinID == Builtin::BIwork_group_commit_write_pipe)
3561      Name = "__work_group_commit_write_pipe";
3562    else if (BuiltinID == Builtin::BIsub_group_commit_read_pipe)
3563      Name = "__sub_group_commit_read_pipe";
3564    else
3565      Name = "__sub_group_commit_write_pipe";
3566
3567    Value *Arg0 = EmitScalarExpr(E->getArg(0)),
3568          *Arg1 = EmitScalarExpr(E->getArg(1));
3569    CGOpenCLRuntime OpenCLRT(CGM);
3570    Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
3571    Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
3572
3573    // Building the generic function prototype.
3574    llvm::Type *ArgTys[] = {Arg0->getType(), Arg1->getType(), Int32Ty, Int32Ty};
3575    llvm::FunctionType *FTy =
3576        llvm::FunctionType::get(llvm::Type::getVoidTy(getLLVMContext()),
3577                                llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3578
3579    return RValue::get(
3580        Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3581                           {Arg0, Arg1, PacketSize, PacketAlign}));
3582  }
3583  // OpenCL v2.0 s6.13.16.4 Built-in pipe query functions
3584  case Builtin::BIget_pipe_num_packets:
3585  case Builtin::BIget_pipe_max_packets: {
3586    const char *BaseName;
3587    const PipeType *PipeTy = E->getArg(0)->getType()->getAs<PipeType>();
3588    if (BuiltinID == Builtin::BIget_pipe_num_packets)
3589      BaseName = "__get_pipe_num_packets";
3590    else
3591      BaseName = "__get_pipe_max_packets";
3592    auto Name = std::string(BaseName) +
3593                std::string(PipeTy->isReadOnly() ? "_ro" : "_wo");
3594
3595    // Building the generic function prototype.
3596    Value *Arg0 = EmitScalarExpr(E->getArg(0));
3597    CGOpenCLRuntime OpenCLRT(CGM);
3598    Value *PacketSize = OpenCLRT.getPipeElemSize(E->getArg(0));
3599    Value *PacketAlign = OpenCLRT.getPipeElemAlign(E->getArg(0));
3600    llvm::Type *ArgTys[] = {Arg0->getType(), Int32Ty, Int32Ty};
3601    llvm::FunctionType *FTy = llvm::FunctionType::get(
3602        Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3603
3604    return RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3605                                          {Arg0, PacketSize, PacketAlign}));
3606  }
3607
3608  // OpenCL v2.0 s6.13.9 - Address space qualifier functions.
3609  case Builtin::BIto_global:
3610  case Builtin::BIto_local:
3611  case Builtin::BIto_private: {
3612    auto Arg0 = EmitScalarExpr(E->getArg(0));
3613    auto NewArgT = llvm::PointerType::get(Int8Ty,
3614      CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic));
3615    auto NewRetT = llvm::PointerType::get(Int8Ty,
3616      CGM.getContext().getTargetAddressSpace(
3617        E->getType()->getPointeeType().getAddressSpace()));
3618    auto FTy = llvm::FunctionType::get(NewRetT, {NewArgT}, false);
3619    llvm::Value *NewArg;
3620    if (Arg0->getType()->getPointerAddressSpace() !=
3621        NewArgT->getPointerAddressSpace())
3622      NewArg = Builder.CreateAddrSpaceCast(Arg0, NewArgT);
3623    else
3624      NewArg = Builder.CreateBitOrPointerCast(Arg0, NewArgT);
3625    auto NewName = std::string("__") + E->getDirectCallee()->getName().str();
3626    auto NewCall =
3627        Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, NewName), {NewArg});
3628    return RValue::get(Builder.CreateBitOrPointerCast(NewCall,
3629      ConvertType(E->getType())));
3630  }
3631
3632  // OpenCL v2.0, s6.13.17 - Enqueue kernel function.
3633  // It contains four different overload formats specified in Table 6.13.17.1.
3634  case Builtin::BIenqueue_kernel: {
3635    StringRef Name; // Generated function call name
3636    unsigned NumArgs = E->getNumArgs();
3637
3638    llvm::Type *QueueTy = ConvertType(getContext().OCLQueueTy);
3639    llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
3640        getContext().getTargetAddressSpace(LangAS::opencl_generic));
3641
3642    llvm::Value *Queue = EmitScalarExpr(E->getArg(0));
3643    llvm::Value *Flags = EmitScalarExpr(E->getArg(1));
3644    LValue NDRangeL = EmitAggExprToLValue(E->getArg(2));
3645    llvm::Value *Range = NDRangeL.getAddress().getPointer();
3646    llvm::Type *RangeTy = NDRangeL.getAddress().getType();
3647
3648    if (NumArgs == 4) {
3649      // The most basic form of the call with parameters:
3650      // queue_t, kernel_enqueue_flags_t, ndrange_t, block(void)
3651      Name = "__enqueue_kernel_basic";
3652      llvm::Type *ArgTys[] = {QueueTy, Int32Ty, RangeTy, GenericVoidPtrTy,
3653                              GenericVoidPtrTy};
3654      llvm::FunctionType *FTy = llvm::FunctionType::get(
3655          Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3656
3657      auto Info =
3658          CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3));
3659      llvm::Value *Kernel =
3660          Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3661      llvm::Value *Block =
3662          Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3663
3664      AttrBuilder B;
3665      B.addAttribute(Attribute::ByVal);
3666      llvm::AttributeList ByValAttrSet =
3667          llvm::AttributeList::get(CGM.getModule().getContext(), 3U, B);
3668
3669      auto RTCall =
3670          Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name, ByValAttrSet),
3671                             {Queue, Flags, Range, Kernel, Block});
3672      RTCall->setAttributes(ByValAttrSet);
3673      return RValue::get(RTCall);
3674    }
3675     (0) . __assert_fail ("NumArgs >= 5 && \"Invalid enqueue_kernel signature\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 3675, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(NumArgs >= 5 && "Invalid enqueue_kernel signature");
3676
3677    // Create a temporary array to hold the sizes of local pointer arguments
3678    // for the block. \p First is the position of the first size argument.
3679    auto CreateArrayForSizeVar = [=](unsigned First)
3680        -> std::tuple<llvm::Value *, llvm::Value *, llvm::Value *> {
3681      llvm::APInt ArraySize(32, NumArgs - First);
3682      QualType SizeArrayTy = getContext().getConstantArrayType(
3683          getContext().getSizeType(), ArraySize, ArrayType::Normal,
3684          /*IndexTypeQuals=*/0);
3685      auto Tmp = CreateMemTemp(SizeArrayTy, "block_sizes");
3686      llvm::Value *TmpPtr = Tmp.getPointer();
3687      llvm::Value *TmpSize = EmitLifetimeStart(
3688          CGM.getDataLayout().getTypeAllocSize(Tmp.getElementType()), TmpPtr);
3689      llvm::Value *ElemPtr;
3690      // Each of the following arguments specifies the size of the corresponding
3691      // argument passed to the enqueued block.
3692      auto *Zero = llvm::ConstantInt::get(IntTy, 0);
3693      for (unsigned I = First; I < NumArgs; ++I) {
3694        auto *Index = llvm::ConstantInt::get(IntTy, I - First);
3695        auto *GEP = Builder.CreateGEP(TmpPtr, {Zero, Index});
3696        if (I == First)
3697          ElemPtr = GEP;
3698        auto *V =
3699            Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(I)), SizeTy);
3700        Builder.CreateAlignedStore(
3701            V, GEP, CGM.getDataLayout().getPrefTypeAlignment(SizeTy));
3702      }
3703      return std::tie(ElemPtr, TmpSize, TmpPtr);
3704    };
3705
3706    // Could have events and/or varargs.
3707    if (E->getArg(3)->getType()->isBlockPointerType()) {
3708      // No events passed, but has variadic arguments.
3709      Name = "__enqueue_kernel_varargs";
3710      auto Info =
3711          CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(3));
3712      llvm::Value *Kernel =
3713          Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3714      auto *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3715      llvm::Value *ElemPtr, *TmpSize, *TmpPtr;
3716      std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(4);
3717
3718      // Create a vector of the arguments, as well as a constant value to
3719      // express to the runtime the number of variadic arguments.
3720      std::vector<llvm::Value *> Args = {
3721          Queue,  Flags, Range,
3722          Kernel, Block, ConstantInt::get(IntTy, NumArgs - 4),
3723          ElemPtr};
3724      std::vector<llvm::Type *> ArgTys = {
3725          QueueTy,          IntTy, RangeTy,           GenericVoidPtrTy,
3726          GenericVoidPtrTy, IntTy, ElemPtr->getType()};
3727
3728      llvm::FunctionType *FTy = llvm::FunctionType::get(
3729          Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3730      auto Call =
3731          RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3732                                         llvm::ArrayRef<llvm::Value *>(Args)));
3733      if (TmpSize)
3734        EmitLifetimeEnd(TmpSize, TmpPtr);
3735      return Call;
3736    }
3737    // Any calls now have event arguments passed.
3738    if (NumArgs >= 7) {
3739      llvm::Type *EventTy = ConvertType(getContext().OCLClkEventTy);
3740      llvm::Type *EventPtrTy = EventTy->getPointerTo(
3741          CGM.getContext().getTargetAddressSpace(LangAS::opencl_generic));
3742
3743      llvm::Value *NumEvents =
3744          Builder.CreateZExtOrTrunc(EmitScalarExpr(E->getArg(3)), Int32Ty);
3745      llvm::Value *EventList =
3746          E->getArg(4)->getType()->isArrayType()
3747              ? EmitArrayToPointerDecay(E->getArg(4)).getPointer()
3748              : EmitScalarExpr(E->getArg(4));
3749      llvm::Value *ClkEvent = EmitScalarExpr(E->getArg(5));
3750      // Convert to generic address space.
3751      EventList = Builder.CreatePointerCast(EventList, EventPtrTy);
3752      ClkEvent = ClkEvent->getType()->isIntegerTy()
3753                   ? Builder.CreateBitOrPointerCast(ClkEvent, EventPtrTy)
3754                   : Builder.CreatePointerCast(ClkEvent, EventPtrTy);
3755      auto Info =
3756          CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(6));
3757      llvm::Value *Kernel =
3758          Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3759      llvm::Value *Block =
3760          Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3761
3762      std::vector<llvm::Type *> ArgTys = {
3763          QueueTy,    Int32Ty,    RangeTy,          Int32Ty,
3764          EventPtrTy, EventPtrTy, GenericVoidPtrTy, GenericVoidPtrTy};
3765
3766      std::vector<llvm::Value *> Args = {Queue,     Flags,    Range,  NumEvents,
3767                                         EventList, ClkEvent, Kernel, Block};
3768
3769      if (NumArgs == 7) {
3770        // Has events but no variadics.
3771        Name = "__enqueue_kernel_basic_events";
3772        llvm::FunctionType *FTy = llvm::FunctionType::get(
3773            Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3774        return RValue::get(
3775            Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3776                               llvm::ArrayRef<llvm::Value *>(Args)));
3777      }
3778      // Has event info and variadics
3779      // Pass the number of variadics to the runtime function too.
3780      Args.push_back(ConstantInt::get(Int32Ty, NumArgs - 7));
3781      ArgTys.push_back(Int32Ty);
3782      Name = "__enqueue_kernel_events_varargs";
3783
3784      llvm::Value *ElemPtr, *TmpSize, *TmpPtr;
3785      std::tie(ElemPtr, TmpSize, TmpPtr) = CreateArrayForSizeVar(7);
3786      Args.push_back(ElemPtr);
3787      ArgTys.push_back(ElemPtr->getType());
3788
3789      llvm::FunctionType *FTy = llvm::FunctionType::get(
3790          Int32Ty, llvm::ArrayRef<llvm::Type *>(ArgTys), false);
3791      auto Call =
3792          RValue::get(Builder.CreateCall(CGM.CreateRuntimeFunction(FTy, Name),
3793                                         llvm::ArrayRef<llvm::Value *>(Args)));
3794      if (TmpSize)
3795        EmitLifetimeEnd(TmpSize, TmpPtr);
3796      return Call;
3797    }
3798    LLVM_FALLTHROUGH;
3799  }
3800  // OpenCL v2.0 s6.13.17.6 - Kernel query functions need bitcast of block
3801  // parameter.
3802  case Builtin::BIget_kernel_work_group_size: {
3803    llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
3804        getContext().getTargetAddressSpace(LangAS::opencl_generic));
3805    auto Info =
3806        CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0));
3807    Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3808    Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3809    return RValue::get(Builder.CreateCall(
3810        CGM.CreateRuntimeFunction(
3811            llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy},
3812                                    false),
3813            "__get_kernel_work_group_size_impl"),
3814        {Kernel, Arg}));
3815  }
3816  case Builtin::BIget_kernel_preferred_work_group_size_multiple: {
3817    llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
3818        getContext().getTargetAddressSpace(LangAS::opencl_generic));
3819    auto Info =
3820        CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(0));
3821    Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3822    Value *Arg = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3823    return RValue::get(Builder.CreateCall(
3824        CGM.CreateRuntimeFunction(
3825            llvm::FunctionType::get(IntTy, {GenericVoidPtrTy, GenericVoidPtrTy},
3826                                    false),
3827            "__get_kernel_preferred_work_group_size_multiple_impl"),
3828        {Kernel, Arg}));
3829  }
3830  case Builtin::BIget_kernel_max_sub_group_size_for_ndrange:
3831  case Builtin::BIget_kernel_sub_group_count_for_ndrange: {
3832    llvm::Type *GenericVoidPtrTy = Builder.getInt8PtrTy(
3833        getContext().getTargetAddressSpace(LangAS::opencl_generic));
3834    LValue NDRangeL = EmitAggExprToLValue(E->getArg(0));
3835    llvm::Value *NDRange = NDRangeL.getAddress().getPointer();
3836    auto Info =
3837        CGM.getOpenCLRuntime().emitOpenCLEnqueuedBlock(*this, E->getArg(1));
3838    Value *Kernel = Builder.CreatePointerCast(Info.Kernel, GenericVoidPtrTy);
3839    Value *Block = Builder.CreatePointerCast(Info.BlockArg, GenericVoidPtrTy);
3840    const char *Name =
3841        BuiltinID == Builtin::BIget_kernel_max_sub_group_size_for_ndrange
3842            ? "__get_kernel_max_sub_group_size_for_ndrange_impl"
3843            : "__get_kernel_sub_group_count_for_ndrange_impl";
3844    return RValue::get(Builder.CreateCall(
3845        CGM.CreateRuntimeFunction(
3846            llvm::FunctionType::get(
3847                IntTy, {NDRange->getType(), GenericVoidPtrTy, GenericVoidPtrTy},
3848                false),
3849            Name),
3850        {NDRange, Kernel, Block}));
3851  }
3852
3853  case Builtin::BI__builtin_store_half:
3854  case Builtin::BI__builtin_store_halff: {
3855    Value *Val = EmitScalarExpr(E->getArg(0));
3856    Address Address = EmitPointerWithAlignment(E->getArg(1));
3857    Value *HalfVal = Builder.CreateFPTrunc(Val, Builder.getHalfTy());
3858    return RValue::get(Builder.CreateStore(HalfVal, Address));
3859  }
3860  case Builtin::BI__builtin_load_half: {
3861    Address Address = EmitPointerWithAlignment(E->getArg(0));
3862    Value *HalfVal = Builder.CreateLoad(Address);
3863    return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getDoubleTy()));
3864  }
3865  case Builtin::BI__builtin_load_halff: {
3866    Address Address = EmitPointerWithAlignment(E->getArg(0));
3867    Value *HalfVal = Builder.CreateLoad(Address);
3868    return RValue::get(Builder.CreateFPExt(HalfVal, Builder.getFloatTy()));
3869  }
3870  case Builtin::BIprintf:
3871    if (getTarget().getTriple().isNVPTX())
3872      return EmitNVPTXDevicePrintfCallExpr(E, ReturnValue);
3873    break;
3874  case Builtin::BI__builtin_canonicalize:
3875  case Builtin::BI__builtin_canonicalizef:
3876  case Builtin::BI__builtin_canonicalizel:
3877    return RValue::get(emitUnaryBuiltin(*this, E, Intrinsic::canonicalize));
3878
3879  case Builtin::BI__builtin_thread_pointer: {
3880    if (!getContext().getTargetInfo().isTLSSupported())
3881      CGM.ErrorUnsupported(E, "__builtin_thread_pointer");
3882    // Fall through - it's already mapped to the intrinsic by GCCBuiltin.
3883    break;
3884  }
3885  case Builtin::BI__builtin_os_log_format:
3886    return emitBuiltinOSLogFormat(*E);
3887
3888  case Builtin::BI__xray_customevent: {
3889    if (!ShouldXRayInstrumentFunction())
3890      return RValue::getIgnored();
3891
3892    if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has(
3893            XRayInstrKind::Custom))
3894      return RValue::getIgnored();
3895
3896    if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>())
3897      if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayCustomEvents())
3898        return RValue::getIgnored();
3899
3900    Function *F = CGM.getIntrinsic(Intrinsic::xray_customevent);
3901    auto FTy = F->getFunctionType();
3902    auto Arg0 = E->getArg(0);
3903    auto Arg0Val = EmitScalarExpr(Arg0);
3904    auto Arg0Ty = Arg0->getType();
3905    auto PTy0 = FTy->getParamType(0);
3906    if (PTy0 != Arg0Val->getType()) {
3907      if (Arg0Ty->isArrayType())
3908        Arg0Val = EmitArrayToPointerDecay(Arg0).getPointer();
3909      else
3910        Arg0Val = Builder.CreatePointerCast(Arg0Val, PTy0);
3911    }
3912    auto Arg1 = EmitScalarExpr(E->getArg(1));
3913    auto PTy1 = FTy->getParamType(1);
3914    if (PTy1 != Arg1->getType())
3915      Arg1 = Builder.CreateTruncOrBitCast(Arg1, PTy1);
3916    return RValue::get(Builder.CreateCall(F, {Arg0Val, Arg1}));
3917  }
3918
3919  case Builtin::BI__xray_typedevent: {
3920    // TODO: There should be a way to always emit events even if the current
3921    // function is not instrumented. Losing events in a stream can cripple
3922    // a trace.
3923    if (!ShouldXRayInstrumentFunction())
3924      return RValue::getIgnored();
3925
3926    if (!CGM.getCodeGenOpts().XRayInstrumentationBundle.has(
3927            XRayInstrKind::Typed))
3928      return RValue::getIgnored();
3929
3930    if (const auto *XRayAttr = CurFuncDecl->getAttr<XRayInstrumentAttr>())
3931      if (XRayAttr->neverXRayInstrument() && !AlwaysEmitXRayTypedEvents())
3932        return RValue::getIgnored();
3933
3934    Function *F = CGM.getIntrinsic(Intrinsic::xray_typedevent);
3935    auto FTy = F->getFunctionType();
3936    auto Arg0 = EmitScalarExpr(E->getArg(0));
3937    auto PTy0 = FTy->getParamType(0);
3938    if (PTy0 != Arg0->getType())
3939      Arg0 = Builder.CreateTruncOrBitCast(Arg0, PTy0);
3940    auto Arg1 = E->getArg(1);
3941    auto Arg1Val = EmitScalarExpr(Arg1);
3942    auto Arg1Ty = Arg1->getType();
3943    auto PTy1 = FTy->getParamType(1);
3944    if (PTy1 != Arg1Val->getType()) {
3945      if (Arg1Ty->isArrayType())
3946        Arg1Val = EmitArrayToPointerDecay(Arg1).getPointer();
3947      else
3948        Arg1Val = Builder.CreatePointerCast(Arg1Val, PTy1);
3949    }
3950    auto Arg2 = EmitScalarExpr(E->getArg(2));
3951    auto PTy2 = FTy->getParamType(2);
3952    if (PTy2 != Arg2->getType())
3953      Arg2 = Builder.CreateTruncOrBitCast(Arg2, PTy2);
3954    return RValue::get(Builder.CreateCall(F, {Arg0, Arg1Val, Arg2}));
3955  }
3956
3957  case Builtin::BI__builtin_ms_va_start:
3958  case Builtin::BI__builtin_ms_va_end:
3959    return RValue::get(
3960        EmitVAStartEnd(EmitMSVAListRef(E->getArg(0)).getPointer(),
3961                       BuiltinID == Builtin::BI__builtin_ms_va_start));
3962
3963  case Builtin::BI__builtin_ms_va_copy: {
3964    // Lower this manually. We can't reliably determine whether or not any
3965    // given va_copy() is for a Win64 va_list from the calling convention
3966    // alone, because it's legal to do this from a System V ABI function.
3967    // With opaque pointer types, we won't have enough information in LLVM
3968    // IR to determine this from the argument types, either. Best to do it
3969    // now, while we have enough information.
3970    Address DestAddr = EmitMSVAListRef(E->getArg(0));
3971    Address SrcAddr = EmitMSVAListRef(E->getArg(1));
3972
3973    llvm::Type *BPP = Int8PtrPtrTy;
3974
3975    DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), BPP, "cp"),
3976                       DestAddr.getAlignment());
3977    SrcAddr = Address(Builder.CreateBitCast(SrcAddr.getPointer(), BPP, "ap"),
3978                      SrcAddr.getAlignment());
3979
3980    Value *ArgPtr = Builder.CreateLoad(SrcAddr, "ap.val");
3981    return RValue::get(Builder.CreateStore(ArgPtr, DestAddr));
3982  }
3983  }
3984
3985  // If this is an alias for a lib function (e.g. __builtin_sin), emit
3986  // the call using the normal call path, but using the unmangled
3987  // version of the function name.
3988  if (getContext().BuiltinInfo.isLibFunction(BuiltinID))
3989    return emitLibraryCall(*this, FD, E,
3990                           CGM.getBuiltinLibFunction(FD, BuiltinID));
3991
3992  // If this is a predefined lib function (e.g. malloc), emit the call
3993  // using exactly the normal call path.
3994  if (getContext().BuiltinInfo.isPredefinedLibFunction(BuiltinID))
3995    return emitLibraryCall(*this, FD, E,
3996                      cast<llvm::Constant>(EmitScalarExpr(E->getCallee())));
3997
3998  // Check that a call to a target specific builtin has the correct target
3999  // features.
4000  // This is down here to avoid non-target specific builtins, however, if
4001  // generic builtins start to require generic target features then we
4002  // can move this up to the beginning of the function.
4003  checkTargetFeatures(E, FD);
4004
4005  if (unsigned VectorWidth = getContext().BuiltinInfo.getRequiredVectorWidth(BuiltinID))
4006    LargestVectorWidth = std::max(LargestVectorWidth, VectorWidth);
4007
4008  // See if we have a target specific intrinsic.
4009  const char *Name = getContext().BuiltinInfo.getName(BuiltinID);
4010  Intrinsic::ID IntrinsicID = Intrinsic::not_intrinsic;
4011  StringRef Prefix =
4012      llvm::Triple::getArchTypePrefix(getTarget().getTriple().getArch());
4013  if (!Prefix.empty()) {
4014    IntrinsicID = Intrinsic::getIntrinsicForGCCBuiltin(Prefix.data(), Name);
4015    // NOTE we don't need to perform a compatibility flag check here since the
4016    // intrinsics are declared in Builtins*.def via LANGBUILTIN which filter the
4017    // MS builtins via ALL_MS_LANGUAGES and are filtered earlier.
4018    if (IntrinsicID == Intrinsic::not_intrinsic)
4019      IntrinsicID = Intrinsic::getIntrinsicForMSBuiltin(Prefix.data(), Name);
4020  }
4021
4022  if (IntrinsicID != Intrinsic::not_intrinsic) {
4023    SmallVector<Value*, 16> Args;
4024
4025    // Find out if any arguments are required to be integer constant
4026    // expressions.
4027    unsigned ICEArguments = 0;
4028    ASTContext::GetBuiltinTypeError Error;
4029    getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
4030     (0) . __assert_fail ("Error == ASTContext..GE_None && \"Should not codegen an error\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 4030, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Error == ASTContext::GE_None && "Should not codegen an error");
4031
4032    Function *F = CGM.getIntrinsic(IntrinsicID);
4033    llvm::FunctionType *FTy = F->getFunctionType();
4034
4035    for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) {
4036      Value *ArgValue;
4037      // If this is a normal argument, just emit it as a scalar.
4038      if ((ICEArguments & (1 << i)) == 0) {
4039        ArgValue = EmitScalarExpr(E->getArg(i));
4040      } else {
4041        // If this is required to be a constant, constant fold it so that we
4042        // know that the generated intrinsic gets a ConstantInt.
4043        llvm::APSInt Result;
4044        bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result,getContext());
4045         (0) . __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 4045, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(IsConst && "Constant arg isn't actually constant?");
4046        (void)IsConst;
4047        ArgValue = llvm::ConstantInt::get(getLLVMContext(), Result);
4048      }
4049
4050      // If the intrinsic arg type is different from the builtin arg type
4051      // we need to do a bit cast.
4052      llvm::Type *PTy = FTy->getParamType(i);
4053      if (PTy != ArgValue->getType()) {
4054        // XXX - vector of pointers?
4055        if (auto *PtrTy = dyn_cast<llvm::PointerType>(PTy)) {
4056          if (PtrTy->getAddressSpace() !=
4057              ArgValue->getType()->getPointerAddressSpace()) {
4058            ArgValue = Builder.CreateAddrSpaceCast(
4059              ArgValue,
4060              ArgValue->getType()->getPointerTo(PtrTy->getAddressSpace()));
4061          }
4062        }
4063
4064         (0) . __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 4065, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) &&
4065 (0) . __assert_fail ("PTy->canLosslesslyBitCastTo(FTy->getParamType(i)) && \"Must be able to losslessly bit cast to param\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 4065, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">               "Must be able to losslessly bit cast to param");
4066        ArgValue = Builder.CreateBitCast(ArgValue, PTy);
4067      }
4068
4069      Args.push_back(ArgValue);
4070    }
4071
4072    Value *V = Builder.CreateCall(F, Args);
4073    QualType BuiltinRetType = E->getType();
4074
4075    llvm::Type *RetTy = VoidTy;
4076    if (!BuiltinRetType->isVoidType())
4077      RetTy = ConvertType(BuiltinRetType);
4078
4079    if (RetTy != V->getType()) {
4080      // XXX - vector of pointers?
4081      if (auto *PtrTy = dyn_cast<llvm::PointerType>(RetTy)) {
4082        if (PtrTy->getAddressSpace() != V->getType()->getPointerAddressSpace()) {
4083          V = Builder.CreateAddrSpaceCast(
4084            V, V->getType()->getPointerTo(PtrTy->getAddressSpace()));
4085        }
4086      }
4087
4088       (0) . __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 4089, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(V->getType()->canLosslesslyBitCastTo(RetTy) &&
4089 (0) . __assert_fail ("V->getType()->canLosslesslyBitCastTo(RetTy) && \"Must be able to losslessly bit cast result type\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 4089, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">             "Must be able to losslessly bit cast result type");
4090      V = Builder.CreateBitCast(V, RetTy);
4091    }
4092
4093    return RValue::get(V);
4094  }
4095
4096  // See if we have a target specific builtin that needs to be lowered.
4097  if (Value *V = EmitTargetBuiltinExpr(BuiltinID, E))
4098    return RValue::get(V);
4099
4100  ErrorUnsupported(E, "builtin function");
4101
4102  // Unknown builtin, for now just dump it out and return undef.
4103  return GetUndefRValue(E->getType());
4104}
4105
4106static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF,
4107                                        unsigned BuiltinID, const CallExpr *E,
4108                                        llvm::Triple::ArchType Arch) {
4109  switch (Arch) {
4110  case llvm::Triple::arm:
4111  case llvm::Triple::armeb:
4112  case llvm::Triple::thumb:
4113  case llvm::Triple::thumbeb:
4114    return CGF->EmitARMBuiltinExpr(BuiltinID, E, Arch);
4115  case llvm::Triple::aarch64:
4116  case llvm::Triple::aarch64_be:
4117    return CGF->EmitAArch64BuiltinExpr(BuiltinID, E, Arch);
4118  case llvm::Triple::x86:
4119  case llvm::Triple::x86_64:
4120    return CGF->EmitX86BuiltinExpr(BuiltinID, E);
4121  case llvm::Triple::ppc:
4122  case llvm::Triple::ppc64:
4123  case llvm::Triple::ppc64le:
4124    return CGF->EmitPPCBuiltinExpr(BuiltinID, E);
4125  case llvm::Triple::r600:
4126  case llvm::Triple::amdgcn:
4127    return CGF->EmitAMDGPUBuiltinExpr(BuiltinID, E);
4128  case llvm::Triple::systemz:
4129    return CGF->EmitSystemZBuiltinExpr(BuiltinID, E);
4130  case llvm::Triple::nvptx:
4131  case llvm::Triple::nvptx64:
4132    return CGF->EmitNVPTXBuiltinExpr(BuiltinID, E);
4133  case llvm::Triple::wasm32:
4134  case llvm::Triple::wasm64:
4135    return CGF->EmitWebAssemblyBuiltinExpr(BuiltinID, E);
4136  case llvm::Triple::hexagon:
4137    return CGF->EmitHexagonBuiltinExpr(BuiltinID, E);
4138  default:
4139    return nullptr;
4140  }
4141}
4142
4143Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID,
4144                                              const CallExpr *E) {
4145  if (getContext().BuiltinInfo.isAuxBuiltinID(BuiltinID)) {
4146     (0) . __assert_fail ("getContext().getAuxTargetInfo() && \"Missing aux target info\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 4146, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(getContext().getAuxTargetInfo() && "Missing aux target info");
4147    return EmitTargetArchBuiltinExpr(
4148        this, getContext().BuiltinInfo.getAuxBuiltinID(BuiltinID), E,
4149        getContext().getAuxTargetInfo()->getTriple().getArch());
4150  }
4151
4152  return EmitTargetArchBuiltinExpr(this, BuiltinID, E,
4153                                   getTarget().getTriple().getArch());
4154}
4155
4156static llvm::VectorType *GetNeonType(CodeGenFunction *CGF,
4157                                     NeonTypeFlags TypeFlags,
4158                                     bool HasLegalHalfType=true,
4159                                     bool V1Ty=false) {
4160  int IsQuad = TypeFlags.isQuad();
4161  switch (TypeFlags.getEltType()) {
4162  case NeonTypeFlags::Int8:
4163  case NeonTypeFlags::Poly8:
4164    return llvm::VectorType::get(CGF->Int8Ty, V1Ty ? 1 : (8 << IsQuad));
4165  case NeonTypeFlags::Int16:
4166  case NeonTypeFlags::Poly16:
4167    return llvm::VectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad));
4168  case NeonTypeFlags::Float16:
4169    if (HasLegalHalfType)
4170      return llvm::VectorType::get(CGF->HalfTy, V1Ty ? 1 : (4 << IsQuad));
4171    else
4172      return llvm::VectorType::get(CGF->Int16Ty, V1Ty ? 1 : (4 << IsQuad));
4173  case NeonTypeFlags::Int32:
4174    return llvm::VectorType::get(CGF->Int32Ty, V1Ty ? 1 : (2 << IsQuad));
4175  case NeonTypeFlags::Int64:
4176  case NeonTypeFlags::Poly64:
4177    return llvm::VectorType::get(CGF->Int64Ty, V1Ty ? 1 : (1 << IsQuad));
4178  case NeonTypeFlags::Poly128:
4179    // FIXME: i128 and f128 doesn't get fully support in Clang and llvm.
4180    // There is a lot of i128 and f128 API missing.
4181    // so we use v16i8 to represent poly128 and get pattern matched.
4182    return llvm::VectorType::get(CGF->Int8Ty, 16);
4183  case NeonTypeFlags::Float32:
4184    return llvm::VectorType::get(CGF->FloatTy, V1Ty ? 1 : (2 << IsQuad));
4185  case NeonTypeFlags::Float64:
4186    return llvm::VectorType::get(CGF->DoubleTy, V1Ty ? 1 : (1 << IsQuad));
4187  }
4188  llvm_unreachable("Unknown vector element type!");
4189}
4190
4191static llvm::VectorType *GetFloatNeonType(CodeGenFunction *CGF,
4192                                          NeonTypeFlags IntTypeFlags) {
4193  int IsQuad = IntTypeFlags.isQuad();
4194  switch (IntTypeFlags.getEltType()) {
4195  case NeonTypeFlags::Int16:
4196    return llvm::VectorType::get(CGF->HalfTy, (4 << IsQuad));
4197  case NeonTypeFlags::Int32:
4198    return llvm::VectorType::get(CGF->FloatTy, (2 << IsQuad));
4199  case NeonTypeFlags::Int64:
4200    return llvm::VectorType::get(CGF->DoubleTy, (1 << IsQuad));
4201  default:
4202    llvm_unreachable("Type can't be converted to floating-point!");
4203  }
4204}
4205
4206Value *CodeGenFunction::EmitNeonSplat(Value *V, Constant *C) {
4207  unsigned nElts = V->getType()->getVectorNumElements();
4208  Value* SV = llvm::ConstantVector::getSplat(nElts, C);
4209  return Builder.CreateShuffleVector(V, V, SV, "lane");
4210}
4211
4212Value *CodeGenFunction::EmitNeonCall(Function *F, SmallVectorImpl<Value*> &Ops,
4213                                     const char *name,
4214                                     unsigned shift, bool rightshift) {
4215  unsigned j = 0;
4216  for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
4217       ai != ae; ++ai, ++j)
4218    if (shift > 0 && shift == j)
4219      Ops[j] = EmitNeonShiftVector(Ops[j], ai->getType(), rightshift);
4220    else
4221      Ops[j] = Builder.CreateBitCast(Ops[j], ai->getType(), name);
4222
4223  return Builder.CreateCall(F, Ops, name);
4224}
4225
4226Value *CodeGenFunction::EmitNeonShiftVector(Value *V, llvm::Type *Ty,
4227                                            bool neg) {
4228  int SV = cast<ConstantInt>(V)->getSExtValue();
4229  return ConstantInt::get(Ty, neg ? -SV : SV);
4230}
4231
4232// Right-shift a vector by a constant.
4233Value *CodeGenFunction::EmitNeonRShiftImm(Value *Vec, Value *Shift,
4234                                          llvm::Type *Ty, bool usgn,
4235                                          const char *name) {
4236  llvm::VectorType *VTy = cast<llvm::VectorType>(Ty);
4237
4238  int ShiftAmt = cast<ConstantInt>(Shift)->getSExtValue();
4239  int EltSize = VTy->getScalarSizeInBits();
4240
4241  Vec = Builder.CreateBitCast(Vec, Ty);
4242
4243  // lshr/ashr are undefined when the shift amount is equal to the vector
4244  // element size.
4245  if (ShiftAmt == EltSize) {
4246    if (usgn) {
4247      // Right-shifting an unsigned value by its size yields 0.
4248      return llvm::ConstantAggregateZero::get(VTy);
4249    } else {
4250      // Right-shifting a signed value by its size is equivalent
4251      // to a shift of size-1.
4252      --ShiftAmt;
4253      Shift = ConstantInt::get(VTy->getElementType(), ShiftAmt);
4254    }
4255  }
4256
4257  Shift = EmitNeonShiftVector(Shift, Ty, false);
4258  if (usgn)
4259    return Builder.CreateLShr(Vec, Shift, name);
4260  else
4261    return Builder.CreateAShr(Vec, Shift, name);
4262}
4263
4264enum {
4265  AddRetType = (1 << 0),
4266  Add1ArgType = (1 << 1),
4267  Add2ArgTypes = (1 << 2),
4268
4269  VectorizeRetType = (1 << 3),
4270  VectorizeArgTypes = (1 << 4),
4271
4272  InventFloatType = (1 << 5),
4273  UnsignedAlts = (1 << 6),
4274
4275  Use64BitVectors = (1 << 7),
4276  Use128BitVectors = (1 << 8),
4277
4278  Vectorize1ArgType = Add1ArgType | VectorizeArgTypes,
4279  VectorRet = AddRetType | VectorizeRetType,
4280  VectorRetGetArgs01 =
4281      AddRetType | Add2ArgTypes | VectorizeRetType | VectorizeArgTypes,
4282  FpCmpzModifiers =
4283      AddRetType | VectorizeRetType | Add1ArgType | InventFloatType
4284};
4285
4286namespace {
4287struct NeonIntrinsicInfo {
4288  const char *NameHint;
4289  unsigned BuiltinID;
4290  unsigned LLVMIntrinsic;
4291  unsigned AltLLVMIntrinsic;
4292  unsigned TypeModifier;
4293
4294  bool operator<(unsigned RHSBuiltinID) const {
4295    return BuiltinID < RHSBuiltinID;
4296  }
4297  bool operator<(const NeonIntrinsicInfo &TE) const {
4298    return BuiltinID < TE.BuiltinID;
4299  }
4300};
4301} // end anonymous namespace
4302
4303#define NEONMAP0(NameBase) \
4304  { #NameBase, NEON::BI__builtin_neon_ ## NameBase, 0, 0, 0 }
4305
4306#define NEONMAP1(NameBase, LLVMIntrinsic, TypeModifier) \
4307  { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \
4308      Intrinsic::LLVMIntrinsic, 0, TypeModifier }
4309
4310#define NEONMAP2(NameBase, LLVMIntrinsic, AltLLVMIntrinsic, TypeModifier) \
4311  { #NameBase, NEON:: BI__builtin_neon_ ## NameBase, \
4312      Intrinsic::LLVMIntrinsic, Intrinsic::AltLLVMIntrinsic, \
4313      TypeModifier }
4314
4315static const NeonIntrinsicInfo ARMSIMDIntrinsicMap [] = {
4316  NEONMAP2(vabd_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts),
4317  NEONMAP2(vabdq_v, arm_neon_vabdu, arm_neon_vabds, Add1ArgType | UnsignedAlts),
4318  NEONMAP1(vabs_v, arm_neon_vabs, 0),
4319  NEONMAP1(vabsq_v, arm_neon_vabs, 0),
4320  NEONMAP0(vaddhn_v),
4321  NEONMAP1(vaesdq_v, arm_neon_aesd, 0),
4322  NEONMAP1(vaeseq_v, arm_neon_aese, 0),
4323  NEONMAP1(vaesimcq_v, arm_neon_aesimc, 0),
4324  NEONMAP1(vaesmcq_v, arm_neon_aesmc, 0),
4325  NEONMAP1(vbsl_v, arm_neon_vbsl, AddRetType),
4326  NEONMAP1(vbslq_v, arm_neon_vbsl, AddRetType),
4327  NEONMAP1(vcage_v, arm_neon_vacge, 0),
4328  NEONMAP1(vcageq_v, arm_neon_vacge, 0),
4329  NEONMAP1(vcagt_v, arm_neon_vacgt, 0),
4330  NEONMAP1(vcagtq_v, arm_neon_vacgt, 0),
4331  NEONMAP1(vcale_v, arm_neon_vacge, 0),
4332  NEONMAP1(vcaleq_v, arm_neon_vacge, 0),
4333  NEONMAP1(vcalt_v, arm_neon_vacgt, 0),
4334  NEONMAP1(vcaltq_v, arm_neon_vacgt, 0),
4335  NEONMAP0(vceqz_v),
4336  NEONMAP0(vceqzq_v),
4337  NEONMAP0(vcgez_v),
4338  NEONMAP0(vcgezq_v),
4339  NEONMAP0(vcgtz_v),
4340  NEONMAP0(vcgtzq_v),
4341  NEONMAP0(vclez_v),
4342  NEONMAP0(vclezq_v),
4343  NEONMAP1(vcls_v, arm_neon_vcls, Add1ArgType),
4344  NEONMAP1(vclsq_v, arm_neon_vcls, Add1ArgType),
4345  NEONMAP0(vcltz_v),
4346  NEONMAP0(vcltzq_v),
4347  NEONMAP1(vclz_v, ctlz, Add1ArgType),
4348  NEONMAP1(vclzq_v, ctlz, Add1ArgType),
4349  NEONMAP1(vcnt_v, ctpop, Add1ArgType),
4350  NEONMAP1(vcntq_v, ctpop, Add1ArgType),
4351  NEONMAP1(vcvt_f16_f32, arm_neon_vcvtfp2hf, 0),
4352  NEONMAP0(vcvt_f16_v),
4353  NEONMAP1(vcvt_f32_f16, arm_neon_vcvthf2fp, 0),
4354  NEONMAP0(vcvt_f32_v),
4355  NEONMAP2(vcvt_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
4356  NEONMAP2(vcvt_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
4357  NEONMAP1(vcvt_n_s16_v, arm_neon_vcvtfp2fxs, 0),
4358  NEONMAP1(vcvt_n_s32_v, arm_neon_vcvtfp2fxs, 0),
4359  NEONMAP1(vcvt_n_s64_v, arm_neon_vcvtfp2fxs, 0),
4360  NEONMAP1(vcvt_n_u16_v, arm_neon_vcvtfp2fxu, 0),
4361  NEONMAP1(vcvt_n_u32_v, arm_neon_vcvtfp2fxu, 0),
4362  NEONMAP1(vcvt_n_u64_v, arm_neon_vcvtfp2fxu, 0),
4363  NEONMAP0(vcvt_s16_v),
4364  NEONMAP0(vcvt_s32_v),
4365  NEONMAP0(vcvt_s64_v),
4366  NEONMAP0(vcvt_u16_v),
4367  NEONMAP0(vcvt_u32_v),
4368  NEONMAP0(vcvt_u64_v),
4369  NEONMAP1(vcvta_s16_v, arm_neon_vcvtas, 0),
4370  NEONMAP1(vcvta_s32_v, arm_neon_vcvtas, 0),
4371  NEONMAP1(vcvta_s64_v, arm_neon_vcvtas, 0),
4372  NEONMAP1(vcvta_u16_v, arm_neon_vcvtau, 0),
4373  NEONMAP1(vcvta_u32_v, arm_neon_vcvtau, 0),
4374  NEONMAP1(vcvta_u64_v, arm_neon_vcvtau, 0),
4375  NEONMAP1(vcvtaq_s16_v, arm_neon_vcvtas, 0),
4376  NEONMAP1(vcvtaq_s32_v, arm_neon_vcvtas, 0),
4377  NEONMAP1(vcvtaq_s64_v, arm_neon_vcvtas, 0),
4378  NEONMAP1(vcvtaq_u16_v, arm_neon_vcvtau, 0),
4379  NEONMAP1(vcvtaq_u32_v, arm_neon_vcvtau, 0),
4380  NEONMAP1(vcvtaq_u64_v, arm_neon_vcvtau, 0),
4381  NEONMAP1(vcvtm_s16_v, arm_neon_vcvtms, 0),
4382  NEONMAP1(vcvtm_s32_v, arm_neon_vcvtms, 0),
4383  NEONMAP1(vcvtm_s64_v, arm_neon_vcvtms, 0),
4384  NEONMAP1(vcvtm_u16_v, arm_neon_vcvtmu, 0),
4385  NEONMAP1(vcvtm_u32_v, arm_neon_vcvtmu, 0),
4386  NEONMAP1(vcvtm_u64_v, arm_neon_vcvtmu, 0),
4387  NEONMAP1(vcvtmq_s16_v, arm_neon_vcvtms, 0),
4388  NEONMAP1(vcvtmq_s32_v, arm_neon_vcvtms, 0),
4389  NEONMAP1(vcvtmq_s64_v, arm_neon_vcvtms, 0),
4390  NEONMAP1(vcvtmq_u16_v, arm_neon_vcvtmu, 0),
4391  NEONMAP1(vcvtmq_u32_v, arm_neon_vcvtmu, 0),
4392  NEONMAP1(vcvtmq_u64_v, arm_neon_vcvtmu, 0),
4393  NEONMAP1(vcvtn_s16_v, arm_neon_vcvtns, 0),
4394  NEONMAP1(vcvtn_s32_v, arm_neon_vcvtns, 0),
4395  NEONMAP1(vcvtn_s64_v, arm_neon_vcvtns, 0),
4396  NEONMAP1(vcvtn_u16_v, arm_neon_vcvtnu, 0),
4397  NEONMAP1(vcvtn_u32_v, arm_neon_vcvtnu, 0),
4398  NEONMAP1(vcvtn_u64_v, arm_neon_vcvtnu, 0),
4399  NEONMAP1(vcvtnq_s16_v, arm_neon_vcvtns, 0),
4400  NEONMAP1(vcvtnq_s32_v, arm_neon_vcvtns, 0),
4401  NEONMAP1(vcvtnq_s64_v, arm_neon_vcvtns, 0),
4402  NEONMAP1(vcvtnq_u16_v, arm_neon_vcvtnu, 0),
4403  NEONMAP1(vcvtnq_u32_v, arm_neon_vcvtnu, 0),
4404  NEONMAP1(vcvtnq_u64_v, arm_neon_vcvtnu, 0),
4405  NEONMAP1(vcvtp_s16_v, arm_neon_vcvtps, 0),
4406  NEONMAP1(vcvtp_s32_v, arm_neon_vcvtps, 0),
4407  NEONMAP1(vcvtp_s64_v, arm_neon_vcvtps, 0),
4408  NEONMAP1(vcvtp_u16_v, arm_neon_vcvtpu, 0),
4409  NEONMAP1(vcvtp_u32_v, arm_neon_vcvtpu, 0),
4410  NEONMAP1(vcvtp_u64_v, arm_neon_vcvtpu, 0),
4411  NEONMAP1(vcvtpq_s16_v, arm_neon_vcvtps, 0),
4412  NEONMAP1(vcvtpq_s32_v, arm_neon_vcvtps, 0),
4413  NEONMAP1(vcvtpq_s64_v, arm_neon_vcvtps, 0),
4414  NEONMAP1(vcvtpq_u16_v, arm_neon_vcvtpu, 0),
4415  NEONMAP1(vcvtpq_u32_v, arm_neon_vcvtpu, 0),
4416  NEONMAP1(vcvtpq_u64_v, arm_neon_vcvtpu, 0),
4417  NEONMAP0(vcvtq_f16_v),
4418  NEONMAP0(vcvtq_f32_v),
4419  NEONMAP2(vcvtq_n_f16_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
4420  NEONMAP2(vcvtq_n_f32_v, arm_neon_vcvtfxu2fp, arm_neon_vcvtfxs2fp, 0),
4421  NEONMAP1(vcvtq_n_s16_v, arm_neon_vcvtfp2fxs, 0),
4422  NEONMAP1(vcvtq_n_s32_v, arm_neon_vcvtfp2fxs, 0),
4423  NEONMAP1(vcvtq_n_s64_v, arm_neon_vcvtfp2fxs, 0),
4424  NEONMAP1(vcvtq_n_u16_v, arm_neon_vcvtfp2fxu, 0),
4425  NEONMAP1(vcvtq_n_u32_v, arm_neon_vcvtfp2fxu, 0),
4426  NEONMAP1(vcvtq_n_u64_v, arm_neon_vcvtfp2fxu, 0),
4427  NEONMAP0(vcvtq_s16_v),
4428  NEONMAP0(vcvtq_s32_v),
4429  NEONMAP0(vcvtq_s64_v),
4430  NEONMAP0(vcvtq_u16_v),
4431  NEONMAP0(vcvtq_u32_v),
4432  NEONMAP0(vcvtq_u64_v),
4433  NEONMAP2(vdot_v, arm_neon_udot, arm_neon_sdot, 0),
4434  NEONMAP2(vdotq_v, arm_neon_udot, arm_neon_sdot, 0),
4435  NEONMAP0(vext_v),
4436  NEONMAP0(vextq_v),
4437  NEONMAP0(vfma_v),
4438  NEONMAP0(vfmaq_v),
4439  NEONMAP2(vhadd_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts),
4440  NEONMAP2(vhaddq_v, arm_neon_vhaddu, arm_neon_vhadds, Add1ArgType | UnsignedAlts),
4441  NEONMAP2(vhsub_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts),
4442  NEONMAP2(vhsubq_v, arm_neon_vhsubu, arm_neon_vhsubs, Add1ArgType | UnsignedAlts),
4443  NEONMAP0(vld1_dup_v),
4444  NEONMAP1(vld1_v, arm_neon_vld1, 0),
4445  NEONMAP1(vld1_x2_v, arm_neon_vld1x2, 0),
4446  NEONMAP1(vld1_x3_v, arm_neon_vld1x3, 0),
4447  NEONMAP1(vld1_x4_v, arm_neon_vld1x4, 0),
4448  NEONMAP0(vld1q_dup_v),
4449  NEONMAP1(vld1q_v, arm_neon_vld1, 0),
4450  NEONMAP1(vld1q_x2_v, arm_neon_vld1x2, 0),
4451  NEONMAP1(vld1q_x3_v, arm_neon_vld1x3, 0),
4452  NEONMAP1(vld1q_x4_v, arm_neon_vld1x4, 0),
4453  NEONMAP1(vld2_dup_v, arm_neon_vld2dup, 0),
4454  NEONMAP1(vld2_lane_v, arm_neon_vld2lane, 0),
4455  NEONMAP1(vld2_v, arm_neon_vld2, 0),
4456  NEONMAP1(vld2q_dup_v, arm_neon_vld2dup, 0),
4457  NEONMAP1(vld2q_lane_v, arm_neon_vld2lane, 0),
4458  NEONMAP1(vld2q_v, arm_neon_vld2, 0),
4459  NEONMAP1(vld3_dup_v, arm_neon_vld3dup, 0),
4460  NEONMAP1(vld3_lane_v, arm_neon_vld3lane, 0),
4461  NEONMAP1(vld3_v, arm_neon_vld3, 0),
4462  NEONMAP1(vld3q_dup_v, arm_neon_vld3dup, 0),
4463  NEONMAP1(vld3q_lane_v, arm_neon_vld3lane, 0),
4464  NEONMAP1(vld3q_v, arm_neon_vld3, 0),
4465  NEONMAP1(vld4_dup_v, arm_neon_vld4dup, 0),
4466  NEONMAP1(vld4_lane_v, arm_neon_vld4lane, 0),
4467  NEONMAP1(vld4_v, arm_neon_vld4, 0),
4468  NEONMAP1(vld4q_dup_v, arm_neon_vld4dup, 0),
4469  NEONMAP1(vld4q_lane_v, arm_neon_vld4lane, 0),
4470  NEONMAP1(vld4q_v, arm_neon_vld4, 0),
4471  NEONMAP2(vmax_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts),
4472  NEONMAP1(vmaxnm_v, arm_neon_vmaxnm, Add1ArgType),
4473  NEONMAP1(vmaxnmq_v, arm_neon_vmaxnm, Add1ArgType),
4474  NEONMAP2(vmaxq_v, arm_neon_vmaxu, arm_neon_vmaxs, Add1ArgType | UnsignedAlts),
4475  NEONMAP2(vmin_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts),
4476  NEONMAP1(vminnm_v, arm_neon_vminnm, Add1ArgType),
4477  NEONMAP1(vminnmq_v, arm_neon_vminnm, Add1ArgType),
4478  NEONMAP2(vminq_v, arm_neon_vminu, arm_neon_vmins, Add1ArgType | UnsignedAlts),
4479  NEONMAP0(vmovl_v),
4480  NEONMAP0(vmovn_v),
4481  NEONMAP1(vmul_v, arm_neon_vmulp, Add1ArgType),
4482  NEONMAP0(vmull_v),
4483  NEONMAP1(vmulq_v, arm_neon_vmulp, Add1ArgType),
4484  NEONMAP2(vpadal_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts),
4485  NEONMAP2(vpadalq_v, arm_neon_vpadalu, arm_neon_vpadals, UnsignedAlts),
4486  NEONMAP1(vpadd_v, arm_neon_vpadd, Add1ArgType),
4487  NEONMAP2(vpaddl_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts),
4488  NEONMAP2(vpaddlq_v, arm_neon_vpaddlu, arm_neon_vpaddls, UnsignedAlts),
4489  NEONMAP1(vpaddq_v, arm_neon_vpadd, Add1ArgType),
4490  NEONMAP2(vpmax_v, arm_neon_vpmaxu, arm_neon_vpmaxs, Add1ArgType | UnsignedAlts),
4491  NEONMAP2(vpmin_v, arm_neon_vpminu, arm_neon_vpmins, Add1ArgType | UnsignedAlts),
4492  NEONMAP1(vqabs_v, arm_neon_vqabs, Add1ArgType),
4493  NEONMAP1(vqabsq_v, arm_neon_vqabs, Add1ArgType),
4494  NEONMAP2(vqadd_v, arm_neon_vqaddu, arm_neon_vqadds, Add1ArgType | UnsignedAlts),
4495  NEONMAP2(vqaddq_v, arm_neon_vqaddu, arm_neon_vqadds, Add1ArgType | UnsignedAlts),
4496  NEONMAP2(vqdmlal_v, arm_neon_vqdmull, arm_neon_vqadds, 0),
4497  NEONMAP2(vqdmlsl_v, arm_neon_vqdmull, arm_neon_vqsubs, 0),
4498  NEONMAP1(vqdmulh_v, arm_neon_vqdmulh, Add1ArgType),
4499  NEONMAP1(vqdmulhq_v, arm_neon_vqdmulh, Add1ArgType),
4500  NEONMAP1(vqdmull_v, arm_neon_vqdmull, Add1ArgType),
4501  NEONMAP2(vqmovn_v, arm_neon_vqmovnu, arm_neon_vqmovns, Add1ArgType | UnsignedAlts),
4502  NEONMAP1(vqmovun_v, arm_neon_vqmovnsu, Add1ArgType),
4503  NEONMAP1(vqneg_v, arm_neon_vqneg, Add1ArgType),
4504  NEONMAP1(vqnegq_v, arm_neon_vqneg, Add1ArgType),
4505  NEONMAP1(vqrdmulh_v, arm_neon_vqrdmulh, Add1ArgType),
4506  NEONMAP1(vqrdmulhq_v, arm_neon_vqrdmulh, Add1ArgType),
4507  NEONMAP2(vqrshl_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts),
4508  NEONMAP2(vqrshlq_v, arm_neon_vqrshiftu, arm_neon_vqrshifts, Add1ArgType | UnsignedAlts),
4509  NEONMAP2(vqshl_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts),
4510  NEONMAP2(vqshl_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts),
4511  NEONMAP2(vqshlq_n_v, arm_neon_vqshiftu, arm_neon_vqshifts, UnsignedAlts),
4512  NEONMAP2(vqshlq_v, arm_neon_vqshiftu, arm_neon_vqshifts, Add1ArgType | UnsignedAlts),
4513  NEONMAP1(vqshlu_n_v, arm_neon_vqshiftsu, 0),
4514  NEONMAP1(vqshluq_n_v, arm_neon_vqshiftsu, 0),
4515  NEONMAP2(vqsub_v, arm_neon_vqsubu, arm_neon_vqsubs, Add1ArgType | UnsignedAlts),
4516  NEONMAP2(vqsubq_v, arm_neon_vqsubu, arm_neon_vqsubs, Add1ArgType | UnsignedAlts),
4517  NEONMAP1(vraddhn_v, arm_neon_vraddhn, Add1ArgType),
4518  NEONMAP2(vrecpe_v, arm_neon_vrecpe, arm_neon_vrecpe, 0),
4519  NEONMAP2(vrecpeq_v, arm_neon_vrecpe, arm_neon_vrecpe, 0),
4520  NEONMAP1(vrecps_v, arm_neon_vrecps, Add1ArgType),
4521  NEONMAP1(vrecpsq_v, arm_neon_vrecps, Add1ArgType),
4522  NEONMAP2(vrhadd_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts),
4523  NEONMAP2(vrhaddq_v, arm_neon_vrhaddu, arm_neon_vrhadds, Add1ArgType | UnsignedAlts),
4524  NEONMAP1(vrnd_v, arm_neon_vrintz, Add1ArgType),
4525  NEONMAP1(vrnda_v, arm_neon_vrinta, Add1ArgType),
4526  NEONMAP1(vrndaq_v, arm_neon_vrinta, Add1ArgType),
4527  NEONMAP0(vrndi_v),
4528  NEONMAP0(vrndiq_v),
4529  NEONMAP1(vrndm_v, arm_neon_vrintm, Add1ArgType),
4530  NEONMAP1(vrndmq_v, arm_neon_vrintm, Add1ArgType),
4531  NEONMAP1(vrndn_v, arm_neon_vrintn, Add1ArgType),
4532  NEONMAP1(vrndnq_v, arm_neon_vrintn, Add1ArgType),
4533  NEONMAP1(vrndp_v, arm_neon_vrintp, Add1ArgType),
4534  NEONMAP1(vrndpq_v, arm_neon_vrintp, Add1ArgType),
4535  NEONMAP1(vrndq_v, arm_neon_vrintz, Add1ArgType),
4536  NEONMAP1(vrndx_v, arm_neon_vrintx, Add1ArgType),
4537  NEONMAP1(vrndxq_v, arm_neon_vrintx, Add1ArgType),
4538  NEONMAP2(vrshl_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts),
4539  NEONMAP2(vrshlq_v, arm_neon_vrshiftu, arm_neon_vrshifts, Add1ArgType | UnsignedAlts),
4540  NEONMAP2(vrshr_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts),
4541  NEONMAP2(vrshrq_n_v, arm_neon_vrshiftu, arm_neon_vrshifts, UnsignedAlts),
4542  NEONMAP2(vrsqrte_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0),
4543  NEONMAP2(vrsqrteq_v, arm_neon_vrsqrte, arm_neon_vrsqrte, 0),
4544  NEONMAP1(vrsqrts_v, arm_neon_vrsqrts, Add1ArgType),
4545  NEONMAP1(vrsqrtsq_v, arm_neon_vrsqrts, Add1ArgType),
4546  NEONMAP1(vrsubhn_v, arm_neon_vrsubhn, Add1ArgType),
4547  NEONMAP1(vsha1su0q_v, arm_neon_sha1su0, 0),
4548  NEONMAP1(vsha1su1q_v, arm_neon_sha1su1, 0),
4549  NEONMAP1(vsha256h2q_v, arm_neon_sha256h2, 0),
4550  NEONMAP1(vsha256hq_v, arm_neon_sha256h, 0),
4551  NEONMAP1(vsha256su0q_v, arm_neon_sha256su0, 0),
4552  NEONMAP1(vsha256su1q_v, arm_neon_sha256su1, 0),
4553  NEONMAP0(vshl_n_v),
4554  NEONMAP2(vshl_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts),
4555  NEONMAP0(vshll_n_v),
4556  NEONMAP0(vshlq_n_v),
4557  NEONMAP2(vshlq_v, arm_neon_vshiftu, arm_neon_vshifts, Add1ArgType | UnsignedAlts),
4558  NEONMAP0(vshr_n_v),
4559  NEONMAP0(vshrn_n_v),
4560  NEONMAP0(vshrq_n_v),
4561  NEONMAP1(vst1_v, arm_neon_vst1, 0),
4562  NEONMAP1(vst1_x2_v, arm_neon_vst1x2, 0),
4563  NEONMAP1(vst1_x3_v, arm_neon_vst1x3, 0),
4564  NEONMAP1(vst1_x4_v, arm_neon_vst1x4, 0),
4565  NEONMAP1(vst1q_v, arm_neon_vst1, 0),
4566  NEONMAP1(vst1q_x2_v, arm_neon_vst1x2, 0),
4567  NEONMAP1(vst1q_x3_v, arm_neon_vst1x3, 0),
4568  NEONMAP1(vst1q_x4_v, arm_neon_vst1x4, 0),
4569  NEONMAP1(vst2_lane_v, arm_neon_vst2lane, 0),
4570  NEONMAP1(vst2_v, arm_neon_vst2, 0),
4571  NEONMAP1(vst2q_lane_v, arm_neon_vst2lane, 0),
4572  NEONMAP1(vst2q_v, arm_neon_vst2, 0),
4573  NEONMAP1(vst3_lane_v, arm_neon_vst3lane, 0),
4574  NEONMAP1(vst3_v, arm_neon_vst3, 0),
4575  NEONMAP1(vst3q_lane_v, arm_neon_vst3lane, 0),
4576  NEONMAP1(vst3q_v, arm_neon_vst3, 0),
4577  NEONMAP1(vst4_lane_v, arm_neon_vst4lane, 0),
4578  NEONMAP1(vst4_v, arm_neon_vst4, 0),
4579  NEONMAP1(vst4q_lane_v, arm_neon_vst4lane, 0),
4580  NEONMAP1(vst4q_v, arm_neon_vst4, 0),
4581  NEONMAP0(vsubhn_v),
4582  NEONMAP0(vtrn_v),
4583  NEONMAP0(vtrnq_v),
4584  NEONMAP0(vtst_v),
4585  NEONMAP0(vtstq_v),
4586  NEONMAP0(vuzp_v),
4587  NEONMAP0(vuzpq_v),
4588  NEONMAP0(vzip_v),
4589  NEONMAP0(vzipq_v)
4590};
4591
4592static const NeonIntrinsicInfo AArch64SIMDIntrinsicMap[] = {
4593  NEONMAP1(vabs_v, aarch64_neon_abs, 0),
4594  NEONMAP1(vabsq_v, aarch64_neon_abs, 0),
4595  NEONMAP0(vaddhn_v),
4596  NEONMAP1(vaesdq_v, aarch64_crypto_aesd, 0),
4597  NEONMAP1(vaeseq_v, aarch64_crypto_aese, 0),
4598  NEONMAP1(vaesimcq_v, aarch64_crypto_aesimc, 0),
4599  NEONMAP1(vaesmcq_v, aarch64_crypto_aesmc, 0),
4600  NEONMAP1(vcage_v, aarch64_neon_facge, 0),
4601  NEONMAP1(vcageq_v, aarch64_neon_facge, 0),
4602  NEONMAP1(vcagt_v, aarch64_neon_facgt, 0),
4603  NEONMAP1(vcagtq_v, aarch64_neon_facgt, 0),
4604  NEONMAP1(vcale_v, aarch64_neon_facge, 0),
4605  NEONMAP1(vcaleq_v, aarch64_neon_facge, 0),
4606  NEONMAP1(vcalt_v, aarch64_neon_facgt, 0),
4607  NEONMAP1(vcaltq_v, aarch64_neon_facgt, 0),
4608  NEONMAP0(vceqz_v),
4609  NEONMAP0(vceqzq_v),
4610  NEONMAP0(vcgez_v),
4611  NEONMAP0(vcgezq_v),
4612  NEONMAP0(vcgtz_v),
4613  NEONMAP0(vcgtzq_v),
4614  NEONMAP0(vclez_v),
4615  NEONMAP0(vclezq_v),
4616  NEONMAP1(vcls_v, aarch64_neon_cls, Add1ArgType),
4617  NEONMAP1(vclsq_v, aarch64_neon_cls, Add1ArgType),
4618  NEONMAP0(vcltz_v),
4619  NEONMAP0(vcltzq_v),
4620  NEONMAP1(vclz_v, ctlz, Add1ArgType),
4621  NEONMAP1(vclzq_v, ctlz, Add1ArgType),
4622  NEONMAP1(vcnt_v, ctpop, Add1ArgType),
4623  NEONMAP1(vcntq_v, ctpop, Add1ArgType),
4624  NEONMAP1(vcvt_f16_f32, aarch64_neon_vcvtfp2hf, 0),
4625  NEONMAP0(vcvt_f16_v),
4626  NEONMAP1(vcvt_f32_f16, aarch64_neon_vcvthf2fp, 0),
4627  NEONMAP0(vcvt_f32_v),
4628  NEONMAP2(vcvt_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4629  NEONMAP2(vcvt_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4630  NEONMAP2(vcvt_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4631  NEONMAP1(vcvt_n_s16_v, aarch64_neon_vcvtfp2fxs, 0),
4632  NEONMAP1(vcvt_n_s32_v, aarch64_neon_vcvtfp2fxs, 0),
4633  NEONMAP1(vcvt_n_s64_v, aarch64_neon_vcvtfp2fxs, 0),
4634  NEONMAP1(vcvt_n_u16_v, aarch64_neon_vcvtfp2fxu, 0),
4635  NEONMAP1(vcvt_n_u32_v, aarch64_neon_vcvtfp2fxu, 0),
4636  NEONMAP1(vcvt_n_u64_v, aarch64_neon_vcvtfp2fxu, 0),
4637  NEONMAP0(vcvtq_f16_v),
4638  NEONMAP0(vcvtq_f32_v),
4639  NEONMAP2(vcvtq_n_f16_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4640  NEONMAP2(vcvtq_n_f32_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4641  NEONMAP2(vcvtq_n_f64_v, aarch64_neon_vcvtfxu2fp, aarch64_neon_vcvtfxs2fp, 0),
4642  NEONMAP1(vcvtq_n_s16_v, aarch64_neon_vcvtfp2fxs, 0),
4643  NEONMAP1(vcvtq_n_s32_v, aarch64_neon_vcvtfp2fxs, 0),
4644  NEONMAP1(vcvtq_n_s64_v, aarch64_neon_vcvtfp2fxs, 0),
4645  NEONMAP1(vcvtq_n_u16_v, aarch64_neon_vcvtfp2fxu, 0),
4646  NEONMAP1(vcvtq_n_u32_v, aarch64_neon_vcvtfp2fxu, 0),
4647  NEONMAP1(vcvtq_n_u64_v, aarch64_neon_vcvtfp2fxu, 0),
4648  NEONMAP1(vcvtx_f32_v, aarch64_neon_fcvtxn, AddRetType | Add1ArgType),
4649  NEONMAP2(vdot_v, aarch64_neon_udot, aarch64_neon_sdot, 0),
4650  NEONMAP2(vdotq_v, aarch64_neon_udot, aarch64_neon_sdot, 0),
4651  NEONMAP0(vext_v),
4652  NEONMAP0(vextq_v),
4653  NEONMAP0(vfma_v),
4654  NEONMAP0(vfmaq_v),
4655  NEONMAP1(vfmlal_high_v, aarch64_neon_fmlal2, 0),
4656  NEONMAP1(vfmlal_low_v, aarch64_neon_fmlal, 0),
4657  NEONMAP1(vfmlalq_high_v, aarch64_neon_fmlal2, 0),
4658  NEONMAP1(vfmlalq_low_v, aarch64_neon_fmlal, 0),
4659  NEONMAP1(vfmlsl_high_v, aarch64_neon_fmlsl2, 0),
4660  NEONMAP1(vfmlsl_low_v, aarch64_neon_fmlsl, 0),
4661  NEONMAP1(vfmlslq_high_v, aarch64_neon_fmlsl2, 0),
4662  NEONMAP1(vfmlslq_low_v, aarch64_neon_fmlsl, 0),
4663  NEONMAP2(vhadd_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts),
4664  NEONMAP2(vhaddq_v, aarch64_neon_uhadd, aarch64_neon_shadd, Add1ArgType | UnsignedAlts),
4665  NEONMAP2(vhsub_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts),
4666  NEONMAP2(vhsubq_v, aarch64_neon_uhsub, aarch64_neon_shsub, Add1ArgType | UnsignedAlts),
4667  NEONMAP1(vld1_x2_v, aarch64_neon_ld1x2, 0),
4668  NEONMAP1(vld1_x3_v, aarch64_neon_ld1x3, 0),
4669  NEONMAP1(vld1_x4_v, aarch64_neon_ld1x4, 0),
4670  NEONMAP1(vld1q_x2_v, aarch64_neon_ld1x2, 0),
4671  NEONMAP1(vld1q_x3_v, aarch64_neon_ld1x3, 0),
4672  NEONMAP1(vld1q_x4_v, aarch64_neon_ld1x4, 0),
4673  NEONMAP0(vmovl_v),
4674  NEONMAP0(vmovn_v),
4675  NEONMAP1(vmul_v, aarch64_neon_pmul, Add1ArgType),
4676  NEONMAP1(vmulq_v, aarch64_neon_pmul, Add1ArgType),
4677  NEONMAP1(vpadd_v, aarch64_neon_addp, Add1ArgType),
4678  NEONMAP2(vpaddl_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts),
4679  NEONMAP2(vpaddlq_v, aarch64_neon_uaddlp, aarch64_neon_saddlp, UnsignedAlts),
4680  NEONMAP1(vpaddq_v, aarch64_neon_addp, Add1ArgType),
4681  NEONMAP1(vqabs_v, aarch64_neon_sqabs, Add1ArgType),
4682  NEONMAP1(vqabsq_v, aarch64_neon_sqabs, Add1ArgType),
4683  NEONMAP2(vqadd_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts),
4684  NEONMAP2(vqaddq_v, aarch64_neon_uqadd, aarch64_neon_sqadd, Add1ArgType | UnsignedAlts),
4685  NEONMAP2(vqdmlal_v, aarch64_neon_sqdmull, aarch64_neon_sqadd, 0),
4686  NEONMAP2(vqdmlsl_v, aarch64_neon_sqdmull, aarch64_neon_sqsub, 0),
4687  NEONMAP1(vqdmulh_v, aarch64_neon_sqdmulh, Add1ArgType),
4688  NEONMAP1(vqdmulhq_v, aarch64_neon_sqdmulh, Add1ArgType),
4689  NEONMAP1(vqdmull_v, aarch64_neon_sqdmull, Add1ArgType),
4690  NEONMAP2(vqmovn_v, aarch64_neon_uqxtn, aarch64_neon_sqxtn, Add1ArgType | UnsignedAlts),
4691  NEONMAP1(vqmovun_v, aarch64_neon_sqxtun, Add1ArgType),
4692  NEONMAP1(vqneg_v, aarch64_neon_sqneg, Add1ArgType),
4693  NEONMAP1(vqnegq_v, aarch64_neon_sqneg, Add1ArgType),
4694  NEONMAP1(vqrdmulh_v, aarch64_neon_sqrdmulh, Add1ArgType),
4695  NEONMAP1(vqrdmulhq_v, aarch64_neon_sqrdmulh, Add1ArgType),
4696  NEONMAP2(vqrshl_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts),
4697  NEONMAP2(vqrshlq_v, aarch64_neon_uqrshl, aarch64_neon_sqrshl, Add1ArgType | UnsignedAlts),
4698  NEONMAP2(vqshl_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl, UnsignedAlts),
4699  NEONMAP2(vqshl_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts),
4700  NEONMAP2(vqshlq_n_v, aarch64_neon_uqshl, aarch64_neon_sqshl,UnsignedAlts),
4701  NEONMAP2(vqshlq_v, aarch64_neon_uqshl, aarch64_neon_sqshl, Add1ArgType | UnsignedAlts),
4702  NEONMAP1(vqshlu_n_v, aarch64_neon_sqshlu, 0),
4703  NEONMAP1(vqshluq_n_v, aarch64_neon_sqshlu, 0),
4704  NEONMAP2(vqsub_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts),
4705  NEONMAP2(vqsubq_v, aarch64_neon_uqsub, aarch64_neon_sqsub, Add1ArgType | UnsignedAlts),
4706  NEONMAP1(vraddhn_v, aarch64_neon_raddhn, Add1ArgType),
4707  NEONMAP2(vrecpe_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0),
4708  NEONMAP2(vrecpeq_v, aarch64_neon_frecpe, aarch64_neon_urecpe, 0),
4709  NEONMAP1(vrecps_v, aarch64_neon_frecps, Add1ArgType),
4710  NEONMAP1(vrecpsq_v, aarch64_neon_frecps, Add1ArgType),
4711  NEONMAP2(vrhadd_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts),
4712  NEONMAP2(vrhaddq_v, aarch64_neon_urhadd, aarch64_neon_srhadd, Add1ArgType | UnsignedAlts),
4713  NEONMAP0(vrndi_v),
4714  NEONMAP0(vrndiq_v),
4715  NEONMAP2(vrshl_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts),
4716  NEONMAP2(vrshlq_v, aarch64_neon_urshl, aarch64_neon_srshl, Add1ArgType | UnsignedAlts),
4717  NEONMAP2(vrshr_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts),
4718  NEONMAP2(vrshrq_n_v, aarch64_neon_urshl, aarch64_neon_srshl, UnsignedAlts),
4719  NEONMAP2(vrsqrte_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0),
4720  NEONMAP2(vrsqrteq_v, aarch64_neon_frsqrte, aarch64_neon_ursqrte, 0),
4721  NEONMAP1(vrsqrts_v, aarch64_neon_frsqrts, Add1ArgType),
4722  NEONMAP1(vrsqrtsq_v, aarch64_neon_frsqrts, Add1ArgType),
4723  NEONMAP1(vrsubhn_v, aarch64_neon_rsubhn, Add1ArgType),
4724  NEONMAP1(vsha1su0q_v, aarch64_crypto_sha1su0, 0),
4725  NEONMAP1(vsha1su1q_v, aarch64_crypto_sha1su1, 0),
4726  NEONMAP1(vsha256h2q_v, aarch64_crypto_sha256h2, 0),
4727  NEONMAP1(vsha256hq_v, aarch64_crypto_sha256h, 0),
4728  NEONMAP1(vsha256su0q_v, aarch64_crypto_sha256su0, 0),
4729  NEONMAP1(vsha256su1q_v, aarch64_crypto_sha256su1, 0),
4730  NEONMAP0(vshl_n_v),
4731  NEONMAP2(vshl_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts),
4732  NEONMAP0(vshll_n_v),
4733  NEONMAP0(vshlq_n_v),
4734  NEONMAP2(vshlq_v, aarch64_neon_ushl, aarch64_neon_sshl, Add1ArgType | UnsignedAlts),
4735  NEONMAP0(vshr_n_v),
4736  NEONMAP0(vshrn_n_v),
4737  NEONMAP0(vshrq_n_v),
4738  NEONMAP1(vst1_x2_v, aarch64_neon_st1x2, 0),
4739  NEONMAP1(vst1_x3_v, aarch64_neon_st1x3, 0),
4740  NEONMAP1(vst1_x4_v, aarch64_neon_st1x4, 0),
4741  NEONMAP1(vst1q_x2_v, aarch64_neon_st1x2, 0),
4742  NEONMAP1(vst1q_x3_v, aarch64_neon_st1x3, 0),
4743  NEONMAP1(vst1q_x4_v, aarch64_neon_st1x4, 0),
4744  NEONMAP0(vsubhn_v),
4745  NEONMAP0(vtst_v),
4746  NEONMAP0(vtstq_v),
4747};
4748
4749static const NeonIntrinsicInfo AArch64SISDIntrinsicMap[] = {
4750  NEONMAP1(vabdd_f64, aarch64_sisd_fabd, Add1ArgType),
4751  NEONMAP1(vabds_f32, aarch64_sisd_fabd, Add1ArgType),
4752  NEONMAP1(vabsd_s64, aarch64_neon_abs, Add1ArgType),
4753  NEONMAP1(vaddlv_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType),
4754  NEONMAP1(vaddlv_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType),
4755  NEONMAP1(vaddlvq_s32, aarch64_neon_saddlv, AddRetType | Add1ArgType),
4756  NEONMAP1(vaddlvq_u32, aarch64_neon_uaddlv, AddRetType | Add1ArgType),
4757  NEONMAP1(vaddv_f32, aarch64_neon_faddv, AddRetType | Add1ArgType),
4758  NEONMAP1(vaddv_s32, aarch64_neon_saddv, AddRetType | Add1ArgType),
4759  NEONMAP1(vaddv_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4760  NEONMAP1(vaddvq_f32, aarch64_neon_faddv, AddRetType | Add1ArgType),
4761  NEONMAP1(vaddvq_f64, aarch64_neon_faddv, AddRetType | Add1ArgType),
4762  NEONMAP1(vaddvq_s32, aarch64_neon_saddv, AddRetType | Add1ArgType),
4763  NEONMAP1(vaddvq_s64, aarch64_neon_saddv, AddRetType | Add1ArgType),
4764  NEONMAP1(vaddvq_u32, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4765  NEONMAP1(vaddvq_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4766  NEONMAP1(vcaged_f64, aarch64_neon_facge, AddRetType | Add1ArgType),
4767  NEONMAP1(vcages_f32, aarch64_neon_facge, AddRetType | Add1ArgType),
4768  NEONMAP1(vcagtd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType),
4769  NEONMAP1(vcagts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType),
4770  NEONMAP1(vcaled_f64, aarch64_neon_facge, AddRetType | Add1ArgType),
4771  NEONMAP1(vcales_f32, aarch64_neon_facge, AddRetType | Add1ArgType),
4772  NEONMAP1(vcaltd_f64, aarch64_neon_facgt, AddRetType | Add1ArgType),
4773  NEONMAP1(vcalts_f32, aarch64_neon_facgt, AddRetType | Add1ArgType),
4774  NEONMAP1(vcvtad_s64_f64, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
4775  NEONMAP1(vcvtad_u64_f64, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
4776  NEONMAP1(vcvtas_s32_f32, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
4777  NEONMAP1(vcvtas_u32_f32, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
4778  NEONMAP1(vcvtd_n_f64_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
4779  NEONMAP1(vcvtd_n_f64_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
4780  NEONMAP1(vcvtd_n_s64_f64, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
4781  NEONMAP1(vcvtd_n_u64_f64, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
4782  NEONMAP1(vcvtmd_s64_f64, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
4783  NEONMAP1(vcvtmd_u64_f64, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
4784  NEONMAP1(vcvtms_s32_f32, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
4785  NEONMAP1(vcvtms_u32_f32, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
4786  NEONMAP1(vcvtnd_s64_f64, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
4787  NEONMAP1(vcvtnd_u64_f64, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
4788  NEONMAP1(vcvtns_s32_f32, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
4789  NEONMAP1(vcvtns_u32_f32, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
4790  NEONMAP1(vcvtpd_s64_f64, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
4791  NEONMAP1(vcvtpd_u64_f64, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
4792  NEONMAP1(vcvtps_s32_f32, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
4793  NEONMAP1(vcvtps_u32_f32, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
4794  NEONMAP1(vcvts_n_f32_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
4795  NEONMAP1(vcvts_n_f32_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
4796  NEONMAP1(vcvts_n_s32_f32, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
4797  NEONMAP1(vcvts_n_u32_f32, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
4798  NEONMAP1(vcvtxd_f32_f64, aarch64_sisd_fcvtxn, 0),
4799  NEONMAP1(vmaxnmv_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4800  NEONMAP1(vmaxnmvq_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4801  NEONMAP1(vmaxnmvq_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4802  NEONMAP1(vmaxv_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4803  NEONMAP1(vmaxv_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType),
4804  NEONMAP1(vmaxv_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType),
4805  NEONMAP1(vmaxvq_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4806  NEONMAP1(vmaxvq_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4807  NEONMAP1(vmaxvq_s32, aarch64_neon_smaxv, AddRetType | Add1ArgType),
4808  NEONMAP1(vmaxvq_u32, aarch64_neon_umaxv, AddRetType | Add1ArgType),
4809  NEONMAP1(vminnmv_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4810  NEONMAP1(vminnmvq_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4811  NEONMAP1(vminnmvq_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4812  NEONMAP1(vminv_f32, aarch64_neon_fminv, AddRetType | Add1ArgType),
4813  NEONMAP1(vminv_s32, aarch64_neon_sminv, AddRetType | Add1ArgType),
4814  NEONMAP1(vminv_u32, aarch64_neon_uminv, AddRetType | Add1ArgType),
4815  NEONMAP1(vminvq_f32, aarch64_neon_fminv, AddRetType | Add1ArgType),
4816  NEONMAP1(vminvq_f64, aarch64_neon_fminv, AddRetType | Add1ArgType),
4817  NEONMAP1(vminvq_s32, aarch64_neon_sminv, AddRetType | Add1ArgType),
4818  NEONMAP1(vminvq_u32, aarch64_neon_uminv, AddRetType | Add1ArgType),
4819  NEONMAP1(vmull_p64, aarch64_neon_pmull64, 0),
4820  NEONMAP1(vmulxd_f64, aarch64_neon_fmulx, Add1ArgType),
4821  NEONMAP1(vmulxs_f32, aarch64_neon_fmulx, Add1ArgType),
4822  NEONMAP1(vpaddd_s64, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4823  NEONMAP1(vpaddd_u64, aarch64_neon_uaddv, AddRetType | Add1ArgType),
4824  NEONMAP1(vpmaxnmqd_f64, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4825  NEONMAP1(vpmaxnms_f32, aarch64_neon_fmaxnmv, AddRetType | Add1ArgType),
4826  NEONMAP1(vpmaxqd_f64, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4827  NEONMAP1(vpmaxs_f32, aarch64_neon_fmaxv, AddRetType | Add1ArgType),
4828  NEONMAP1(vpminnmqd_f64, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4829  NEONMAP1(vpminnms_f32, aarch64_neon_fminnmv, AddRetType | Add1ArgType),
4830  NEONMAP1(vpminqd_f64, aarch64_neon_fminv, AddRetType | Add1ArgType),
4831  NEONMAP1(vpmins_f32, aarch64_neon_fminv, AddRetType | Add1ArgType),
4832  NEONMAP1(vqabsb_s8, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors),
4833  NEONMAP1(vqabsd_s64, aarch64_neon_sqabs, Add1ArgType),
4834  NEONMAP1(vqabsh_s16, aarch64_neon_sqabs, Vectorize1ArgType | Use64BitVectors),
4835  NEONMAP1(vqabss_s32, aarch64_neon_sqabs, Add1ArgType),
4836  NEONMAP1(vqaddb_s8, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors),
4837  NEONMAP1(vqaddb_u8, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors),
4838  NEONMAP1(vqaddd_s64, aarch64_neon_sqadd, Add1ArgType),
4839  NEONMAP1(vqaddd_u64, aarch64_neon_uqadd, Add1ArgType),
4840  NEONMAP1(vqaddh_s16, aarch64_neon_sqadd, Vectorize1ArgType | Use64BitVectors),
4841  NEONMAP1(vqaddh_u16, aarch64_neon_uqadd, Vectorize1ArgType | Use64BitVectors),
4842  NEONMAP1(vqadds_s32, aarch64_neon_sqadd, Add1ArgType),
4843  NEONMAP1(vqadds_u32, aarch64_neon_uqadd, Add1ArgType),
4844  NEONMAP1(vqdmulhh_s16, aarch64_neon_sqdmulh, Vectorize1ArgType | Use64BitVectors),
4845  NEONMAP1(vqdmulhs_s32, aarch64_neon_sqdmulh, Add1ArgType),
4846  NEONMAP1(vqdmullh_s16, aarch64_neon_sqdmull, VectorRet | Use128BitVectors),
4847  NEONMAP1(vqdmulls_s32, aarch64_neon_sqdmulls_scalar, 0),
4848  NEONMAP1(vqmovnd_s64, aarch64_neon_scalar_sqxtn, AddRetType | Add1ArgType),
4849  NEONMAP1(vqmovnd_u64, aarch64_neon_scalar_uqxtn, AddRetType | Add1ArgType),
4850  NEONMAP1(vqmovnh_s16, aarch64_neon_sqxtn, VectorRet | Use64BitVectors),
4851  NEONMAP1(vqmovnh_u16, aarch64_neon_uqxtn, VectorRet | Use64BitVectors),
4852  NEONMAP1(vqmovns_s32, aarch64_neon_sqxtn, VectorRet | Use64BitVectors),
4853  NEONMAP1(vqmovns_u32, aarch64_neon_uqxtn, VectorRet | Use64BitVectors),
4854  NEONMAP1(vqmovund_s64, aarch64_neon_scalar_sqxtun, AddRetType | Add1ArgType),
4855  NEONMAP1(vqmovunh_s16, aarch64_neon_sqxtun, VectorRet | Use64BitVectors),
4856  NEONMAP1(vqmovuns_s32, aarch64_neon_sqxtun, VectorRet | Use64BitVectors),
4857  NEONMAP1(vqnegb_s8, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors),
4858  NEONMAP1(vqnegd_s64, aarch64_neon_sqneg, Add1ArgType),
4859  NEONMAP1(vqnegh_s16, aarch64_neon_sqneg, Vectorize1ArgType | Use64BitVectors),
4860  NEONMAP1(vqnegs_s32, aarch64_neon_sqneg, Add1ArgType),
4861  NEONMAP1(vqrdmulhh_s16, aarch64_neon_sqrdmulh, Vectorize1ArgType | Use64BitVectors),
4862  NEONMAP1(vqrdmulhs_s32, aarch64_neon_sqrdmulh, Add1ArgType),
4863  NEONMAP1(vqrshlb_s8, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors),
4864  NEONMAP1(vqrshlb_u8, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors),
4865  NEONMAP1(vqrshld_s64, aarch64_neon_sqrshl, Add1ArgType),
4866  NEONMAP1(vqrshld_u64, aarch64_neon_uqrshl, Add1ArgType),
4867  NEONMAP1(vqrshlh_s16, aarch64_neon_sqrshl, Vectorize1ArgType | Use64BitVectors),
4868  NEONMAP1(vqrshlh_u16, aarch64_neon_uqrshl, Vectorize1ArgType | Use64BitVectors),
4869  NEONMAP1(vqrshls_s32, aarch64_neon_sqrshl, Add1ArgType),
4870  NEONMAP1(vqrshls_u32, aarch64_neon_uqrshl, Add1ArgType),
4871  NEONMAP1(vqrshrnd_n_s64, aarch64_neon_sqrshrn, AddRetType),
4872  NEONMAP1(vqrshrnd_n_u64, aarch64_neon_uqrshrn, AddRetType),
4873  NEONMAP1(vqrshrnh_n_s16, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors),
4874  NEONMAP1(vqrshrnh_n_u16, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors),
4875  NEONMAP1(vqrshrns_n_s32, aarch64_neon_sqrshrn, VectorRet | Use64BitVectors),
4876  NEONMAP1(vqrshrns_n_u32, aarch64_neon_uqrshrn, VectorRet | Use64BitVectors),
4877  NEONMAP1(vqrshrund_n_s64, aarch64_neon_sqrshrun, AddRetType),
4878  NEONMAP1(vqrshrunh_n_s16, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors),
4879  NEONMAP1(vqrshruns_n_s32, aarch64_neon_sqrshrun, VectorRet | Use64BitVectors),
4880  NEONMAP1(vqshlb_n_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
4881  NEONMAP1(vqshlb_n_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
4882  NEONMAP1(vqshlb_s8, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
4883  NEONMAP1(vqshlb_u8, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
4884  NEONMAP1(vqshld_s64, aarch64_neon_sqshl, Add1ArgType),
4885  NEONMAP1(vqshld_u64, aarch64_neon_uqshl, Add1ArgType),
4886  NEONMAP1(vqshlh_n_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
4887  NEONMAP1(vqshlh_n_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
4888  NEONMAP1(vqshlh_s16, aarch64_neon_sqshl, Vectorize1ArgType | Use64BitVectors),
4889  NEONMAP1(vqshlh_u16, aarch64_neon_uqshl, Vectorize1ArgType | Use64BitVectors),
4890  NEONMAP1(vqshls_n_s32, aarch64_neon_sqshl, Add1ArgType),
4891  NEONMAP1(vqshls_n_u32, aarch64_neon_uqshl, Add1ArgType),
4892  NEONMAP1(vqshls_s32, aarch64_neon_sqshl, Add1ArgType),
4893  NEONMAP1(vqshls_u32, aarch64_neon_uqshl, Add1ArgType),
4894  NEONMAP1(vqshlub_n_s8, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors),
4895  NEONMAP1(vqshluh_n_s16, aarch64_neon_sqshlu, Vectorize1ArgType | Use64BitVectors),
4896  NEONMAP1(vqshlus_n_s32, aarch64_neon_sqshlu, Add1ArgType),
4897  NEONMAP1(vqshrnd_n_s64, aarch64_neon_sqshrn, AddRetType),
4898  NEONMAP1(vqshrnd_n_u64, aarch64_neon_uqshrn, AddRetType),
4899  NEONMAP1(vqshrnh_n_s16, aarch64_neon_sqshrn, VectorRet | Use64BitVectors),
4900  NEONMAP1(vqshrnh_n_u16, aarch64_neon_uqshrn, VectorRet | Use64BitVectors),
4901  NEONMAP1(vqshrns_n_s32, aarch64_neon_sqshrn, VectorRet | Use64BitVectors),
4902  NEONMAP1(vqshrns_n_u32, aarch64_neon_uqshrn, VectorRet | Use64BitVectors),
4903  NEONMAP1(vqshrund_n_s64, aarch64_neon_sqshrun, AddRetType),
4904  NEONMAP1(vqshrunh_n_s16, aarch64_neon_sqshrun, VectorRet | Use64BitVectors),
4905  NEONMAP1(vqshruns_n_s32, aarch64_neon_sqshrun, VectorRet | Use64BitVectors),
4906  NEONMAP1(vqsubb_s8, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors),
4907  NEONMAP1(vqsubb_u8, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors),
4908  NEONMAP1(vqsubd_s64, aarch64_neon_sqsub, Add1ArgType),
4909  NEONMAP1(vqsubd_u64, aarch64_neon_uqsub, Add1ArgType),
4910  NEONMAP1(vqsubh_s16, aarch64_neon_sqsub, Vectorize1ArgType | Use64BitVectors),
4911  NEONMAP1(vqsubh_u16, aarch64_neon_uqsub, Vectorize1ArgType | Use64BitVectors),
4912  NEONMAP1(vqsubs_s32, aarch64_neon_sqsub, Add1ArgType),
4913  NEONMAP1(vqsubs_u32, aarch64_neon_uqsub, Add1ArgType),
4914  NEONMAP1(vrecped_f64, aarch64_neon_frecpe, Add1ArgType),
4915  NEONMAP1(vrecpes_f32, aarch64_neon_frecpe, Add1ArgType),
4916  NEONMAP1(vrecpxd_f64, aarch64_neon_frecpx, Add1ArgType),
4917  NEONMAP1(vrecpxs_f32, aarch64_neon_frecpx, Add1ArgType),
4918  NEONMAP1(vrshld_s64, aarch64_neon_srshl, Add1ArgType),
4919  NEONMAP1(vrshld_u64, aarch64_neon_urshl, Add1ArgType),
4920  NEONMAP1(vrsqrted_f64, aarch64_neon_frsqrte, Add1ArgType),
4921  NEONMAP1(vrsqrtes_f32, aarch64_neon_frsqrte, Add1ArgType),
4922  NEONMAP1(vrsqrtsd_f64, aarch64_neon_frsqrts, Add1ArgType),
4923  NEONMAP1(vrsqrtss_f32, aarch64_neon_frsqrts, Add1ArgType),
4924  NEONMAP1(vsha1cq_u32, aarch64_crypto_sha1c, 0),
4925  NEONMAP1(vsha1h_u32, aarch64_crypto_sha1h, 0),
4926  NEONMAP1(vsha1mq_u32, aarch64_crypto_sha1m, 0),
4927  NEONMAP1(vsha1pq_u32, aarch64_crypto_sha1p, 0),
4928  NEONMAP1(vshld_s64, aarch64_neon_sshl, Add1ArgType),
4929  NEONMAP1(vshld_u64, aarch64_neon_ushl, Add1ArgType),
4930  NEONMAP1(vslid_n_s64, aarch64_neon_vsli, Vectorize1ArgType),
4931  NEONMAP1(vslid_n_u64, aarch64_neon_vsli, Vectorize1ArgType),
4932  NEONMAP1(vsqaddb_u8, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors),
4933  NEONMAP1(vsqaddd_u64, aarch64_neon_usqadd, Add1ArgType),
4934  NEONMAP1(vsqaddh_u16, aarch64_neon_usqadd, Vectorize1ArgType | Use64BitVectors),
4935  NEONMAP1(vsqadds_u32, aarch64_neon_usqadd, Add1ArgType),
4936  NEONMAP1(vsrid_n_s64, aarch64_neon_vsri, Vectorize1ArgType),
4937  NEONMAP1(vsrid_n_u64, aarch64_neon_vsri, Vectorize1ArgType),
4938  NEONMAP1(vuqaddb_s8, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors),
4939  NEONMAP1(vuqaddd_s64, aarch64_neon_suqadd, Add1ArgType),
4940  NEONMAP1(vuqaddh_s16, aarch64_neon_suqadd, Vectorize1ArgType | Use64BitVectors),
4941  NEONMAP1(vuqadds_s32, aarch64_neon_suqadd, Add1ArgType),
4942  // FP16 scalar intrinisics go here.
4943  NEONMAP1(vabdh_f16, aarch64_sisd_fabd, Add1ArgType),
4944  NEONMAP1(vcvtah_s32_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
4945  NEONMAP1(vcvtah_s64_f16, aarch64_neon_fcvtas, AddRetType | Add1ArgType),
4946  NEONMAP1(vcvtah_u32_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
4947  NEONMAP1(vcvtah_u64_f16, aarch64_neon_fcvtau, AddRetType | Add1ArgType),
4948  NEONMAP1(vcvth_n_f16_s32, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
4949  NEONMAP1(vcvth_n_f16_s64, aarch64_neon_vcvtfxs2fp, AddRetType | Add1ArgType),
4950  NEONMAP1(vcvth_n_f16_u32, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
4951  NEONMAP1(vcvth_n_f16_u64, aarch64_neon_vcvtfxu2fp, AddRetType | Add1ArgType),
4952  NEONMAP1(vcvth_n_s32_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
4953  NEONMAP1(vcvth_n_s64_f16, aarch64_neon_vcvtfp2fxs, AddRetType | Add1ArgType),
4954  NEONMAP1(vcvth_n_u32_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
4955  NEONMAP1(vcvth_n_u64_f16, aarch64_neon_vcvtfp2fxu, AddRetType | Add1ArgType),
4956  NEONMAP1(vcvtmh_s32_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
4957  NEONMAP1(vcvtmh_s64_f16, aarch64_neon_fcvtms, AddRetType | Add1ArgType),
4958  NEONMAP1(vcvtmh_u32_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
4959  NEONMAP1(vcvtmh_u64_f16, aarch64_neon_fcvtmu, AddRetType | Add1ArgType),
4960  NEONMAP1(vcvtnh_s32_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
4961  NEONMAP1(vcvtnh_s64_f16, aarch64_neon_fcvtns, AddRetType | Add1ArgType),
4962  NEONMAP1(vcvtnh_u32_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
4963  NEONMAP1(vcvtnh_u64_f16, aarch64_neon_fcvtnu, AddRetType | Add1ArgType),
4964  NEONMAP1(vcvtph_s32_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
4965  NEONMAP1(vcvtph_s64_f16, aarch64_neon_fcvtps, AddRetType | Add1ArgType),
4966  NEONMAP1(vcvtph_u32_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
4967  NEONMAP1(vcvtph_u64_f16, aarch64_neon_fcvtpu, AddRetType | Add1ArgType),
4968  NEONMAP1(vmulxh_f16, aarch64_neon_fmulx, Add1ArgType),
4969  NEONMAP1(vrecpeh_f16, aarch64_neon_frecpe, Add1ArgType),
4970  NEONMAP1(vrecpxh_f16, aarch64_neon_frecpx, Add1ArgType),
4971  NEONMAP1(vrsqrteh_f16, aarch64_neon_frsqrte, Add1ArgType),
4972  NEONMAP1(vrsqrtsh_f16, aarch64_neon_frsqrts, Add1ArgType),
4973};
4974
4975#undef NEONMAP0
4976#undef NEONMAP1
4977#undef NEONMAP2
4978
4979static bool NEONSIMDIntrinsicsProvenSorted = false;
4980
4981static bool AArch64SIMDIntrinsicsProvenSorted = false;
4982static bool AArch64SISDIntrinsicsProvenSorted = false;
4983
4984
4985static const NeonIntrinsicInfo *
4986findNeonIntrinsicInMap(ArrayRef<NeonIntrinsicInfo> IntrinsicMap,
4987                       unsigned BuiltinID, bool &MapProvenSorted) {
4988
4989#ifndef NDEBUG
4990  if (!MapProvenSorted) {
4991    assert(std::is_sorted(std::begin(IntrinsicMap), std::end(IntrinsicMap)));
4992    MapProvenSorted = true;
4993  }
4994#endif
4995
4996  const NeonIntrinsicInfo *Builtin =
4997      std::lower_bound(IntrinsicMap.begin(), IntrinsicMap.end(), BuiltinID);
4998
4999  if (Builtin != IntrinsicMap.end() && Builtin->BuiltinID == BuiltinID)
5000    return Builtin;
5001
5002  return nullptr;
5003}
5004
5005Function *CodeGenFunction::LookupNeonLLVMIntrinsic(unsigned IntrinsicID,
5006                                                   unsigned Modifier,
5007                                                   llvm::Type *ArgType,
5008                                                   const CallExpr *E) {
5009  int VectorSize = 0;
5010  if (Modifier & Use64BitVectors)
5011    VectorSize = 64;
5012  else if (Modifier & Use128BitVectors)
5013    VectorSize = 128;
5014
5015  // Return type.
5016  SmallVector<llvm::Type *, 3> Tys;
5017  if (Modifier & AddRetType) {
5018    llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext()));
5019    if (Modifier & VectorizeRetType)
5020      Ty = llvm::VectorType::get(
5021          Ty, VectorSize ? VectorSize / Ty->getPrimitiveSizeInBits() : 1);
5022
5023    Tys.push_back(Ty);
5024  }
5025
5026  // Arguments.
5027  if (Modifier & VectorizeArgTypes) {
5028    int Elts = VectorSize ? VectorSize / ArgType->getPrimitiveSizeInBits() : 1;
5029    ArgType = llvm::VectorType::get(ArgType, Elts);
5030  }
5031
5032  if (Modifier & (Add1ArgType | Add2ArgTypes))
5033    Tys.push_back(ArgType);
5034
5035  if (Modifier & Add2ArgTypes)
5036    Tys.push_back(ArgType);
5037
5038  if (Modifier & InventFloatType)
5039    Tys.push_back(FloatTy);
5040
5041  return CGM.getIntrinsic(IntrinsicID, Tys);
5042}
5043
5044static Value *EmitCommonNeonSISDBuiltinExpr(CodeGenFunction &CGF,
5045                                            const NeonIntrinsicInfo &SISDInfo,
5046                                            SmallVectorImpl<Value *> &Ops,
5047                                            const CallExpr *E) {
5048  unsigned BuiltinID = SISDInfo.BuiltinID;
5049  unsigned int Int = SISDInfo.LLVMIntrinsic;
5050  unsigned Modifier = SISDInfo.TypeModifier;
5051  const char *s = SISDInfo.NameHint;
5052
5053  switch (BuiltinID) {
5054  case NEON::BI__builtin_neon_vcled_s64:
5055  case NEON::BI__builtin_neon_vcled_u64:
5056  case NEON::BI__builtin_neon_vcles_f32:
5057  case NEON::BI__builtin_neon_vcled_f64:
5058  case NEON::BI__builtin_neon_vcltd_s64:
5059  case NEON::BI__builtin_neon_vcltd_u64:
5060  case NEON::BI__builtin_neon_vclts_f32:
5061  case NEON::BI__builtin_neon_vcltd_f64:
5062  case NEON::BI__builtin_neon_vcales_f32:
5063  case NEON::BI__builtin_neon_vcaled_f64:
5064  case NEON::BI__builtin_neon_vcalts_f32:
5065  case NEON::BI__builtin_neon_vcaltd_f64:
5066    // Only one direction of comparisons actually exist, cmle is actually a cmge
5067    // with swapped operands. The table gives us the right intrinsic but we
5068    // still need to do the swap.
5069    std::swap(Ops[0], Ops[1]);
5070    break;
5071  }
5072
5073   (0) . __assert_fail ("Int && \"Generic code assumes a valid intrinsic\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 5073, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Int && "Generic code assumes a valid intrinsic");
5074
5075  // Determine the type(s) of this overloaded AArch64 intrinsic.
5076  const Expr *Arg = E->getArg(0);
5077  llvm::Type *ArgTy = CGF.ConvertType(Arg->getType());
5078  Function *F = CGF.LookupNeonLLVMIntrinsic(Int, Modifier, ArgTy, E);
5079
5080  int j = 0;
5081  ConstantInt *C0 = ConstantInt::get(CGF.SizeTy, 0);
5082  for (Function::const_arg_iterator ai = F->arg_begin(), ae = F->arg_end();
5083       ai != ae; ++ai, ++j) {
5084    llvm::Type *ArgTy = ai->getType();
5085    if (Ops[j]->getType()->getPrimitiveSizeInBits() ==
5086             ArgTy->getPrimitiveSizeInBits())
5087      continue;
5088
5089    isVectorTy() && !Ops[j]->getType()->isVectorTy()", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 5089, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(ArgTy->isVectorTy() && !Ops[j]->getType()->isVectorTy());
5090    // The constant argument to an _n_ intrinsic always has Int32Ty, so truncate
5091    // it before inserting.
5092    Ops[j] =
5093        CGF.Builder.CreateTruncOrBitCast(Ops[j], ArgTy->getVectorElementType());
5094    Ops[j] =
5095        CGF.Builder.CreateInsertElement(UndefValue::get(ArgTy), Ops[j], C0);
5096  }
5097
5098  Value *Result = CGF.EmitNeonCall(F, Ops, s);
5099  llvm::Type *ResultType = CGF.ConvertType(E->getType());
5100  if (ResultType->getPrimitiveSizeInBits() <
5101      Result->getType()->getPrimitiveSizeInBits())
5102    return CGF.Builder.CreateExtractElement(Result, C0);
5103
5104  return CGF.Builder.CreateBitCast(Result, ResultType, s);
5105}
5106
5107Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
5108    unsigned BuiltinID, unsigned LLVMIntrinsic, unsigned AltLLVMIntrinsic,
5109    const char *NameHint, unsigned Modifier, const CallExpr *E,
5110    SmallVectorImpl<llvm::Value *> &Ops, Address PtrOp0, Address PtrOp1,
5111    llvm::Triple::ArchType Arch) {
5112  // Get the last argument, which specifies the vector type.
5113  llvm::APSInt NeonTypeConst;
5114  const Expr *Arg = E->getArg(E->getNumArgs() - 1);
5115  if (!Arg->isIntegerConstantExpr(NeonTypeConst, getContext()))
5116    return nullptr;
5117
5118  // Determine the type of this overloaded NEON intrinsic.
5119  NeonTypeFlags Type(NeonTypeConst.getZExtValue());
5120  bool Usgn = Type.isUnsigned();
5121  bool Quad = Type.isQuad();
5122  const bool HasLegalHalfType = getTarget().hasLegalHalfType();
5123
5124  llvm::VectorType *VTy = GetNeonType(this, Type, HasLegalHalfType);
5125  llvm::Type *Ty = VTy;
5126  if (!Ty)
5127    return nullptr;
5128
5129  auto getAlignmentValue32 = [&](Address addr) -> Value* {
5130    return Builder.getInt32(addr.getAlignment().getQuantity());
5131  };
5132
5133  unsigned Int = LLVMIntrinsic;
5134  if ((Modifier & UnsignedAlts) && !Usgn)
5135    Int = AltLLVMIntrinsic;
5136
5137  switch (BuiltinID) {
5138  default: break;
5139  case NEON::BI__builtin_neon_vpadd_v:
5140  case NEON::BI__builtin_neon_vpaddq_v:
5141    // We don't allow fp/int overloading of intrinsics.
5142    if (VTy->getElementType()->isFloatingPointTy() &&
5143        Int == Intrinsic::aarch64_neon_addp)
5144      Int = Intrinsic::aarch64_neon_faddp;
5145    break;
5146  case NEON::BI__builtin_neon_vabs_v:
5147  case NEON::BI__builtin_neon_vabsq_v:
5148    if (VTy->getElementType()->isFloatingPointTy())
5149      return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, Ty), Ops, "vabs");
5150    return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), Ops, "vabs");
5151  case NEON::BI__builtin_neon_vaddhn_v: {
5152    llvm::VectorType *SrcTy =
5153        llvm::VectorType::getExtendedElementVectorType(VTy);
5154
5155    // %sum = add <4 x i32> %lhs, %rhs
5156    Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
5157    Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy);
5158    Ops[0] = Builder.CreateAdd(Ops[0], Ops[1], "vaddhn");
5159
5160    // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16>
5161    Constant *ShiftAmt =
5162        ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2);
5163    Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vaddhn");
5164
5165    // %res = trunc <4 x i32> %high to <4 x i16>
5166    return Builder.CreateTrunc(Ops[0], VTy, "vaddhn");
5167  }
5168  case NEON::BI__builtin_neon_vcale_v:
5169  case NEON::BI__builtin_neon_vcaleq_v:
5170  case NEON::BI__builtin_neon_vcalt_v:
5171  case NEON::BI__builtin_neon_vcaltq_v:
5172    std::swap(Ops[0], Ops[1]);
5173    LLVM_FALLTHROUGH;
5174  case NEON::BI__builtin_neon_vcage_v:
5175  case NEON::BI__builtin_neon_vcageq_v:
5176  case NEON::BI__builtin_neon_vcagt_v:
5177  case NEON::BI__builtin_neon_vcagtq_v: {
5178    llvm::Type *Ty;
5179    switch (VTy->getScalarSizeInBits()) {
5180    default: llvm_unreachable("unexpected type");
5181    case 32:
5182      Ty = FloatTy;
5183      break;
5184    case 64:
5185      Ty = DoubleTy;
5186      break;
5187    case 16:
5188      Ty = HalfTy;
5189      break;
5190    }
5191    llvm::Type *VecFlt = llvm::VectorType::get(Ty, VTy->getNumElements());
5192    llvm::Type *Tys[] = { VTy, VecFlt };
5193    Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
5194    return EmitNeonCall(F, Ops, NameHint);
5195  }
5196  case NEON::BI__builtin_neon_vceqz_v:
5197  case NEON::BI__builtin_neon_vceqzq_v:
5198    return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OEQ,
5199                                         ICmpInst::ICMP_EQ, "vceqz");
5200  case NEON::BI__builtin_neon_vcgez_v:
5201  case NEON::BI__builtin_neon_vcgezq_v:
5202    return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGE,
5203                                         ICmpInst::ICMP_SGE, "vcgez");
5204  case NEON::BI__builtin_neon_vclez_v:
5205  case NEON::BI__builtin_neon_vclezq_v:
5206    return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLE,
5207                                         ICmpInst::ICMP_SLE, "vclez");
5208  case NEON::BI__builtin_neon_vcgtz_v:
5209  case NEON::BI__builtin_neon_vcgtzq_v:
5210    return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OGT,
5211                                         ICmpInst::ICMP_SGT, "vcgtz");
5212  case NEON::BI__builtin_neon_vcltz_v:
5213  case NEON::BI__builtin_neon_vcltzq_v:
5214    return EmitAArch64CompareBuiltinExpr(Ops[0], Ty, ICmpInst::FCMP_OLT,
5215                                         ICmpInst::ICMP_SLT, "vcltz");
5216  case NEON::BI__builtin_neon_vclz_v:
5217  case NEON::BI__builtin_neon_vclzq_v:
5218    // We generate target-independent intrinsic, which needs a second argument
5219    // for whether or not clz of zero is undefined; on ARM it isn't.
5220    Ops.push_back(Builder.getInt1(getTarget().isCLZForZeroUndef()));
5221    break;
5222  case NEON::BI__builtin_neon_vcvt_f32_v:
5223  case NEON::BI__builtin_neon_vcvtq_f32_v:
5224    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5225    Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float32, false, Quad),
5226                     HasLegalHalfType);
5227    return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
5228                : Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
5229  case NEON::BI__builtin_neon_vcvt_f16_v:
5230  case NEON::BI__builtin_neon_vcvtq_f16_v:
5231    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5232    Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float16, false, Quad),
5233                     HasLegalHalfType);
5234    return Usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
5235                : Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
5236  case NEON::BI__builtin_neon_vcvt_n_f16_v:
5237  case NEON::BI__builtin_neon_vcvt_n_f32_v:
5238  case NEON::BI__builtin_neon_vcvt_n_f64_v:
5239  case NEON::BI__builtin_neon_vcvtq_n_f16_v:
5240  case NEON::BI__builtin_neon_vcvtq_n_f32_v:
5241  case NEON::BI__builtin_neon_vcvtq_n_f64_v: {
5242    llvm::Type *Tys[2] = { GetFloatNeonType(this, Type), Ty };
5243    Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic;
5244    Function *F = CGM.getIntrinsic(Int, Tys);
5245    return EmitNeonCall(F, Ops, "vcvt_n");
5246  }
5247  case NEON::BI__builtin_neon_vcvt_n_s16_v:
5248  case NEON::BI__builtin_neon_vcvt_n_s32_v:
5249  case NEON::BI__builtin_neon_vcvt_n_u16_v:
5250  case NEON::BI__builtin_neon_vcvt_n_u32_v:
5251  case NEON::BI__builtin_neon_vcvt_n_s64_v:
5252  case NEON::BI__builtin_neon_vcvt_n_u64_v:
5253  case NEON::BI__builtin_neon_vcvtq_n_s16_v:
5254  case NEON::BI__builtin_neon_vcvtq_n_s32_v:
5255  case NEON::BI__builtin_neon_vcvtq_n_u16_v:
5256  case NEON::BI__builtin_neon_vcvtq_n_u32_v:
5257  case NEON::BI__builtin_neon_vcvtq_n_s64_v:
5258  case NEON::BI__builtin_neon_vcvtq_n_u64_v: {
5259    llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
5260    Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
5261    return EmitNeonCall(F, Ops, "vcvt_n");
5262  }
5263  case NEON::BI__builtin_neon_vcvt_s32_v:
5264  case NEON::BI__builtin_neon_vcvt_u32_v:
5265  case NEON::BI__builtin_neon_vcvt_s64_v:
5266  case NEON::BI__builtin_neon_vcvt_u64_v:
5267  case NEON::BI__builtin_neon_vcvt_s16_v:
5268  case NEON::BI__builtin_neon_vcvt_u16_v:
5269  case NEON::BI__builtin_neon_vcvtq_s32_v:
5270  case NEON::BI__builtin_neon_vcvtq_u32_v:
5271  case NEON::BI__builtin_neon_vcvtq_s64_v:
5272  case NEON::BI__builtin_neon_vcvtq_u64_v:
5273  case NEON::BI__builtin_neon_vcvtq_s16_v:
5274  case NEON::BI__builtin_neon_vcvtq_u16_v: {
5275    Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type));
5276    return Usgn ? Builder.CreateFPToUI(Ops[0], Ty, "vcvt")
5277                : Builder.CreateFPToSI(Ops[0], Ty, "vcvt");
5278  }
5279  case NEON::BI__builtin_neon_vcvta_s16_v:
5280  case NEON::BI__builtin_neon_vcvta_s32_v:
5281  case NEON::BI__builtin_neon_vcvta_s64_v:
5282  case NEON::BI__builtin_neon_vcvta_u16_v:
5283  case NEON::BI__builtin_neon_vcvta_u32_v:
5284  case NEON::BI__builtin_neon_vcvta_u64_v:
5285  case NEON::BI__builtin_neon_vcvtaq_s16_v:
5286  case NEON::BI__builtin_neon_vcvtaq_s32_v:
5287  case NEON::BI__builtin_neon_vcvtaq_s64_v:
5288  case NEON::BI__builtin_neon_vcvtaq_u16_v:
5289  case NEON::BI__builtin_neon_vcvtaq_u32_v:
5290  case NEON::BI__builtin_neon_vcvtaq_u64_v:
5291  case NEON::BI__builtin_neon_vcvtn_s16_v:
5292  case NEON::BI__builtin_neon_vcvtn_s32_v:
5293  case NEON::BI__builtin_neon_vcvtn_s64_v:
5294  case NEON::BI__builtin_neon_vcvtn_u16_v:
5295  case NEON::BI__builtin_neon_vcvtn_u32_v:
5296  case NEON::BI__builtin_neon_vcvtn_u64_v:
5297  case NEON::BI__builtin_neon_vcvtnq_s16_v:
5298  case NEON::BI__builtin_neon_vcvtnq_s32_v:
5299  case NEON::BI__builtin_neon_vcvtnq_s64_v:
5300  case NEON::BI__builtin_neon_vcvtnq_u16_v:
5301  case NEON::BI__builtin_neon_vcvtnq_u32_v:
5302  case NEON::BI__builtin_neon_vcvtnq_u64_v:
5303  case NEON::BI__builtin_neon_vcvtp_s16_v:
5304  case NEON::BI__builtin_neon_vcvtp_s32_v:
5305  case NEON::BI__builtin_neon_vcvtp_s64_v:
5306  case NEON::BI__builtin_neon_vcvtp_u16_v:
5307  case NEON::BI__builtin_neon_vcvtp_u32_v:
5308  case NEON::BI__builtin_neon_vcvtp_u64_v:
5309  case NEON::BI__builtin_neon_vcvtpq_s16_v:
5310  case NEON::BI__builtin_neon_vcvtpq_s32_v:
5311  case NEON::BI__builtin_neon_vcvtpq_s64_v:
5312  case NEON::BI__builtin_neon_vcvtpq_u16_v:
5313  case NEON::BI__builtin_neon_vcvtpq_u32_v:
5314  case NEON::BI__builtin_neon_vcvtpq_u64_v:
5315  case NEON::BI__builtin_neon_vcvtm_s16_v:
5316  case NEON::BI__builtin_neon_vcvtm_s32_v:
5317  case NEON::BI__builtin_neon_vcvtm_s64_v:
5318  case NEON::BI__builtin_neon_vcvtm_u16_v:
5319  case NEON::BI__builtin_neon_vcvtm_u32_v:
5320  case NEON::BI__builtin_neon_vcvtm_u64_v:
5321  case NEON::BI__builtin_neon_vcvtmq_s16_v:
5322  case NEON::BI__builtin_neon_vcvtmq_s32_v:
5323  case NEON::BI__builtin_neon_vcvtmq_s64_v:
5324  case NEON::BI__builtin_neon_vcvtmq_u16_v:
5325  case NEON::BI__builtin_neon_vcvtmq_u32_v:
5326  case NEON::BI__builtin_neon_vcvtmq_u64_v: {
5327    llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
5328    return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, NameHint);
5329  }
5330  case NEON::BI__builtin_neon_vext_v:
5331  case NEON::BI__builtin_neon_vextq_v: {
5332    int CV = cast<ConstantInt>(Ops[2])->getSExtValue();
5333    SmallVector<uint32_t, 16> Indices;
5334    for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
5335      Indices.push_back(i+CV);
5336
5337    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5338    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5339    return Builder.CreateShuffleVector(Ops[0], Ops[1], Indices, "vext");
5340  }
5341  case NEON::BI__builtin_neon_vfma_v:
5342  case NEON::BI__builtin_neon_vfmaq_v: {
5343    Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
5344    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5345    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5346    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
5347
5348    // NEON intrinsic puts accumulator first, unlike the LLVM fma.
5349    return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]});
5350  }
5351  case NEON::BI__builtin_neon_vld1_v:
5352  case NEON::BI__builtin_neon_vld1q_v: {
5353    llvm::Type *Tys[] = {Ty, Int8PtrTy};
5354    Ops.push_back(getAlignmentValue32(PtrOp0));
5355    return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "vld1");
5356  }
5357  case NEON::BI__builtin_neon_vld1_x2_v:
5358  case NEON::BI__builtin_neon_vld1q_x2_v:
5359  case NEON::BI__builtin_neon_vld1_x3_v:
5360  case NEON::BI__builtin_neon_vld1q_x3_v:
5361  case NEON::BI__builtin_neon_vld1_x4_v:
5362  case NEON::BI__builtin_neon_vld1q_x4_v: {
5363    llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getVectorElementType());
5364    Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
5365    llvm::Type *Tys[2] = { VTy, PTy };
5366    Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
5367    Ops[1] = Builder.CreateCall(F, Ops[1], "vld1xN");
5368    Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
5369    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5370    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
5371  }
5372  case NEON::BI__builtin_neon_vld2_v:
5373  case NEON::BI__builtin_neon_vld2q_v:
5374  case NEON::BI__builtin_neon_vld3_v:
5375  case NEON::BI__builtin_neon_vld3q_v:
5376  case NEON::BI__builtin_neon_vld4_v:
5377  case NEON::BI__builtin_neon_vld4q_v:
5378  case NEON::BI__builtin_neon_vld2_dup_v:
5379  case NEON::BI__builtin_neon_vld2q_dup_v:
5380  case NEON::BI__builtin_neon_vld3_dup_v:
5381  case NEON::BI__builtin_neon_vld3q_dup_v:
5382  case NEON::BI__builtin_neon_vld4_dup_v:
5383  case NEON::BI__builtin_neon_vld4q_dup_v: {
5384    llvm::Type *Tys[] = {Ty, Int8PtrTy};
5385    Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
5386    Value *Align = getAlignmentValue32(PtrOp1);
5387    Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, NameHint);
5388    Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
5389    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5390    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
5391  }
5392  case NEON::BI__builtin_neon_vld1_dup_v:
5393  case NEON::BI__builtin_neon_vld1q_dup_v: {
5394    Value *V = UndefValue::get(Ty);
5395    Ty = llvm::PointerType::getUnqual(VTy->getElementType());
5396    PtrOp0 = Builder.CreateBitCast(PtrOp0, Ty);
5397    LoadInst *Ld = Builder.CreateLoad(PtrOp0);
5398    llvm::Constant *CI = ConstantInt::get(SizeTy, 0);
5399    Ops[0] = Builder.CreateInsertElement(V, Ld, CI);
5400    return EmitNeonSplat(Ops[0], CI);
5401  }
5402  case NEON::BI__builtin_neon_vld2_lane_v:
5403  case NEON::BI__builtin_neon_vld2q_lane_v:
5404  case NEON::BI__builtin_neon_vld3_lane_v:
5405  case NEON::BI__builtin_neon_vld3q_lane_v:
5406  case NEON::BI__builtin_neon_vld4_lane_v:
5407  case NEON::BI__builtin_neon_vld4q_lane_v: {
5408    llvm::Type *Tys[] = {Ty, Int8PtrTy};
5409    Function *F = CGM.getIntrinsic(LLVMIntrinsic, Tys);
5410    for (unsigned I = 2; I < Ops.size() - 1; ++I)
5411      Ops[I] = Builder.CreateBitCast(Ops[I], Ty);
5412    Ops.push_back(getAlignmentValue32(PtrOp1));
5413    Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), NameHint);
5414    Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
5415    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5416    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
5417  }
5418  case NEON::BI__builtin_neon_vmovl_v: {
5419    llvm::Type *DTy =llvm::VectorType::getTruncatedElementVectorType(VTy);
5420    Ops[0] = Builder.CreateBitCast(Ops[0], DTy);
5421    if (Usgn)
5422      return Builder.CreateZExt(Ops[0], Ty, "vmovl");
5423    return Builder.CreateSExt(Ops[0], Ty, "vmovl");
5424  }
5425  case NEON::BI__builtin_neon_vmovn_v: {
5426    llvm::Type *QTy = llvm::VectorType::getExtendedElementVectorType(VTy);
5427    Ops[0] = Builder.CreateBitCast(Ops[0], QTy);
5428    return Builder.CreateTrunc(Ops[0], Ty, "vmovn");
5429  }
5430  case NEON::BI__builtin_neon_vmull_v:
5431    // FIXME: the integer vmull operations could be emitted in terms of pure
5432    // LLVM IR (2 exts followed by a mul). Unfortunately LLVM has a habit of
5433    // hoisting the exts outside loops. Until global ISel comes along that can
5434    // see through such movement this leads to bad CodeGen. So we need an
5435    // intrinsic for now.
5436    Int = Usgn ? Intrinsic::arm_neon_vmullu : Intrinsic::arm_neon_vmulls;
5437    Int = Type.isPoly() ? (unsigned)Intrinsic::arm_neon_vmullp : Int;
5438    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull");
5439  case NEON::BI__builtin_neon_vpadal_v:
5440  case NEON::BI__builtin_neon_vpadalq_v: {
5441    // The source operand type has twice as many elements of half the size.
5442    unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits();
5443    llvm::Type *EltTy =
5444      llvm::IntegerType::get(getLLVMContext(), EltBits / 2);
5445    llvm::Type *NarrowTy =
5446      llvm::VectorType::get(EltTy, VTy->getNumElements() * 2);
5447    llvm::Type *Tys[2] = { Ty, NarrowTy };
5448    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, NameHint);
5449  }
5450  case NEON::BI__builtin_neon_vpaddl_v:
5451  case NEON::BI__builtin_neon_vpaddlq_v: {
5452    // The source operand type has twice as many elements of half the size.
5453    unsigned EltBits = VTy->getElementType()->getPrimitiveSizeInBits();
5454    llvm::Type *EltTy = llvm::IntegerType::get(getLLVMContext(), EltBits / 2);
5455    llvm::Type *NarrowTy =
5456      llvm::VectorType::get(EltTy, VTy->getNumElements() * 2);
5457    llvm::Type *Tys[2] = { Ty, NarrowTy };
5458    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vpaddl");
5459  }
5460  case NEON::BI__builtin_neon_vqdmlal_v:
5461  case NEON::BI__builtin_neon_vqdmlsl_v: {
5462    SmallVector<Value *, 2> MulOps(Ops.begin() + 1, Ops.end());
5463    Ops[1] =
5464        EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Ty), MulOps, "vqdmlal");
5465    Ops.resize(2);
5466    return EmitNeonCall(CGM.getIntrinsic(AltLLVMIntrinsic, Ty), Ops, NameHint);
5467  }
5468  case NEON::BI__builtin_neon_vqshl_n_v:
5469  case NEON::BI__builtin_neon_vqshlq_n_v:
5470    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshl_n",
5471                        1, false);
5472  case NEON::BI__builtin_neon_vqshlu_n_v:
5473  case NEON::BI__builtin_neon_vqshluq_n_v:
5474    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshlu_n",
5475                        1, false);
5476  case NEON::BI__builtin_neon_vrecpe_v:
5477  case NEON::BI__builtin_neon_vrecpeq_v:
5478  case NEON::BI__builtin_neon_vrsqrte_v:
5479  case NEON::BI__builtin_neon_vrsqrteq_v:
5480    Int = Ty->isFPOrFPVectorTy() ? LLVMIntrinsic : AltLLVMIntrinsic;
5481    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint);
5482  case NEON::BI__builtin_neon_vrndi_v:
5483  case NEON::BI__builtin_neon_vrndiq_v:
5484    Int = Intrinsic::nearbyint;
5485    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, NameHint);
5486  case NEON::BI__builtin_neon_vrshr_n_v:
5487  case NEON::BI__builtin_neon_vrshrq_n_v:
5488    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshr_n",
5489                        1, true);
5490  case NEON::BI__builtin_neon_vshl_n_v:
5491  case NEON::BI__builtin_neon_vshlq_n_v:
5492    Ops[1] = EmitNeonShiftVector(Ops[1], Ty, false);
5493    return Builder.CreateShl(Builder.CreateBitCast(Ops[0],Ty), Ops[1],
5494                             "vshl_n");
5495  case NEON::BI__builtin_neon_vshll_n_v: {
5496    llvm::Type *SrcTy = llvm::VectorType::getTruncatedElementVectorType(VTy);
5497    Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
5498    if (Usgn)
5499      Ops[0] = Builder.CreateZExt(Ops[0], VTy);
5500    else
5501      Ops[0] = Builder.CreateSExt(Ops[0], VTy);
5502    Ops[1] = EmitNeonShiftVector(Ops[1], VTy, false);
5503    return Builder.CreateShl(Ops[0], Ops[1], "vshll_n");
5504  }
5505  case NEON::BI__builtin_neon_vshrn_n_v: {
5506    llvm::Type *SrcTy = llvm::VectorType::getExtendedElementVectorType(VTy);
5507    Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
5508    Ops[1] = EmitNeonShiftVector(Ops[1], SrcTy, false);
5509    if (Usgn)
5510      Ops[0] = Builder.CreateLShr(Ops[0], Ops[1]);
5511    else
5512      Ops[0] = Builder.CreateAShr(Ops[0], Ops[1]);
5513    return Builder.CreateTrunc(Ops[0], Ty, "vshrn_n");
5514  }
5515  case NEON::BI__builtin_neon_vshr_n_v:
5516  case NEON::BI__builtin_neon_vshrq_n_v:
5517    return EmitNeonRShiftImm(Ops[0], Ops[1], Ty, Usgn, "vshr_n");
5518  case NEON::BI__builtin_neon_vst1_v:
5519  case NEON::BI__builtin_neon_vst1q_v:
5520  case NEON::BI__builtin_neon_vst2_v:
5521  case NEON::BI__builtin_neon_vst2q_v:
5522  case NEON::BI__builtin_neon_vst3_v:
5523  case NEON::BI__builtin_neon_vst3q_v:
5524  case NEON::BI__builtin_neon_vst4_v:
5525  case NEON::BI__builtin_neon_vst4q_v:
5526  case NEON::BI__builtin_neon_vst2_lane_v:
5527  case NEON::BI__builtin_neon_vst2q_lane_v:
5528  case NEON::BI__builtin_neon_vst3_lane_v:
5529  case NEON::BI__builtin_neon_vst3q_lane_v:
5530  case NEON::BI__builtin_neon_vst4_lane_v:
5531  case NEON::BI__builtin_neon_vst4q_lane_v: {
5532    llvm::Type *Tys[] = {Int8PtrTy, Ty};
5533    Ops.push_back(getAlignmentValue32(PtrOp0));
5534    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "");
5535  }
5536  case NEON::BI__builtin_neon_vst1_x2_v:
5537  case NEON::BI__builtin_neon_vst1q_x2_v:
5538  case NEON::BI__builtin_neon_vst1_x3_v:
5539  case NEON::BI__builtin_neon_vst1q_x3_v:
5540  case NEON::BI__builtin_neon_vst1_x4_v:
5541  case NEON::BI__builtin_neon_vst1q_x4_v: {
5542    llvm::Type *PTy = llvm::PointerType::getUnqual(VTy->getVectorElementType());
5543    // TODO: Currently in AArch32 mode the pointer operand comes first, whereas
5544    // in AArch64 it comes last. We may want to stick to one or another.
5545    if (Arch == llvm::Triple::aarch64 || Arch == llvm::Triple::aarch64_be) {
5546      llvm::Type *Tys[2] = { VTy, PTy };
5547      std::rotate(Ops.begin(), Ops.begin() + 1, Ops.end());
5548      return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "");
5549    }
5550    llvm::Type *Tys[2] = { PTy, VTy };
5551    return EmitNeonCall(CGM.getIntrinsic(LLVMIntrinsic, Tys), Ops, "");
5552  }
5553  case NEON::BI__builtin_neon_vsubhn_v: {
5554    llvm::VectorType *SrcTy =
5555        llvm::VectorType::getExtendedElementVectorType(VTy);
5556
5557    // %sum = add <4 x i32> %lhs, %rhs
5558    Ops[0] = Builder.CreateBitCast(Ops[0], SrcTy);
5559    Ops[1] = Builder.CreateBitCast(Ops[1], SrcTy);
5560    Ops[0] = Builder.CreateSub(Ops[0], Ops[1], "vsubhn");
5561
5562    // %high = lshr <4 x i32> %sum, <i32 16, i32 16, i32 16, i32 16>
5563    Constant *ShiftAmt =
5564        ConstantInt::get(SrcTy, SrcTy->getScalarSizeInBits() / 2);
5565    Ops[0] = Builder.CreateLShr(Ops[0], ShiftAmt, "vsubhn");
5566
5567    // %res = trunc <4 x i32> %high to <4 x i16>
5568    return Builder.CreateTrunc(Ops[0], VTy, "vsubhn");
5569  }
5570  case NEON::BI__builtin_neon_vtrn_v:
5571  case NEON::BI__builtin_neon_vtrnq_v: {
5572    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
5573    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5574    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
5575    Value *SV = nullptr;
5576
5577    for (unsigned vi = 0; vi != 2; ++vi) {
5578      SmallVector<uint32_t, 16> Indices;
5579      for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
5580        Indices.push_back(i+vi);
5581        Indices.push_back(i+e+vi);
5582      }
5583      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
5584      SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn");
5585      SV = Builder.CreateDefaultAlignedStore(SV, Addr);
5586    }
5587    return SV;
5588  }
5589  case NEON::BI__builtin_neon_vtst_v:
5590  case NEON::BI__builtin_neon_vtstq_v: {
5591    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
5592    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5593    Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]);
5594    Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0],
5595                                ConstantAggregateZero::get(Ty));
5596    return Builder.CreateSExt(Ops[0], Ty, "vtst");
5597  }
5598  case NEON::BI__builtin_neon_vuzp_v:
5599  case NEON::BI__builtin_neon_vuzpq_v: {
5600    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
5601    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5602    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
5603    Value *SV = nullptr;
5604
5605    for (unsigned vi = 0; vi != 2; ++vi) {
5606      SmallVector<uint32_t, 16> Indices;
5607      for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
5608        Indices.push_back(2*i+vi);
5609
5610      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
5611      SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp");
5612      SV = Builder.CreateDefaultAlignedStore(SV, Addr);
5613    }
5614    return SV;
5615  }
5616  case NEON::BI__builtin_neon_vzip_v:
5617  case NEON::BI__builtin_neon_vzipq_v: {
5618    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
5619    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
5620    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
5621    Value *SV = nullptr;
5622
5623    for (unsigned vi = 0; vi != 2; ++vi) {
5624      SmallVector<uint32_t, 16> Indices;
5625      for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
5626        Indices.push_back((i + vi*e) >> 1);
5627        Indices.push_back(((i + vi*e) >> 1)+e);
5628      }
5629      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
5630      SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip");
5631      SV = Builder.CreateDefaultAlignedStore(SV, Addr);
5632    }
5633    return SV;
5634  }
5635  case NEON::BI__builtin_neon_vdot_v:
5636  case NEON::BI__builtin_neon_vdotq_v: {
5637    llvm::Type *InputTy =
5638        llvm::VectorType::get(Int8Ty, Ty->getPrimitiveSizeInBits() / 8);
5639    llvm::Type *Tys[2] = { Ty, InputTy };
5640    Int = Usgn ? LLVMIntrinsic : AltLLVMIntrinsic;
5641    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vdot");
5642  }
5643  case NEON::BI__builtin_neon_vfmlal_low_v:
5644  case NEON::BI__builtin_neon_vfmlalq_low_v: {
5645    llvm::Type *InputTy =
5646        llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
5647    llvm::Type *Tys[2] = { Ty, InputTy };
5648    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_low");
5649  }
5650  case NEON::BI__builtin_neon_vfmlsl_low_v:
5651  case NEON::BI__builtin_neon_vfmlslq_low_v: {
5652    llvm::Type *InputTy =
5653        llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
5654    llvm::Type *Tys[2] = { Ty, InputTy };
5655    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_low");
5656  }
5657  case NEON::BI__builtin_neon_vfmlal_high_v:
5658  case NEON::BI__builtin_neon_vfmlalq_high_v: {
5659    llvm::Type *InputTy =
5660           llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
5661    llvm::Type *Tys[2] = { Ty, InputTy };
5662    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlal_high");
5663  }
5664  case NEON::BI__builtin_neon_vfmlsl_high_v:
5665  case NEON::BI__builtin_neon_vfmlslq_high_v: {
5666    llvm::Type *InputTy =
5667           llvm::VectorType::get(HalfTy, Ty->getPrimitiveSizeInBits() / 16);
5668    llvm::Type *Tys[2] = { Ty, InputTy };
5669    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vfmlsl_high");
5670  }
5671  }
5672
5673   (0) . __assert_fail ("Int && \"Expected valid intrinsic number\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 5673, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Int && "Expected valid intrinsic number");
5674
5675  // Determine the type(s) of this overloaded AArch64 intrinsic.
5676  Function *F = LookupNeonLLVMIntrinsic(Int, Modifier, Ty, E);
5677
5678  Value *Result = EmitNeonCall(F, Ops, NameHint);
5679  llvm::Type *ResultType = ConvertType(E->getType());
5680  // AArch64 intrinsic one-element vector type cast to
5681  // scalar type expected by the builtin
5682  return Builder.CreateBitCast(Result, ResultType, NameHint);
5683}
5684
5685Value *CodeGenFunction::EmitAArch64CompareBuiltinExpr(
5686    Value *Op, llvm::Type *Ty, const CmpInst::Predicate Fp,
5687    const CmpInst::Predicate Ip, const Twine &Name) {
5688  llvm::Type *OTy = Op->getType();
5689
5690  // FIXME: this is utterly horrific. We should not be looking at previous
5691  // codegen context to find out what needs doing. Unfortunately TableGen
5692  // currently gives us exactly the same calls for vceqz_f32 and vceqz_s32
5693  // (etc).
5694  if (BitCastInst *BI = dyn_cast<BitCastInst>(Op))
5695    OTy = BI->getOperand(0)->getType();
5696
5697  Op = Builder.CreateBitCast(Op, OTy);
5698  if (OTy->getScalarType()->isFloatingPointTy()) {
5699    Op = Builder.CreateFCmp(Fp, Op, Constant::getNullValue(OTy));
5700  } else {
5701    Op = Builder.CreateICmp(Ip, Op, Constant::getNullValue(OTy));
5702  }
5703  return Builder.CreateSExt(Op, Ty, Name);
5704}
5705
5706static Value *packTBLDVectorList(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
5707                                 Value *ExtOp, Value *IndexOp,
5708                                 llvm::Type *ResTy, unsigned IntID,
5709                                 const char *Name) {
5710  SmallVector<Value *, 2> TblOps;
5711  if (ExtOp)
5712    TblOps.push_back(ExtOp);
5713
5714  // Build a vector containing sequential number like (0, 1, 2, ..., 15)
5715  SmallVector<uint32_t, 16> Indices;
5716  llvm::VectorType *TblTy = cast<llvm::VectorType>(Ops[0]->getType());
5717  for (unsigned i = 0, e = TblTy->getNumElements(); i != e; ++i) {
5718    Indices.push_back(2*i);
5719    Indices.push_back(2*i+1);
5720  }
5721
5722  int PairPos = 0, End = Ops.size() - 1;
5723  while (PairPos < End) {
5724    TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos],
5725                                                     Ops[PairPos+1], Indices,
5726                                                     Name));
5727    PairPos += 2;
5728  }
5729
5730  // If there's an odd number of 64-bit lookup table, fill the high 64-bit
5731  // of the 128-bit lookup table with zero.
5732  if (PairPos == End) {
5733    Value *ZeroTbl = ConstantAggregateZero::get(TblTy);
5734    TblOps.push_back(CGF.Builder.CreateShuffleVector(Ops[PairPos],
5735                                                     ZeroTbl, Indices, Name));
5736  }
5737
5738  Function *TblF;
5739  TblOps.push_back(IndexOp);
5740  TblF = CGF.CGM.getIntrinsic(IntID, ResTy);
5741
5742  return CGF.EmitNeonCall(TblF, TblOps, Name);
5743}
5744
5745Value *CodeGenFunction::GetValueForARMHint(unsigned BuiltinID) {
5746  unsigned Value;
5747  switch (BuiltinID) {
5748  default:
5749    return nullptr;
5750  case ARM::BI__builtin_arm_nop:
5751    Value = 0;
5752    break;
5753  case ARM::BI__builtin_arm_yield:
5754  case ARM::BI__yield:
5755    Value = 1;
5756    break;
5757  case ARM::BI__builtin_arm_wfe:
5758  case ARM::BI__wfe:
5759    Value = 2;
5760    break;
5761  case ARM::BI__builtin_arm_wfi:
5762  case ARM::BI__wfi:
5763    Value = 3;
5764    break;
5765  case ARM::BI__builtin_arm_sev:
5766  case ARM::BI__sev:
5767    Value = 4;
5768    break;
5769  case ARM::BI__builtin_arm_sevl:
5770  case ARM::BI__sevl:
5771    Value = 5;
5772    break;
5773  }
5774
5775  return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_hint),
5776                            llvm::ConstantInt::get(Int32Ty, Value));
5777}
5778
5779// Generates the IR for the read/write special register builtin,
5780// ValueType is the type of the value that is to be written or read,
5781// RegisterType is the type of the register being written to or read from.
5782static Value *EmitSpecialRegisterBuiltin(CodeGenFunction &CGF,
5783                                         const CallExpr *E,
5784                                         llvm::Type *RegisterType,
5785                                         llvm::Type *ValueType,
5786                                         bool IsRead,
5787                                         StringRef SysReg = "") {
5788  // write and register intrinsics only support 32 and 64 bit operations.
5789   (0) . __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 5790, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert((RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64))
5790 (0) . __assert_fail ("(RegisterType->isIntegerTy(32) || RegisterType->isIntegerTy(64)) && \"Unsupported size for register.\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 5790, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">          && "Unsupported size for register.");
5791
5792  CodeGen::CGBuilderTy &Builder = CGF.Builder;
5793  CodeGen::CodeGenModule &CGM = CGF.CGM;
5794  LLVMContext &Context = CGM.getLLVMContext();
5795
5796  if (SysReg.empty()) {
5797    const Expr *SysRegStrExpr = E->getArg(0)->IgnoreParenCasts();
5798    SysReg = cast<clang::StringLiteral>(SysRegStrExpr)->getString();
5799  }
5800
5801  llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysReg) };
5802  llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops);
5803  llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName);
5804
5805  llvm::Type *Types[] = { RegisterType };
5806
5807  bool MixedTypes = RegisterType->isIntegerTy(64) && ValueType->isIntegerTy(32);
5808   (0) . __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 5809, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64))
5809 (0) . __assert_fail ("!(RegisterType->isIntegerTy(32) && ValueType->isIntegerTy(64)) && \"Can't fit 64-bit value in 32-bit register\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 5809, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">            && "Can't fit 64-bit value in 32-bit register");
5810
5811  if (IsRead) {
5812    llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types);
5813    llvm::Value *Call = Builder.CreateCall(F, Metadata);
5814
5815    if (MixedTypes)
5816      // Read into 64 bit register and then truncate result to 32 bit.
5817      return Builder.CreateTrunc(Call, ValueType);
5818
5819    if (ValueType->isPointerTy())
5820      // Have i32/i64 result (Call) but want to return a VoidPtrTy (i8*).
5821      return Builder.CreateIntToPtr(Call, ValueType);
5822
5823    return Call;
5824  }
5825
5826  llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types);
5827  llvm::Value *ArgValue = CGF.EmitScalarExpr(E->getArg(1));
5828  if (MixedTypes) {
5829    // Extend 32 bit write value to 64 bit to pass to write.
5830    ArgValue = Builder.CreateZExt(ArgValue, RegisterType);
5831    return Builder.CreateCall(F, { Metadata, ArgValue });
5832  }
5833
5834  if (ValueType->isPointerTy()) {
5835    // Have VoidPtrTy ArgValue but want to return an i32/i64.
5836    ArgValue = Builder.CreatePtrToInt(ArgValue, RegisterType);
5837    return Builder.CreateCall(F, { Metadata, ArgValue });
5838  }
5839
5840  return Builder.CreateCall(F, { Metadata, ArgValue });
5841}
5842
5843/// Return true if BuiltinID is an overloaded Neon intrinsic with an extra
5844/// argument that specifies the vector type.
5845static bool HasExtraNeonArgument(unsigned BuiltinID) {
5846  switch (BuiltinID) {
5847  default: break;
5848  case NEON::BI__builtin_neon_vget_lane_i8:
5849  case NEON::BI__builtin_neon_vget_lane_i16:
5850  case NEON::BI__builtin_neon_vget_lane_i32:
5851  case NEON::BI__builtin_neon_vget_lane_i64:
5852  case NEON::BI__builtin_neon_vget_lane_f32:
5853  case NEON::BI__builtin_neon_vgetq_lane_i8:
5854  case NEON::BI__builtin_neon_vgetq_lane_i16:
5855  case NEON::BI__builtin_neon_vgetq_lane_i32:
5856  case NEON::BI__builtin_neon_vgetq_lane_i64:
5857  case NEON::BI__builtin_neon_vgetq_lane_f32:
5858  case NEON::BI__builtin_neon_vset_lane_i8:
5859  case NEON::BI__builtin_neon_vset_lane_i16:
5860  case NEON::BI__builtin_neon_vset_lane_i32:
5861  case NEON::BI__builtin_neon_vset_lane_i64:
5862  case NEON::BI__builtin_neon_vset_lane_f32:
5863  case NEON::BI__builtin_neon_vsetq_lane_i8:
5864  case NEON::BI__builtin_neon_vsetq_lane_i16:
5865  case NEON::BI__builtin_neon_vsetq_lane_i32:
5866  case NEON::BI__builtin_neon_vsetq_lane_i64:
5867  case NEON::BI__builtin_neon_vsetq_lane_f32:
5868  case NEON::BI__builtin_neon_vsha1h_u32:
5869  case NEON::BI__builtin_neon_vsha1cq_u32:
5870  case NEON::BI__builtin_neon_vsha1pq_u32:
5871  case NEON::BI__builtin_neon_vsha1mq_u32:
5872  case clang::ARM::BI_MoveToCoprocessor:
5873  case clang::ARM::BI_MoveToCoprocessor2:
5874    return false;
5875  }
5876  return true;
5877}
5878
5879Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
5880                                           const CallExpr *E,
5881                                           llvm::Triple::ArchType Arch) {
5882  if (auto Hint = GetValueForARMHint(BuiltinID))
5883    return Hint;
5884
5885  if (BuiltinID == ARM::BI__emit) {
5886    bool IsThumb = getTarget().getTriple().getArch() == llvm::Triple::thumb;
5887    llvm::FunctionType *FTy =
5888        llvm::FunctionType::get(VoidTy, /*Variadic=*/false);
5889
5890    Expr::EvalResult Result;
5891    if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext()))
5892      llvm_unreachable("Sema will ensure that the parameter is constant");
5893
5894    llvm::APSInt Value = Result.Val.getInt();
5895    uint64_t ZExtValue = Value.zextOrTrunc(IsThumb ? 16 : 32).getZExtValue();
5896
5897    llvm::InlineAsm *Emit =
5898        IsThumb ? InlineAsm::get(FTy, ".inst.n 0x" + utohexstr(ZExtValue), "",
5899                                 /*SideEffects=*/true)
5900                : InlineAsm::get(FTy, ".inst 0x" + utohexstr(ZExtValue), "",
5901                                 /*SideEffects=*/true);
5902
5903    return Builder.CreateCall(Emit);
5904  }
5905
5906  if (BuiltinID == ARM::BI__builtin_arm_dbg) {
5907    Value *Option = EmitScalarExpr(E->getArg(0));
5908    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_dbg), Option);
5909  }
5910
5911  if (BuiltinID == ARM::BI__builtin_arm_prefetch) {
5912    Value *Address = EmitScalarExpr(E->getArg(0));
5913    Value *RW      = EmitScalarExpr(E->getArg(1));
5914    Value *IsData  = EmitScalarExpr(E->getArg(2));
5915
5916    // Locality is not supported on ARM target
5917    Value *Locality = llvm::ConstantInt::get(Int32Ty, 3);
5918
5919    Function *F = CGM.getIntrinsic(Intrinsic::prefetch);
5920    return Builder.CreateCall(F, {Address, RW, Locality, IsData});
5921  }
5922
5923  if (BuiltinID == ARM::BI__builtin_arm_rbit) {
5924    llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
5925    return Builder.CreateCall(
5926        CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
5927  }
5928
5929  if (BuiltinID == ARM::BI__clear_cache) {
5930     (0) . __assert_fail ("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 5930, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments");
5931    const FunctionDecl *FD = E->getDirectCallee();
5932    Value *Ops[2];
5933    for (unsigned i = 0; i < 2; i++)
5934      Ops[i] = EmitScalarExpr(E->getArg(i));
5935    llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType());
5936    llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty);
5937    StringRef Name = FD->getName();
5938    return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops);
5939  }
5940
5941  if (BuiltinID == ARM::BI__builtin_arm_mcrr ||
5942      BuiltinID == ARM::BI__builtin_arm_mcrr2) {
5943    Function *F;
5944
5945    switch (BuiltinID) {
5946    default: llvm_unreachable("unexpected builtin");
5947    case ARM::BI__builtin_arm_mcrr:
5948      F = CGM.getIntrinsic(Intrinsic::arm_mcrr);
5949      break;
5950    case ARM::BI__builtin_arm_mcrr2:
5951      F = CGM.getIntrinsic(Intrinsic::arm_mcrr2);
5952      break;
5953    }
5954
5955    // MCRR{2} instruction has 5 operands but
5956    // the intrinsic has 4 because Rt and Rt2
5957    // are represented as a single unsigned 64
5958    // bit integer in the intrinsic definition
5959    // but internally it's represented as 2 32
5960    // bit integers.
5961
5962    Value *Coproc = EmitScalarExpr(E->getArg(0));
5963    Value *Opc1 = EmitScalarExpr(E->getArg(1));
5964    Value *RtAndRt2 = EmitScalarExpr(E->getArg(2));
5965    Value *CRm = EmitScalarExpr(E->getArg(3));
5966
5967    Value *C1 = llvm::ConstantInt::get(Int64Ty, 32);
5968    Value *Rt = Builder.CreateTruncOrBitCast(RtAndRt2, Int32Ty);
5969    Value *Rt2 = Builder.CreateLShr(RtAndRt2, C1);
5970    Rt2 = Builder.CreateTruncOrBitCast(Rt2, Int32Ty);
5971
5972    return Builder.CreateCall(F, {Coproc, Opc1, Rt, Rt2, CRm});
5973  }
5974
5975  if (BuiltinID == ARM::BI__builtin_arm_mrrc ||
5976      BuiltinID == ARM::BI__builtin_arm_mrrc2) {
5977    Function *F;
5978
5979    switch (BuiltinID) {
5980    default: llvm_unreachable("unexpected builtin");
5981    case ARM::BI__builtin_arm_mrrc:
5982      F = CGM.getIntrinsic(Intrinsic::arm_mrrc);
5983      break;
5984    case ARM::BI__builtin_arm_mrrc2:
5985      F = CGM.getIntrinsic(Intrinsic::arm_mrrc2);
5986      break;
5987    }
5988
5989    Value *Coproc = EmitScalarExpr(E->getArg(0));
5990    Value *Opc1 = EmitScalarExpr(E->getArg(1));
5991    Value *CRm  = EmitScalarExpr(E->getArg(2));
5992    Value *RtAndRt2 = Builder.CreateCall(F, {Coproc, Opc1, CRm});
5993
5994    // Returns an unsigned 64 bit integer, represented
5995    // as two 32 bit integers.
5996
5997    Value *Rt = Builder.CreateExtractValue(RtAndRt2, 1);
5998    Value *Rt1 = Builder.CreateExtractValue(RtAndRt2, 0);
5999    Rt = Builder.CreateZExt(Rt, Int64Ty);
6000    Rt1 = Builder.CreateZExt(Rt1, Int64Ty);
6001
6002    Value *ShiftCast = llvm::ConstantInt::get(Int64Ty, 32);
6003    RtAndRt2 = Builder.CreateShl(Rt, ShiftCast, "shl", true);
6004    RtAndRt2 = Builder.CreateOr(RtAndRt2, Rt1);
6005
6006    return Builder.CreateBitCast(RtAndRt2, ConvertType(E->getType()));
6007  }
6008
6009  if (BuiltinID == ARM::BI__builtin_arm_ldrexd ||
6010      ((BuiltinID == ARM::BI__builtin_arm_ldrex ||
6011        BuiltinID == ARM::BI__builtin_arm_ldaex) &&
6012       getContext().getTypeSize(E->getType()) == 64) ||
6013      BuiltinID == ARM::BI__ldrexd) {
6014    Function *F;
6015
6016    switch (BuiltinID) {
6017    default: llvm_unreachable("unexpected builtin");
6018    case ARM::BI__builtin_arm_ldaex:
6019      F = CGM.getIntrinsic(Intrinsic::arm_ldaexd);
6020      break;
6021    case ARM::BI__builtin_arm_ldrexd:
6022    case ARM::BI__builtin_arm_ldrex:
6023    case ARM::BI__ldrexd:
6024      F = CGM.getIntrinsic(Intrinsic::arm_ldrexd);
6025      break;
6026    }
6027
6028    Value *LdPtr = EmitScalarExpr(E->getArg(0));
6029    Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy),
6030                                    "ldrexd");
6031
6032    Value *Val0 = Builder.CreateExtractValue(Val, 1);
6033    Value *Val1 = Builder.CreateExtractValue(Val, 0);
6034    Val0 = Builder.CreateZExt(Val0, Int64Ty);
6035    Val1 = Builder.CreateZExt(Val1, Int64Ty);
6036
6037    Value *ShiftCst = llvm::ConstantInt::get(Int64Ty, 32);
6038    Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */);
6039    Val = Builder.CreateOr(Val, Val1);
6040    return Builder.CreateBitCast(Val, ConvertType(E->getType()));
6041  }
6042
6043  if (BuiltinID == ARM::BI__builtin_arm_ldrex ||
6044      BuiltinID == ARM::BI__builtin_arm_ldaex) {
6045    Value *LoadAddr = EmitScalarExpr(E->getArg(0));
6046
6047    QualType Ty = E->getType();
6048    llvm::Type *RealResTy = ConvertType(Ty);
6049    llvm::Type *PtrTy = llvm::IntegerType::get(
6050        getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo();
6051    LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy);
6052
6053    Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_ldaex
6054                                       ? Intrinsic::arm_ldaex
6055                                       : Intrinsic::arm_ldrex,
6056                                   PtrTy);
6057    Value *Val = Builder.CreateCall(F, LoadAddr, "ldrex");
6058
6059    if (RealResTy->isPointerTy())
6060      return Builder.CreateIntToPtr(Val, RealResTy);
6061    else {
6062      llvm::Type *IntResTy = llvm::IntegerType::get(
6063          getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy));
6064      Val = Builder.CreateTruncOrBitCast(Val, IntResTy);
6065      return Builder.CreateBitCast(Val, RealResTy);
6066    }
6067  }
6068
6069  if (BuiltinID == ARM::BI__builtin_arm_strexd ||
6070      ((BuiltinID == ARM::BI__builtin_arm_stlex ||
6071        BuiltinID == ARM::BI__builtin_arm_strex) &&
6072       getContext().getTypeSize(E->getArg(0)->getType()) == 64)) {
6073    Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex
6074                                       ? Intrinsic::arm_stlexd
6075                                       : Intrinsic::arm_strexd);
6076    llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty);
6077
6078    Address Tmp = CreateMemTemp(E->getArg(0)->getType());
6079    Value *Val = EmitScalarExpr(E->getArg(0));
6080    Builder.CreateStore(Val, Tmp);
6081
6082    Address LdPtr = Builder.CreateBitCast(Tmp,llvm::PointerType::getUnqual(STy));
6083    Val = Builder.CreateLoad(LdPtr);
6084
6085    Value *Arg0 = Builder.CreateExtractValue(Val, 0);
6086    Value *Arg1 = Builder.CreateExtractValue(Val, 1);
6087    Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), Int8PtrTy);
6088    return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "strexd");
6089  }
6090
6091  if (BuiltinID == ARM::BI__builtin_arm_strex ||
6092      BuiltinID == ARM::BI__builtin_arm_stlex) {
6093    Value *StoreVal = EmitScalarExpr(E->getArg(0));
6094    Value *StoreAddr = EmitScalarExpr(E->getArg(1));
6095
6096    QualType Ty = E->getArg(0)->getType();
6097    llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(),
6098                                                 getContext().getTypeSize(Ty));
6099    StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo());
6100
6101    if (StoreVal->getType()->isPointerTy())
6102      StoreVal = Builder.CreatePtrToInt(StoreVal, Int32Ty);
6103    else {
6104      llvm::Type *IntTy = llvm::IntegerType::get(
6105          getLLVMContext(),
6106          CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType()));
6107      StoreVal = Builder.CreateBitCast(StoreVal, IntTy);
6108      StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int32Ty);
6109    }
6110
6111    Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI__builtin_arm_stlex
6112                                       ? Intrinsic::arm_stlex
6113                                       : Intrinsic::arm_strex,
6114                                   StoreAddr->getType());
6115    return Builder.CreateCall(F, {StoreVal, StoreAddr}, "strex");
6116  }
6117
6118  if (BuiltinID == ARM::BI__builtin_arm_clrex) {
6119    Function *F = CGM.getIntrinsic(Intrinsic::arm_clrex);
6120    return Builder.CreateCall(F);
6121  }
6122
6123  // CRC32
6124  Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic;
6125  switch (BuiltinID) {
6126  case ARM::BI__builtin_arm_crc32b:
6127    CRCIntrinsicID = Intrinsic::arm_crc32b; break;
6128  case ARM::BI__builtin_arm_crc32cb:
6129    CRCIntrinsicID = Intrinsic::arm_crc32cb; break;
6130  case ARM::BI__builtin_arm_crc32h:
6131    CRCIntrinsicID = Intrinsic::arm_crc32h; break;
6132  case ARM::BI__builtin_arm_crc32ch:
6133    CRCIntrinsicID = Intrinsic::arm_crc32ch; break;
6134  case ARM::BI__builtin_arm_crc32w:
6135  case ARM::BI__builtin_arm_crc32d:
6136    CRCIntrinsicID = Intrinsic::arm_crc32w; break;
6137  case ARM::BI__builtin_arm_crc32cw:
6138  case ARM::BI__builtin_arm_crc32cd:
6139    CRCIntrinsicID = Intrinsic::arm_crc32cw; break;
6140  }
6141
6142  if (CRCIntrinsicID != Intrinsic::not_intrinsic) {
6143    Value *Arg0 = EmitScalarExpr(E->getArg(0));
6144    Value *Arg1 = EmitScalarExpr(E->getArg(1));
6145
6146    // crc32{c,}d intrinsics are implemnted as two calls to crc32{c,}w
6147    // intrinsics, hence we need different codegen for these cases.
6148    if (BuiltinID == ARM::BI__builtin_arm_crc32d ||
6149        BuiltinID == ARM::BI__builtin_arm_crc32cd) {
6150      Value *C1 = llvm::ConstantInt::get(Int64Ty, 32);
6151      Value *Arg1a = Builder.CreateTruncOrBitCast(Arg1, Int32Ty);
6152      Value *Arg1b = Builder.CreateLShr(Arg1, C1);
6153      Arg1b = Builder.CreateTruncOrBitCast(Arg1b, Int32Ty);
6154
6155      Function *F = CGM.getIntrinsic(CRCIntrinsicID);
6156      Value *Res = Builder.CreateCall(F, {Arg0, Arg1a});
6157      return Builder.CreateCall(F, {Res, Arg1b});
6158    } else {
6159      Arg1 = Builder.CreateZExtOrBitCast(Arg1, Int32Ty);
6160
6161      Function *F = CGM.getIntrinsic(CRCIntrinsicID);
6162      return Builder.CreateCall(F, {Arg0, Arg1});
6163    }
6164  }
6165
6166  if (BuiltinID == ARM::BI__builtin_arm_rsr ||
6167      BuiltinID == ARM::BI__builtin_arm_rsr64 ||
6168      BuiltinID == ARM::BI__builtin_arm_rsrp ||
6169      BuiltinID == ARM::BI__builtin_arm_wsr ||
6170      BuiltinID == ARM::BI__builtin_arm_wsr64 ||
6171      BuiltinID == ARM::BI__builtin_arm_wsrp) {
6172
6173    bool IsRead = BuiltinID == ARM::BI__builtin_arm_rsr ||
6174                  BuiltinID == ARM::BI__builtin_arm_rsr64 ||
6175                  BuiltinID == ARM::BI__builtin_arm_rsrp;
6176
6177    bool IsPointerBuiltin = BuiltinID == ARM::BI__builtin_arm_rsrp ||
6178                            BuiltinID == ARM::BI__builtin_arm_wsrp;
6179
6180    bool Is64Bit = BuiltinID == ARM::BI__builtin_arm_rsr64 ||
6181                   BuiltinID == ARM::BI__builtin_arm_wsr64;
6182
6183    llvm::Type *ValueType;
6184    llvm::Type *RegisterType;
6185    if (IsPointerBuiltin) {
6186      ValueType = VoidPtrTy;
6187      RegisterType = Int32Ty;
6188    } else if (Is64Bit) {
6189      ValueType = RegisterType = Int64Ty;
6190    } else {
6191      ValueType = RegisterType = Int32Ty;
6192    }
6193
6194    return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, IsRead);
6195  }
6196
6197  // Find out if any arguments are required to be integer constant
6198  // expressions.
6199  unsigned ICEArguments = 0;
6200  ASTContext::GetBuiltinTypeError Error;
6201  getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
6202   (0) . __assert_fail ("Error == ASTContext..GE_None && \"Should not codegen an error\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 6202, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Error == ASTContext::GE_None && "Should not codegen an error");
6203
6204  auto getAlignmentValue32 = [&](Address addr) -> Value* {
6205    return Builder.getInt32(addr.getAlignment().getQuantity());
6206  };
6207
6208  Address PtrOp0 = Address::invalid();
6209  Address PtrOp1 = Address::invalid();
6210  SmallVector<Value*, 4> Ops;
6211  bool HasExtraArg = HasExtraNeonArgument(BuiltinID);
6212  unsigned NumArgs = E->getNumArgs() - (HasExtraArg ? 1 : 0);
6213  for (unsigned i = 0, e = NumArgs; i != e; i++) {
6214    if (i == 0) {
6215      switch (BuiltinID) {
6216      case NEON::BI__builtin_neon_vld1_v:
6217      case NEON::BI__builtin_neon_vld1q_v:
6218      case NEON::BI__builtin_neon_vld1q_lane_v:
6219      case NEON::BI__builtin_neon_vld1_lane_v:
6220      case NEON::BI__builtin_neon_vld1_dup_v:
6221      case NEON::BI__builtin_neon_vld1q_dup_v:
6222      case NEON::BI__builtin_neon_vst1_v:
6223      case NEON::BI__builtin_neon_vst1q_v:
6224      case NEON::BI__builtin_neon_vst1q_lane_v:
6225      case NEON::BI__builtin_neon_vst1_lane_v:
6226      case NEON::BI__builtin_neon_vst2_v:
6227      case NEON::BI__builtin_neon_vst2q_v:
6228      case NEON::BI__builtin_neon_vst2_lane_v:
6229      case NEON::BI__builtin_neon_vst2q_lane_v:
6230      case NEON::BI__builtin_neon_vst3_v:
6231      case NEON::BI__builtin_neon_vst3q_v:
6232      case NEON::BI__builtin_neon_vst3_lane_v:
6233      case NEON::BI__builtin_neon_vst3q_lane_v:
6234      case NEON::BI__builtin_neon_vst4_v:
6235      case NEON::BI__builtin_neon_vst4q_v:
6236      case NEON::BI__builtin_neon_vst4_lane_v:
6237      case NEON::BI__builtin_neon_vst4q_lane_v:
6238        // Get the alignment for the argument in addition to the value;
6239        // we'll use it later.
6240        PtrOp0 = EmitPointerWithAlignment(E->getArg(0));
6241        Ops.push_back(PtrOp0.getPointer());
6242        continue;
6243      }
6244    }
6245    if (i == 1) {
6246      switch (BuiltinID) {
6247      case NEON::BI__builtin_neon_vld2_v:
6248      case NEON::BI__builtin_neon_vld2q_v:
6249      case NEON::BI__builtin_neon_vld3_v:
6250      case NEON::BI__builtin_neon_vld3q_v:
6251      case NEON::BI__builtin_neon_vld4_v:
6252      case NEON::BI__builtin_neon_vld4q_v:
6253      case NEON::BI__builtin_neon_vld2_lane_v:
6254      case NEON::BI__builtin_neon_vld2q_lane_v:
6255      case NEON::BI__builtin_neon_vld3_lane_v:
6256      case NEON::BI__builtin_neon_vld3q_lane_v:
6257      case NEON::BI__builtin_neon_vld4_lane_v:
6258      case NEON::BI__builtin_neon_vld4q_lane_v:
6259      case NEON::BI__builtin_neon_vld2_dup_v:
6260      case NEON::BI__builtin_neon_vld2q_dup_v:
6261      case NEON::BI__builtin_neon_vld3_dup_v:
6262      case NEON::BI__builtin_neon_vld3q_dup_v:
6263      case NEON::BI__builtin_neon_vld4_dup_v:
6264      case NEON::BI__builtin_neon_vld4q_dup_v:
6265        // Get the alignment for the argument in addition to the value;
6266        // we'll use it later.
6267        PtrOp1 = EmitPointerWithAlignment(E->getArg(1));
6268        Ops.push_back(PtrOp1.getPointer());
6269        continue;
6270      }
6271    }
6272
6273    if ((ICEArguments & (1 << i)) == 0) {
6274      Ops.push_back(EmitScalarExpr(E->getArg(i)));
6275    } else {
6276      // If this is required to be a constant, constant fold it so that we know
6277      // that the generated intrinsic gets a ConstantInt.
6278      llvm::APSInt Result;
6279      bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext());
6280       (0) . __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 6280, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(IsConst && "Constant arg isn't actually constant?"); (void)IsConst;
6281      Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result));
6282    }
6283  }
6284
6285  switch (BuiltinID) {
6286  default: break;
6287
6288  case NEON::BI__builtin_neon_vget_lane_i8:
6289  case NEON::BI__builtin_neon_vget_lane_i16:
6290  case NEON::BI__builtin_neon_vget_lane_i32:
6291  case NEON::BI__builtin_neon_vget_lane_i64:
6292  case NEON::BI__builtin_neon_vget_lane_f32:
6293  case NEON::BI__builtin_neon_vgetq_lane_i8:
6294  case NEON::BI__builtin_neon_vgetq_lane_i16:
6295  case NEON::BI__builtin_neon_vgetq_lane_i32:
6296  case NEON::BI__builtin_neon_vgetq_lane_i64:
6297  case NEON::BI__builtin_neon_vgetq_lane_f32:
6298    return Builder.CreateExtractElement(Ops[0], Ops[1], "vget_lane");
6299
6300  case NEON::BI__builtin_neon_vrndns_f32: {
6301    Value *Arg = EmitScalarExpr(E->getArg(0));
6302    llvm::Type *Tys[] = {Arg->getType()};
6303    Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vrintn, Tys);
6304    return Builder.CreateCall(F, {Arg}, "vrndn"); }
6305
6306  case NEON::BI__builtin_neon_vset_lane_i8:
6307  case NEON::BI__builtin_neon_vset_lane_i16:
6308  case NEON::BI__builtin_neon_vset_lane_i32:
6309  case NEON::BI__builtin_neon_vset_lane_i64:
6310  case NEON::BI__builtin_neon_vset_lane_f32:
6311  case NEON::BI__builtin_neon_vsetq_lane_i8:
6312  case NEON::BI__builtin_neon_vsetq_lane_i16:
6313  case NEON::BI__builtin_neon_vsetq_lane_i32:
6314  case NEON::BI__builtin_neon_vsetq_lane_i64:
6315  case NEON::BI__builtin_neon_vsetq_lane_f32:
6316    return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
6317
6318  case NEON::BI__builtin_neon_vsha1h_u32:
6319    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1h), Ops,
6320                        "vsha1h");
6321  case NEON::BI__builtin_neon_vsha1cq_u32:
6322    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1c), Ops,
6323                        "vsha1h");
6324  case NEON::BI__builtin_neon_vsha1pq_u32:
6325    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1p), Ops,
6326                        "vsha1h");
6327  case NEON::BI__builtin_neon_vsha1mq_u32:
6328    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_sha1m), Ops,
6329                        "vsha1h");
6330
6331  // The ARM _MoveToCoprocessor builtins put the input register value as
6332  // the first argument, but the LLVM intrinsic expects it as the third one.
6333  case ARM::BI_MoveToCoprocessor:
6334  case ARM::BI_MoveToCoprocessor2: {
6335    Function *F = CGM.getIntrinsic(BuiltinID == ARM::BI_MoveToCoprocessor ?
6336                                   Intrinsic::arm_mcr : Intrinsic::arm_mcr2);
6337    return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0],
6338                                  Ops[3], Ops[4], Ops[5]});
6339  }
6340  case ARM::BI_BitScanForward:
6341  case ARM::BI_BitScanForward64:
6342    return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E);
6343  case ARM::BI_BitScanReverse:
6344  case ARM::BI_BitScanReverse64:
6345    return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E);
6346
6347  case ARM::BI_InterlockedAnd64:
6348    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E);
6349  case ARM::BI_InterlockedExchange64:
6350    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E);
6351  case ARM::BI_InterlockedExchangeAdd64:
6352    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E);
6353  case ARM::BI_InterlockedExchangeSub64:
6354    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E);
6355  case ARM::BI_InterlockedOr64:
6356    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E);
6357  case ARM::BI_InterlockedXor64:
6358    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E);
6359  case ARM::BI_InterlockedDecrement64:
6360    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E);
6361  case ARM::BI_InterlockedIncrement64:
6362    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E);
6363  case ARM::BI_InterlockedExchangeAdd8_acq:
6364  case ARM::BI_InterlockedExchangeAdd16_acq:
6365  case ARM::BI_InterlockedExchangeAdd_acq:
6366  case ARM::BI_InterlockedExchangeAdd64_acq:
6367    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_acq, E);
6368  case ARM::BI_InterlockedExchangeAdd8_rel:
6369  case ARM::BI_InterlockedExchangeAdd16_rel:
6370  case ARM::BI_InterlockedExchangeAdd_rel:
6371  case ARM::BI_InterlockedExchangeAdd64_rel:
6372    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_rel, E);
6373  case ARM::BI_InterlockedExchangeAdd8_nf:
6374  case ARM::BI_InterlockedExchangeAdd16_nf:
6375  case ARM::BI_InterlockedExchangeAdd_nf:
6376  case ARM::BI_InterlockedExchangeAdd64_nf:
6377    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_nf, E);
6378  case ARM::BI_InterlockedExchange8_acq:
6379  case ARM::BI_InterlockedExchange16_acq:
6380  case ARM::BI_InterlockedExchange_acq:
6381  case ARM::BI_InterlockedExchange64_acq:
6382    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_acq, E);
6383  case ARM::BI_InterlockedExchange8_rel:
6384  case ARM::BI_InterlockedExchange16_rel:
6385  case ARM::BI_InterlockedExchange_rel:
6386  case ARM::BI_InterlockedExchange64_rel:
6387    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_rel, E);
6388  case ARM::BI_InterlockedExchange8_nf:
6389  case ARM::BI_InterlockedExchange16_nf:
6390  case ARM::BI_InterlockedExchange_nf:
6391  case ARM::BI_InterlockedExchange64_nf:
6392    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_nf, E);
6393  case ARM::BI_InterlockedCompareExchange8_acq:
6394  case ARM::BI_InterlockedCompareExchange16_acq:
6395  case ARM::BI_InterlockedCompareExchange_acq:
6396  case ARM::BI_InterlockedCompareExchange64_acq:
6397    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_acq, E);
6398  case ARM::BI_InterlockedCompareExchange8_rel:
6399  case ARM::BI_InterlockedCompareExchange16_rel:
6400  case ARM::BI_InterlockedCompareExchange_rel:
6401  case ARM::BI_InterlockedCompareExchange64_rel:
6402    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_rel, E);
6403  case ARM::BI_InterlockedCompareExchange8_nf:
6404  case ARM::BI_InterlockedCompareExchange16_nf:
6405  case ARM::BI_InterlockedCompareExchange_nf:
6406  case ARM::BI_InterlockedCompareExchange64_nf:
6407    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_nf, E);
6408  case ARM::BI_InterlockedOr8_acq:
6409  case ARM::BI_InterlockedOr16_acq:
6410  case ARM::BI_InterlockedOr_acq:
6411  case ARM::BI_InterlockedOr64_acq:
6412    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_acq, E);
6413  case ARM::BI_InterlockedOr8_rel:
6414  case ARM::BI_InterlockedOr16_rel:
6415  case ARM::BI_InterlockedOr_rel:
6416  case ARM::BI_InterlockedOr64_rel:
6417    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_rel, E);
6418  case ARM::BI_InterlockedOr8_nf:
6419  case ARM::BI_InterlockedOr16_nf:
6420  case ARM::BI_InterlockedOr_nf:
6421  case ARM::BI_InterlockedOr64_nf:
6422    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_nf, E);
6423  case ARM::BI_InterlockedXor8_acq:
6424  case ARM::BI_InterlockedXor16_acq:
6425  case ARM::BI_InterlockedXor_acq:
6426  case ARM::BI_InterlockedXor64_acq:
6427    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_acq, E);
6428  case ARM::BI_InterlockedXor8_rel:
6429  case ARM::BI_InterlockedXor16_rel:
6430  case ARM::BI_InterlockedXor_rel:
6431  case ARM::BI_InterlockedXor64_rel:
6432    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_rel, E);
6433  case ARM::BI_InterlockedXor8_nf:
6434  case ARM::BI_InterlockedXor16_nf:
6435  case ARM::BI_InterlockedXor_nf:
6436  case ARM::BI_InterlockedXor64_nf:
6437    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_nf, E);
6438  case ARM::BI_InterlockedAnd8_acq:
6439  case ARM::BI_InterlockedAnd16_acq:
6440  case ARM::BI_InterlockedAnd_acq:
6441  case ARM::BI_InterlockedAnd64_acq:
6442    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_acq, E);
6443  case ARM::BI_InterlockedAnd8_rel:
6444  case ARM::BI_InterlockedAnd16_rel:
6445  case ARM::BI_InterlockedAnd_rel:
6446  case ARM::BI_InterlockedAnd64_rel:
6447    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_rel, E);
6448  case ARM::BI_InterlockedAnd8_nf:
6449  case ARM::BI_InterlockedAnd16_nf:
6450  case ARM::BI_InterlockedAnd_nf:
6451  case ARM::BI_InterlockedAnd64_nf:
6452    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_nf, E);
6453  case ARM::BI_InterlockedIncrement16_acq:
6454  case ARM::BI_InterlockedIncrement_acq:
6455  case ARM::BI_InterlockedIncrement64_acq:
6456    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_acq, E);
6457  case ARM::BI_InterlockedIncrement16_rel:
6458  case ARM::BI_InterlockedIncrement_rel:
6459  case ARM::BI_InterlockedIncrement64_rel:
6460    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_rel, E);
6461  case ARM::BI_InterlockedIncrement16_nf:
6462  case ARM::BI_InterlockedIncrement_nf:
6463  case ARM::BI_InterlockedIncrement64_nf:
6464    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_nf, E);
6465  case ARM::BI_InterlockedDecrement16_acq:
6466  case ARM::BI_InterlockedDecrement_acq:
6467  case ARM::BI_InterlockedDecrement64_acq:
6468    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_acq, E);
6469  case ARM::BI_InterlockedDecrement16_rel:
6470  case ARM::BI_InterlockedDecrement_rel:
6471  case ARM::BI_InterlockedDecrement64_rel:
6472    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_rel, E);
6473  case ARM::BI_InterlockedDecrement16_nf:
6474  case ARM::BI_InterlockedDecrement_nf:
6475  case ARM::BI_InterlockedDecrement64_nf:
6476    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_nf, E);
6477  }
6478
6479  // Get the last argument, which specifies the vector type.
6480  assert(HasExtraArg);
6481  llvm::APSInt Result;
6482  const Expr *Arg = E->getArg(E->getNumArgs()-1);
6483  if (!Arg->isIntegerConstantExpr(Result, getContext()))
6484    return nullptr;
6485
6486  if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f ||
6487      BuiltinID == ARM::BI__builtin_arm_vcvtr_d) {
6488    // Determine the overloaded type of this builtin.
6489    llvm::Type *Ty;
6490    if (BuiltinID == ARM::BI__builtin_arm_vcvtr_f)
6491      Ty = FloatTy;
6492    else
6493      Ty = DoubleTy;
6494
6495    // Determine whether this is an unsigned conversion or not.
6496    bool usgn = Result.getZExtValue() == 1;
6497    unsigned Int = usgn ? Intrinsic::arm_vcvtru : Intrinsic::arm_vcvtr;
6498
6499    // Call the appropriate intrinsic.
6500    Function *F = CGM.getIntrinsic(Int, Ty);
6501    return Builder.CreateCall(F, Ops, "vcvtr");
6502  }
6503
6504  // Determine the type of this overloaded NEON intrinsic.
6505  NeonTypeFlags Type(Result.getZExtValue());
6506  bool usgn = Type.isUnsigned();
6507  bool rightShift = false;
6508
6509  llvm::VectorType *VTy = GetNeonType(this, Type,
6510                                      getTarget().hasLegalHalfType());
6511  llvm::Type *Ty = VTy;
6512  if (!Ty)
6513    return nullptr;
6514
6515  // Many NEON builtins have identical semantics and uses in ARM and
6516  // AArch64. Emit these in a single function.
6517  auto IntrinsicMap = makeArrayRef(ARMSIMDIntrinsicMap);
6518  const NeonIntrinsicInfo *Builtin = findNeonIntrinsicInMap(
6519      IntrinsicMap, BuiltinID, NEONSIMDIntrinsicsProvenSorted);
6520  if (Builtin)
6521    return EmitCommonNeonBuiltinExpr(
6522        Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic,
6523        Builtin->NameHint, Builtin->TypeModifier, E, Ops, PtrOp0, PtrOp1, Arch);
6524
6525  unsigned Int;
6526  switch (BuiltinID) {
6527  default: return nullptr;
6528  case NEON::BI__builtin_neon_vld1q_lane_v:
6529    // Handle 64-bit integer elements as a special case.  Use shuffles of
6530    // one-element vectors to avoid poor code for i64 in the backend.
6531    if (VTy->getElementType()->isIntegerTy(64)) {
6532      // Extract the other lane.
6533      Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6534      uint32_t Lane = cast<ConstantInt>(Ops[2])->getZExtValue();
6535      Value *SV = llvm::ConstantVector::get(ConstantInt::get(Int32Ty, 1-Lane));
6536      Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV);
6537      // Load the value as a one-element vector.
6538      Ty = llvm::VectorType::get(VTy->getElementType(), 1);
6539      llvm::Type *Tys[] = {Ty, Int8PtrTy};
6540      Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vld1, Tys);
6541      Value *Align = getAlignmentValue32(PtrOp0);
6542      Value *Ld = Builder.CreateCall(F, {Ops[0], Align});
6543      // Combine them.
6544      uint32_t Indices[] = {1 - Lane, Lane};
6545      SV = llvm::ConstantDataVector::get(getLLVMContext(), Indices);
6546      return Builder.CreateShuffleVector(Ops[1], Ld, SV, "vld1q_lane");
6547    }
6548    LLVM_FALLTHROUGH;
6549  case NEON::BI__builtin_neon_vld1_lane_v: {
6550    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6551    PtrOp0 = Builder.CreateElementBitCast(PtrOp0, VTy->getElementType());
6552    Value *Ld = Builder.CreateLoad(PtrOp0);
6553    return Builder.CreateInsertElement(Ops[1], Ld, Ops[2], "vld1_lane");
6554  }
6555  case NEON::BI__builtin_neon_vqrshrn_n_v:
6556    Int =
6557      usgn ? Intrinsic::arm_neon_vqrshiftnu : Intrinsic::arm_neon_vqrshiftns;
6558    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n",
6559                        1, true);
6560  case NEON::BI__builtin_neon_vqrshrun_n_v:
6561    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqrshiftnsu, Ty),
6562                        Ops, "vqrshrun_n", 1, true);
6563  case NEON::BI__builtin_neon_vqshrn_n_v:
6564    Int = usgn ? Intrinsic::arm_neon_vqshiftnu : Intrinsic::arm_neon_vqshiftns;
6565    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n",
6566                        1, true);
6567  case NEON::BI__builtin_neon_vqshrun_n_v:
6568    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vqshiftnsu, Ty),
6569                        Ops, "vqshrun_n", 1, true);
6570  case NEON::BI__builtin_neon_vrecpe_v:
6571  case NEON::BI__builtin_neon_vrecpeq_v:
6572    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrecpe, Ty),
6573                        Ops, "vrecpe");
6574  case NEON::BI__builtin_neon_vrshrn_n_v:
6575    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vrshiftn, Ty),
6576                        Ops, "vrshrn_n", 1, true);
6577  case NEON::BI__builtin_neon_vrsra_n_v:
6578  case NEON::BI__builtin_neon_vrsraq_n_v:
6579    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6580    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6581    Ops[2] = EmitNeonShiftVector(Ops[2], Ty, true);
6582    Int = usgn ? Intrinsic::arm_neon_vrshiftu : Intrinsic::arm_neon_vrshifts;
6583    Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Ty), {Ops[1], Ops[2]});
6584    return Builder.CreateAdd(Ops[0], Ops[1], "vrsra_n");
6585  case NEON::BI__builtin_neon_vsri_n_v:
6586  case NEON::BI__builtin_neon_vsriq_n_v:
6587    rightShift = true;
6588    LLVM_FALLTHROUGH;
6589  case NEON::BI__builtin_neon_vsli_n_v:
6590  case NEON::BI__builtin_neon_vsliq_n_v:
6591    Ops[2] = EmitNeonShiftVector(Ops[2], Ty, rightShift);
6592    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vshiftins, Ty),
6593                        Ops, "vsli_n");
6594  case NEON::BI__builtin_neon_vsra_n_v:
6595  case NEON::BI__builtin_neon_vsraq_n_v:
6596    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
6597    Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n");
6598    return Builder.CreateAdd(Ops[0], Ops[1]);
6599  case NEON::BI__builtin_neon_vst1q_lane_v:
6600    // Handle 64-bit integer elements as a special case.  Use a shuffle to get
6601    // a one-element vector and avoid poor code for i64 in the backend.
6602    if (VTy->getElementType()->isIntegerTy(64)) {
6603      Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6604      Value *SV = llvm::ConstantVector::get(cast<llvm::Constant>(Ops[2]));
6605      Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV);
6606      Ops[2] = getAlignmentValue32(PtrOp0);
6607      llvm::Type *Tys[] = {Int8PtrTy, Ops[1]->getType()};
6608      return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::arm_neon_vst1,
6609                                                 Tys), Ops);
6610    }
6611    LLVM_FALLTHROUGH;
6612  case NEON::BI__builtin_neon_vst1_lane_v: {
6613    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
6614    Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]);
6615    Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
6616    auto St = Builder.CreateStore(Ops[1], Builder.CreateBitCast(PtrOp0, Ty));
6617    return St;
6618  }
6619  case NEON::BI__builtin_neon_vtbl1_v:
6620    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl1),
6621                        Ops, "vtbl1");
6622  case NEON::BI__builtin_neon_vtbl2_v:
6623    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl2),
6624                        Ops, "vtbl2");
6625  case NEON::BI__builtin_neon_vtbl3_v:
6626    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl3),
6627                        Ops, "vtbl3");
6628  case NEON::BI__builtin_neon_vtbl4_v:
6629    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbl4),
6630                        Ops, "vtbl4");
6631  case NEON::BI__builtin_neon_vtbx1_v:
6632    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx1),
6633                        Ops, "vtbx1");
6634  case NEON::BI__builtin_neon_vtbx2_v:
6635    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx2),
6636                        Ops, "vtbx2");
6637  case NEON::BI__builtin_neon_vtbx3_v:
6638    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx3),
6639                        Ops, "vtbx3");
6640  case NEON::BI__builtin_neon_vtbx4_v:
6641    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::arm_neon_vtbx4),
6642                        Ops, "vtbx4");
6643  }
6644}
6645
6646static Value *EmitAArch64TblBuiltinExpr(CodeGenFunction &CGF, unsigned BuiltinID,
6647                                      const CallExpr *E,
6648                                      SmallVectorImpl<Value *> &Ops,
6649                                      llvm::Triple::ArchType Arch) {
6650  unsigned int Int = 0;
6651  const char *s = nullptr;
6652
6653  switch (BuiltinID) {
6654  default:
6655    return nullptr;
6656  case NEON::BI__builtin_neon_vtbl1_v:
6657  case NEON::BI__builtin_neon_vqtbl1_v:
6658  case NEON::BI__builtin_neon_vqtbl1q_v:
6659  case NEON::BI__builtin_neon_vtbl2_v:
6660  case NEON::BI__builtin_neon_vqtbl2_v:
6661  case NEON::BI__builtin_neon_vqtbl2q_v:
6662  case NEON::BI__builtin_neon_vtbl3_v:
6663  case NEON::BI__builtin_neon_vqtbl3_v:
6664  case NEON::BI__builtin_neon_vqtbl3q_v:
6665  case NEON::BI__builtin_neon_vtbl4_v:
6666  case NEON::BI__builtin_neon_vqtbl4_v:
6667  case NEON::BI__builtin_neon_vqtbl4q_v:
6668    break;
6669  case NEON::BI__builtin_neon_vtbx1_v:
6670  case NEON::BI__builtin_neon_vqtbx1_v:
6671  case NEON::BI__builtin_neon_vqtbx1q_v:
6672  case NEON::BI__builtin_neon_vtbx2_v:
6673  case NEON::BI__builtin_neon_vqtbx2_v:
6674  case NEON::BI__builtin_neon_vqtbx2q_v:
6675  case NEON::BI__builtin_neon_vtbx3_v:
6676  case NEON::BI__builtin_neon_vqtbx3_v:
6677  case NEON::BI__builtin_neon_vqtbx3q_v:
6678  case NEON::BI__builtin_neon_vtbx4_v:
6679  case NEON::BI__builtin_neon_vqtbx4_v:
6680  case NEON::BI__builtin_neon_vqtbx4q_v:
6681    break;
6682  }
6683
6684  getNumArgs() >= 3", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 6684, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(E->getNumArgs() >= 3);
6685
6686  // Get the last argument, which specifies the vector type.
6687  llvm::APSInt Result;
6688  const Expr *Arg = E->getArg(E->getNumArgs() - 1);
6689  if (!Arg->isIntegerConstantExpr(Result, CGF.getContext()))
6690    return nullptr;
6691
6692  // Determine the type of this overloaded NEON intrinsic.
6693  NeonTypeFlags Type(Result.getZExtValue());
6694  llvm::VectorType *Ty = GetNeonType(&CGF, Type);
6695  if (!Ty)
6696    return nullptr;
6697
6698  CodeGen::CGBuilderTy &Builder = CGF.Builder;
6699
6700  // AArch64 scalar builtins are not overloaded, they do not have an extra
6701  // argument that specifies the vector type, need to handle each case.
6702  switch (BuiltinID) {
6703  case NEON::BI__builtin_neon_vtbl1_v: {
6704    return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 1), nullptr,
6705                              Ops[1], Ty, Intrinsic::aarch64_neon_tbl1,
6706                              "vtbl1");
6707  }
6708  case NEON::BI__builtin_neon_vtbl2_v: {
6709    return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 2), nullptr,
6710                              Ops[2], Ty, Intrinsic::aarch64_neon_tbl1,
6711                              "vtbl1");
6712  }
6713  case NEON::BI__builtin_neon_vtbl3_v: {
6714    return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 3), nullptr,
6715                              Ops[3], Ty, Intrinsic::aarch64_neon_tbl2,
6716                              "vtbl2");
6717  }
6718  case NEON::BI__builtin_neon_vtbl4_v: {
6719    return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(0, 4), nullptr,
6720                              Ops[4], Ty, Intrinsic::aarch64_neon_tbl2,
6721                              "vtbl2");
6722  }
6723  case NEON::BI__builtin_neon_vtbx1_v: {
6724    Value *TblRes =
6725        packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 1), nullptr, Ops[2],
6726                           Ty, Intrinsic::aarch64_neon_tbl1, "vtbl1");
6727
6728    llvm::Constant *EightV = ConstantInt::get(Ty, 8);
6729    Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[2], EightV);
6730    CmpRes = Builder.CreateSExt(CmpRes, Ty);
6731
6732    Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]);
6733    Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes);
6734    return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx");
6735  }
6736  case NEON::BI__builtin_neon_vtbx2_v: {
6737    return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 2), Ops[0],
6738                              Ops[3], Ty, Intrinsic::aarch64_neon_tbx1,
6739                              "vtbx1");
6740  }
6741  case NEON::BI__builtin_neon_vtbx3_v: {
6742    Value *TblRes =
6743        packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 3), nullptr, Ops[4],
6744                           Ty, Intrinsic::aarch64_neon_tbl2, "vtbl2");
6745
6746    llvm::Constant *TwentyFourV = ConstantInt::get(Ty, 24);
6747    Value *CmpRes = Builder.CreateICmp(ICmpInst::ICMP_UGE, Ops[4],
6748                                           TwentyFourV);
6749    CmpRes = Builder.CreateSExt(CmpRes, Ty);
6750
6751    Value *EltsFromInput = Builder.CreateAnd(CmpRes, Ops[0]);
6752    Value *EltsFromTbl = Builder.CreateAnd(Builder.CreateNot(CmpRes), TblRes);
6753    return Builder.CreateOr(EltsFromInput, EltsFromTbl, "vtbx");
6754  }
6755  case NEON::BI__builtin_neon_vtbx4_v: {
6756    return packTBLDVectorList(CGF, makeArrayRef(Ops).slice(1, 4), Ops[0],
6757                              Ops[5], Ty, Intrinsic::aarch64_neon_tbx2,
6758                              "vtbx2");
6759  }
6760  case NEON::BI__builtin_neon_vqtbl1_v:
6761  case NEON::BI__builtin_neon_vqtbl1q_v:
6762    Int = Intrinsic::aarch64_neon_tbl1; s = "vtbl1"; break;
6763  case NEON::BI__builtin_neon_vqtbl2_v:
6764  case NEON::BI__builtin_neon_vqtbl2q_v: {
6765    Int = Intrinsic::aarch64_neon_tbl2; s = "vtbl2"; break;
6766  case NEON::BI__builtin_neon_vqtbl3_v:
6767  case NEON::BI__builtin_neon_vqtbl3q_v:
6768    Int = Intrinsic::aarch64_neon_tbl3; s = "vtbl3"; break;
6769  case NEON::BI__builtin_neon_vqtbl4_v:
6770  case NEON::BI__builtin_neon_vqtbl4q_v:
6771    Int = Intrinsic::aarch64_neon_tbl4; s = "vtbl4"; break;
6772  case NEON::BI__builtin_neon_vqtbx1_v:
6773  case NEON::BI__builtin_neon_vqtbx1q_v:
6774    Int = Intrinsic::aarch64_neon_tbx1; s = "vtbx1"; break;
6775  case NEON::BI__builtin_neon_vqtbx2_v:
6776  case NEON::BI__builtin_neon_vqtbx2q_v:
6777    Int = Intrinsic::aarch64_neon_tbx2; s = "vtbx2"; break;
6778  case NEON::BI__builtin_neon_vqtbx3_v:
6779  case NEON::BI__builtin_neon_vqtbx3q_v:
6780    Int = Intrinsic::aarch64_neon_tbx3; s = "vtbx3"; break;
6781  case NEON::BI__builtin_neon_vqtbx4_v:
6782  case NEON::BI__builtin_neon_vqtbx4q_v:
6783    Int = Intrinsic::aarch64_neon_tbx4; s = "vtbx4"; break;
6784  }
6785  }
6786
6787  if (!Int)
6788    return nullptr;
6789
6790  Function *F = CGF.CGM.getIntrinsic(Int, Ty);
6791  return CGF.EmitNeonCall(F, Ops, s);
6792}
6793
6794Value *CodeGenFunction::vectorWrapScalar16(Value *Op) {
6795  llvm::Type *VTy = llvm::VectorType::get(Int16Ty, 4);
6796  Op = Builder.CreateBitCast(Op, Int16Ty);
6797  Value *V = UndefValue::get(VTy);
6798  llvm::Constant *CI = ConstantInt::get(SizeTy, 0);
6799  Op = Builder.CreateInsertElement(V, Op, CI);
6800  return Op;
6801}
6802
6803Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
6804                                               const CallExpr *E,
6805                                               llvm::Triple::ArchType Arch) {
6806  unsigned HintID = static_cast<unsigned>(-1);
6807  switch (BuiltinID) {
6808  default: break;
6809  case AArch64::BI__builtin_arm_nop:
6810    HintID = 0;
6811    break;
6812  case AArch64::BI__builtin_arm_yield:
6813  case AArch64::BI__yield:
6814    HintID = 1;
6815    break;
6816  case AArch64::BI__builtin_arm_wfe:
6817  case AArch64::BI__wfe:
6818    HintID = 2;
6819    break;
6820  case AArch64::BI__builtin_arm_wfi:
6821  case AArch64::BI__wfi:
6822    HintID = 3;
6823    break;
6824  case AArch64::BI__builtin_arm_sev:
6825  case AArch64::BI__sev:
6826    HintID = 4;
6827    break;
6828  case AArch64::BI__builtin_arm_sevl:
6829  case AArch64::BI__sevl:
6830    HintID = 5;
6831    break;
6832  }
6833
6834  if (HintID != static_cast<unsigned>(-1)) {
6835    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_hint);
6836    return Builder.CreateCall(F, llvm::ConstantInt::get(Int32Ty, HintID));
6837  }
6838
6839  if (BuiltinID == AArch64::BI__builtin_arm_prefetch) {
6840    Value *Address         = EmitScalarExpr(E->getArg(0));
6841    Value *RW              = EmitScalarExpr(E->getArg(1));
6842    Value *CacheLevel      = EmitScalarExpr(E->getArg(2));
6843    Value *RetentionPolicy = EmitScalarExpr(E->getArg(3));
6844    Value *IsData          = EmitScalarExpr(E->getArg(4));
6845
6846    Value *Locality = nullptr;
6847    if (cast<llvm::ConstantInt>(RetentionPolicy)->isZero()) {
6848      // Temporal fetch, needs to convert cache level to locality.
6849      Locality = llvm::ConstantInt::get(Int32Ty,
6850        -cast<llvm::ConstantInt>(CacheLevel)->getValue() + 3);
6851    } else {
6852      // Streaming fetch.
6853      Locality = llvm::ConstantInt::get(Int32Ty, 0);
6854    }
6855
6856    // FIXME: We need AArch64 specific LLVM intrinsic if we want to specify
6857    // PLDL3STRM or PLDL2STRM.
6858    Function *F = CGM.getIntrinsic(Intrinsic::prefetch);
6859    return Builder.CreateCall(F, {Address, RW, Locality, IsData});
6860  }
6861
6862  if (BuiltinID == AArch64::BI__builtin_arm_rbit) {
6863     (0) . __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 6864, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert((getContext().getTypeSize(E->getType()) == 32) &&
6864 (0) . __assert_fail ("(getContext().getTypeSize(E->getType()) == 32) && \"rbit of unusual size!\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 6864, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">           "rbit of unusual size!");
6865    llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
6866    return Builder.CreateCall(
6867        CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
6868  }
6869  if (BuiltinID == AArch64::BI__builtin_arm_rbit64) {
6870     (0) . __assert_fail ("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 6871, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert((getContext().getTypeSize(E->getType()) == 64) &&
6871 (0) . __assert_fail ("(getContext().getTypeSize(E->getType()) == 64) && \"rbit of unusual size!\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 6871, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">           "rbit of unusual size!");
6872    llvm::Value *Arg = EmitScalarExpr(E->getArg(0));
6873    return Builder.CreateCall(
6874        CGM.getIntrinsic(Intrinsic::bitreverse, Arg->getType()), Arg, "rbit");
6875  }
6876
6877  if (BuiltinID == AArch64::BI__clear_cache) {
6878     (0) . __assert_fail ("E->getNumArgs() == 2 && \"__clear_cache takes 2 arguments\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 6878, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(E->getNumArgs() == 2 && "__clear_cache takes 2 arguments");
6879    const FunctionDecl *FD = E->getDirectCallee();
6880    Value *Ops[2];
6881    for (unsigned i = 0; i < 2; i++)
6882      Ops[i] = EmitScalarExpr(E->getArg(i));
6883    llvm::Type *Ty = CGM.getTypes().ConvertType(FD->getType());
6884    llvm::FunctionType *FTy = cast<llvm::FunctionType>(Ty);
6885    StringRef Name = FD->getName();
6886    return EmitNounwindRuntimeCall(CGM.CreateRuntimeFunction(FTy, Name), Ops);
6887  }
6888
6889  if ((BuiltinID == AArch64::BI__builtin_arm_ldrex ||
6890      BuiltinID == AArch64::BI__builtin_arm_ldaex) &&
6891      getContext().getTypeSize(E->getType()) == 128) {
6892    Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex
6893                                       ? Intrinsic::aarch64_ldaxp
6894                                       : Intrinsic::aarch64_ldxp);
6895
6896    Value *LdPtr = EmitScalarExpr(E->getArg(0));
6897    Value *Val = Builder.CreateCall(F, Builder.CreateBitCast(LdPtr, Int8PtrTy),
6898                                    "ldxp");
6899
6900    Value *Val0 = Builder.CreateExtractValue(Val, 1);
6901    Value *Val1 = Builder.CreateExtractValue(Val, 0);
6902    llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128);
6903    Val0 = Builder.CreateZExt(Val0, Int128Ty);
6904    Val1 = Builder.CreateZExt(Val1, Int128Ty);
6905
6906    Value *ShiftCst = llvm::ConstantInt::get(Int128Ty, 64);
6907    Val = Builder.CreateShl(Val0, ShiftCst, "shl", true /* nuw */);
6908    Val = Builder.CreateOr(Val, Val1);
6909    return Builder.CreateBitCast(Val, ConvertType(E->getType()));
6910  } else if (BuiltinID == AArch64::BI__builtin_arm_ldrex ||
6911             BuiltinID == AArch64::BI__builtin_arm_ldaex) {
6912    Value *LoadAddr = EmitScalarExpr(E->getArg(0));
6913
6914    QualType Ty = E->getType();
6915    llvm::Type *RealResTy = ConvertType(Ty);
6916    llvm::Type *PtrTy = llvm::IntegerType::get(
6917        getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo();
6918    LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy);
6919
6920    Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex
6921                                       ? Intrinsic::aarch64_ldaxr
6922                                       : Intrinsic::aarch64_ldxr,
6923                                   PtrTy);
6924    Value *Val = Builder.CreateCall(F, LoadAddr, "ldxr");
6925
6926    if (RealResTy->isPointerTy())
6927      return Builder.CreateIntToPtr(Val, RealResTy);
6928
6929    llvm::Type *IntResTy = llvm::IntegerType::get(
6930        getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy));
6931    Val = Builder.CreateTruncOrBitCast(Val, IntResTy);
6932    return Builder.CreateBitCast(Val, RealResTy);
6933  }
6934
6935  if ((BuiltinID == AArch64::BI__builtin_arm_strex ||
6936       BuiltinID == AArch64::BI__builtin_arm_stlex) &&
6937      getContext().getTypeSize(E->getArg(0)->getType()) == 128) {
6938    Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex
6939                                       ? Intrinsic::aarch64_stlxp
6940                                       : Intrinsic::aarch64_stxp);
6941    llvm::Type *STy = llvm::StructType::get(Int64Ty, Int64Ty);
6942
6943    Address Tmp = CreateMemTemp(E->getArg(0)->getType());
6944    EmitAnyExprToMem(E->getArg(0), Tmp, Qualifiers(), /*init*/ true);
6945
6946    Tmp = Builder.CreateBitCast(Tmp, llvm::PointerType::getUnqual(STy));
6947    llvm::Value *Val = Builder.CreateLoad(Tmp);
6948
6949    Value *Arg0 = Builder.CreateExtractValue(Val, 0);
6950    Value *Arg1 = Builder.CreateExtractValue(Val, 1);
6951    Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)),
6952                                         Int8PtrTy);
6953    return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "stxp");
6954  }
6955
6956  if (BuiltinID == AArch64::BI__builtin_arm_strex ||
6957      BuiltinID == AArch64::BI__builtin_arm_stlex) {
6958    Value *StoreVal = EmitScalarExpr(E->getArg(0));
6959    Value *StoreAddr = EmitScalarExpr(E->getArg(1));
6960
6961    QualType Ty = E->getArg(0)->getType();
6962    llvm::Type *StoreTy = llvm::IntegerType::get(getLLVMContext(),
6963                                                 getContext().getTypeSize(Ty));
6964    StoreAddr = Builder.CreateBitCast(StoreAddr, StoreTy->getPointerTo());
6965
6966    if (StoreVal->getType()->isPointerTy())
6967      StoreVal = Builder.CreatePtrToInt(StoreVal, Int64Ty);
6968    else {
6969      llvm::Type *IntTy = llvm::IntegerType::get(
6970          getLLVMContext(),
6971          CGM.getDataLayout().getTypeSizeInBits(StoreVal->getType()));
6972      StoreVal = Builder.CreateBitCast(StoreVal, IntTy);
6973      StoreVal = Builder.CreateZExtOrBitCast(StoreVal, Int64Ty);
6974    }
6975
6976    Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_stlex
6977                                       ? Intrinsic::aarch64_stlxr
6978                                       : Intrinsic::aarch64_stxr,
6979                                   StoreAddr->getType());
6980    return Builder.CreateCall(F, {StoreVal, StoreAddr}, "stxr");
6981  }
6982
6983  if (BuiltinID == AArch64::BI__getReg) {
6984    Expr::EvalResult Result;
6985    if (!E->getArg(0)->EvaluateAsInt(Result, CGM.getContext()))
6986      llvm_unreachable("Sema will ensure that the parameter is constant");
6987
6988    llvm::APSInt Value = Result.Val.getInt();
6989    LLVMContext &Context = CGM.getLLVMContext();
6990    std::string Reg = Value == 31 ? "sp" : "x" + Value.toString(10);
6991
6992    llvm::Metadata *Ops[] = {llvm::MDString::get(Context, Reg)};
6993    llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops);
6994    llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName);
6995
6996    llvm::Function *F =
6997        CGM.getIntrinsic(llvm::Intrinsic::read_register, {Int64Ty});
6998    return Builder.CreateCall(F, Metadata);
6999  }
7000
7001  if (BuiltinID == AArch64::BI__builtin_arm_clrex) {
7002    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_clrex);
7003    return Builder.CreateCall(F);
7004  }
7005
7006  if (BuiltinID == AArch64::BI_ReadWriteBarrier)
7007    return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent,
7008                               llvm::SyncScope::SingleThread);
7009
7010  // CRC32
7011  Intrinsic::ID CRCIntrinsicID = Intrinsic::not_intrinsic;
7012  switch (BuiltinID) {
7013  case AArch64::BI__builtin_arm_crc32b:
7014    CRCIntrinsicID = Intrinsic::aarch64_crc32b; break;
7015  case AArch64::BI__builtin_arm_crc32cb:
7016    CRCIntrinsicID = Intrinsic::aarch64_crc32cb; break;
7017  case AArch64::BI__builtin_arm_crc32h:
7018    CRCIntrinsicID = Intrinsic::aarch64_crc32h; break;
7019  case AArch64::BI__builtin_arm_crc32ch:
7020    CRCIntrinsicID = Intrinsic::aarch64_crc32ch; break;
7021  case AArch64::BI__builtin_arm_crc32w:
7022    CRCIntrinsicID = Intrinsic::aarch64_crc32w; break;
7023  case AArch64::BI__builtin_arm_crc32cw:
7024    CRCIntrinsicID = Intrinsic::aarch64_crc32cw; break;
7025  case AArch64::BI__builtin_arm_crc32d:
7026    CRCIntrinsicID = Intrinsic::aarch64_crc32x; break;
7027  case AArch64::BI__builtin_arm_crc32cd:
7028    CRCIntrinsicID = Intrinsic::aarch64_crc32cx; break;
7029  }
7030
7031  if (CRCIntrinsicID != Intrinsic::not_intrinsic) {
7032    Value *Arg0 = EmitScalarExpr(E->getArg(0));
7033    Value *Arg1 = EmitScalarExpr(E->getArg(1));
7034    Function *F = CGM.getIntrinsic(CRCIntrinsicID);
7035
7036    llvm::Type *DataTy = F->getFunctionType()->getParamType(1);
7037    Arg1 = Builder.CreateZExtOrBitCast(Arg1, DataTy);
7038
7039    return Builder.CreateCall(F, {Arg0, Arg1});
7040  }
7041
7042  if (BuiltinID == AArch64::BI__builtin_arm_rsr ||
7043      BuiltinID == AArch64::BI__builtin_arm_rsr64 ||
7044      BuiltinID == AArch64::BI__builtin_arm_rsrp ||
7045      BuiltinID == AArch64::BI__builtin_arm_wsr ||
7046      BuiltinID == AArch64::BI__builtin_arm_wsr64 ||
7047      BuiltinID == AArch64::BI__builtin_arm_wsrp) {
7048
7049    bool IsRead = BuiltinID == AArch64::BI__builtin_arm_rsr ||
7050                  BuiltinID == AArch64::BI__builtin_arm_rsr64 ||
7051                  BuiltinID == AArch64::BI__builtin_arm_rsrp;
7052
7053    bool IsPointerBuiltin = BuiltinID == AArch64::BI__builtin_arm_rsrp ||
7054                            BuiltinID == AArch64::BI__builtin_arm_wsrp;
7055
7056    bool Is64Bit = BuiltinID != AArch64::BI__builtin_arm_rsr &&
7057                   BuiltinID != AArch64::BI__builtin_arm_wsr;
7058
7059    llvm::Type *ValueType;
7060    llvm::Type *RegisterType = Int64Ty;
7061    if (IsPointerBuiltin) {
7062      ValueType = VoidPtrTy;
7063    } else if (Is64Bit) {
7064      ValueType = Int64Ty;
7065    } else {
7066      ValueType = Int32Ty;
7067    }
7068
7069    return EmitSpecialRegisterBuiltin(*this, E, RegisterType, ValueType, IsRead);
7070  }
7071
7072  if (BuiltinID == AArch64::BI_ReadStatusReg ||
7073      BuiltinID == AArch64::BI_WriteStatusReg) {
7074    LLVMContext &Context = CGM.getLLVMContext();
7075
7076    unsigned SysReg =
7077      E->getArg(0)->EvaluateKnownConstInt(getContext()).getZExtValue();
7078
7079    std::string SysRegStr;
7080    llvm::raw_string_ostream(SysRegStr) <<
7081                       ((1 << 1) | ((SysReg >> 14) & 1))  << ":" <<
7082                       ((SysReg >> 11) & 7)               << ":" <<
7083                       ((SysReg >> 7)  & 15)              << ":" <<
7084                       ((SysReg >> 3)  & 15)              << ":" <<
7085                       ( SysReg        & 7);
7086
7087    llvm::Metadata *Ops[] = { llvm::MDString::get(Context, SysRegStr) };
7088    llvm::MDNode *RegName = llvm::MDNode::get(Context, Ops);
7089    llvm::Value *Metadata = llvm::MetadataAsValue::get(Context, RegName);
7090
7091    llvm::Type *RegisterType = Int64Ty;
7092    llvm::Type *Types[] = { RegisterType };
7093
7094    if (BuiltinID == AArch64::BI_ReadStatusReg) {
7095      llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::read_register, Types);
7096
7097      return Builder.CreateCall(F, Metadata);
7098    }
7099
7100    llvm::Function *F = CGM.getIntrinsic(llvm::Intrinsic::write_register, Types);
7101    llvm::Value *ArgValue = EmitScalarExpr(E->getArg(1));
7102
7103    return Builder.CreateCall(F, { Metadata, ArgValue });
7104  }
7105
7106  if (BuiltinID == AArch64::BI_AddressOfReturnAddress) {
7107    llvm::Function *F = CGM.getIntrinsic(Intrinsic::addressofreturnaddress);
7108    return Builder.CreateCall(F);
7109  }
7110
7111  // Find out if any arguments are required to be integer constant
7112  // expressions.
7113  unsigned ICEArguments = 0;
7114  ASTContext::GetBuiltinTypeError Error;
7115  getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
7116   (0) . __assert_fail ("Error == ASTContext..GE_None && \"Should not codegen an error\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 7116, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Error == ASTContext::GE_None && "Should not codegen an error");
7117
7118  llvm::SmallVector<Value*, 4> Ops;
7119  for (unsigned i = 0, e = E->getNumArgs() - 1; i != e; i++) {
7120    if ((ICEArguments & (1 << i)) == 0) {
7121      Ops.push_back(EmitScalarExpr(E->getArg(i)));
7122    } else {
7123      // If this is required to be a constant, constant fold it so that we know
7124      // that the generated intrinsic gets a ConstantInt.
7125      llvm::APSInt Result;
7126      bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext());
7127       (0) . __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 7127, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(IsConst && "Constant arg isn't actually constant?");
7128      (void)IsConst;
7129      Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result));
7130    }
7131  }
7132
7133  auto SISDMap = makeArrayRef(AArch64SISDIntrinsicMap);
7134  const NeonIntrinsicInfo *Builtin = findNeonIntrinsicInMap(
7135      SISDMap, BuiltinID, AArch64SISDIntrinsicsProvenSorted);
7136
7137  if (Builtin) {
7138    Ops.push_back(EmitScalarExpr(E->getArg(E->getNumArgs() - 1)));
7139    Value *Result = EmitCommonNeonSISDBuiltinExpr(*this, *Builtin, Ops, E);
7140     (0) . __assert_fail ("Result && \"SISD intrinsic should have been handled\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 7140, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Result && "SISD intrinsic should have been handled");
7141    return Result;
7142  }
7143
7144  llvm::APSInt Result;
7145  const Expr *Arg = E->getArg(E->getNumArgs()-1);
7146  NeonTypeFlags Type(0);
7147  if (Arg->isIntegerConstantExpr(Result, getContext()))
7148    // Determine the type of this overloaded NEON intrinsic.
7149    Type = NeonTypeFlags(Result.getZExtValue());
7150
7151  bool usgn = Type.isUnsigned();
7152  bool quad = Type.isQuad();
7153
7154  // Handle non-overloaded intrinsics first.
7155  switch (BuiltinID) {
7156  default: break;
7157  case NEON::BI__builtin_neon_vabsh_f16:
7158    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7159    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::fabs, HalfTy), Ops, "vabs");
7160  case NEON::BI__builtin_neon_vldrq_p128: {
7161    llvm::Type *Int128Ty = llvm::Type::getIntNTy(getLLVMContext(), 128);
7162    llvm::Type *Int128PTy = llvm::PointerType::get(Int128Ty, 0);
7163    Value *Ptr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int128PTy);
7164    return Builder.CreateAlignedLoad(Int128Ty, Ptr,
7165                                     CharUnits::fromQuantity(16));
7166  }
7167  case NEON::BI__builtin_neon_vstrq_p128: {
7168    llvm::Type *Int128PTy = llvm::Type::getIntNPtrTy(getLLVMContext(), 128);
7169    Value *Ptr = Builder.CreateBitCast(Ops[0], Int128PTy);
7170    return Builder.CreateDefaultAlignedStore(EmitScalarExpr(E->getArg(1)), Ptr);
7171  }
7172  case NEON::BI__builtin_neon_vcvts_u32_f32:
7173  case NEON::BI__builtin_neon_vcvtd_u64_f64:
7174    usgn = true;
7175    LLVM_FALLTHROUGH;
7176  case NEON::BI__builtin_neon_vcvts_s32_f32:
7177  case NEON::BI__builtin_neon_vcvtd_s64_f64: {
7178    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7179    bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64;
7180    llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty;
7181    llvm::Type *FTy = Is64 ? DoubleTy : FloatTy;
7182    Ops[0] = Builder.CreateBitCast(Ops[0], FTy);
7183    if (usgn)
7184      return Builder.CreateFPToUI(Ops[0], InTy);
7185    return Builder.CreateFPToSI(Ops[0], InTy);
7186  }
7187  case NEON::BI__builtin_neon_vcvts_f32_u32:
7188  case NEON::BI__builtin_neon_vcvtd_f64_u64:
7189    usgn = true;
7190    LLVM_FALLTHROUGH;
7191  case NEON::BI__builtin_neon_vcvts_f32_s32:
7192  case NEON::BI__builtin_neon_vcvtd_f64_s64: {
7193    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7194    bool Is64 = Ops[0]->getType()->getPrimitiveSizeInBits() == 64;
7195    llvm::Type *InTy = Is64 ? Int64Ty : Int32Ty;
7196    llvm::Type *FTy = Is64 ? DoubleTy : FloatTy;
7197    Ops[0] = Builder.CreateBitCast(Ops[0], InTy);
7198    if (usgn)
7199      return Builder.CreateUIToFP(Ops[0], FTy);
7200    return Builder.CreateSIToFP(Ops[0], FTy);
7201  }
7202  case NEON::BI__builtin_neon_vcvth_f16_u16:
7203  case NEON::BI__builtin_neon_vcvth_f16_u32:
7204  case NEON::BI__builtin_neon_vcvth_f16_u64:
7205    usgn = true;
7206    LLVM_FALLTHROUGH;
7207  case NEON::BI__builtin_neon_vcvth_f16_s16:
7208  case NEON::BI__builtin_neon_vcvth_f16_s32:
7209  case NEON::BI__builtin_neon_vcvth_f16_s64: {
7210    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7211    llvm::Type *FTy = HalfTy;
7212    llvm::Type *InTy;
7213    if (Ops[0]->getType()->getPrimitiveSizeInBits() == 64)
7214      InTy = Int64Ty;
7215    else if (Ops[0]->getType()->getPrimitiveSizeInBits() == 32)
7216      InTy = Int32Ty;
7217    else
7218      InTy = Int16Ty;
7219    Ops[0] = Builder.CreateBitCast(Ops[0], InTy);
7220    if (usgn)
7221      return Builder.CreateUIToFP(Ops[0], FTy);
7222    return Builder.CreateSIToFP(Ops[0], FTy);
7223  }
7224  case NEON::BI__builtin_neon_vcvth_u16_f16:
7225    usgn = true;
7226    LLVM_FALLTHROUGH;
7227  case NEON::BI__builtin_neon_vcvth_s16_f16: {
7228    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7229    Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy);
7230    if (usgn)
7231      return Builder.CreateFPToUI(Ops[0], Int16Ty);
7232    return Builder.CreateFPToSI(Ops[0], Int16Ty);
7233  }
7234  case NEON::BI__builtin_neon_vcvth_u32_f16:
7235    usgn = true;
7236    LLVM_FALLTHROUGH;
7237  case NEON::BI__builtin_neon_vcvth_s32_f16: {
7238    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7239    Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy);
7240    if (usgn)
7241      return Builder.CreateFPToUI(Ops[0], Int32Ty);
7242    return Builder.CreateFPToSI(Ops[0], Int32Ty);
7243  }
7244  case NEON::BI__builtin_neon_vcvth_u64_f16:
7245    usgn = true;
7246    LLVM_FALLTHROUGH;
7247  case NEON::BI__builtin_neon_vcvth_s64_f16: {
7248    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7249    Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy);
7250    if (usgn)
7251      return Builder.CreateFPToUI(Ops[0], Int64Ty);
7252    return Builder.CreateFPToSI(Ops[0], Int64Ty);
7253  }
7254  case NEON::BI__builtin_neon_vcvtah_u16_f16:
7255  case NEON::BI__builtin_neon_vcvtmh_u16_f16:
7256  case NEON::BI__builtin_neon_vcvtnh_u16_f16:
7257  case NEON::BI__builtin_neon_vcvtph_u16_f16:
7258  case NEON::BI__builtin_neon_vcvtah_s16_f16:
7259  case NEON::BI__builtin_neon_vcvtmh_s16_f16:
7260  case NEON::BI__builtin_neon_vcvtnh_s16_f16:
7261  case NEON::BI__builtin_neon_vcvtph_s16_f16: {
7262    unsigned Int;
7263    llvm::Type* InTy = Int32Ty;
7264    llvm::Type* FTy  = HalfTy;
7265    llvm::Type *Tys[2] = {InTy, FTy};
7266    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7267    switch (BuiltinID) {
7268    default: llvm_unreachable("missing builtin ID in switch!");
7269    case NEON::BI__builtin_neon_vcvtah_u16_f16:
7270      Int = Intrinsic::aarch64_neon_fcvtau; break;
7271    case NEON::BI__builtin_neon_vcvtmh_u16_f16:
7272      Int = Intrinsic::aarch64_neon_fcvtmu; break;
7273    case NEON::BI__builtin_neon_vcvtnh_u16_f16:
7274      Int = Intrinsic::aarch64_neon_fcvtnu; break;
7275    case NEON::BI__builtin_neon_vcvtph_u16_f16:
7276      Int = Intrinsic::aarch64_neon_fcvtpu; break;
7277    case NEON::BI__builtin_neon_vcvtah_s16_f16:
7278      Int = Intrinsic::aarch64_neon_fcvtas; break;
7279    case NEON::BI__builtin_neon_vcvtmh_s16_f16:
7280      Int = Intrinsic::aarch64_neon_fcvtms; break;
7281    case NEON::BI__builtin_neon_vcvtnh_s16_f16:
7282      Int = Intrinsic::aarch64_neon_fcvtns; break;
7283    case NEON::BI__builtin_neon_vcvtph_s16_f16:
7284      Int = Intrinsic::aarch64_neon_fcvtps; break;
7285    }
7286    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvt");
7287    return Builder.CreateTrunc(Ops[0], Int16Ty);
7288  }
7289  case NEON::BI__builtin_neon_vcaleh_f16:
7290  case NEON::BI__builtin_neon_vcalth_f16:
7291  case NEON::BI__builtin_neon_vcageh_f16:
7292  case NEON::BI__builtin_neon_vcagth_f16: {
7293    unsigned Int;
7294    llvm::Type* InTy = Int32Ty;
7295    llvm::Type* FTy  = HalfTy;
7296    llvm::Type *Tys[2] = {InTy, FTy};
7297    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7298    switch (BuiltinID) {
7299    default: llvm_unreachable("missing builtin ID in switch!");
7300    case NEON::BI__builtin_neon_vcageh_f16:
7301      Int = Intrinsic::aarch64_neon_facge; break;
7302    case NEON::BI__builtin_neon_vcagth_f16:
7303      Int = Intrinsic::aarch64_neon_facgt; break;
7304    case NEON::BI__builtin_neon_vcaleh_f16:
7305      Int = Intrinsic::aarch64_neon_facge; std::swap(Ops[0], Ops[1]); break;
7306    case NEON::BI__builtin_neon_vcalth_f16:
7307      Int = Intrinsic::aarch64_neon_facgt; std::swap(Ops[0], Ops[1]); break;
7308    }
7309    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "facg");
7310    return Builder.CreateTrunc(Ops[0], Int16Ty);
7311  }
7312  case NEON::BI__builtin_neon_vcvth_n_s16_f16:
7313  case NEON::BI__builtin_neon_vcvth_n_u16_f16: {
7314    unsigned Int;
7315    llvm::Type* InTy = Int32Ty;
7316    llvm::Type* FTy  = HalfTy;
7317    llvm::Type *Tys[2] = {InTy, FTy};
7318    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7319    switch (BuiltinID) {
7320    default: llvm_unreachable("missing builtin ID in switch!");
7321    case NEON::BI__builtin_neon_vcvth_n_s16_f16:
7322      Int = Intrinsic::aarch64_neon_vcvtfp2fxs; break;
7323    case NEON::BI__builtin_neon_vcvth_n_u16_f16:
7324      Int = Intrinsic::aarch64_neon_vcvtfp2fxu; break;
7325    }
7326    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n");
7327    return Builder.CreateTrunc(Ops[0], Int16Ty);
7328  }
7329  case NEON::BI__builtin_neon_vcvth_n_f16_s16:
7330  case NEON::BI__builtin_neon_vcvth_n_f16_u16: {
7331    unsigned Int;
7332    llvm::Type* FTy  = HalfTy;
7333    llvm::Type* InTy = Int32Ty;
7334    llvm::Type *Tys[2] = {FTy, InTy};
7335    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7336    switch (BuiltinID) {
7337    default: llvm_unreachable("missing builtin ID in switch!");
7338    case NEON::BI__builtin_neon_vcvth_n_f16_s16:
7339      Int = Intrinsic::aarch64_neon_vcvtfxs2fp;
7340      Ops[0] = Builder.CreateSExt(Ops[0], InTy, "sext");
7341      break;
7342    case NEON::BI__builtin_neon_vcvth_n_f16_u16:
7343      Int = Intrinsic::aarch64_neon_vcvtfxu2fp;
7344      Ops[0] = Builder.CreateZExt(Ops[0], InTy);
7345      break;
7346    }
7347    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "fcvth_n");
7348  }
7349  case NEON::BI__builtin_neon_vpaddd_s64: {
7350    llvm::Type *Ty = llvm::VectorType::get(Int64Ty, 2);
7351    Value *Vec = EmitScalarExpr(E->getArg(0));
7352    // The vector is v2f64, so make sure it's bitcast to that.
7353    Vec = Builder.CreateBitCast(Vec, Ty, "v2i64");
7354    llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0);
7355    llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1);
7356    Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0");
7357    Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1");
7358    // Pairwise addition of a v2f64 into a scalar f64.
7359    return Builder.CreateAdd(Op0, Op1, "vpaddd");
7360  }
7361  case NEON::BI__builtin_neon_vpaddd_f64: {
7362    llvm::Type *Ty =
7363      llvm::VectorType::get(DoubleTy, 2);
7364    Value *Vec = EmitScalarExpr(E->getArg(0));
7365    // The vector is v2f64, so make sure it's bitcast to that.
7366    Vec = Builder.CreateBitCast(Vec, Ty, "v2f64");
7367    llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0);
7368    llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1);
7369    Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0");
7370    Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1");
7371    // Pairwise addition of a v2f64 into a scalar f64.
7372    return Builder.CreateFAdd(Op0, Op1, "vpaddd");
7373  }
7374  case NEON::BI__builtin_neon_vpadds_f32: {
7375    llvm::Type *Ty =
7376      llvm::VectorType::get(FloatTy, 2);
7377    Value *Vec = EmitScalarExpr(E->getArg(0));
7378    // The vector is v2f32, so make sure it's bitcast to that.
7379    Vec = Builder.CreateBitCast(Vec, Ty, "v2f32");
7380    llvm::Value *Idx0 = llvm::ConstantInt::get(SizeTy, 0);
7381    llvm::Value *Idx1 = llvm::ConstantInt::get(SizeTy, 1);
7382    Value *Op0 = Builder.CreateExtractElement(Vec, Idx0, "lane0");
7383    Value *Op1 = Builder.CreateExtractElement(Vec, Idx1, "lane1");
7384    // Pairwise addition of a v2f32 into a scalar f32.
7385    return Builder.CreateFAdd(Op0, Op1, "vpaddd");
7386  }
7387  case NEON::BI__builtin_neon_vceqzd_s64:
7388  case NEON::BI__builtin_neon_vceqzd_f64:
7389  case NEON::BI__builtin_neon_vceqzs_f32:
7390  case NEON::BI__builtin_neon_vceqzh_f16:
7391    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7392    return EmitAArch64CompareBuiltinExpr(
7393        Ops[0], ConvertType(E->getCallReturnType(getContext())),
7394        ICmpInst::FCMP_OEQ, ICmpInst::ICMP_EQ, "vceqz");
7395  case NEON::BI__builtin_neon_vcgezd_s64:
7396  case NEON::BI__builtin_neon_vcgezd_f64:
7397  case NEON::BI__builtin_neon_vcgezs_f32:
7398  case NEON::BI__builtin_neon_vcgezh_f16:
7399    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7400    return EmitAArch64CompareBuiltinExpr(
7401        Ops[0], ConvertType(E->getCallReturnType(getContext())),
7402        ICmpInst::FCMP_OGE, ICmpInst::ICMP_SGE, "vcgez");
7403  case NEON::BI__builtin_neon_vclezd_s64:
7404  case NEON::BI__builtin_neon_vclezd_f64:
7405  case NEON::BI__builtin_neon_vclezs_f32:
7406  case NEON::BI__builtin_neon_vclezh_f16:
7407    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7408    return EmitAArch64CompareBuiltinExpr(
7409        Ops[0], ConvertType(E->getCallReturnType(getContext())),
7410        ICmpInst::FCMP_OLE, ICmpInst::ICMP_SLE, "vclez");
7411  case NEON::BI__builtin_neon_vcgtzd_s64:
7412  case NEON::BI__builtin_neon_vcgtzd_f64:
7413  case NEON::BI__builtin_neon_vcgtzs_f32:
7414  case NEON::BI__builtin_neon_vcgtzh_f16:
7415    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7416    return EmitAArch64CompareBuiltinExpr(
7417        Ops[0], ConvertType(E->getCallReturnType(getContext())),
7418        ICmpInst::FCMP_OGT, ICmpInst::ICMP_SGT, "vcgtz");
7419  case NEON::BI__builtin_neon_vcltzd_s64:
7420  case NEON::BI__builtin_neon_vcltzd_f64:
7421  case NEON::BI__builtin_neon_vcltzs_f32:
7422  case NEON::BI__builtin_neon_vcltzh_f16:
7423    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7424    return EmitAArch64CompareBuiltinExpr(
7425        Ops[0], ConvertType(E->getCallReturnType(getContext())),
7426        ICmpInst::FCMP_OLT, ICmpInst::ICMP_SLT, "vcltz");
7427
7428  case NEON::BI__builtin_neon_vceqzd_u64: {
7429    Ops.push_back(EmitScalarExpr(E->getArg(0)));
7430    Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty);
7431    Ops[0] =
7432        Builder.CreateICmpEQ(Ops[0], llvm::Constant::getNullValue(Int64Ty));
7433    return Builder.CreateSExt(Ops[0], Int64Ty, "vceqzd");
7434  }
7435  case NEON::BI__builtin_neon_vceqd_f64:
7436  case NEON::BI__builtin_neon_vcled_f64:
7437  case NEON::BI__builtin_neon_vcltd_f64:
7438  case NEON::BI__builtin_neon_vcged_f64:
7439  case NEON::BI__builtin_neon_vcgtd_f64: {
7440    llvm::CmpInst::Predicate P;
7441    switch (BuiltinID) {
7442    default: llvm_unreachable("missing builtin ID in switch!");
7443    case NEON::BI__builtin_neon_vceqd_f64: P = llvm::FCmpInst::FCMP_OEQ; break;
7444    case NEON::BI__builtin_neon_vcled_f64: P = llvm::FCmpInst::FCMP_OLE; break;
7445    case NEON::BI__builtin_neon_vcltd_f64: P = llvm::FCmpInst::FCMP_OLT; break;
7446    case NEON::BI__builtin_neon_vcged_f64: P = llvm::FCmpInst::FCMP_OGE; break;
7447    case NEON::BI__builtin_neon_vcgtd_f64: P = llvm::FCmpInst::FCMP_OGT; break;
7448    }
7449    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7450    Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
7451    Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy);
7452    Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]);
7453    return Builder.CreateSExt(Ops[0], Int64Ty, "vcmpd");
7454  }
7455  case NEON::BI__builtin_neon_vceqs_f32:
7456  case NEON::BI__builtin_neon_vcles_f32:
7457  case NEON::BI__builtin_neon_vclts_f32:
7458  case NEON::BI__builtin_neon_vcges_f32:
7459  case NEON::BI__builtin_neon_vcgts_f32: {
7460    llvm::CmpInst::Predicate P;
7461    switch (BuiltinID) {
7462    default: llvm_unreachable("missing builtin ID in switch!");
7463    case NEON::BI__builtin_neon_vceqs_f32: P = llvm::FCmpInst::FCMP_OEQ; break;
7464    case NEON::BI__builtin_neon_vcles_f32: P = llvm::FCmpInst::FCMP_OLE; break;
7465    case NEON::BI__builtin_neon_vclts_f32: P = llvm::FCmpInst::FCMP_OLT; break;
7466    case NEON::BI__builtin_neon_vcges_f32: P = llvm::FCmpInst::FCMP_OGE; break;
7467    case NEON::BI__builtin_neon_vcgts_f32: P = llvm::FCmpInst::FCMP_OGT; break;
7468    }
7469    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7470    Ops[0] = Builder.CreateBitCast(Ops[0], FloatTy);
7471    Ops[1] = Builder.CreateBitCast(Ops[1], FloatTy);
7472    Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]);
7473    return Builder.CreateSExt(Ops[0], Int32Ty, "vcmpd");
7474  }
7475  case NEON::BI__builtin_neon_vceqh_f16:
7476  case NEON::BI__builtin_neon_vcleh_f16:
7477  case NEON::BI__builtin_neon_vclth_f16:
7478  case NEON::BI__builtin_neon_vcgeh_f16:
7479  case NEON::BI__builtin_neon_vcgth_f16: {
7480    llvm::CmpInst::Predicate P;
7481    switch (BuiltinID) {
7482    default: llvm_unreachable("missing builtin ID in switch!");
7483    case NEON::BI__builtin_neon_vceqh_f16: P = llvm::FCmpInst::FCMP_OEQ; break;
7484    case NEON::BI__builtin_neon_vcleh_f16: P = llvm::FCmpInst::FCMP_OLE; break;
7485    case NEON::BI__builtin_neon_vclth_f16: P = llvm::FCmpInst::FCMP_OLT; break;
7486    case NEON::BI__builtin_neon_vcgeh_f16: P = llvm::FCmpInst::FCMP_OGE; break;
7487    case NEON::BI__builtin_neon_vcgth_f16: P = llvm::FCmpInst::FCMP_OGT; break;
7488    }
7489    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7490    Ops[0] = Builder.CreateBitCast(Ops[0], HalfTy);
7491    Ops[1] = Builder.CreateBitCast(Ops[1], HalfTy);
7492    Ops[0] = Builder.CreateFCmp(P, Ops[0], Ops[1]);
7493    return Builder.CreateSExt(Ops[0], Int16Ty, "vcmpd");
7494  }
7495  case NEON::BI__builtin_neon_vceqd_s64:
7496  case NEON::BI__builtin_neon_vceqd_u64:
7497  case NEON::BI__builtin_neon_vcgtd_s64:
7498  case NEON::BI__builtin_neon_vcgtd_u64:
7499  case NEON::BI__builtin_neon_vcltd_s64:
7500  case NEON::BI__builtin_neon_vcltd_u64:
7501  case NEON::BI__builtin_neon_vcged_u64:
7502  case NEON::BI__builtin_neon_vcged_s64:
7503  case NEON::BI__builtin_neon_vcled_u64:
7504  case NEON::BI__builtin_neon_vcled_s64: {
7505    llvm::CmpInst::Predicate P;
7506    switch (BuiltinID) {
7507    default: llvm_unreachable("missing builtin ID in switch!");
7508    case NEON::BI__builtin_neon_vceqd_s64:
7509    case NEON::BI__builtin_neon_vceqd_u64:P = llvm::ICmpInst::ICMP_EQ;break;
7510    case NEON::BI__builtin_neon_vcgtd_s64:P = llvm::ICmpInst::ICMP_SGT;break;
7511    case NEON::BI__builtin_neon_vcgtd_u64:P = llvm::ICmpInst::ICMP_UGT;break;
7512    case NEON::BI__builtin_neon_vcltd_s64:P = llvm::ICmpInst::ICMP_SLT;break;
7513    case NEON::BI__builtin_neon_vcltd_u64:P = llvm::ICmpInst::ICMP_ULT;break;
7514    case NEON::BI__builtin_neon_vcged_u64:P = llvm::ICmpInst::ICMP_UGE;break;
7515    case NEON::BI__builtin_neon_vcged_s64:P = llvm::ICmpInst::ICMP_SGE;break;
7516    case NEON::BI__builtin_neon_vcled_u64:P = llvm::ICmpInst::ICMP_ULE;break;
7517    case NEON::BI__builtin_neon_vcled_s64:P = llvm::ICmpInst::ICMP_SLE;break;
7518    }
7519    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7520    Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty);
7521    Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
7522    Ops[0] = Builder.CreateICmp(P, Ops[0], Ops[1]);
7523    return Builder.CreateSExt(Ops[0], Int64Ty, "vceqd");
7524  }
7525  case NEON::BI__builtin_neon_vtstd_s64:
7526  case NEON::BI__builtin_neon_vtstd_u64: {
7527    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7528    Ops[0] = Builder.CreateBitCast(Ops[0], Int64Ty);
7529    Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
7530    Ops[0] = Builder.CreateAnd(Ops[0], Ops[1]);
7531    Ops[0] = Builder.CreateICmp(ICmpInst::ICMP_NE, Ops[0],
7532                                llvm::Constant::getNullValue(Int64Ty));
7533    return Builder.CreateSExt(Ops[0], Int64Ty, "vtstd");
7534  }
7535  case NEON::BI__builtin_neon_vset_lane_i8:
7536  case NEON::BI__builtin_neon_vset_lane_i16:
7537  case NEON::BI__builtin_neon_vset_lane_i32:
7538  case NEON::BI__builtin_neon_vset_lane_i64:
7539  case NEON::BI__builtin_neon_vset_lane_f32:
7540  case NEON::BI__builtin_neon_vsetq_lane_i8:
7541  case NEON::BI__builtin_neon_vsetq_lane_i16:
7542  case NEON::BI__builtin_neon_vsetq_lane_i32:
7543  case NEON::BI__builtin_neon_vsetq_lane_i64:
7544  case NEON::BI__builtin_neon_vsetq_lane_f32:
7545    Ops.push_back(EmitScalarExpr(E->getArg(2)));
7546    return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
7547  case NEON::BI__builtin_neon_vset_lane_f64:
7548    // The vector type needs a cast for the v1f64 variant.
7549    Ops[1] = Builder.CreateBitCast(Ops[1],
7550                                   llvm::VectorType::get(DoubleTy, 1));
7551    Ops.push_back(EmitScalarExpr(E->getArg(2)));
7552    return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
7553  case NEON::BI__builtin_neon_vsetq_lane_f64:
7554    // The vector type needs a cast for the v2f64 variant.
7555    Ops[1] = Builder.CreateBitCast(Ops[1],
7556        llvm::VectorType::get(DoubleTy, 2));
7557    Ops.push_back(EmitScalarExpr(E->getArg(2)));
7558    return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vset_lane");
7559
7560  case NEON::BI__builtin_neon_vget_lane_i8:
7561  case NEON::BI__builtin_neon_vdupb_lane_i8:
7562    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int8Ty, 8));
7563    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7564                                        "vget_lane");
7565  case NEON::BI__builtin_neon_vgetq_lane_i8:
7566  case NEON::BI__builtin_neon_vdupb_laneq_i8:
7567    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int8Ty, 16));
7568    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7569                                        "vgetq_lane");
7570  case NEON::BI__builtin_neon_vget_lane_i16:
7571  case NEON::BI__builtin_neon_vduph_lane_i16:
7572    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int16Ty, 4));
7573    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7574                                        "vget_lane");
7575  case NEON::BI__builtin_neon_vgetq_lane_i16:
7576  case NEON::BI__builtin_neon_vduph_laneq_i16:
7577    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int16Ty, 8));
7578    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7579                                        "vgetq_lane");
7580  case NEON::BI__builtin_neon_vget_lane_i32:
7581  case NEON::BI__builtin_neon_vdups_lane_i32:
7582    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 2));
7583    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7584                                        "vget_lane");
7585  case NEON::BI__builtin_neon_vdups_lane_f32:
7586    Ops[0] = Builder.CreateBitCast(Ops[0],
7587        llvm::VectorType::get(FloatTy, 2));
7588    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7589                                        "vdups_lane");
7590  case NEON::BI__builtin_neon_vgetq_lane_i32:
7591  case NEON::BI__builtin_neon_vdups_laneq_i32:
7592    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4));
7593    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7594                                        "vgetq_lane");
7595  case NEON::BI__builtin_neon_vget_lane_i64:
7596  case NEON::BI__builtin_neon_vdupd_lane_i64:
7597    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 1));
7598    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7599                                        "vget_lane");
7600  case NEON::BI__builtin_neon_vdupd_lane_f64:
7601    Ops[0] = Builder.CreateBitCast(Ops[0],
7602        llvm::VectorType::get(DoubleTy, 1));
7603    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7604                                        "vdupd_lane");
7605  case NEON::BI__builtin_neon_vgetq_lane_i64:
7606  case NEON::BI__builtin_neon_vdupd_laneq_i64:
7607    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2));
7608    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7609                                        "vgetq_lane");
7610  case NEON::BI__builtin_neon_vget_lane_f32:
7611    Ops[0] = Builder.CreateBitCast(Ops[0],
7612        llvm::VectorType::get(FloatTy, 2));
7613    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7614                                        "vget_lane");
7615  case NEON::BI__builtin_neon_vget_lane_f64:
7616    Ops[0] = Builder.CreateBitCast(Ops[0],
7617        llvm::VectorType::get(DoubleTy, 1));
7618    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7619                                        "vget_lane");
7620  case NEON::BI__builtin_neon_vgetq_lane_f32:
7621  case NEON::BI__builtin_neon_vdups_laneq_f32:
7622    Ops[0] = Builder.CreateBitCast(Ops[0],
7623        llvm::VectorType::get(FloatTy, 4));
7624    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7625                                        "vgetq_lane");
7626  case NEON::BI__builtin_neon_vgetq_lane_f64:
7627  case NEON::BI__builtin_neon_vdupd_laneq_f64:
7628    Ops[0] = Builder.CreateBitCast(Ops[0],
7629        llvm::VectorType::get(DoubleTy, 2));
7630    return Builder.CreateExtractElement(Ops[0], EmitScalarExpr(E->getArg(1)),
7631                                        "vgetq_lane");
7632  case NEON::BI__builtin_neon_vaddh_f16:
7633    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7634    return Builder.CreateFAdd(Ops[0], Ops[1], "vaddh");
7635  case NEON::BI__builtin_neon_vsubh_f16:
7636    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7637    return Builder.CreateFSub(Ops[0], Ops[1], "vsubh");
7638  case NEON::BI__builtin_neon_vmulh_f16:
7639    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7640    return Builder.CreateFMul(Ops[0], Ops[1], "vmulh");
7641  case NEON::BI__builtin_neon_vdivh_f16:
7642    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7643    return Builder.CreateFDiv(Ops[0], Ops[1], "vdivh");
7644  case NEON::BI__builtin_neon_vfmah_f16: {
7645    Function *F = CGM.getIntrinsic(Intrinsic::fma, HalfTy);
7646    // NEON intrinsic puts accumulator first, unlike the LLVM fma.
7647    return Builder.CreateCall(F,
7648      {EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)), Ops[0]});
7649  }
7650  case NEON::BI__builtin_neon_vfmsh_f16: {
7651    Function *F = CGM.getIntrinsic(Intrinsic::fma, HalfTy);
7652    Value *Zero = llvm::ConstantFP::getZeroValueForNegation(HalfTy);
7653    Value* Sub = Builder.CreateFSub(Zero, EmitScalarExpr(E->getArg(1)), "vsubh");
7654    // NEON intrinsic puts accumulator first, unlike the LLVM fma.
7655    return Builder.CreateCall(F, {Sub, EmitScalarExpr(E->getArg(2)), Ops[0]});
7656  }
7657  case NEON::BI__builtin_neon_vaddd_s64:
7658  case NEON::BI__builtin_neon_vaddd_u64:
7659    return Builder.CreateAdd(Ops[0], EmitScalarExpr(E->getArg(1)), "vaddd");
7660  case NEON::BI__builtin_neon_vsubd_s64:
7661  case NEON::BI__builtin_neon_vsubd_u64:
7662    return Builder.CreateSub(Ops[0], EmitScalarExpr(E->getArg(1)), "vsubd");
7663  case NEON::BI__builtin_neon_vqdmlalh_s16:
7664  case NEON::BI__builtin_neon_vqdmlslh_s16: {
7665    SmallVector<Value *, 2> ProductOps;
7666    ProductOps.push_back(vectorWrapScalar16(Ops[1]));
7667    ProductOps.push_back(vectorWrapScalar16(EmitScalarExpr(E->getArg(2))));
7668    llvm::Type *VTy = llvm::VectorType::get(Int32Ty, 4);
7669    Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy),
7670                          ProductOps, "vqdmlXl");
7671    Constant *CI = ConstantInt::get(SizeTy, 0);
7672    Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0");
7673
7674    unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlalh_s16
7675                                        ? Intrinsic::aarch64_neon_sqadd
7676                                        : Intrinsic::aarch64_neon_sqsub;
7677    return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int32Ty), Ops, "vqdmlXl");
7678  }
7679  case NEON::BI__builtin_neon_vqshlud_n_s64: {
7680    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7681    Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty);
7682    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqshlu, Int64Ty),
7683                        Ops, "vqshlu_n");
7684  }
7685  case NEON::BI__builtin_neon_vqshld_n_u64:
7686  case NEON::BI__builtin_neon_vqshld_n_s64: {
7687    unsigned Int = BuiltinID == NEON::BI__builtin_neon_vqshld_n_u64
7688                                   ? Intrinsic::aarch64_neon_uqshl
7689                                   : Intrinsic::aarch64_neon_sqshl;
7690    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7691    Ops[1] = Builder.CreateZExt(Ops[1], Int64Ty);
7692    return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vqshl_n");
7693  }
7694  case NEON::BI__builtin_neon_vrshrd_n_u64:
7695  case NEON::BI__builtin_neon_vrshrd_n_s64: {
7696    unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrshrd_n_u64
7697                                   ? Intrinsic::aarch64_neon_urshl
7698                                   : Intrinsic::aarch64_neon_srshl;
7699    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7700    int SV = cast<ConstantInt>(Ops[1])->getSExtValue();
7701    Ops[1] = ConstantInt::get(Int64Ty, -SV);
7702    return EmitNeonCall(CGM.getIntrinsic(Int, Int64Ty), Ops, "vrshr_n");
7703  }
7704  case NEON::BI__builtin_neon_vrsrad_n_u64:
7705  case NEON::BI__builtin_neon_vrsrad_n_s64: {
7706    unsigned Int = BuiltinID == NEON::BI__builtin_neon_vrsrad_n_u64
7707                                   ? Intrinsic::aarch64_neon_urshl
7708                                   : Intrinsic::aarch64_neon_srshl;
7709    Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
7710    Ops.push_back(Builder.CreateNeg(EmitScalarExpr(E->getArg(2))));
7711    Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Int64Ty),
7712                                {Ops[1], Builder.CreateSExt(Ops[2], Int64Ty)});
7713    return Builder.CreateAdd(Ops[0], Builder.CreateBitCast(Ops[1], Int64Ty));
7714  }
7715  case NEON::BI__builtin_neon_vshld_n_s64:
7716  case NEON::BI__builtin_neon_vshld_n_u64: {
7717    llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
7718    return Builder.CreateShl(
7719        Ops[0], ConstantInt::get(Int64Ty, Amt->getZExtValue()), "shld_n");
7720  }
7721  case NEON::BI__builtin_neon_vshrd_n_s64: {
7722    llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
7723    return Builder.CreateAShr(
7724        Ops[0], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
7725                                                   Amt->getZExtValue())),
7726        "shrd_n");
7727  }
7728  case NEON::BI__builtin_neon_vshrd_n_u64: {
7729    llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(1)));
7730    uint64_t ShiftAmt = Amt->getZExtValue();
7731    // Right-shifting an unsigned value by its size yields 0.
7732    if (ShiftAmt == 64)
7733      return ConstantInt::get(Int64Ty, 0);
7734    return Builder.CreateLShr(Ops[0], ConstantInt::get(Int64Ty, ShiftAmt),
7735                              "shrd_n");
7736  }
7737  case NEON::BI__builtin_neon_vsrad_n_s64: {
7738    llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2)));
7739    Ops[1] = Builder.CreateAShr(
7740        Ops[1], ConstantInt::get(Int64Ty, std::min(static_cast<uint64_t>(63),
7741                                                   Amt->getZExtValue())),
7742        "shrd_n");
7743    return Builder.CreateAdd(Ops[0], Ops[1]);
7744  }
7745  case NEON::BI__builtin_neon_vsrad_n_u64: {
7746    llvm::ConstantInt *Amt = cast<ConstantInt>(EmitScalarExpr(E->getArg(2)));
7747    uint64_t ShiftAmt = Amt->getZExtValue();
7748    // Right-shifting an unsigned value by its size yields 0.
7749    // As Op + 0 = Op, return Ops[0] directly.
7750    if (ShiftAmt == 64)
7751      return Ops[0];
7752    Ops[1] = Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, ShiftAmt),
7753                                "shrd_n");
7754    return Builder.CreateAdd(Ops[0], Ops[1]);
7755  }
7756  case NEON::BI__builtin_neon_vqdmlalh_lane_s16:
7757  case NEON::BI__builtin_neon_vqdmlalh_laneq_s16:
7758  case NEON::BI__builtin_neon_vqdmlslh_lane_s16:
7759  case NEON::BI__builtin_neon_vqdmlslh_laneq_s16: {
7760    Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)),
7761                                          "lane");
7762    SmallVector<Value *, 2> ProductOps;
7763    ProductOps.push_back(vectorWrapScalar16(Ops[1]));
7764    ProductOps.push_back(vectorWrapScalar16(Ops[2]));
7765    llvm::Type *VTy = llvm::VectorType::get(Int32Ty, 4);
7766    Ops[1] = EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmull, VTy),
7767                          ProductOps, "vqdmlXl");
7768    Constant *CI = ConstantInt::get(SizeTy, 0);
7769    Ops[1] = Builder.CreateExtractElement(Ops[1], CI, "lane0");
7770    Ops.pop_back();
7771
7772    unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlalh_lane_s16 ||
7773                       BuiltinID == NEON::BI__builtin_neon_vqdmlalh_laneq_s16)
7774                          ? Intrinsic::aarch64_neon_sqadd
7775                          : Intrinsic::aarch64_neon_sqsub;
7776    return EmitNeonCall(CGM.getIntrinsic(AccInt, Int32Ty), Ops, "vqdmlXl");
7777  }
7778  case NEON::BI__builtin_neon_vqdmlals_s32:
7779  case NEON::BI__builtin_neon_vqdmlsls_s32: {
7780    SmallVector<Value *, 2> ProductOps;
7781    ProductOps.push_back(Ops[1]);
7782    ProductOps.push_back(EmitScalarExpr(E->getArg(2)));
7783    Ops[1] =
7784        EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar),
7785                     ProductOps, "vqdmlXl");
7786
7787    unsigned AccumInt = BuiltinID == NEON::BI__builtin_neon_vqdmlals_s32
7788                                        ? Intrinsic::aarch64_neon_sqadd
7789                                        : Intrinsic::aarch64_neon_sqsub;
7790    return EmitNeonCall(CGM.getIntrinsic(AccumInt, Int64Ty), Ops, "vqdmlXl");
7791  }
7792  case NEON::BI__builtin_neon_vqdmlals_lane_s32:
7793  case NEON::BI__builtin_neon_vqdmlals_laneq_s32:
7794  case NEON::BI__builtin_neon_vqdmlsls_lane_s32:
7795  case NEON::BI__builtin_neon_vqdmlsls_laneq_s32: {
7796    Ops[2] = Builder.CreateExtractElement(Ops[2], EmitScalarExpr(E->getArg(3)),
7797                                          "lane");
7798    SmallVector<Value *, 2> ProductOps;
7799    ProductOps.push_back(Ops[1]);
7800    ProductOps.push_back(Ops[2]);
7801    Ops[1] =
7802        EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_sqdmulls_scalar),
7803                     ProductOps, "vqdmlXl");
7804    Ops.pop_back();
7805
7806    unsigned AccInt = (BuiltinID == NEON::BI__builtin_neon_vqdmlals_lane_s32 ||
7807                       BuiltinID == NEON::BI__builtin_neon_vqdmlals_laneq_s32)
7808                          ? Intrinsic::aarch64_neon_sqadd
7809                          : Intrinsic::aarch64_neon_sqsub;
7810    return EmitNeonCall(CGM.getIntrinsic(AccInt, Int64Ty), Ops, "vqdmlXl");
7811  }
7812  }
7813
7814  llvm::VectorType *VTy = GetNeonType(this, Type);
7815  llvm::Type *Ty = VTy;
7816  if (!Ty)
7817    return nullptr;
7818
7819  // Not all intrinsics handled by the common case work for AArch64 yet, so only
7820  // defer to common code if it's been added to our special map.
7821  Builtin = findNeonIntrinsicInMap(AArch64SIMDIntrinsicMap, BuiltinID,
7822                                   AArch64SIMDIntrinsicsProvenSorted);
7823
7824  if (Builtin)
7825    return EmitCommonNeonBuiltinExpr(
7826        Builtin->BuiltinID, Builtin->LLVMIntrinsic, Builtin->AltLLVMIntrinsic,
7827        Builtin->NameHint, Builtin->TypeModifier, E, Ops,
7828        /*never use addresses*/ Address::invalid(), Address::invalid(), Arch);
7829
7830  if (Value *V = EmitAArch64TblBuiltinExpr(*this, BuiltinID, E, Ops, Arch))
7831    return V;
7832
7833  unsigned Int;
7834  switch (BuiltinID) {
7835  default: return nullptr;
7836  case NEON::BI__builtin_neon_vbsl_v:
7837  case NEON::BI__builtin_neon_vbslq_v: {
7838    llvm::Type *BitTy = llvm::VectorType::getInteger(VTy);
7839    Ops[0] = Builder.CreateBitCast(Ops[0], BitTy, "vbsl");
7840    Ops[1] = Builder.CreateBitCast(Ops[1], BitTy, "vbsl");
7841    Ops[2] = Builder.CreateBitCast(Ops[2], BitTy, "vbsl");
7842
7843    Ops[1] = Builder.CreateAnd(Ops[0], Ops[1], "vbsl");
7844    Ops[2] = Builder.CreateAnd(Builder.CreateNot(Ops[0]), Ops[2], "vbsl");
7845    Ops[0] = Builder.CreateOr(Ops[1], Ops[2], "vbsl");
7846    return Builder.CreateBitCast(Ops[0], Ty);
7847  }
7848  case NEON::BI__builtin_neon_vfma_lane_v:
7849  case NEON::BI__builtin_neon_vfmaq_lane_v: { // Only used for FP types
7850    // The ARM builtins (and instructions) have the addend as the first
7851    // operand, but the 'fma' intrinsics have it last. Swap it around here.
7852    Value *Addend = Ops[0];
7853    Value *Multiplicand = Ops[1];
7854    Value *LaneSource = Ops[2];
7855    Ops[0] = Multiplicand;
7856    Ops[1] = LaneSource;
7857    Ops[2] = Addend;
7858
7859    // Now adjust things to handle the lane access.
7860    llvm::Type *SourceTy = BuiltinID == NEON::BI__builtin_neon_vfmaq_lane_v ?
7861      llvm::VectorType::get(VTy->getElementType(), VTy->getNumElements() / 2) :
7862      VTy;
7863    llvm::Constant *cst = cast<Constant>(Ops[3]);
7864    Value *SV = llvm::ConstantVector::getSplat(VTy->getNumElements(), cst);
7865    Ops[1] = Builder.CreateBitCast(Ops[1], SourceTy);
7866    Ops[1] = Builder.CreateShuffleVector(Ops[1], Ops[1], SV, "lane");
7867
7868    Ops.pop_back();
7869    Int = Intrinsic::fma;
7870    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "fmla");
7871  }
7872  case NEON::BI__builtin_neon_vfma_laneq_v: {
7873    llvm::VectorType *VTy = cast<llvm::VectorType>(Ty);
7874    // v1f64 fma should be mapped to Neon scalar f64 fma
7875    if (VTy && VTy->getElementType() == DoubleTy) {
7876      Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
7877      Ops[1] = Builder.CreateBitCast(Ops[1], DoubleTy);
7878      llvm::Type *VTy = GetNeonType(this,
7879        NeonTypeFlags(NeonTypeFlags::Float64, false, true));
7880      Ops[2] = Builder.CreateBitCast(Ops[2], VTy);
7881      Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract");
7882      Function *F = CGM.getIntrinsic(Intrinsic::fma, DoubleTy);
7883      Value *Result = Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]});
7884      return Builder.CreateBitCast(Result, Ty);
7885    }
7886    Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
7887    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
7888    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7889
7890    llvm::Type *STy = llvm::VectorType::get(VTy->getElementType(),
7891                                            VTy->getNumElements() * 2);
7892    Ops[2] = Builder.CreateBitCast(Ops[2], STy);
7893    Value* SV = llvm::ConstantVector::getSplat(VTy->getNumElements(),
7894                                               cast<ConstantInt>(Ops[3]));
7895    Ops[2] = Builder.CreateShuffleVector(Ops[2], Ops[2], SV, "lane");
7896
7897    return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]});
7898  }
7899  case NEON::BI__builtin_neon_vfmaq_laneq_v: {
7900    Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
7901    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
7902    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
7903
7904    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
7905    Ops[2] = EmitNeonSplat(Ops[2], cast<ConstantInt>(Ops[3]));
7906    return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]});
7907  }
7908  case NEON::BI__builtin_neon_vfmah_lane_f16:
7909  case NEON::BI__builtin_neon_vfmas_lane_f32:
7910  case NEON::BI__builtin_neon_vfmah_laneq_f16:
7911  case NEON::BI__builtin_neon_vfmas_laneq_f32:
7912  case NEON::BI__builtin_neon_vfmad_lane_f64:
7913  case NEON::BI__builtin_neon_vfmad_laneq_f64: {
7914    Ops.push_back(EmitScalarExpr(E->getArg(3)));
7915    llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext()));
7916    Function *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
7917    Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract");
7918    return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]});
7919  }
7920  case NEON::BI__builtin_neon_vmull_v:
7921    // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7922    Int = usgn ? Intrinsic::aarch64_neon_umull : Intrinsic::aarch64_neon_smull;
7923    if (Type.isPoly()) Int = Intrinsic::aarch64_neon_pmull;
7924    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmull");
7925  case NEON::BI__builtin_neon_vmax_v:
7926  case NEON::BI__builtin_neon_vmaxq_v:
7927    // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7928    Int = usgn ? Intrinsic::aarch64_neon_umax : Intrinsic::aarch64_neon_smax;
7929    if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmax;
7930    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmax");
7931  case NEON::BI__builtin_neon_vmaxh_f16: {
7932    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7933    Int = Intrinsic::aarch64_neon_fmax;
7934    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmax");
7935  }
7936  case NEON::BI__builtin_neon_vmin_v:
7937  case NEON::BI__builtin_neon_vminq_v:
7938    // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7939    Int = usgn ? Intrinsic::aarch64_neon_umin : Intrinsic::aarch64_neon_smin;
7940    if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmin;
7941    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmin");
7942  case NEON::BI__builtin_neon_vminh_f16: {
7943    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7944    Int = Intrinsic::aarch64_neon_fmin;
7945    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmin");
7946  }
7947  case NEON::BI__builtin_neon_vabd_v:
7948  case NEON::BI__builtin_neon_vabdq_v:
7949    // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7950    Int = usgn ? Intrinsic::aarch64_neon_uabd : Intrinsic::aarch64_neon_sabd;
7951    if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fabd;
7952    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vabd");
7953  case NEON::BI__builtin_neon_vpadal_v:
7954  case NEON::BI__builtin_neon_vpadalq_v: {
7955    unsigned ArgElts = VTy->getNumElements();
7956    llvm::IntegerType *EltTy = cast<IntegerType>(VTy->getElementType());
7957    unsigned BitWidth = EltTy->getBitWidth();
7958    llvm::Type *ArgTy = llvm::VectorType::get(
7959        llvm::IntegerType::get(getLLVMContext(), BitWidth/2), 2*ArgElts);
7960    llvm::Type* Tys[2] = { VTy, ArgTy };
7961    Int = usgn ? Intrinsic::aarch64_neon_uaddlp : Intrinsic::aarch64_neon_saddlp;
7962    SmallVector<llvm::Value*, 1> TmpOps;
7963    TmpOps.push_back(Ops[1]);
7964    Function *F = CGM.getIntrinsic(Int, Tys);
7965    llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vpadal");
7966    llvm::Value *addend = Builder.CreateBitCast(Ops[0], tmp->getType());
7967    return Builder.CreateAdd(tmp, addend);
7968  }
7969  case NEON::BI__builtin_neon_vpmin_v:
7970  case NEON::BI__builtin_neon_vpminq_v:
7971    // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7972    Int = usgn ? Intrinsic::aarch64_neon_uminp : Intrinsic::aarch64_neon_sminp;
7973    if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fminp;
7974    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmin");
7975  case NEON::BI__builtin_neon_vpmax_v:
7976  case NEON::BI__builtin_neon_vpmaxq_v:
7977    // FIXME: improve sharing scheme to cope with 3 alternative LLVM intrinsics.
7978    Int = usgn ? Intrinsic::aarch64_neon_umaxp : Intrinsic::aarch64_neon_smaxp;
7979    if (Ty->isFPOrFPVectorTy()) Int = Intrinsic::aarch64_neon_fmaxp;
7980    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmax");
7981  case NEON::BI__builtin_neon_vminnm_v:
7982  case NEON::BI__builtin_neon_vminnmq_v:
7983    Int = Intrinsic::aarch64_neon_fminnm;
7984    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vminnm");
7985  case NEON::BI__builtin_neon_vminnmh_f16:
7986    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7987    Int = Intrinsic::aarch64_neon_fminnm;
7988    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vminnm");
7989  case NEON::BI__builtin_neon_vmaxnm_v:
7990  case NEON::BI__builtin_neon_vmaxnmq_v:
7991    Int = Intrinsic::aarch64_neon_fmaxnm;
7992    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmaxnm");
7993  case NEON::BI__builtin_neon_vmaxnmh_f16:
7994    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7995    Int = Intrinsic::aarch64_neon_fmaxnm;
7996    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmaxnm");
7997  case NEON::BI__builtin_neon_vrecpss_f32: {
7998    Ops.push_back(EmitScalarExpr(E->getArg(1)));
7999    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, FloatTy),
8000                        Ops, "vrecps");
8001  }
8002  case NEON::BI__builtin_neon_vrecpsd_f64:
8003    Ops.push_back(EmitScalarExpr(E->getArg(1)));
8004    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, DoubleTy),
8005                        Ops, "vrecps");
8006  case NEON::BI__builtin_neon_vrecpsh_f16:
8007    Ops.push_back(EmitScalarExpr(E->getArg(1)));
8008    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_frecps, HalfTy),
8009                        Ops, "vrecps");
8010  case NEON::BI__builtin_neon_vqshrun_n_v:
8011    Int = Intrinsic::aarch64_neon_sqshrun;
8012    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrun_n");
8013  case NEON::BI__builtin_neon_vqrshrun_n_v:
8014    Int = Intrinsic::aarch64_neon_sqrshrun;
8015    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrun_n");
8016  case NEON::BI__builtin_neon_vqshrn_n_v:
8017    Int = usgn ? Intrinsic::aarch64_neon_uqshrn : Intrinsic::aarch64_neon_sqshrn;
8018    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqshrn_n");
8019  case NEON::BI__builtin_neon_vrshrn_n_v:
8020    Int = Intrinsic::aarch64_neon_rshrn;
8021    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrshrn_n");
8022  case NEON::BI__builtin_neon_vqrshrn_n_v:
8023    Int = usgn ? Intrinsic::aarch64_neon_uqrshrn : Intrinsic::aarch64_neon_sqrshrn;
8024    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vqrshrn_n");
8025  case NEON::BI__builtin_neon_vrndah_f16: {
8026    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8027    Int = Intrinsic::round;
8028    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrnda");
8029  }
8030  case NEON::BI__builtin_neon_vrnda_v:
8031  case NEON::BI__builtin_neon_vrndaq_v: {
8032    Int = Intrinsic::round;
8033    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrnda");
8034  }
8035  case NEON::BI__builtin_neon_vrndih_f16: {
8036    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8037    Int = Intrinsic::nearbyint;
8038    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndi");
8039  }
8040  case NEON::BI__builtin_neon_vrndmh_f16: {
8041    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8042    Int = Intrinsic::floor;
8043    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndm");
8044  }
8045  case NEON::BI__builtin_neon_vrndm_v:
8046  case NEON::BI__builtin_neon_vrndmq_v: {
8047    Int = Intrinsic::floor;
8048    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndm");
8049  }
8050  case NEON::BI__builtin_neon_vrndnh_f16: {
8051    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8052    Int = Intrinsic::aarch64_neon_frintn;
8053    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndn");
8054  }
8055  case NEON::BI__builtin_neon_vrndn_v:
8056  case NEON::BI__builtin_neon_vrndnq_v: {
8057    Int = Intrinsic::aarch64_neon_frintn;
8058    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndn");
8059  }
8060  case NEON::BI__builtin_neon_vrndns_f32: {
8061    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8062    Int = Intrinsic::aarch64_neon_frintn;
8063    return EmitNeonCall(CGM.getIntrinsic(Int, FloatTy), Ops, "vrndn");
8064  }
8065  case NEON::BI__builtin_neon_vrndph_f16: {
8066    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8067    Int = Intrinsic::ceil;
8068    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndp");
8069  }
8070  case NEON::BI__builtin_neon_vrndp_v:
8071  case NEON::BI__builtin_neon_vrndpq_v: {
8072    Int = Intrinsic::ceil;
8073    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndp");
8074  }
8075  case NEON::BI__builtin_neon_vrndxh_f16: {
8076    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8077    Int = Intrinsic::rint;
8078    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndx");
8079  }
8080  case NEON::BI__builtin_neon_vrndx_v:
8081  case NEON::BI__builtin_neon_vrndxq_v: {
8082    Int = Intrinsic::rint;
8083    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndx");
8084  }
8085  case NEON::BI__builtin_neon_vrndh_f16: {
8086    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8087    Int = Intrinsic::trunc;
8088    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vrndz");
8089  }
8090  case NEON::BI__builtin_neon_vrnd_v:
8091  case NEON::BI__builtin_neon_vrndq_v: {
8092    Int = Intrinsic::trunc;
8093    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrndz");
8094  }
8095  case NEON::BI__builtin_neon_vcvt_f64_v:
8096  case NEON::BI__builtin_neon_vcvtq_f64_v:
8097    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8098    Ty = GetNeonType(this, NeonTypeFlags(NeonTypeFlags::Float64, false, quad));
8099    return usgn ? Builder.CreateUIToFP(Ops[0], Ty, "vcvt")
8100                : Builder.CreateSIToFP(Ops[0], Ty, "vcvt");
8101  case NEON::BI__builtin_neon_vcvt_f64_f32: {
8102     (0) . __assert_fail ("Type.getEltType() == NeonTypeFlags..Float64 && quad && \"unexpected vcvt_f64_f32 builtin\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 8103, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Type.getEltType() == NeonTypeFlags::Float64 && quad &&
8103 (0) . __assert_fail ("Type.getEltType() == NeonTypeFlags..Float64 && quad && \"unexpected vcvt_f64_f32 builtin\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 8103, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">           "unexpected vcvt_f64_f32 builtin");
8104    NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float32, false, false);
8105    Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag));
8106
8107    return Builder.CreateFPExt(Ops[0], Ty, "vcvt");
8108  }
8109  case NEON::BI__builtin_neon_vcvt_f32_f64: {
8110     (0) . __assert_fail ("Type.getEltType() == NeonTypeFlags..Float32 && \"unexpected vcvt_f32_f64 builtin\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 8111, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Type.getEltType() == NeonTypeFlags::Float32 &&
8111 (0) . __assert_fail ("Type.getEltType() == NeonTypeFlags..Float32 && \"unexpected vcvt_f32_f64 builtin\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 8111, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">           "unexpected vcvt_f32_f64 builtin");
8112    NeonTypeFlags SrcFlag = NeonTypeFlags(NeonTypeFlags::Float64, false, true);
8113    Ops[0] = Builder.CreateBitCast(Ops[0], GetNeonType(this, SrcFlag));
8114
8115    return Builder.CreateFPTrunc(Ops[0], Ty, "vcvt");
8116  }
8117  case NEON::BI__builtin_neon_vcvt_s32_v:
8118  case NEON::BI__builtin_neon_vcvt_u32_v:
8119  case NEON::BI__builtin_neon_vcvt_s64_v:
8120  case NEON::BI__builtin_neon_vcvt_u64_v:
8121  case NEON::BI__builtin_neon_vcvt_s16_v:
8122  case NEON::BI__builtin_neon_vcvt_u16_v:
8123  case NEON::BI__builtin_neon_vcvtq_s32_v:
8124  case NEON::BI__builtin_neon_vcvtq_u32_v:
8125  case NEON::BI__builtin_neon_vcvtq_s64_v:
8126  case NEON::BI__builtin_neon_vcvtq_u64_v:
8127  case NEON::BI__builtin_neon_vcvtq_s16_v:
8128  case NEON::BI__builtin_neon_vcvtq_u16_v: {
8129    Ops[0] = Builder.CreateBitCast(Ops[0], GetFloatNeonType(this, Type));
8130    if (usgn)
8131      return Builder.CreateFPToUI(Ops[0], Ty);
8132    return Builder.CreateFPToSI(Ops[0], Ty);
8133  }
8134  case NEON::BI__builtin_neon_vcvta_s16_v:
8135  case NEON::BI__builtin_neon_vcvta_u16_v:
8136  case NEON::BI__builtin_neon_vcvta_s32_v:
8137  case NEON::BI__builtin_neon_vcvtaq_s16_v:
8138  case NEON::BI__builtin_neon_vcvtaq_s32_v:
8139  case NEON::BI__builtin_neon_vcvta_u32_v:
8140  case NEON::BI__builtin_neon_vcvtaq_u16_v:
8141  case NEON::BI__builtin_neon_vcvtaq_u32_v:
8142  case NEON::BI__builtin_neon_vcvta_s64_v:
8143  case NEON::BI__builtin_neon_vcvtaq_s64_v:
8144  case NEON::BI__builtin_neon_vcvta_u64_v:
8145  case NEON::BI__builtin_neon_vcvtaq_u64_v: {
8146    Int = usgn ? Intrinsic::aarch64_neon_fcvtau : Intrinsic::aarch64_neon_fcvtas;
8147    llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
8148    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvta");
8149  }
8150  case NEON::BI__builtin_neon_vcvtm_s16_v:
8151  case NEON::BI__builtin_neon_vcvtm_s32_v:
8152  case NEON::BI__builtin_neon_vcvtmq_s16_v:
8153  case NEON::BI__builtin_neon_vcvtmq_s32_v:
8154  case NEON::BI__builtin_neon_vcvtm_u16_v:
8155  case NEON::BI__builtin_neon_vcvtm_u32_v:
8156  case NEON::BI__builtin_neon_vcvtmq_u16_v:
8157  case NEON::BI__builtin_neon_vcvtmq_u32_v:
8158  case NEON::BI__builtin_neon_vcvtm_s64_v:
8159  case NEON::BI__builtin_neon_vcvtmq_s64_v:
8160  case NEON::BI__builtin_neon_vcvtm_u64_v:
8161  case NEON::BI__builtin_neon_vcvtmq_u64_v: {
8162    Int = usgn ? Intrinsic::aarch64_neon_fcvtmu : Intrinsic::aarch64_neon_fcvtms;
8163    llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
8164    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtm");
8165  }
8166  case NEON::BI__builtin_neon_vcvtn_s16_v:
8167  case NEON::BI__builtin_neon_vcvtn_s32_v:
8168  case NEON::BI__builtin_neon_vcvtnq_s16_v:
8169  case NEON::BI__builtin_neon_vcvtnq_s32_v:
8170  case NEON::BI__builtin_neon_vcvtn_u16_v:
8171  case NEON::BI__builtin_neon_vcvtn_u32_v:
8172  case NEON::BI__builtin_neon_vcvtnq_u16_v:
8173  case NEON::BI__builtin_neon_vcvtnq_u32_v:
8174  case NEON::BI__builtin_neon_vcvtn_s64_v:
8175  case NEON::BI__builtin_neon_vcvtnq_s64_v:
8176  case NEON::BI__builtin_neon_vcvtn_u64_v:
8177  case NEON::BI__builtin_neon_vcvtnq_u64_v: {
8178    Int = usgn ? Intrinsic::aarch64_neon_fcvtnu : Intrinsic::aarch64_neon_fcvtns;
8179    llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
8180    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtn");
8181  }
8182  case NEON::BI__builtin_neon_vcvtp_s16_v:
8183  case NEON::BI__builtin_neon_vcvtp_s32_v:
8184  case NEON::BI__builtin_neon_vcvtpq_s16_v:
8185  case NEON::BI__builtin_neon_vcvtpq_s32_v:
8186  case NEON::BI__builtin_neon_vcvtp_u16_v:
8187  case NEON::BI__builtin_neon_vcvtp_u32_v:
8188  case NEON::BI__builtin_neon_vcvtpq_u16_v:
8189  case NEON::BI__builtin_neon_vcvtpq_u32_v:
8190  case NEON::BI__builtin_neon_vcvtp_s64_v:
8191  case NEON::BI__builtin_neon_vcvtpq_s64_v:
8192  case NEON::BI__builtin_neon_vcvtp_u64_v:
8193  case NEON::BI__builtin_neon_vcvtpq_u64_v: {
8194    Int = usgn ? Intrinsic::aarch64_neon_fcvtpu : Intrinsic::aarch64_neon_fcvtps;
8195    llvm::Type *Tys[2] = { Ty, GetFloatNeonType(this, Type) };
8196    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vcvtp");
8197  }
8198  case NEON::BI__builtin_neon_vmulx_v:
8199  case NEON::BI__builtin_neon_vmulxq_v: {
8200    Int = Intrinsic::aarch64_neon_fmulx;
8201    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vmulx");
8202  }
8203  case NEON::BI__builtin_neon_vmulxh_lane_f16:
8204  case NEON::BI__builtin_neon_vmulxh_laneq_f16: {
8205    // vmulx_lane should be mapped to Neon scalar mulx after
8206    // extracting the scalar element
8207    Ops.push_back(EmitScalarExpr(E->getArg(2)));
8208    Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract");
8209    Ops.pop_back();
8210    Int = Intrinsic::aarch64_neon_fmulx;
8211    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vmulx");
8212  }
8213  case NEON::BI__builtin_neon_vmul_lane_v:
8214  case NEON::BI__builtin_neon_vmul_laneq_v: {
8215    // v1f64 vmul_lane should be mapped to Neon scalar mul lane
8216    bool Quad = false;
8217    if (BuiltinID == NEON::BI__builtin_neon_vmul_laneq_v)
8218      Quad = true;
8219    Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
8220    llvm::Type *VTy = GetNeonType(this,
8221      NeonTypeFlags(NeonTypeFlags::Float64, false, Quad));
8222    Ops[1] = Builder.CreateBitCast(Ops[1], VTy);
8223    Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2], "extract");
8224    Value *Result = Builder.CreateFMul(Ops[0], Ops[1]);
8225    return Builder.CreateBitCast(Result, Ty);
8226  }
8227  case NEON::BI__builtin_neon_vnegd_s64:
8228    return Builder.CreateNeg(EmitScalarExpr(E->getArg(0)), "vnegd");
8229  case NEON::BI__builtin_neon_vnegh_f16:
8230    return Builder.CreateFNeg(EmitScalarExpr(E->getArg(0)), "vnegh");
8231  case NEON::BI__builtin_neon_vpmaxnm_v:
8232  case NEON::BI__builtin_neon_vpmaxnmq_v: {
8233    Int = Intrinsic::aarch64_neon_fmaxnmp;
8234    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpmaxnm");
8235  }
8236  case NEON::BI__builtin_neon_vpminnm_v:
8237  case NEON::BI__builtin_neon_vpminnmq_v: {
8238    Int = Intrinsic::aarch64_neon_fminnmp;
8239    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vpminnm");
8240  }
8241  case NEON::BI__builtin_neon_vsqrth_f16: {
8242    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8243    Int = Intrinsic::sqrt;
8244    return EmitNeonCall(CGM.getIntrinsic(Int, HalfTy), Ops, "vsqrt");
8245  }
8246  case NEON::BI__builtin_neon_vsqrt_v:
8247  case NEON::BI__builtin_neon_vsqrtq_v: {
8248    Int = Intrinsic::sqrt;
8249    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8250    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqrt");
8251  }
8252  case NEON::BI__builtin_neon_vrbit_v:
8253  case NEON::BI__builtin_neon_vrbitq_v: {
8254    Int = Intrinsic::aarch64_neon_rbit;
8255    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vrbit");
8256  }
8257  case NEON::BI__builtin_neon_vaddv_u8:
8258    // FIXME: These are handled by the AArch64 scalar code.
8259    usgn = true;
8260    LLVM_FALLTHROUGH;
8261  case NEON::BI__builtin_neon_vaddv_s8: {
8262    Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
8263    Ty = Int32Ty;
8264    VTy = llvm::VectorType::get(Int8Ty, 8);
8265    llvm::Type *Tys[2] = { Ty, VTy };
8266    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8267    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
8268    return Builder.CreateTrunc(Ops[0], Int8Ty);
8269  }
8270  case NEON::BI__builtin_neon_vaddv_u16:
8271    usgn = true;
8272    LLVM_FALLTHROUGH;
8273  case NEON::BI__builtin_neon_vaddv_s16: {
8274    Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
8275    Ty = Int32Ty;
8276    VTy = llvm::VectorType::get(Int16Ty, 4);
8277    llvm::Type *Tys[2] = { Ty, VTy };
8278    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8279    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
8280    return Builder.CreateTrunc(Ops[0], Int16Ty);
8281  }
8282  case NEON::BI__builtin_neon_vaddvq_u8:
8283    usgn = true;
8284    LLVM_FALLTHROUGH;
8285  case NEON::BI__builtin_neon_vaddvq_s8: {
8286    Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
8287    Ty = Int32Ty;
8288    VTy = llvm::VectorType::get(Int8Ty, 16);
8289    llvm::Type *Tys[2] = { Ty, VTy };
8290    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8291    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
8292    return Builder.CreateTrunc(Ops[0], Int8Ty);
8293  }
8294  case NEON::BI__builtin_neon_vaddvq_u16:
8295    usgn = true;
8296    LLVM_FALLTHROUGH;
8297  case NEON::BI__builtin_neon_vaddvq_s16: {
8298    Int = usgn ? Intrinsic::aarch64_neon_uaddv : Intrinsic::aarch64_neon_saddv;
8299    Ty = Int32Ty;
8300    VTy = llvm::VectorType::get(Int16Ty, 8);
8301    llvm::Type *Tys[2] = { Ty, VTy };
8302    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8303    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddv");
8304    return Builder.CreateTrunc(Ops[0], Int16Ty);
8305  }
8306  case NEON::BI__builtin_neon_vmaxv_u8: {
8307    Int = Intrinsic::aarch64_neon_umaxv;
8308    Ty = Int32Ty;
8309    VTy = llvm::VectorType::get(Int8Ty, 8);
8310    llvm::Type *Tys[2] = { Ty, VTy };
8311    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8312    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8313    return Builder.CreateTrunc(Ops[0], Int8Ty);
8314  }
8315  case NEON::BI__builtin_neon_vmaxv_u16: {
8316    Int = Intrinsic::aarch64_neon_umaxv;
8317    Ty = Int32Ty;
8318    VTy = llvm::VectorType::get(Int16Ty, 4);
8319    llvm::Type *Tys[2] = { Ty, VTy };
8320    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8321    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8322    return Builder.CreateTrunc(Ops[0], Int16Ty);
8323  }
8324  case NEON::BI__builtin_neon_vmaxvq_u8: {
8325    Int = Intrinsic::aarch64_neon_umaxv;
8326    Ty = Int32Ty;
8327    VTy = llvm::VectorType::get(Int8Ty, 16);
8328    llvm::Type *Tys[2] = { Ty, VTy };
8329    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8330    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8331    return Builder.CreateTrunc(Ops[0], Int8Ty);
8332  }
8333  case NEON::BI__builtin_neon_vmaxvq_u16: {
8334    Int = Intrinsic::aarch64_neon_umaxv;
8335    Ty = Int32Ty;
8336    VTy = llvm::VectorType::get(Int16Ty, 8);
8337    llvm::Type *Tys[2] = { Ty, VTy };
8338    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8339    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8340    return Builder.CreateTrunc(Ops[0], Int16Ty);
8341  }
8342  case NEON::BI__builtin_neon_vmaxv_s8: {
8343    Int = Intrinsic::aarch64_neon_smaxv;
8344    Ty = Int32Ty;
8345    VTy = llvm::VectorType::get(Int8Ty, 8);
8346    llvm::Type *Tys[2] = { Ty, VTy };
8347    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8348    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8349    return Builder.CreateTrunc(Ops[0], Int8Ty);
8350  }
8351  case NEON::BI__builtin_neon_vmaxv_s16: {
8352    Int = Intrinsic::aarch64_neon_smaxv;
8353    Ty = Int32Ty;
8354    VTy = llvm::VectorType::get(Int16Ty, 4);
8355    llvm::Type *Tys[2] = { Ty, VTy };
8356    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8357    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8358    return Builder.CreateTrunc(Ops[0], Int16Ty);
8359  }
8360  case NEON::BI__builtin_neon_vmaxvq_s8: {
8361    Int = Intrinsic::aarch64_neon_smaxv;
8362    Ty = Int32Ty;
8363    VTy = llvm::VectorType::get(Int8Ty, 16);
8364    llvm::Type *Tys[2] = { Ty, VTy };
8365    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8366    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8367    return Builder.CreateTrunc(Ops[0], Int8Ty);
8368  }
8369  case NEON::BI__builtin_neon_vmaxvq_s16: {
8370    Int = Intrinsic::aarch64_neon_smaxv;
8371    Ty = Int32Ty;
8372    VTy = llvm::VectorType::get(Int16Ty, 8);
8373    llvm::Type *Tys[2] = { Ty, VTy };
8374    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8375    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8376    return Builder.CreateTrunc(Ops[0], Int16Ty);
8377  }
8378  case NEON::BI__builtin_neon_vmaxv_f16: {
8379    Int = Intrinsic::aarch64_neon_fmaxv;
8380    Ty = HalfTy;
8381    VTy = llvm::VectorType::get(HalfTy, 4);
8382    llvm::Type *Tys[2] = { Ty, VTy };
8383    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8384    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8385    return Builder.CreateTrunc(Ops[0], HalfTy);
8386  }
8387  case NEON::BI__builtin_neon_vmaxvq_f16: {
8388    Int = Intrinsic::aarch64_neon_fmaxv;
8389    Ty = HalfTy;
8390    VTy = llvm::VectorType::get(HalfTy, 8);
8391    llvm::Type *Tys[2] = { Ty, VTy };
8392    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8393    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxv");
8394    return Builder.CreateTrunc(Ops[0], HalfTy);
8395  }
8396  case NEON::BI__builtin_neon_vminv_u8: {
8397    Int = Intrinsic::aarch64_neon_uminv;
8398    Ty = Int32Ty;
8399    VTy = llvm::VectorType::get(Int8Ty, 8);
8400    llvm::Type *Tys[2] = { Ty, VTy };
8401    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8402    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8403    return Builder.CreateTrunc(Ops[0], Int8Ty);
8404  }
8405  case NEON::BI__builtin_neon_vminv_u16: {
8406    Int = Intrinsic::aarch64_neon_uminv;
8407    Ty = Int32Ty;
8408    VTy = llvm::VectorType::get(Int16Ty, 4);
8409    llvm::Type *Tys[2] = { Ty, VTy };
8410    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8411    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8412    return Builder.CreateTrunc(Ops[0], Int16Ty);
8413  }
8414  case NEON::BI__builtin_neon_vminvq_u8: {
8415    Int = Intrinsic::aarch64_neon_uminv;
8416    Ty = Int32Ty;
8417    VTy = llvm::VectorType::get(Int8Ty, 16);
8418    llvm::Type *Tys[2] = { Ty, VTy };
8419    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8420    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8421    return Builder.CreateTrunc(Ops[0], Int8Ty);
8422  }
8423  case NEON::BI__builtin_neon_vminvq_u16: {
8424    Int = Intrinsic::aarch64_neon_uminv;
8425    Ty = Int32Ty;
8426    VTy = llvm::VectorType::get(Int16Ty, 8);
8427    llvm::Type *Tys[2] = { Ty, VTy };
8428    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8429    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8430    return Builder.CreateTrunc(Ops[0], Int16Ty);
8431  }
8432  case NEON::BI__builtin_neon_vminv_s8: {
8433    Int = Intrinsic::aarch64_neon_sminv;
8434    Ty = Int32Ty;
8435    VTy = llvm::VectorType::get(Int8Ty, 8);
8436    llvm::Type *Tys[2] = { Ty, VTy };
8437    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8438    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8439    return Builder.CreateTrunc(Ops[0], Int8Ty);
8440  }
8441  case NEON::BI__builtin_neon_vminv_s16: {
8442    Int = Intrinsic::aarch64_neon_sminv;
8443    Ty = Int32Ty;
8444    VTy = llvm::VectorType::get(Int16Ty, 4);
8445    llvm::Type *Tys[2] = { Ty, VTy };
8446    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8447    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8448    return Builder.CreateTrunc(Ops[0], Int16Ty);
8449  }
8450  case NEON::BI__builtin_neon_vminvq_s8: {
8451    Int = Intrinsic::aarch64_neon_sminv;
8452    Ty = Int32Ty;
8453    VTy = llvm::VectorType::get(Int8Ty, 16);
8454    llvm::Type *Tys[2] = { Ty, VTy };
8455    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8456    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8457    return Builder.CreateTrunc(Ops[0], Int8Ty);
8458  }
8459  case NEON::BI__builtin_neon_vminvq_s16: {
8460    Int = Intrinsic::aarch64_neon_sminv;
8461    Ty = Int32Ty;
8462    VTy = llvm::VectorType::get(Int16Ty, 8);
8463    llvm::Type *Tys[2] = { Ty, VTy };
8464    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8465    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8466    return Builder.CreateTrunc(Ops[0], Int16Ty);
8467  }
8468  case NEON::BI__builtin_neon_vminv_f16: {
8469    Int = Intrinsic::aarch64_neon_fminv;
8470    Ty = HalfTy;
8471    VTy = llvm::VectorType::get(HalfTy, 4);
8472    llvm::Type *Tys[2] = { Ty, VTy };
8473    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8474    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8475    return Builder.CreateTrunc(Ops[0], HalfTy);
8476  }
8477  case NEON::BI__builtin_neon_vminvq_f16: {
8478    Int = Intrinsic::aarch64_neon_fminv;
8479    Ty = HalfTy;
8480    VTy = llvm::VectorType::get(HalfTy, 8);
8481    llvm::Type *Tys[2] = { Ty, VTy };
8482    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8483    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminv");
8484    return Builder.CreateTrunc(Ops[0], HalfTy);
8485  }
8486  case NEON::BI__builtin_neon_vmaxnmv_f16: {
8487    Int = Intrinsic::aarch64_neon_fmaxnmv;
8488    Ty = HalfTy;
8489    VTy = llvm::VectorType::get(HalfTy, 4);
8490    llvm::Type *Tys[2] = { Ty, VTy };
8491    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8492    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv");
8493    return Builder.CreateTrunc(Ops[0], HalfTy);
8494  }
8495  case NEON::BI__builtin_neon_vmaxnmvq_f16: {
8496    Int = Intrinsic::aarch64_neon_fmaxnmv;
8497    Ty = HalfTy;
8498    VTy = llvm::VectorType::get(HalfTy, 8);
8499    llvm::Type *Tys[2] = { Ty, VTy };
8500    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8501    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vmaxnmv");
8502    return Builder.CreateTrunc(Ops[0], HalfTy);
8503  }
8504  case NEON::BI__builtin_neon_vminnmv_f16: {
8505    Int = Intrinsic::aarch64_neon_fminnmv;
8506    Ty = HalfTy;
8507    VTy = llvm::VectorType::get(HalfTy, 4);
8508    llvm::Type *Tys[2] = { Ty, VTy };
8509    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8510    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv");
8511    return Builder.CreateTrunc(Ops[0], HalfTy);
8512  }
8513  case NEON::BI__builtin_neon_vminnmvq_f16: {
8514    Int = Intrinsic::aarch64_neon_fminnmv;
8515    Ty = HalfTy;
8516    VTy = llvm::VectorType::get(HalfTy, 8);
8517    llvm::Type *Tys[2] = { Ty, VTy };
8518    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8519    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vminnmv");
8520    return Builder.CreateTrunc(Ops[0], HalfTy);
8521  }
8522  case NEON::BI__builtin_neon_vmul_n_f64: {
8523    Ops[0] = Builder.CreateBitCast(Ops[0], DoubleTy);
8524    Value *RHS = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), DoubleTy);
8525    return Builder.CreateFMul(Ops[0], RHS);
8526  }
8527  case NEON::BI__builtin_neon_vaddlv_u8: {
8528    Int = Intrinsic::aarch64_neon_uaddlv;
8529    Ty = Int32Ty;
8530    VTy = llvm::VectorType::get(Int8Ty, 8);
8531    llvm::Type *Tys[2] = { Ty, VTy };
8532    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8533    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8534    return Builder.CreateTrunc(Ops[0], Int16Ty);
8535  }
8536  case NEON::BI__builtin_neon_vaddlv_u16: {
8537    Int = Intrinsic::aarch64_neon_uaddlv;
8538    Ty = Int32Ty;
8539    VTy = llvm::VectorType::get(Int16Ty, 4);
8540    llvm::Type *Tys[2] = { Ty, VTy };
8541    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8542    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8543  }
8544  case NEON::BI__builtin_neon_vaddlvq_u8: {
8545    Int = Intrinsic::aarch64_neon_uaddlv;
8546    Ty = Int32Ty;
8547    VTy = llvm::VectorType::get(Int8Ty, 16);
8548    llvm::Type *Tys[2] = { Ty, VTy };
8549    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8550    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8551    return Builder.CreateTrunc(Ops[0], Int16Ty);
8552  }
8553  case NEON::BI__builtin_neon_vaddlvq_u16: {
8554    Int = Intrinsic::aarch64_neon_uaddlv;
8555    Ty = Int32Ty;
8556    VTy = llvm::VectorType::get(Int16Ty, 8);
8557    llvm::Type *Tys[2] = { Ty, VTy };
8558    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8559    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8560  }
8561  case NEON::BI__builtin_neon_vaddlv_s8: {
8562    Int = Intrinsic::aarch64_neon_saddlv;
8563    Ty = Int32Ty;
8564    VTy = llvm::VectorType::get(Int8Ty, 8);
8565    llvm::Type *Tys[2] = { Ty, VTy };
8566    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8567    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8568    return Builder.CreateTrunc(Ops[0], Int16Ty);
8569  }
8570  case NEON::BI__builtin_neon_vaddlv_s16: {
8571    Int = Intrinsic::aarch64_neon_saddlv;
8572    Ty = Int32Ty;
8573    VTy = llvm::VectorType::get(Int16Ty, 4);
8574    llvm::Type *Tys[2] = { Ty, VTy };
8575    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8576    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8577  }
8578  case NEON::BI__builtin_neon_vaddlvq_s8: {
8579    Int = Intrinsic::aarch64_neon_saddlv;
8580    Ty = Int32Ty;
8581    VTy = llvm::VectorType::get(Int8Ty, 16);
8582    llvm::Type *Tys[2] = { Ty, VTy };
8583    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8584    Ops[0] = EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8585    return Builder.CreateTrunc(Ops[0], Int16Ty);
8586  }
8587  case NEON::BI__builtin_neon_vaddlvq_s16: {
8588    Int = Intrinsic::aarch64_neon_saddlv;
8589    Ty = Int32Ty;
8590    VTy = llvm::VectorType::get(Int16Ty, 8);
8591    llvm::Type *Tys[2] = { Ty, VTy };
8592    Ops.push_back(EmitScalarExpr(E->getArg(0)));
8593    return EmitNeonCall(CGM.getIntrinsic(Int, Tys), Ops, "vaddlv");
8594  }
8595  case NEON::BI__builtin_neon_vsri_n_v:
8596  case NEON::BI__builtin_neon_vsriq_n_v: {
8597    Int = Intrinsic::aarch64_neon_vsri;
8598    llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty);
8599    return EmitNeonCall(Intrin, Ops, "vsri_n");
8600  }
8601  case NEON::BI__builtin_neon_vsli_n_v:
8602  case NEON::BI__builtin_neon_vsliq_n_v: {
8603    Int = Intrinsic::aarch64_neon_vsli;
8604    llvm::Function *Intrin = CGM.getIntrinsic(Int, Ty);
8605    return EmitNeonCall(Intrin, Ops, "vsli_n");
8606  }
8607  case NEON::BI__builtin_neon_vsra_n_v:
8608  case NEON::BI__builtin_neon_vsraq_n_v:
8609    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8610    Ops[1] = EmitNeonRShiftImm(Ops[1], Ops[2], Ty, usgn, "vsra_n");
8611    return Builder.CreateAdd(Ops[0], Ops[1]);
8612  case NEON::BI__builtin_neon_vrsra_n_v:
8613  case NEON::BI__builtin_neon_vrsraq_n_v: {
8614    Int = usgn ? Intrinsic::aarch64_neon_urshl : Intrinsic::aarch64_neon_srshl;
8615    SmallVector<llvm::Value*,2> TmpOps;
8616    TmpOps.push_back(Ops[1]);
8617    TmpOps.push_back(Ops[2]);
8618    Function* F = CGM.getIntrinsic(Int, Ty);
8619    llvm::Value *tmp = EmitNeonCall(F, TmpOps, "vrshr_n", 1, true);
8620    Ops[0] = Builder.CreateBitCast(Ops[0], VTy);
8621    return Builder.CreateAdd(Ops[0], tmp);
8622  }
8623  case NEON::BI__builtin_neon_vld1_v:
8624  case NEON::BI__builtin_neon_vld1q_v: {
8625    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy));
8626    auto Alignment = CharUnits::fromQuantity(
8627        BuiltinID == NEON::BI__builtin_neon_vld1_v ? 8 : 16);
8628    return Builder.CreateAlignedLoad(VTy, Ops[0], Alignment);
8629  }
8630  case NEON::BI__builtin_neon_vst1_v:
8631  case NEON::BI__builtin_neon_vst1q_v:
8632    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(VTy));
8633    Ops[1] = Builder.CreateBitCast(Ops[1], VTy);
8634    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8635  case NEON::BI__builtin_neon_vld1_lane_v:
8636  case NEON::BI__builtin_neon_vld1q_lane_v: {
8637    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8638    Ty = llvm::PointerType::getUnqual(VTy->getElementType());
8639    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8640    auto Alignment = CharUnits::fromQuantity(
8641        BuiltinID == NEON::BI__builtin_neon_vld1_lane_v ? 8 : 16);
8642    Ops[0] =
8643        Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment);
8644    return Builder.CreateInsertElement(Ops[1], Ops[0], Ops[2], "vld1_lane");
8645  }
8646  case NEON::BI__builtin_neon_vld1_dup_v:
8647  case NEON::BI__builtin_neon_vld1q_dup_v: {
8648    Value *V = UndefValue::get(Ty);
8649    Ty = llvm::PointerType::getUnqual(VTy->getElementType());
8650    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8651    auto Alignment = CharUnits::fromQuantity(
8652        BuiltinID == NEON::BI__builtin_neon_vld1_dup_v ? 8 : 16);
8653    Ops[0] =
8654        Builder.CreateAlignedLoad(VTy->getElementType(), Ops[0], Alignment);
8655    llvm::Constant *CI = ConstantInt::get(Int32Ty, 0);
8656    Ops[0] = Builder.CreateInsertElement(V, Ops[0], CI);
8657    return EmitNeonSplat(Ops[0], CI);
8658  }
8659  case NEON::BI__builtin_neon_vst1_lane_v:
8660  case NEON::BI__builtin_neon_vst1q_lane_v:
8661    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8662    Ops[1] = Builder.CreateExtractElement(Ops[1], Ops[2]);
8663    Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
8664    return Builder.CreateDefaultAlignedStore(Ops[1],
8665                                             Builder.CreateBitCast(Ops[0], Ty));
8666  case NEON::BI__builtin_neon_vld2_v:
8667  case NEON::BI__builtin_neon_vld2q_v: {
8668    llvm::Type *PTy = llvm::PointerType::getUnqual(VTy);
8669    Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8670    llvm::Type *Tys[2] = { VTy, PTy };
8671    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2, Tys);
8672    Ops[1] = Builder.CreateCall(F, Ops[1], "vld2");
8673    Ops[0] = Builder.CreateBitCast(Ops[0],
8674                llvm::PointerType::getUnqual(Ops[1]->getType()));
8675    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8676  }
8677  case NEON::BI__builtin_neon_vld3_v:
8678  case NEON::BI__builtin_neon_vld3q_v: {
8679    llvm::Type *PTy = llvm::PointerType::getUnqual(VTy);
8680    Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8681    llvm::Type *Tys[2] = { VTy, PTy };
8682    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3, Tys);
8683    Ops[1] = Builder.CreateCall(F, Ops[1], "vld3");
8684    Ops[0] = Builder.CreateBitCast(Ops[0],
8685                llvm::PointerType::getUnqual(Ops[1]->getType()));
8686    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8687  }
8688  case NEON::BI__builtin_neon_vld4_v:
8689  case NEON::BI__builtin_neon_vld4q_v: {
8690    llvm::Type *PTy = llvm::PointerType::getUnqual(VTy);
8691    Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8692    llvm::Type *Tys[2] = { VTy, PTy };
8693    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4, Tys);
8694    Ops[1] = Builder.CreateCall(F, Ops[1], "vld4");
8695    Ops[0] = Builder.CreateBitCast(Ops[0],
8696                llvm::PointerType::getUnqual(Ops[1]->getType()));
8697    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8698  }
8699  case NEON::BI__builtin_neon_vld2_dup_v:
8700  case NEON::BI__builtin_neon_vld2q_dup_v: {
8701    llvm::Type *PTy =
8702      llvm::PointerType::getUnqual(VTy->getElementType());
8703    Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8704    llvm::Type *Tys[2] = { VTy, PTy };
8705    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2r, Tys);
8706    Ops[1] = Builder.CreateCall(F, Ops[1], "vld2");
8707    Ops[0] = Builder.CreateBitCast(Ops[0],
8708                llvm::PointerType::getUnqual(Ops[1]->getType()));
8709    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8710  }
8711  case NEON::BI__builtin_neon_vld3_dup_v:
8712  case NEON::BI__builtin_neon_vld3q_dup_v: {
8713    llvm::Type *PTy =
8714      llvm::PointerType::getUnqual(VTy->getElementType());
8715    Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8716    llvm::Type *Tys[2] = { VTy, PTy };
8717    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3r, Tys);
8718    Ops[1] = Builder.CreateCall(F, Ops[1], "vld3");
8719    Ops[0] = Builder.CreateBitCast(Ops[0],
8720                llvm::PointerType::getUnqual(Ops[1]->getType()));
8721    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8722  }
8723  case NEON::BI__builtin_neon_vld4_dup_v:
8724  case NEON::BI__builtin_neon_vld4q_dup_v: {
8725    llvm::Type *PTy =
8726      llvm::PointerType::getUnqual(VTy->getElementType());
8727    Ops[1] = Builder.CreateBitCast(Ops[1], PTy);
8728    llvm::Type *Tys[2] = { VTy, PTy };
8729    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4r, Tys);
8730    Ops[1] = Builder.CreateCall(F, Ops[1], "vld4");
8731    Ops[0] = Builder.CreateBitCast(Ops[0],
8732                llvm::PointerType::getUnqual(Ops[1]->getType()));
8733    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8734  }
8735  case NEON::BI__builtin_neon_vld2_lane_v:
8736  case NEON::BI__builtin_neon_vld2q_lane_v: {
8737    llvm::Type *Tys[2] = { VTy, Ops[1]->getType() };
8738    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld2lane, Tys);
8739    Ops.push_back(Ops[1]);
8740    Ops.erase(Ops.begin()+1);
8741    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8742    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8743    Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty);
8744    Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld2_lane");
8745    Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
8746    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8747    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8748  }
8749  case NEON::BI__builtin_neon_vld3_lane_v:
8750  case NEON::BI__builtin_neon_vld3q_lane_v: {
8751    llvm::Type *Tys[2] = { VTy, Ops[1]->getType() };
8752    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld3lane, Tys);
8753    Ops.push_back(Ops[1]);
8754    Ops.erase(Ops.begin()+1);
8755    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8756    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8757    Ops[3] = Builder.CreateBitCast(Ops[3], Ty);
8758    Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty);
8759    Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld3_lane");
8760    Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
8761    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8762    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8763  }
8764  case NEON::BI__builtin_neon_vld4_lane_v:
8765  case NEON::BI__builtin_neon_vld4q_lane_v: {
8766    llvm::Type *Tys[2] = { VTy, Ops[1]->getType() };
8767    Function *F = CGM.getIntrinsic(Intrinsic::aarch64_neon_ld4lane, Tys);
8768    Ops.push_back(Ops[1]);
8769    Ops.erase(Ops.begin()+1);
8770    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8771    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8772    Ops[3] = Builder.CreateBitCast(Ops[3], Ty);
8773    Ops[4] = Builder.CreateBitCast(Ops[4], Ty);
8774    Ops[5] = Builder.CreateZExt(Ops[5], Int64Ty);
8775    Ops[1] = Builder.CreateCall(F, makeArrayRef(Ops).slice(1), "vld4_lane");
8776    Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
8777    Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
8778    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
8779  }
8780  case NEON::BI__builtin_neon_vst2_v:
8781  case NEON::BI__builtin_neon_vst2q_v: {
8782    Ops.push_back(Ops[0]);
8783    Ops.erase(Ops.begin());
8784    llvm::Type *Tys[2] = { VTy, Ops[2]->getType() };
8785    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2, Tys),
8786                        Ops, "");
8787  }
8788  case NEON::BI__builtin_neon_vst2_lane_v:
8789  case NEON::BI__builtin_neon_vst2q_lane_v: {
8790    Ops.push_back(Ops[0]);
8791    Ops.erase(Ops.begin());
8792    Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty);
8793    llvm::Type *Tys[2] = { VTy, Ops[3]->getType() };
8794    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st2lane, Tys),
8795                        Ops, "");
8796  }
8797  case NEON::BI__builtin_neon_vst3_v:
8798  case NEON::BI__builtin_neon_vst3q_v: {
8799    Ops.push_back(Ops[0]);
8800    Ops.erase(Ops.begin());
8801    llvm::Type *Tys[2] = { VTy, Ops[3]->getType() };
8802    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3, Tys),
8803                        Ops, "");
8804  }
8805  case NEON::BI__builtin_neon_vst3_lane_v:
8806  case NEON::BI__builtin_neon_vst3q_lane_v: {
8807    Ops.push_back(Ops[0]);
8808    Ops.erase(Ops.begin());
8809    Ops[3] = Builder.CreateZExt(Ops[3], Int64Ty);
8810    llvm::Type *Tys[2] = { VTy, Ops[4]->getType() };
8811    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st3lane, Tys),
8812                        Ops, "");
8813  }
8814  case NEON::BI__builtin_neon_vst4_v:
8815  case NEON::BI__builtin_neon_vst4q_v: {
8816    Ops.push_back(Ops[0]);
8817    Ops.erase(Ops.begin());
8818    llvm::Type *Tys[2] = { VTy, Ops[4]->getType() };
8819    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4, Tys),
8820                        Ops, "");
8821  }
8822  case NEON::BI__builtin_neon_vst4_lane_v:
8823  case NEON::BI__builtin_neon_vst4q_lane_v: {
8824    Ops.push_back(Ops[0]);
8825    Ops.erase(Ops.begin());
8826    Ops[4] = Builder.CreateZExt(Ops[4], Int64Ty);
8827    llvm::Type *Tys[2] = { VTy, Ops[5]->getType() };
8828    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_st4lane, Tys),
8829                        Ops, "");
8830  }
8831  case NEON::BI__builtin_neon_vtrn_v:
8832  case NEON::BI__builtin_neon_vtrnq_v: {
8833    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
8834    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8835    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8836    Value *SV = nullptr;
8837
8838    for (unsigned vi = 0; vi != 2; ++vi) {
8839      SmallVector<uint32_t, 16> Indices;
8840      for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
8841        Indices.push_back(i+vi);
8842        Indices.push_back(i+e+vi);
8843      }
8844      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
8845      SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vtrn");
8846      SV = Builder.CreateDefaultAlignedStore(SV, Addr);
8847    }
8848    return SV;
8849  }
8850  case NEON::BI__builtin_neon_vuzp_v:
8851  case NEON::BI__builtin_neon_vuzpq_v: {
8852    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
8853    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8854    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8855    Value *SV = nullptr;
8856
8857    for (unsigned vi = 0; vi != 2; ++vi) {
8858      SmallVector<uint32_t, 16> Indices;
8859      for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
8860        Indices.push_back(2*i+vi);
8861
8862      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
8863      SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vuzp");
8864      SV = Builder.CreateDefaultAlignedStore(SV, Addr);
8865    }
8866    return SV;
8867  }
8868  case NEON::BI__builtin_neon_vzip_v:
8869  case NEON::BI__builtin_neon_vzipq_v: {
8870    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::PointerType::getUnqual(Ty));
8871    Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
8872    Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
8873    Value *SV = nullptr;
8874
8875    for (unsigned vi = 0; vi != 2; ++vi) {
8876      SmallVector<uint32_t, 16> Indices;
8877      for (unsigned i = 0, e = VTy->getNumElements(); i != e; i += 2) {
8878        Indices.push_back((i + vi*e) >> 1);
8879        Indices.push_back(((i + vi*e) >> 1)+e);
8880      }
8881      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
8882      SV = Builder.CreateShuffleVector(Ops[1], Ops[2], Indices, "vzip");
8883      SV = Builder.CreateDefaultAlignedStore(SV, Addr);
8884    }
8885    return SV;
8886  }
8887  case NEON::BI__builtin_neon_vqtbl1q_v: {
8888    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl1, Ty),
8889                        Ops, "vtbl1");
8890  }
8891  case NEON::BI__builtin_neon_vqtbl2q_v: {
8892    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl2, Ty),
8893                        Ops, "vtbl2");
8894  }
8895  case NEON::BI__builtin_neon_vqtbl3q_v: {
8896    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl3, Ty),
8897                        Ops, "vtbl3");
8898  }
8899  case NEON::BI__builtin_neon_vqtbl4q_v: {
8900    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbl4, Ty),
8901                        Ops, "vtbl4");
8902  }
8903  case NEON::BI__builtin_neon_vqtbx1q_v: {
8904    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx1, Ty),
8905                        Ops, "vtbx1");
8906  }
8907  case NEON::BI__builtin_neon_vqtbx2q_v: {
8908    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx2, Ty),
8909                        Ops, "vtbx2");
8910  }
8911  case NEON::BI__builtin_neon_vqtbx3q_v: {
8912    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx3, Ty),
8913                        Ops, "vtbx3");
8914  }
8915  case NEON::BI__builtin_neon_vqtbx4q_v: {
8916    return EmitNeonCall(CGM.getIntrinsic(Intrinsic::aarch64_neon_tbx4, Ty),
8917                        Ops, "vtbx4");
8918  }
8919  case NEON::BI__builtin_neon_vsqadd_v:
8920  case NEON::BI__builtin_neon_vsqaddq_v: {
8921    Int = Intrinsic::aarch64_neon_usqadd;
8922    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vsqadd");
8923  }
8924  case NEON::BI__builtin_neon_vuqadd_v:
8925  case NEON::BI__builtin_neon_vuqaddq_v: {
8926    Int = Intrinsic::aarch64_neon_suqadd;
8927    return EmitNeonCall(CGM.getIntrinsic(Int, Ty), Ops, "vuqadd");
8928  }
8929  case AArch64::BI_BitScanForward:
8930  case AArch64::BI_BitScanForward64:
8931    return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E);
8932  case AArch64::BI_BitScanReverse:
8933  case AArch64::BI_BitScanReverse64:
8934    return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E);
8935  case AArch64::BI_InterlockedAnd64:
8936    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E);
8937  case AArch64::BI_InterlockedExchange64:
8938    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E);
8939  case AArch64::BI_InterlockedExchangeAdd64:
8940    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E);
8941  case AArch64::BI_InterlockedExchangeSub64:
8942    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E);
8943  case AArch64::BI_InterlockedOr64:
8944    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E);
8945  case AArch64::BI_InterlockedXor64:
8946    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E);
8947  case AArch64::BI_InterlockedDecrement64:
8948    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E);
8949  case AArch64::BI_InterlockedIncrement64:
8950    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E);
8951  case AArch64::BI_InterlockedExchangeAdd8_acq:
8952  case AArch64::BI_InterlockedExchangeAdd16_acq:
8953  case AArch64::BI_InterlockedExchangeAdd_acq:
8954  case AArch64::BI_InterlockedExchangeAdd64_acq:
8955    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_acq, E);
8956  case AArch64::BI_InterlockedExchangeAdd8_rel:
8957  case AArch64::BI_InterlockedExchangeAdd16_rel:
8958  case AArch64::BI_InterlockedExchangeAdd_rel:
8959  case AArch64::BI_InterlockedExchangeAdd64_rel:
8960    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_rel, E);
8961  case AArch64::BI_InterlockedExchangeAdd8_nf:
8962  case AArch64::BI_InterlockedExchangeAdd16_nf:
8963  case AArch64::BI_InterlockedExchangeAdd_nf:
8964  case AArch64::BI_InterlockedExchangeAdd64_nf:
8965    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd_nf, E);
8966  case AArch64::BI_InterlockedExchange8_acq:
8967  case AArch64::BI_InterlockedExchange16_acq:
8968  case AArch64::BI_InterlockedExchange_acq:
8969  case AArch64::BI_InterlockedExchange64_acq:
8970    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_acq, E);
8971  case AArch64::BI_InterlockedExchange8_rel:
8972  case AArch64::BI_InterlockedExchange16_rel:
8973  case AArch64::BI_InterlockedExchange_rel:
8974  case AArch64::BI_InterlockedExchange64_rel:
8975    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_rel, E);
8976  case AArch64::BI_InterlockedExchange8_nf:
8977  case AArch64::BI_InterlockedExchange16_nf:
8978  case AArch64::BI_InterlockedExchange_nf:
8979  case AArch64::BI_InterlockedExchange64_nf:
8980    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange_nf, E);
8981  case AArch64::BI_InterlockedCompareExchange8_acq:
8982  case AArch64::BI_InterlockedCompareExchange16_acq:
8983  case AArch64::BI_InterlockedCompareExchange_acq:
8984  case AArch64::BI_InterlockedCompareExchange64_acq:
8985    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_acq, E);
8986  case AArch64::BI_InterlockedCompareExchange8_rel:
8987  case AArch64::BI_InterlockedCompareExchange16_rel:
8988  case AArch64::BI_InterlockedCompareExchange_rel:
8989  case AArch64::BI_InterlockedCompareExchange64_rel:
8990    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_rel, E);
8991  case AArch64::BI_InterlockedCompareExchange8_nf:
8992  case AArch64::BI_InterlockedCompareExchange16_nf:
8993  case AArch64::BI_InterlockedCompareExchange_nf:
8994  case AArch64::BI_InterlockedCompareExchange64_nf:
8995    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedCompareExchange_nf, E);
8996  case AArch64::BI_InterlockedOr8_acq:
8997  case AArch64::BI_InterlockedOr16_acq:
8998  case AArch64::BI_InterlockedOr_acq:
8999  case AArch64::BI_InterlockedOr64_acq:
9000    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_acq, E);
9001  case AArch64::BI_InterlockedOr8_rel:
9002  case AArch64::BI_InterlockedOr16_rel:
9003  case AArch64::BI_InterlockedOr_rel:
9004  case AArch64::BI_InterlockedOr64_rel:
9005    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_rel, E);
9006  case AArch64::BI_InterlockedOr8_nf:
9007  case AArch64::BI_InterlockedOr16_nf:
9008  case AArch64::BI_InterlockedOr_nf:
9009  case AArch64::BI_InterlockedOr64_nf:
9010    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr_nf, E);
9011  case AArch64::BI_InterlockedXor8_acq:
9012  case AArch64::BI_InterlockedXor16_acq:
9013  case AArch64::BI_InterlockedXor_acq:
9014  case AArch64::BI_InterlockedXor64_acq:
9015    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_acq, E);
9016  case AArch64::BI_InterlockedXor8_rel:
9017  case AArch64::BI_InterlockedXor16_rel:
9018  case AArch64::BI_InterlockedXor_rel:
9019  case AArch64::BI_InterlockedXor64_rel:
9020    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_rel, E);
9021  case AArch64::BI_InterlockedXor8_nf:
9022  case AArch64::BI_InterlockedXor16_nf:
9023  case AArch64::BI_InterlockedXor_nf:
9024  case AArch64::BI_InterlockedXor64_nf:
9025    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor_nf, E);
9026  case AArch64::BI_InterlockedAnd8_acq:
9027  case AArch64::BI_InterlockedAnd16_acq:
9028  case AArch64::BI_InterlockedAnd_acq:
9029  case AArch64::BI_InterlockedAnd64_acq:
9030    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_acq, E);
9031  case AArch64::BI_InterlockedAnd8_rel:
9032  case AArch64::BI_InterlockedAnd16_rel:
9033  case AArch64::BI_InterlockedAnd_rel:
9034  case AArch64::BI_InterlockedAnd64_rel:
9035    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_rel, E);
9036  case AArch64::BI_InterlockedAnd8_nf:
9037  case AArch64::BI_InterlockedAnd16_nf:
9038  case AArch64::BI_InterlockedAnd_nf:
9039  case AArch64::BI_InterlockedAnd64_nf:
9040    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd_nf, E);
9041  case AArch64::BI_InterlockedIncrement16_acq:
9042  case AArch64::BI_InterlockedIncrement_acq:
9043  case AArch64::BI_InterlockedIncrement64_acq:
9044    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_acq, E);
9045  case AArch64::BI_InterlockedIncrement16_rel:
9046  case AArch64::BI_InterlockedIncrement_rel:
9047  case AArch64::BI_InterlockedIncrement64_rel:
9048    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_rel, E);
9049  case AArch64::BI_InterlockedIncrement16_nf:
9050  case AArch64::BI_InterlockedIncrement_nf:
9051  case AArch64::BI_InterlockedIncrement64_nf:
9052    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement_nf, E);
9053  case AArch64::BI_InterlockedDecrement16_acq:
9054  case AArch64::BI_InterlockedDecrement_acq:
9055  case AArch64::BI_InterlockedDecrement64_acq:
9056    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_acq, E);
9057  case AArch64::BI_InterlockedDecrement16_rel:
9058  case AArch64::BI_InterlockedDecrement_rel:
9059  case AArch64::BI_InterlockedDecrement64_rel:
9060    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_rel, E);
9061  case AArch64::BI_InterlockedDecrement16_nf:
9062  case AArch64::BI_InterlockedDecrement_nf:
9063  case AArch64::BI_InterlockedDecrement64_nf:
9064    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement_nf, E);
9065
9066  case AArch64::BI_InterlockedAdd: {
9067    Value *Arg0 = EmitScalarExpr(E->getArg(0));
9068    Value *Arg1 = EmitScalarExpr(E->getArg(1));
9069    AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
9070      AtomicRMWInst::Add, Arg0, Arg1,
9071      llvm::AtomicOrdering::SequentiallyConsistent);
9072    return Builder.CreateAdd(RMWI, Arg1);
9073  }
9074  }
9075}
9076
9077llvm::Value *CodeGenFunction::
9078BuildVector(ArrayRef<llvm::Value*> Ops) {
9079   (0) . __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 9080, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert((Ops.size() & (Ops.size() - 1)) == 0 &&
9080 (0) . __assert_fail ("(Ops.size() & (Ops.size() - 1)) == 0 && \"Not a power-of-two sized vector!\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 9080, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">         "Not a power-of-two sized vector!");
9081  bool AllConstants = true;
9082  for (unsigned i = 0, e = Ops.size(); i != e && AllConstants; ++i)
9083    AllConstants &= isa<Constant>(Ops[i]);
9084
9085  // If this is a constant vector, create a ConstantVector.
9086  if (AllConstants) {
9087    SmallVector<llvm::Constant*, 16> CstOps;
9088    for (unsigned i = 0, e = Ops.size(); i != e; ++i)
9089      CstOps.push_back(cast<Constant>(Ops[i]));
9090    return llvm::ConstantVector::get(CstOps);
9091  }
9092
9093  // Otherwise, insertelement the values to build the vector.
9094  Value *Result =
9095    llvm::UndefValue::get(llvm::VectorType::get(Ops[0]->getType(), Ops.size()));
9096
9097  for (unsigned i = 0, e = Ops.size(); i != e; ++i)
9098    Result = Builder.CreateInsertElement(Result, Ops[i], Builder.getInt32(i));
9099
9100  return Result;
9101}
9102
9103// Convert the mask from an integer type to a vector of i1.
9104static Value *getMaskVecValue(CodeGenFunction &CGF, Value *Mask,
9105                              unsigned NumElts) {
9106
9107  llvm::VectorType *MaskTy = llvm::VectorType::get(CGF.Builder.getInt1Ty(),
9108                         cast<IntegerType>(Mask->getType())->getBitWidth());
9109  Value *MaskVec = CGF.Builder.CreateBitCast(Mask, MaskTy);
9110
9111  // If we have less than 8 elements, then the starting mask was an i8 and
9112  // we need to extract down to the right number of elements.
9113  if (NumElts < 8) {
9114    uint32_t Indices[4];
9115    for (unsigned i = 0; i != NumElts; ++i)
9116      Indices[i] = i;
9117    MaskVec = CGF.Builder.CreateShuffleVector(MaskVec, MaskVec,
9118                                             makeArrayRef(Indices, NumElts),
9119                                             "extract");
9120  }
9121  return MaskVec;
9122}
9123
9124static Value *EmitX86MaskedStore(CodeGenFunction &CGF,
9125                                 ArrayRef<Value *> Ops,
9126                                 unsigned Align) {
9127  // Cast the pointer to right type.
9128  Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
9129                               llvm::PointerType::getUnqual(Ops[1]->getType()));
9130
9131  Value *MaskVec = getMaskVecValue(CGF, Ops[2],
9132                                   Ops[1]->getType()->getVectorNumElements());
9133
9134  return CGF.Builder.CreateMaskedStore(Ops[1], Ptr, Align, MaskVec);
9135}
9136
9137static Value *EmitX86MaskedLoad(CodeGenFunction &CGF,
9138                                ArrayRef<Value *> Ops, unsigned Align) {
9139  // Cast the pointer to right type.
9140  Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
9141                               llvm::PointerType::getUnqual(Ops[1]->getType()));
9142
9143  Value *MaskVec = getMaskVecValue(CGF, Ops[2],
9144                                   Ops[1]->getType()->getVectorNumElements());
9145
9146  return CGF.Builder.CreateMaskedLoad(Ptr, Align, MaskVec, Ops[1]);
9147}
9148
9149static Value *EmitX86ExpandLoad(CodeGenFunction &CGF,
9150                                ArrayRef<Value *> Ops) {
9151  llvm::Type *ResultTy = Ops[1]->getType();
9152  llvm::Type *PtrTy = ResultTy->getVectorElementType();
9153
9154  // Cast the pointer to element type.
9155  Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
9156                                         llvm::PointerType::getUnqual(PtrTy));
9157
9158  Value *MaskVec = getMaskVecValue(CGF, Ops[2],
9159                                   ResultTy->getVectorNumElements());
9160
9161  llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_expandload,
9162                                           ResultTy);
9163  return CGF.Builder.CreateCall(F, { Ptr, MaskVec, Ops[1] });
9164}
9165
9166static Value *EmitX86CompressExpand(CodeGenFunction &CGF,
9167                                    ArrayRef<Value *> Ops,
9168                                    bool IsCompress) {
9169  llvm::Type *ResultTy = Ops[1]->getType();
9170
9171  Value *MaskVec = getMaskVecValue(CGF, Ops[2],
9172                                   ResultTy->getVectorNumElements());
9173
9174  Intrinsic::ID IID = IsCompress ? Intrinsic::x86_avx512_mask_compress
9175                                 : Intrinsic::x86_avx512_mask_expand;
9176  llvm::Function *F = CGF.CGM.getIntrinsic(IID, ResultTy);
9177  return CGF.Builder.CreateCall(F, { Ops[0], Ops[1], MaskVec });
9178}
9179
9180static Value *EmitX86CompressStore(CodeGenFunction &CGF,
9181                                   ArrayRef<Value *> Ops) {
9182  llvm::Type *ResultTy = Ops[1]->getType();
9183  llvm::Type *PtrTy = ResultTy->getVectorElementType();
9184
9185  // Cast the pointer to element type.
9186  Value *Ptr = CGF.Builder.CreateBitCast(Ops[0],
9187                                         llvm::PointerType::getUnqual(PtrTy));
9188
9189  Value *MaskVec = getMaskVecValue(CGF, Ops[2],
9190                                   ResultTy->getVectorNumElements());
9191
9192  llvm::Function *F = CGF.CGM.getIntrinsic(Intrinsic::masked_compressstore,
9193                                           ResultTy);
9194  return CGF.Builder.CreateCall(F, { Ops[1], Ptr, MaskVec });
9195}
9196
9197static Value *EmitX86MaskLogic(CodeGenFunction &CGF, Instruction::BinaryOps Opc,
9198                              ArrayRef<Value *> Ops,
9199                              bool InvertLHS = false) {
9200  unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
9201  Value *LHS = getMaskVecValue(CGF, Ops[0], NumElts);
9202  Value *RHS = getMaskVecValue(CGF, Ops[1], NumElts);
9203
9204  if (InvertLHS)
9205    LHS = CGF.Builder.CreateNot(LHS);
9206
9207  return CGF.Builder.CreateBitCast(CGF.Builder.CreateBinOp(Opc, LHS, RHS),
9208                                   Ops[0]->getType());
9209}
9210
9211static Value *EmitX86FunnelShift(CodeGenFunction &CGF, Value *Op0, Value *Op1,
9212                                 Value *Amt, bool IsRight) {
9213  llvm::Type *Ty = Op0->getType();
9214
9215  // Amount may be scalar immediate, in which case create a splat vector.
9216  // Funnel shifts amounts are treated as modulo and types are all power-of-2 so
9217  // we only care about the lowest log2 bits anyway.
9218  if (Amt->getType() != Ty) {
9219    unsigned NumElts = Ty->getVectorNumElements();
9220    Amt = CGF.Builder.CreateIntCast(Amt, Ty->getScalarType(), false);
9221    Amt = CGF.Builder.CreateVectorSplat(NumElts, Amt);
9222  }
9223
9224  unsigned IID = IsRight ? Intrinsic::fshr : Intrinsic::fshl;
9225  Function *F = CGF.CGM.getIntrinsic(IID, Ty);
9226  return CGF.Builder.CreateCall(F, {Op0, Op1, Amt});
9227}
9228
9229static Value *EmitX86vpcom(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
9230                           bool IsSigned) {
9231  Value *Op0 = Ops[0];
9232  Value *Op1 = Ops[1];
9233  llvm::Type *Ty = Op0->getType();
9234  uint64_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
9235
9236  CmpInst::Predicate Pred;
9237  switch (Imm) {
9238  case 0x0:
9239    Pred = IsSigned ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT;
9240    break;
9241  case 0x1:
9242    Pred = IsSigned ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE;
9243    break;
9244  case 0x2:
9245    Pred = IsSigned ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT;
9246    break;
9247  case 0x3:
9248    Pred = IsSigned ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE;
9249    break;
9250  case 0x4:
9251    Pred = ICmpInst::ICMP_EQ;
9252    break;
9253  case 0x5:
9254    Pred = ICmpInst::ICMP_NE;
9255    break;
9256  case 0x6:
9257    return llvm::Constant::getNullValue(Ty); // FALSE
9258  case 0x7:
9259    return llvm::Constant::getAllOnesValue(Ty); // TRUE
9260  default:
9261    llvm_unreachable("Unexpected XOP vpcom/vpcomu predicate");
9262  }
9263
9264  Value *Cmp = CGF.Builder.CreateICmp(Pred, Op0, Op1);
9265  Value *Res = CGF.Builder.CreateSExt(Cmp, Ty);
9266  return Res;
9267}
9268
9269static Value *EmitX86Select(CodeGenFunction &CGF,
9270                            Value *Mask, Value *Op0, Value *Op1) {
9271
9272  // If the mask is all ones just return first argument.
9273  if (const auto *C = dyn_cast<Constant>(Mask))
9274    if (C->isAllOnesValue())
9275      return Op0;
9276
9277  Mask = getMaskVecValue(CGF, Mask, Op0->getType()->getVectorNumElements());
9278
9279  return CGF.Builder.CreateSelect(Mask, Op0, Op1);
9280}
9281
9282static Value *EmitX86ScalarSelect(CodeGenFunction &CGF,
9283                                  Value *Mask, Value *Op0, Value *Op1) {
9284  // If the mask is all ones just return first argument.
9285  if (const auto *C = dyn_cast<Constant>(Mask))
9286    if (C->isAllOnesValue())
9287      return Op0;
9288
9289  llvm::VectorType *MaskTy =
9290    llvm::VectorType::get(CGF.Builder.getInt1Ty(),
9291                          Mask->getType()->getIntegerBitWidth());
9292  Mask = CGF.Builder.CreateBitCast(Mask, MaskTy);
9293  Mask = CGF.Builder.CreateExtractElement(Mask, (uint64_t)0);
9294  return CGF.Builder.CreateSelect(Mask, Op0, Op1);
9295}
9296
9297static Value *EmitX86MaskedCompareResult(CodeGenFunction &CGF, Value *Cmp,
9298                                         unsigned NumElts, Value *MaskIn) {
9299  if (MaskIn) {
9300    const auto *C = dyn_cast<Constant>(MaskIn);
9301    if (!C || !C->isAllOnesValue())
9302      Cmp = CGF.Builder.CreateAnd(Cmp, getMaskVecValue(CGF, MaskIn, NumElts));
9303  }
9304
9305  if (NumElts < 8) {
9306    uint32_t Indices[8];
9307    for (unsigned i = 0; i != NumElts; ++i)
9308      Indices[i] = i;
9309    for (unsigned i = NumElts; i != 8; ++i)
9310      Indices[i] = i % NumElts + NumElts;
9311    Cmp = CGF.Builder.CreateShuffleVector(
9312        Cmp, llvm::Constant::getNullValue(Cmp->getType()), Indices);
9313  }
9314
9315  return CGF.Builder.CreateBitCast(Cmp,
9316                                   IntegerType::get(CGF.getLLVMContext(),
9317                                                    std::max(NumElts, 8U)));
9318}
9319
9320static Value *EmitX86MaskedCompare(CodeGenFunction &CGF, unsigned CC,
9321                                   bool Signed, ArrayRef<Value *> Ops) {
9322   (0) . __assert_fail ("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 9323, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert((Ops.size() == 2 || Ops.size() == 4) &&
9323 (0) . __assert_fail ("(Ops.size() == 2 || Ops.size() == 4) && \"Unexpected number of arguments\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 9323, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">         "Unexpected number of arguments");
9324  unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
9325  Value *Cmp;
9326
9327  if (CC == 3) {
9328    Cmp = Constant::getNullValue(
9329                       llvm::VectorType::get(CGF.Builder.getInt1Ty(), NumElts));
9330  } else if (CC == 7) {
9331    Cmp = Constant::getAllOnesValue(
9332                       llvm::VectorType::get(CGF.Builder.getInt1Ty(), NumElts));
9333  } else {
9334    ICmpInst::Predicate Pred;
9335    switch (CC) {
9336    default: llvm_unreachable("Unknown condition code");
9337    case 0: Pred = ICmpInst::ICMP_EQ;  break;
9338    case 1: Pred = Signed ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; break;
9339    case 2: Pred = Signed ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; break;
9340    case 4: Pred = ICmpInst::ICMP_NE;  break;
9341    case 5: Pred = Signed ? ICmpInst::ICMP_SGE : ICmpInst::ICMP_UGE; break;
9342    case 6: Pred = Signed ? ICmpInst::ICMP_SGT : ICmpInst::ICMP_UGT; break;
9343    }
9344    Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]);
9345  }
9346
9347  Value *MaskIn = nullptr;
9348  if (Ops.size() == 4)
9349    MaskIn = Ops[3];
9350
9351  return EmitX86MaskedCompareResult(CGF, Cmp, NumElts, MaskIn);
9352}
9353
9354static Value *EmitX86ConvertToMask(CodeGenFunction &CGF, Value *In) {
9355  Value *Zero = Constant::getNullValue(In->getType());
9356  return EmitX86MaskedCompare(CGF, 1, true, { In, Zero });
9357}
9358
9359static Value *EmitX86ConvertIntToFp(CodeGenFunction &CGF,
9360                                    ArrayRef<Value *> Ops, bool IsSigned) {
9361  unsigned Rnd = cast<llvm::ConstantInt>(Ops[3])->getZExtValue();
9362  llvm::Type *Ty = Ops[1]->getType();
9363
9364  Value *Res;
9365  if (Rnd != 4) {
9366    Intrinsic::ID IID = IsSigned ? Intrinsic::x86_avx512_sitofp_round
9367                                 : Intrinsic::x86_avx512_uitofp_round;
9368    Function *F = CGF.CGM.getIntrinsic(IID, { Ty, Ops[0]->getType() });
9369    Res = CGF.Builder.CreateCall(F, { Ops[0], Ops[3] });
9370  } else {
9371    Res = IsSigned ? CGF.Builder.CreateSIToFP(Ops[0], Ty)
9372                   : CGF.Builder.CreateUIToFP(Ops[0], Ty);
9373  }
9374
9375  return EmitX86Select(CGF, Ops[2], Res, Ops[1]);
9376}
9377
9378static Value *EmitX86Abs(CodeGenFunction &CGF, ArrayRef<Value *> Ops) {
9379
9380  llvm::Type *Ty = Ops[0]->getType();
9381  Value *Zero = llvm::Constant::getNullValue(Ty);
9382  Value *Sub = CGF.Builder.CreateSub(Zero, Ops[0]);
9383  Value *Cmp = CGF.Builder.CreateICmp(ICmpInst::ICMP_SGT, Ops[0], Zero);
9384  Value *Res = CGF.Builder.CreateSelect(Cmp, Ops[0], Sub);
9385  return Res;
9386}
9387
9388static Value *EmitX86MinMax(CodeGenFunction &CGF, ICmpInst::Predicate Pred,
9389                            ArrayRef<Value *> Ops) {
9390  Value *Cmp = CGF.Builder.CreateICmp(Pred, Ops[0], Ops[1]);
9391  Value *Res = CGF.Builder.CreateSelect(Cmp, Ops[0], Ops[1]);
9392
9393  assert(Ops.size() == 2);
9394  return Res;
9395}
9396
9397// Lowers X86 FMA intrinsics to IR.
9398static Value *EmitX86FMAExpr(CodeGenFunction &CGF, ArrayRef<Value *> Ops,
9399                             unsigned BuiltinID, bool IsAddSub) {
9400
9401  bool Subtract = false;
9402  Intrinsic::ID IID = Intrinsic::not_intrinsic;
9403  switch (BuiltinID) {
9404  default: break;
9405  case clang::X86::BI__builtin_ia32_vfmsubps512_mask3:
9406    Subtract = true;
9407    LLVM_FALLTHROUGH;
9408  case clang::X86::BI__builtin_ia32_vfmaddps512_mask:
9409  case clang::X86::BI__builtin_ia32_vfmaddps512_maskz:
9410  case clang::X86::BI__builtin_ia32_vfmaddps512_mask3:
9411    IID = llvm::Intrinsic::x86_avx512_vfmadd_ps_512; break;
9412  case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3:
9413    Subtract = true;
9414    LLVM_FALLTHROUGH;
9415  case clang::X86::BI__builtin_ia32_vfmaddpd512_mask:
9416  case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz:
9417  case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3:
9418    IID = llvm::Intrinsic::x86_avx512_vfmadd_pd_512; break;
9419  case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3:
9420    Subtract = true;
9421    LLVM_FALLTHROUGH;
9422  case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask:
9423  case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz:
9424  case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3:
9425    IID = llvm::Intrinsic::x86_avx512_vfmaddsub_ps_512;
9426    break;
9427  case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3:
9428    Subtract = true;
9429    LLVM_FALLTHROUGH;
9430  case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask:
9431  case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz:
9432  case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3:
9433    IID = llvm::Intrinsic::x86_avx512_vfmaddsub_pd_512;
9434    break;
9435  }
9436
9437  Value *A = Ops[0];
9438  Value *B = Ops[1];
9439  Value *C = Ops[2];
9440
9441  if (Subtract)
9442    C = CGF.Builder.CreateFNeg(C);
9443
9444  Value *Res;
9445
9446  // Only handle in case of _MM_FROUND_CUR_DIRECTION/4 (no rounding).
9447  if (IID != Intrinsic::not_intrinsic &&
9448      cast<llvm::ConstantInt>(Ops.back())->getZExtValue() != (uint64_t)4) {
9449    Function *Intr = CGF.CGM.getIntrinsic(IID);
9450    Res = CGF.Builder.CreateCall(Intr, {A, B, C, Ops.back() });
9451  } else {
9452    llvm::Type *Ty = A->getType();
9453    Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ty);
9454    Res = CGF.Builder.CreateCall(FMA, {A, B, C} );
9455
9456    if (IsAddSub) {
9457      // Negate even elts in C using a mask.
9458      unsigned NumElts = Ty->getVectorNumElements();
9459      SmallVector<uint32_t, 16> Indices(NumElts);
9460      for (unsigned i = 0; i != NumElts; ++i)
9461        Indices[i] = i + (i % 2) * NumElts;
9462
9463      Value *NegC = CGF.Builder.CreateFNeg(C);
9464      Value *FMSub = CGF.Builder.CreateCall(FMA, {A, B, NegC} );
9465      Res = CGF.Builder.CreateShuffleVector(FMSub, Res, Indices);
9466    }
9467  }
9468
9469  // Handle any required masking.
9470  Value *MaskFalseVal = nullptr;
9471  switch (BuiltinID) {
9472  case clang::X86::BI__builtin_ia32_vfmaddps512_mask:
9473  case clang::X86::BI__builtin_ia32_vfmaddpd512_mask:
9474  case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask:
9475  case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask:
9476    MaskFalseVal = Ops[0];
9477    break;
9478  case clang::X86::BI__builtin_ia32_vfmaddps512_maskz:
9479  case clang::X86::BI__builtin_ia32_vfmaddpd512_maskz:
9480  case clang::X86::BI__builtin_ia32_vfmaddsubps512_maskz:
9481  case clang::X86::BI__builtin_ia32_vfmaddsubpd512_maskz:
9482    MaskFalseVal = Constant::getNullValue(Ops[0]->getType());
9483    break;
9484  case clang::X86::BI__builtin_ia32_vfmsubps512_mask3:
9485  case clang::X86::BI__builtin_ia32_vfmaddps512_mask3:
9486  case clang::X86::BI__builtin_ia32_vfmsubpd512_mask3:
9487  case clang::X86::BI__builtin_ia32_vfmaddpd512_mask3:
9488  case clang::X86::BI__builtin_ia32_vfmsubaddps512_mask3:
9489  case clang::X86::BI__builtin_ia32_vfmaddsubps512_mask3:
9490  case clang::X86::BI__builtin_ia32_vfmsubaddpd512_mask3:
9491  case clang::X86::BI__builtin_ia32_vfmaddsubpd512_mask3:
9492    MaskFalseVal = Ops[2];
9493    break;
9494  }
9495
9496  if (MaskFalseVal)
9497    return EmitX86Select(CGF, Ops[3], Res, MaskFalseVal);
9498
9499  return Res;
9500}
9501
9502static Value *
9503EmitScalarFMAExpr(CodeGenFunction &CGF, MutableArrayRef<Value *> Ops,
9504                  Value *Upper, bool ZeroMask = false, unsigned PTIdx = 0,
9505                  bool NegAcc = false) {
9506  unsigned Rnd = 4;
9507  if (Ops.size() > 4)
9508    Rnd = cast<llvm::ConstantInt>(Ops[4])->getZExtValue();
9509
9510  if (NegAcc)
9511    Ops[2] = CGF.Builder.CreateFNeg(Ops[2]);
9512
9513  Ops[0] = CGF.Builder.CreateExtractElement(Ops[0], (uint64_t)0);
9514  Ops[1] = CGF.Builder.CreateExtractElement(Ops[1], (uint64_t)0);
9515  Ops[2] = CGF.Builder.CreateExtractElement(Ops[2], (uint64_t)0);
9516  Value *Res;
9517  if (Rnd != 4) {
9518    Intrinsic::ID IID = Ops[0]->getType()->getPrimitiveSizeInBits() == 32 ?
9519                        Intrinsic::x86_avx512_vfmadd_f32 :
9520                        Intrinsic::x86_avx512_vfmadd_f64;
9521    Res = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID),
9522                                 {Ops[0], Ops[1], Ops[2], Ops[4]});
9523  } else {
9524    Function *FMA = CGF.CGM.getIntrinsic(Intrinsic::fma, Ops[0]->getType());
9525    Res = CGF.Builder.CreateCall(FMA, Ops.slice(0, 3));
9526  }
9527  // If we have more than 3 arguments, we need to do masking.
9528  if (Ops.size() > 3) {
9529    Value *PassThru = ZeroMask ? Constant::getNullValue(Res->getType())
9530                               : Ops[PTIdx];
9531
9532    // If we negated the accumulator and the its the PassThru value we need to
9533    // bypass the negate. Conveniently Upper should be the same thing in this
9534    // case.
9535    if (NegAcc && PTIdx == 2)
9536      PassThru = CGF.Builder.CreateExtractElement(Upper, (uint64_t)0);
9537
9538    Res = EmitX86ScalarSelect(CGF, Ops[3], Res, PassThru);
9539  }
9540  return CGF.Builder.CreateInsertElement(Upper, Res, (uint64_t)0);
9541}
9542
9543static Value *EmitX86Muldq(CodeGenFunction &CGF, bool IsSigned,
9544                           ArrayRef<Value *> Ops) {
9545  llvm::Type *Ty = Ops[0]->getType();
9546  // Arguments have a vXi32 type so cast to vXi64.
9547  Ty = llvm::VectorType::get(CGF.Int64Ty,
9548                             Ty->getPrimitiveSizeInBits() / 64);
9549  Value *LHS = CGF.Builder.CreateBitCast(Ops[0], Ty);
9550  Value *RHS = CGF.Builder.CreateBitCast(Ops[1], Ty);
9551
9552  if (IsSigned) {
9553    // Shift left then arithmetic shift right.
9554    Constant *ShiftAmt = ConstantInt::get(Ty, 32);
9555    LHS = CGF.Builder.CreateShl(LHS, ShiftAmt);
9556    LHS = CGF.Builder.CreateAShr(LHS, ShiftAmt);
9557    RHS = CGF.Builder.CreateShl(RHS, ShiftAmt);
9558    RHS = CGF.Builder.CreateAShr(RHS, ShiftAmt);
9559  } else {
9560    // Clear the upper bits.
9561    Constant *Mask = ConstantInt::get(Ty, 0xffffffff);
9562    LHS = CGF.Builder.CreateAnd(LHS, Mask);
9563    RHS = CGF.Builder.CreateAnd(RHS, Mask);
9564  }
9565
9566  return CGF.Builder.CreateMul(LHS, RHS);
9567}
9568
9569// Emit a masked pternlog intrinsic. This only exists because the header has to
9570// use a macro and we aren't able to pass the input argument to a pternlog
9571// builtin and a select builtin without evaluating it twice.
9572static Value *EmitX86Ternlog(CodeGenFunction &CGF, bool ZeroMask,
9573                             ArrayRef<Value *> Ops) {
9574  llvm::Type *Ty = Ops[0]->getType();
9575
9576  unsigned VecWidth = Ty->getPrimitiveSizeInBits();
9577  unsigned EltWidth = Ty->getScalarSizeInBits();
9578  Intrinsic::ID IID;
9579  if (VecWidth == 128 && EltWidth == 32)
9580    IID = Intrinsic::x86_avx512_pternlog_d_128;
9581  else if (VecWidth == 256 && EltWidth == 32)
9582    IID = Intrinsic::x86_avx512_pternlog_d_256;
9583  else if (VecWidth == 512 && EltWidth == 32)
9584    IID = Intrinsic::x86_avx512_pternlog_d_512;
9585  else if (VecWidth == 128 && EltWidth == 64)
9586    IID = Intrinsic::x86_avx512_pternlog_q_128;
9587  else if (VecWidth == 256 && EltWidth == 64)
9588    IID = Intrinsic::x86_avx512_pternlog_q_256;
9589  else if (VecWidth == 512 && EltWidth == 64)
9590    IID = Intrinsic::x86_avx512_pternlog_q_512;
9591  else
9592    llvm_unreachable("Unexpected intrinsic");
9593
9594  Value *Ternlog = CGF.Builder.CreateCall(CGF.CGM.getIntrinsic(IID),
9595                                          Ops.drop_back());
9596  Value *PassThru = ZeroMask ? ConstantAggregateZero::get(Ty) : Ops[0];
9597  return EmitX86Select(CGF, Ops[4], Ternlog, PassThru);
9598}
9599
9600static Value *EmitX86SExtMask(CodeGenFunction &CGF, Value *Op,
9601                              llvm::Type *DstTy) {
9602  unsigned NumberOfElements = DstTy->getVectorNumElements();
9603  Value *Mask = getMaskVecValue(CGF, Op, NumberOfElements);
9604  return CGF.Builder.CreateSExt(Mask, DstTy, "vpmovm2");
9605}
9606
9607// Emit addition or subtraction with signed/unsigned saturation.
9608static Value *EmitX86AddSubSatExpr(CodeGenFunction &CGF,
9609                                   ArrayRef<Value *> Ops, bool IsSigned,
9610                                   bool IsAddition) {
9611  Intrinsic::ID IID =
9612      IsSigned ? (IsAddition ? Intrinsic::sadd_sat : Intrinsic::ssub_sat)
9613               : (IsAddition ? Intrinsic::uadd_sat : Intrinsic::usub_sat);
9614  llvm::Function *F = CGF.CGM.getIntrinsic(IID, Ops[0]->getType());
9615  return CGF.Builder.CreateCall(F, {Ops[0], Ops[1]});
9616}
9617
9618Value *CodeGenFunction::EmitX86CpuIs(const CallExpr *E) {
9619  const Expr *CPUExpr = E->getArg(0)->IgnoreParenCasts();
9620  StringRef CPUStr = cast<clang::StringLiteral>(CPUExpr)->getString();
9621  return EmitX86CpuIs(CPUStr);
9622}
9623
9624Value *CodeGenFunction::EmitX86CpuIs(StringRef CPUStr) {
9625
9626  llvm::Type *Int32Ty = Builder.getInt32Ty();
9627
9628  // Matching the struct layout from the compiler-rt/libgcc structure that is
9629  // filled in:
9630  // unsigned int __cpu_vendor;
9631  // unsigned int __cpu_type;
9632  // unsigned int __cpu_subtype;
9633  // unsigned int __cpu_features[1];
9634  llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty,
9635                                          llvm::ArrayType::get(Int32Ty, 1));
9636
9637  // Grab the global __cpu_model.
9638  llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model");
9639  cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true);
9640
9641  // Calculate the index needed to access the correct field based on the
9642  // range. Also adjust the expected value.
9643  unsigned Index;
9644  unsigned Value;
9645  std::tie(Index, Value) = StringSwitch<std::pair<unsigned, unsigned>>(CPUStr)
9646#define X86_VENDOR(ENUM, STRING)                                               \
9647  .Case(STRING, {0u, static_cast<unsigned>(llvm::X86::ENUM)})
9648#define X86_CPU_TYPE_COMPAT_WITH_ALIAS(ARCHNAME, ENUM, STR, ALIAS)             \
9649  .Cases(STR, ALIAS, {1u, static_cast<unsigned>(llvm::X86::ENUM)})
9650#define X86_CPU_TYPE_COMPAT(ARCHNAME, ENUM, STR)                               \
9651  .Case(STR, {1u, static_cast<unsigned>(llvm::X86::ENUM)})
9652#define X86_CPU_SUBTYPE_COMPAT(ARCHNAME, ENUM, STR)                            \
9653  .Case(STR, {2u, static_cast<unsigned>(llvm::X86::ENUM)})
9654#include "llvm/Support/X86TargetParser.def"
9655                               .Default({0, 0});
9656   (0) . __assert_fail ("Value != 0 && \"Invalid CPUStr passed to CpuIs\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 9656, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Value != 0 && "Invalid CPUStr passed to CpuIs");
9657
9658  // Grab the appropriate field from __cpu_model.
9659  llvm::Value *Idxs[] = {ConstantInt::get(Int32Ty, 0),
9660                         ConstantInt::get(Int32Ty, Index)};
9661  llvm::Value *CpuValue = Builder.CreateGEP(STy, CpuModel, Idxs);
9662  CpuValue = Builder.CreateAlignedLoad(CpuValue, CharUnits::fromQuantity(4));
9663
9664  // Check the value of the field against the requested value.
9665  return Builder.CreateICmpEQ(CpuValue,
9666                                  llvm::ConstantInt::get(Int32Ty, Value));
9667}
9668
9669Value *CodeGenFunction::EmitX86CpuSupports(const CallExpr *E) {
9670  const Expr *FeatureExpr = E->getArg(0)->IgnoreParenCasts();
9671  StringRef FeatureStr = cast<StringLiteral>(FeatureExpr)->getString();
9672  return EmitX86CpuSupports(FeatureStr);
9673}
9674
9675uint64_t
9676CodeGenFunction::GetX86CpuSupportsMask(ArrayRef<StringRef> FeatureStrs) {
9677  // Processor features and mapping to processor feature value.
9678  uint64_t FeaturesMask = 0;
9679  for (const StringRef &FeatureStr : FeatureStrs) {
9680    unsigned Feature =
9681        StringSwitch<unsigned>(FeatureStr)
9682#define X86_FEATURE_COMPAT(VAL, ENUM, STR) .Case(STR, VAL)
9683#include "llvm/Support/X86TargetParser.def"
9684        ;
9685    FeaturesMask |= (1ULL << Feature);
9686  }
9687  return FeaturesMask;
9688}
9689
9690Value *CodeGenFunction::EmitX86CpuSupports(ArrayRef<StringRef> FeatureStrs) {
9691  return EmitX86CpuSupports(GetX86CpuSupportsMask(FeatureStrs));
9692}
9693
9694llvm::Value *CodeGenFunction::EmitX86CpuSupports(uint64_t FeaturesMask) {
9695  uint32_t Features1 = Lo_32(FeaturesMask);
9696  uint32_t Features2 = Hi_32(FeaturesMask);
9697
9698  Value *Result = Builder.getTrue();
9699
9700  if (Features1 != 0) {
9701    // Matching the struct layout from the compiler-rt/libgcc structure that is
9702    // filled in:
9703    // unsigned int __cpu_vendor;
9704    // unsigned int __cpu_type;
9705    // unsigned int __cpu_subtype;
9706    // unsigned int __cpu_features[1];
9707    llvm::Type *STy = llvm::StructType::get(Int32Ty, Int32Ty, Int32Ty,
9708                                            llvm::ArrayType::get(Int32Ty, 1));
9709
9710    // Grab the global __cpu_model.
9711    llvm::Constant *CpuModel = CGM.CreateRuntimeVariable(STy, "__cpu_model");
9712    cast<llvm::GlobalValue>(CpuModel)->setDSOLocal(true);
9713
9714    // Grab the first (0th) element from the field __cpu_features off of the
9715    // global in the struct STy.
9716    Value *Idxs[] = {Builder.getInt32(0), Builder.getInt32(3),
9717                     Builder.getInt32(0)};
9718    Value *CpuFeatures = Builder.CreateGEP(STy, CpuModel, Idxs);
9719    Value *Features =
9720        Builder.CreateAlignedLoad(CpuFeatures, CharUnits::fromQuantity(4));
9721
9722    // Check the value of the bit corresponding to the feature requested.
9723    Value *Mask = Builder.getInt32(Features1);
9724    Value *Bitset = Builder.CreateAnd(Features, Mask);
9725    Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask);
9726    Result = Builder.CreateAnd(Result, Cmp);
9727  }
9728
9729  if (Features2 != 0) {
9730    llvm::Constant *CpuFeatures2 = CGM.CreateRuntimeVariable(Int32Ty,
9731                                                             "__cpu_features2");
9732    cast<llvm::GlobalValue>(CpuFeatures2)->setDSOLocal(true);
9733
9734    Value *Features =
9735        Builder.CreateAlignedLoad(CpuFeatures2, CharUnits::fromQuantity(4));
9736
9737    // Check the value of the bit corresponding to the feature requested.
9738    Value *Mask = Builder.getInt32(Features2);
9739    Value *Bitset = Builder.CreateAnd(Features, Mask);
9740    Value *Cmp = Builder.CreateICmpEQ(Bitset, Mask);
9741    Result = Builder.CreateAnd(Result, Cmp);
9742  }
9743
9744  return Result;
9745}
9746
9747Value *CodeGenFunction::EmitX86CpuInit() {
9748  llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy,
9749                                                    /*Variadic*/ false);
9750  llvm::FunctionCallee Func =
9751      CGM.CreateRuntimeFunction(FTy, "__cpu_indicator_init");
9752  cast<llvm::GlobalValue>(Func.getCallee())->setDSOLocal(true);
9753  cast<llvm::GlobalValue>(Func.getCallee())
9754      ->setDLLStorageClass(llvm::GlobalValue::DefaultStorageClass);
9755  return Builder.CreateCall(Func);
9756}
9757
9758Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
9759                                           const CallExpr *E) {
9760  if (BuiltinID == X86::BI__builtin_cpu_is)
9761    return EmitX86CpuIs(E);
9762  if (BuiltinID == X86::BI__builtin_cpu_supports)
9763    return EmitX86CpuSupports(E);
9764  if (BuiltinID == X86::BI__builtin_cpu_init)
9765    return EmitX86CpuInit();
9766
9767  SmallVector<Value*, 4> Ops;
9768
9769  // Find out if any arguments are required to be integer constant expressions.
9770  unsigned ICEArguments = 0;
9771  ASTContext::GetBuiltinTypeError Error;
9772  getContext().GetBuiltinType(BuiltinID, Error, &ICEArguments);
9773   (0) . __assert_fail ("Error == ASTContext..GE_None && \"Should not codegen an error\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 9773, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(Error == ASTContext::GE_None && "Should not codegen an error");
9774
9775  for (unsigned i = 0, e = E->getNumArgs(); i != e; i++) {
9776    // If this is a normal argument, just emit it as a scalar.
9777    if ((ICEArguments & (1 << i)) == 0) {
9778      Ops.push_back(EmitScalarExpr(E->getArg(i)));
9779      continue;
9780    }
9781
9782    // If this is required to be a constant, constant fold it so that we know
9783    // that the generated intrinsic gets a ConstantInt.
9784    llvm::APSInt Result;
9785    bool IsConst = E->getArg(i)->isIntegerConstantExpr(Result, getContext());
9786     (0) . __assert_fail ("IsConst && \"Constant arg isn't actually constant?\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 9786, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(IsConst && "Constant arg isn't actually constant?"); (void)IsConst;
9787    Ops.push_back(llvm::ConstantInt::get(getLLVMContext(), Result));
9788  }
9789
9790  // These exist so that the builtin that takes an immediate can be bounds
9791  // checked by clang to avoid passing bad immediates to the backend. Since
9792  // AVX has a larger immediate than SSE we would need separate builtins to
9793  // do the different bounds checking. Rather than create a clang specific
9794  // SSE only builtin, this implements eight separate builtins to match gcc
9795  // implementation.
9796  auto getCmpIntrinsicCall = [this, &Ops](Intrinsic::ID ID, unsigned Imm) {
9797    Ops.push_back(llvm::ConstantInt::get(Int8Ty, Imm));
9798    llvm::Function *F = CGM.getIntrinsic(ID);
9799    return Builder.CreateCall(F, Ops);
9800  };
9801
9802  // For the vector forms of FP comparisons, translate the builtins directly to
9803  // IR.
9804  // TODO: The builtins could be removed if the SSE header files used vector
9805  // extension comparisons directly (vector ordered/unordered may need
9806  // additional support via __builtin_isnan()).
9807  auto getVectorFCmpIR = [this, &Ops](CmpInst::Predicate Pred) {
9808    Value *Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]);
9809    llvm::VectorType *FPVecTy = cast<llvm::VectorType>(Ops[0]->getType());
9810    llvm::VectorType *IntVecTy = llvm::VectorType::getInteger(FPVecTy);
9811    Value *Sext = Builder.CreateSExt(Cmp, IntVecTy);
9812    return Builder.CreateBitCast(Sext, FPVecTy);
9813  };
9814
9815  switch (BuiltinID) {
9816  default: return nullptr;
9817  case X86::BI_mm_prefetch: {
9818    Value *Address = Ops[0];
9819    ConstantInt *C = cast<ConstantInt>(Ops[1]);
9820    Value *RW = ConstantInt::get(Int32Ty, (C->getZExtValue() >> 2) & 0x1);
9821    Value *Locality = ConstantInt::get(Int32Ty, C->getZExtValue() & 0x3);
9822    Value *Data = ConstantInt::get(Int32Ty, 1);
9823    Function *F = CGM.getIntrinsic(Intrinsic::prefetch);
9824    return Builder.CreateCall(F, {Address, RW, Locality, Data});
9825  }
9826  case X86::BI_mm_clflush: {
9827    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_clflush),
9828                              Ops[0]);
9829  }
9830  case X86::BI_mm_lfence: {
9831    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_lfence));
9832  }
9833  case X86::BI_mm_mfence: {
9834    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_mfence));
9835  }
9836  case X86::BI_mm_sfence: {
9837    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_sfence));
9838  }
9839  case X86::BI_mm_pause: {
9840    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse2_pause));
9841  }
9842  case X86::BI__rdtsc: {
9843    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtsc));
9844  }
9845  case X86::BI__builtin_ia32_rdtscp: {
9846    Value *Call = Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_rdtscp));
9847    Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1),
9848                                      Ops[0]);
9849    return Builder.CreateExtractValue(Call, 0);
9850  }
9851  case X86::BI__builtin_ia32_lzcnt_u16:
9852  case X86::BI__builtin_ia32_lzcnt_u32:
9853  case X86::BI__builtin_ia32_lzcnt_u64: {
9854    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType());
9855    return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)});
9856  }
9857  case X86::BI__builtin_ia32_tzcnt_u16:
9858  case X86::BI__builtin_ia32_tzcnt_u32:
9859  case X86::BI__builtin_ia32_tzcnt_u64: {
9860    Function *F = CGM.getIntrinsic(Intrinsic::cttz, Ops[0]->getType());
9861    return Builder.CreateCall(F, {Ops[0], Builder.getInt1(false)});
9862  }
9863  case X86::BI__builtin_ia32_undef128:
9864  case X86::BI__builtin_ia32_undef256:
9865  case X86::BI__builtin_ia32_undef512:
9866    // The x86 definition of "undef" is not the same as the LLVM definition
9867    // (PR32176). We leave optimizing away an unnecessary zero constant to the
9868    // IR optimizer and backend.
9869    // TODO: If we had a "freeze" IR instruction to generate a fixed undef
9870    // value, we should use that here instead of a zero.
9871    return llvm::Constant::getNullValue(ConvertType(E->getType()));
9872  case X86::BI__builtin_ia32_vec_init_v8qi:
9873  case X86::BI__builtin_ia32_vec_init_v4hi:
9874  case X86::BI__builtin_ia32_vec_init_v2si:
9875    return Builder.CreateBitCast(BuildVector(Ops),
9876                                 llvm::Type::getX86_MMXTy(getLLVMContext()));
9877  case X86::BI__builtin_ia32_vec_ext_v2si:
9878  case X86::BI__builtin_ia32_vec_ext_v16qi:
9879  case X86::BI__builtin_ia32_vec_ext_v8hi:
9880  case X86::BI__builtin_ia32_vec_ext_v4si:
9881  case X86::BI__builtin_ia32_vec_ext_v4sf:
9882  case X86::BI__builtin_ia32_vec_ext_v2di:
9883  case X86::BI__builtin_ia32_vec_ext_v32qi:
9884  case X86::BI__builtin_ia32_vec_ext_v16hi:
9885  case X86::BI__builtin_ia32_vec_ext_v8si:
9886  case X86::BI__builtin_ia32_vec_ext_v4di: {
9887    unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
9888    uint64_t Index = cast<ConstantInt>(Ops[1])->getZExtValue();
9889    Index &= NumElts - 1;
9890    // These builtins exist so we can ensure the index is an ICE and in range.
9891    // Otherwise we could just do this in the header file.
9892    return Builder.CreateExtractElement(Ops[0], Index);
9893  }
9894  case X86::BI__builtin_ia32_vec_set_v16qi:
9895  case X86::BI__builtin_ia32_vec_set_v8hi:
9896  case X86::BI__builtin_ia32_vec_set_v4si:
9897  case X86::BI__builtin_ia32_vec_set_v2di:
9898  case X86::BI__builtin_ia32_vec_set_v32qi:
9899  case X86::BI__builtin_ia32_vec_set_v16hi:
9900  case X86::BI__builtin_ia32_vec_set_v8si:
9901  case X86::BI__builtin_ia32_vec_set_v4di: {
9902    unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
9903    unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue();
9904    Index &= NumElts - 1;
9905    // These builtins exist so we can ensure the index is an ICE and in range.
9906    // Otherwise we could just do this in the header file.
9907    return Builder.CreateInsertElement(Ops[0], Ops[1], Index);
9908  }
9909  case X86::BI_mm_setcsr:
9910  case X86::BI__builtin_ia32_ldmxcsr: {
9911    Address Tmp = CreateMemTemp(E->getArg(0)->getType());
9912    Builder.CreateStore(Ops[0], Tmp);
9913    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_ldmxcsr),
9914                          Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy));
9915  }
9916  case X86::BI_mm_getcsr:
9917  case X86::BI__builtin_ia32_stmxcsr: {
9918    Address Tmp = CreateMemTemp(E->getType());
9919    Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_sse_stmxcsr),
9920                       Builder.CreateBitCast(Tmp.getPointer(), Int8PtrTy));
9921    return Builder.CreateLoad(Tmp, "stmxcsr");
9922  }
9923  case X86::BI__builtin_ia32_xsave:
9924  case X86::BI__builtin_ia32_xsave64:
9925  case X86::BI__builtin_ia32_xrstor:
9926  case X86::BI__builtin_ia32_xrstor64:
9927  case X86::BI__builtin_ia32_xsaveopt:
9928  case X86::BI__builtin_ia32_xsaveopt64:
9929  case X86::BI__builtin_ia32_xrstors:
9930  case X86::BI__builtin_ia32_xrstors64:
9931  case X86::BI__builtin_ia32_xsavec:
9932  case X86::BI__builtin_ia32_xsavec64:
9933  case X86::BI__builtin_ia32_xsaves:
9934  case X86::BI__builtin_ia32_xsaves64:
9935  case X86::BI__builtin_ia32_xsetbv:
9936  case X86::BI_xsetbv: {
9937    Intrinsic::ID ID;
9938#define INTRINSIC_X86_XSAVE_ID(NAME) \
9939    case X86::BI__builtin_ia32_##NAME: \
9940      ID = Intrinsic::x86_##NAME; \
9941      break
9942    switch (BuiltinID) {
9943    default: llvm_unreachable("Unsupported intrinsic!");
9944    INTRINSIC_X86_XSAVE_ID(xsave);
9945    INTRINSIC_X86_XSAVE_ID(xsave64);
9946    INTRINSIC_X86_XSAVE_ID(xrstor);
9947    INTRINSIC_X86_XSAVE_ID(xrstor64);
9948    INTRINSIC_X86_XSAVE_ID(xsaveopt);
9949    INTRINSIC_X86_XSAVE_ID(xsaveopt64);
9950    INTRINSIC_X86_XSAVE_ID(xrstors);
9951    INTRINSIC_X86_XSAVE_ID(xrstors64);
9952    INTRINSIC_X86_XSAVE_ID(xsavec);
9953    INTRINSIC_X86_XSAVE_ID(xsavec64);
9954    INTRINSIC_X86_XSAVE_ID(xsaves);
9955    INTRINSIC_X86_XSAVE_ID(xsaves64);
9956    INTRINSIC_X86_XSAVE_ID(xsetbv);
9957    case X86::BI_xsetbv:
9958      ID = Intrinsic::x86_xsetbv;
9959      break;
9960    }
9961#undef INTRINSIC_X86_XSAVE_ID
9962    Value *Mhi = Builder.CreateTrunc(
9963      Builder.CreateLShr(Ops[1], ConstantInt::get(Int64Ty, 32)), Int32Ty);
9964    Value *Mlo = Builder.CreateTrunc(Ops[1], Int32Ty);
9965    Ops[1] = Mhi;
9966    Ops.push_back(Mlo);
9967    return Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
9968  }
9969  case X86::BI__builtin_ia32_xgetbv:
9970  case X86::BI_xgetbv:
9971    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::x86_xgetbv), Ops);
9972  case X86::BI__builtin_ia32_storedqudi128_mask:
9973  case X86::BI__builtin_ia32_storedqusi128_mask:
9974  case X86::BI__builtin_ia32_storedquhi128_mask:
9975  case X86::BI__builtin_ia32_storedquqi128_mask:
9976  case X86::BI__builtin_ia32_storeupd128_mask:
9977  case X86::BI__builtin_ia32_storeups128_mask:
9978  case X86::BI__builtin_ia32_storedqudi256_mask:
9979  case X86::BI__builtin_ia32_storedqusi256_mask:
9980  case X86::BI__builtin_ia32_storedquhi256_mask:
9981  case X86::BI__builtin_ia32_storedquqi256_mask:
9982  case X86::BI__builtin_ia32_storeupd256_mask:
9983  case X86::BI__builtin_ia32_storeups256_mask:
9984  case X86::BI__builtin_ia32_storedqudi512_mask:
9985  case X86::BI__builtin_ia32_storedqusi512_mask:
9986  case X86::BI__builtin_ia32_storedquhi512_mask:
9987  case X86::BI__builtin_ia32_storedquqi512_mask:
9988  case X86::BI__builtin_ia32_storeupd512_mask:
9989  case X86::BI__builtin_ia32_storeups512_mask:
9990    return EmitX86MaskedStore(*this, Ops, 1);
9991
9992  case X86::BI__builtin_ia32_storess128_mask:
9993  case X86::BI__builtin_ia32_storesd128_mask: {
9994    return EmitX86MaskedStore(*this, Ops, 1);
9995  }
9996  case X86::BI__builtin_ia32_vpopcntb_128:
9997  case X86::BI__builtin_ia32_vpopcntd_128:
9998  case X86::BI__builtin_ia32_vpopcntq_128:
9999  case X86::BI__builtin_ia32_vpopcntw_128:
10000  case X86::BI__builtin_ia32_vpopcntb_256:
10001  case X86::BI__builtin_ia32_vpopcntd_256:
10002  case X86::BI__builtin_ia32_vpopcntq_256:
10003  case X86::BI__builtin_ia32_vpopcntw_256:
10004  case X86::BI__builtin_ia32_vpopcntb_512:
10005  case X86::BI__builtin_ia32_vpopcntd_512:
10006  case X86::BI__builtin_ia32_vpopcntq_512:
10007  case X86::BI__builtin_ia32_vpopcntw_512: {
10008    llvm::Type *ResultType = ConvertType(E->getType());
10009    llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
10010    return Builder.CreateCall(F, Ops);
10011  }
10012  case X86::BI__builtin_ia32_cvtmask2b128:
10013  case X86::BI__builtin_ia32_cvtmask2b256:
10014  case X86::BI__builtin_ia32_cvtmask2b512:
10015  case X86::BI__builtin_ia32_cvtmask2w128:
10016  case X86::BI__builtin_ia32_cvtmask2w256:
10017  case X86::BI__builtin_ia32_cvtmask2w512:
10018  case X86::BI__builtin_ia32_cvtmask2d128:
10019  case X86::BI__builtin_ia32_cvtmask2d256:
10020  case X86::BI__builtin_ia32_cvtmask2d512:
10021  case X86::BI__builtin_ia32_cvtmask2q128:
10022  case X86::BI__builtin_ia32_cvtmask2q256:
10023  case X86::BI__builtin_ia32_cvtmask2q512:
10024    return EmitX86SExtMask(*this, Ops[0], ConvertType(E->getType()));
10025
10026  case X86::BI__builtin_ia32_cvtb2mask128:
10027  case X86::BI__builtin_ia32_cvtb2mask256:
10028  case X86::BI__builtin_ia32_cvtb2mask512:
10029  case X86::BI__builtin_ia32_cvtw2mask128:
10030  case X86::BI__builtin_ia32_cvtw2mask256:
10031  case X86::BI__builtin_ia32_cvtw2mask512:
10032  case X86::BI__builtin_ia32_cvtd2mask128:
10033  case X86::BI__builtin_ia32_cvtd2mask256:
10034  case X86::BI__builtin_ia32_cvtd2mask512:
10035  case X86::BI__builtin_ia32_cvtq2mask128:
10036  case X86::BI__builtin_ia32_cvtq2mask256:
10037  case X86::BI__builtin_ia32_cvtq2mask512:
10038    return EmitX86ConvertToMask(*this, Ops[0]);
10039
10040  case X86::BI__builtin_ia32_cvtdq2ps512_mask:
10041  case X86::BI__builtin_ia32_cvtqq2ps512_mask:
10042  case X86::BI__builtin_ia32_cvtqq2pd512_mask:
10043    return EmitX86ConvertIntToFp(*this, Ops, /*IsSigned*/true);
10044  case X86::BI__builtin_ia32_cvtudq2ps512_mask:
10045  case X86::BI__builtin_ia32_cvtuqq2ps512_mask:
10046  case X86::BI__builtin_ia32_cvtuqq2pd512_mask:
10047    return EmitX86ConvertIntToFp(*this, Ops, /*IsSigned*/false);
10048
10049  case X86::BI__builtin_ia32_vfmaddss3:
10050  case X86::BI__builtin_ia32_vfmaddsd3:
10051  case X86::BI__builtin_ia32_vfmaddss3_mask:
10052  case X86::BI__builtin_ia32_vfmaddsd3_mask:
10053    return EmitScalarFMAExpr(*this, Ops, Ops[0]);
10054  case X86::BI__builtin_ia32_vfmaddss:
10055  case X86::BI__builtin_ia32_vfmaddsd:
10056    return EmitScalarFMAExpr(*this, Ops,
10057                             Constant::getNullValue(Ops[0]->getType()));
10058  case X86::BI__builtin_ia32_vfmaddss3_maskz:
10059  case X86::BI__builtin_ia32_vfmaddsd3_maskz:
10060    return EmitScalarFMAExpr(*this, Ops, Ops[0], /*ZeroMask*/true);
10061  case X86::BI__builtin_ia32_vfmaddss3_mask3:
10062  case X86::BI__builtin_ia32_vfmaddsd3_mask3:
10063    return EmitScalarFMAExpr(*this, Ops, Ops[2], /*ZeroMask*/false, 2);
10064  case X86::BI__builtin_ia32_vfmsubss3_mask3:
10065  case X86::BI__builtin_ia32_vfmsubsd3_mask3:
10066    return EmitScalarFMAExpr(*this, Ops, Ops[2], /*ZeroMask*/false, 2,
10067                             /*NegAcc*/true);
10068  case X86::BI__builtin_ia32_vfmaddps:
10069  case X86::BI__builtin_ia32_vfmaddpd:
10070  case X86::BI__builtin_ia32_vfmaddps256:
10071  case X86::BI__builtin_ia32_vfmaddpd256:
10072  case X86::BI__builtin_ia32_vfmaddps512_mask:
10073  case X86::BI__builtin_ia32_vfmaddps512_maskz:
10074  case X86::BI__builtin_ia32_vfmaddps512_mask3:
10075  case X86::BI__builtin_ia32_vfmsubps512_mask3:
10076  case X86::BI__builtin_ia32_vfmaddpd512_mask:
10077  case X86::BI__builtin_ia32_vfmaddpd512_maskz:
10078  case X86::BI__builtin_ia32_vfmaddpd512_mask3:
10079  case X86::BI__builtin_ia32_vfmsubpd512_mask3:
10080    return EmitX86FMAExpr(*this, Ops, BuiltinID, /*IsAddSub*/false);
10081  case X86::BI__builtin_ia32_vfmaddsubps:
10082  case X86::BI__builtin_ia32_vfmaddsubpd:
10083  case X86::BI__builtin_ia32_vfmaddsubps256:
10084  case X86::BI__builtin_ia32_vfmaddsubpd256:
10085  case X86::BI__builtin_ia32_vfmaddsubps512_mask:
10086  case X86::BI__builtin_ia32_vfmaddsubps512_maskz:
10087  case X86::BI__builtin_ia32_vfmaddsubps512_mask3:
10088  case X86::BI__builtin_ia32_vfmsubaddps512_mask3:
10089  case X86::BI__builtin_ia32_vfmaddsubpd512_mask:
10090  case X86::BI__builtin_ia32_vfmaddsubpd512_maskz:
10091  case X86::BI__builtin_ia32_vfmaddsubpd512_mask3:
10092  case X86::BI__builtin_ia32_vfmsubaddpd512_mask3:
10093    return EmitX86FMAExpr(*this, Ops, BuiltinID, /*IsAddSub*/true);
10094
10095  case X86::BI__builtin_ia32_movdqa32store128_mask:
10096  case X86::BI__builtin_ia32_movdqa64store128_mask:
10097  case X86::BI__builtin_ia32_storeaps128_mask:
10098  case X86::BI__builtin_ia32_storeapd128_mask:
10099  case X86::BI__builtin_ia32_movdqa32store256_mask:
10100  case X86::BI__builtin_ia32_movdqa64store256_mask:
10101  case X86::BI__builtin_ia32_storeaps256_mask:
10102  case X86::BI__builtin_ia32_storeapd256_mask:
10103  case X86::BI__builtin_ia32_movdqa32store512_mask:
10104  case X86::BI__builtin_ia32_movdqa64store512_mask:
10105  case X86::BI__builtin_ia32_storeaps512_mask:
10106  case X86::BI__builtin_ia32_storeapd512_mask: {
10107    unsigned Align =
10108      getContext().getTypeAlignInChars(E->getArg(1)->getType()).getQuantity();
10109    return EmitX86MaskedStore(*this, Ops, Align);
10110  }
10111  case X86::BI__builtin_ia32_loadups128_mask:
10112  case X86::BI__builtin_ia32_loadups256_mask:
10113  case X86::BI__builtin_ia32_loadups512_mask:
10114  case X86::BI__builtin_ia32_loadupd128_mask:
10115  case X86::BI__builtin_ia32_loadupd256_mask:
10116  case X86::BI__builtin_ia32_loadupd512_mask:
10117  case X86::BI__builtin_ia32_loaddquqi128_mask:
10118  case X86::BI__builtin_ia32_loaddquqi256_mask:
10119  case X86::BI__builtin_ia32_loaddquqi512_mask:
10120  case X86::BI__builtin_ia32_loaddquhi128_mask:
10121  case X86::BI__builtin_ia32_loaddquhi256_mask:
10122  case X86::BI__builtin_ia32_loaddquhi512_mask:
10123  case X86::BI__builtin_ia32_loaddqusi128_mask:
10124  case X86::BI__builtin_ia32_loaddqusi256_mask:
10125  case X86::BI__builtin_ia32_loaddqusi512_mask:
10126  case X86::BI__builtin_ia32_loaddqudi128_mask:
10127  case X86::BI__builtin_ia32_loaddqudi256_mask:
10128  case X86::BI__builtin_ia32_loaddqudi512_mask:
10129    return EmitX86MaskedLoad(*this, Ops, 1);
10130
10131  case X86::BI__builtin_ia32_loadss128_mask:
10132  case X86::BI__builtin_ia32_loadsd128_mask:
10133    return EmitX86MaskedLoad(*this, Ops, 1);
10134
10135  case X86::BI__builtin_ia32_loadaps128_mask:
10136  case X86::BI__builtin_ia32_loadaps256_mask:
10137  case X86::BI__builtin_ia32_loadaps512_mask:
10138  case X86::BI__builtin_ia32_loadapd128_mask:
10139  case X86::BI__builtin_ia32_loadapd256_mask:
10140  case X86::BI__builtin_ia32_loadapd512_mask:
10141  case X86::BI__builtin_ia32_movdqa32load128_mask:
10142  case X86::BI__builtin_ia32_movdqa32load256_mask:
10143  case X86::BI__builtin_ia32_movdqa32load512_mask:
10144  case X86::BI__builtin_ia32_movdqa64load128_mask:
10145  case X86::BI__builtin_ia32_movdqa64load256_mask:
10146  case X86::BI__builtin_ia32_movdqa64load512_mask: {
10147    unsigned Align =
10148      getContext().getTypeAlignInChars(E->getArg(1)->getType()).getQuantity();
10149    return EmitX86MaskedLoad(*this, Ops, Align);
10150  }
10151
10152  case X86::BI__builtin_ia32_expandloaddf128_mask:
10153  case X86::BI__builtin_ia32_expandloaddf256_mask:
10154  case X86::BI__builtin_ia32_expandloaddf512_mask:
10155  case X86::BI__builtin_ia32_expandloadsf128_mask:
10156  case X86::BI__builtin_ia32_expandloadsf256_mask:
10157  case X86::BI__builtin_ia32_expandloadsf512_mask:
10158  case X86::BI__builtin_ia32_expandloaddi128_mask:
10159  case X86::BI__builtin_ia32_expandloaddi256_mask:
10160  case X86::BI__builtin_ia32_expandloaddi512_mask:
10161  case X86::BI__builtin_ia32_expandloadsi128_mask:
10162  case X86::BI__builtin_ia32_expandloadsi256_mask:
10163  case X86::BI__builtin_ia32_expandloadsi512_mask:
10164  case X86::BI__builtin_ia32_expandloadhi128_mask:
10165  case X86::BI__builtin_ia32_expandloadhi256_mask:
10166  case X86::BI__builtin_ia32_expandloadhi512_mask:
10167  case X86::BI__builtin_ia32_expandloadqi128_mask:
10168  case X86::BI__builtin_ia32_expandloadqi256_mask:
10169  case X86::BI__builtin_ia32_expandloadqi512_mask:
10170    return EmitX86ExpandLoad(*this, Ops);
10171
10172  case X86::BI__builtin_ia32_compressstoredf128_mask:
10173  case X86::BI__builtin_ia32_compressstoredf256_mask:
10174  case X86::BI__builtin_ia32_compressstoredf512_mask:
10175  case X86::BI__builtin_ia32_compressstoresf128_mask:
10176  case X86::BI__builtin_ia32_compressstoresf256_mask:
10177  case X86::BI__builtin_ia32_compressstoresf512_mask:
10178  case X86::BI__builtin_ia32_compressstoredi128_mask:
10179  case X86::BI__builtin_ia32_compressstoredi256_mask:
10180  case X86::BI__builtin_ia32_compressstoredi512_mask:
10181  case X86::BI__builtin_ia32_compressstoresi128_mask:
10182  case X86::BI__builtin_ia32_compressstoresi256_mask:
10183  case X86::BI__builtin_ia32_compressstoresi512_mask:
10184  case X86::BI__builtin_ia32_compressstorehi128_mask:
10185  case X86::BI__builtin_ia32_compressstorehi256_mask:
10186  case X86::BI__builtin_ia32_compressstorehi512_mask:
10187  case X86::BI__builtin_ia32_compressstoreqi128_mask:
10188  case X86::BI__builtin_ia32_compressstoreqi256_mask:
10189  case X86::BI__builtin_ia32_compressstoreqi512_mask:
10190    return EmitX86CompressStore(*this, Ops);
10191
10192  case X86::BI__builtin_ia32_expanddf128_mask:
10193  case X86::BI__builtin_ia32_expanddf256_mask:
10194  case X86::BI__builtin_ia32_expanddf512_mask:
10195  case X86::BI__builtin_ia32_expandsf128_mask:
10196  case X86::BI__builtin_ia32_expandsf256_mask:
10197  case X86::BI__builtin_ia32_expandsf512_mask:
10198  case X86::BI__builtin_ia32_expanddi128_mask:
10199  case X86::BI__builtin_ia32_expanddi256_mask:
10200  case X86::BI__builtin_ia32_expanddi512_mask:
10201  case X86::BI__builtin_ia32_expandsi128_mask:
10202  case X86::BI__builtin_ia32_expandsi256_mask:
10203  case X86::BI__builtin_ia32_expandsi512_mask:
10204  case X86::BI__builtin_ia32_expandhi128_mask:
10205  case X86::BI__builtin_ia32_expandhi256_mask:
10206  case X86::BI__builtin_ia32_expandhi512_mask:
10207  case X86::BI__builtin_ia32_expandqi128_mask:
10208  case X86::BI__builtin_ia32_expandqi256_mask:
10209  case X86::BI__builtin_ia32_expandqi512_mask:
10210    return EmitX86CompressExpand(*this, Ops, /*IsCompress*/false);
10211
10212  case X86::BI__builtin_ia32_compressdf128_mask:
10213  case X86::BI__builtin_ia32_compressdf256_mask:
10214  case X86::BI__builtin_ia32_compressdf512_mask:
10215  case X86::BI__builtin_ia32_compresssf128_mask:
10216  case X86::BI__builtin_ia32_compresssf256_mask:
10217  case X86::BI__builtin_ia32_compresssf512_mask:
10218  case X86::BI__builtin_ia32_compressdi128_mask:
10219  case X86::BI__builtin_ia32_compressdi256_mask:
10220  case X86::BI__builtin_ia32_compressdi512_mask:
10221  case X86::BI__builtin_ia32_compresssi128_mask:
10222  case X86::BI__builtin_ia32_compresssi256_mask:
10223  case X86::BI__builtin_ia32_compresssi512_mask:
10224  case X86::BI__builtin_ia32_compresshi128_mask:
10225  case X86::BI__builtin_ia32_compresshi256_mask:
10226  case X86::BI__builtin_ia32_compresshi512_mask:
10227  case X86::BI__builtin_ia32_compressqi128_mask:
10228  case X86::BI__builtin_ia32_compressqi256_mask:
10229  case X86::BI__builtin_ia32_compressqi512_mask:
10230    return EmitX86CompressExpand(*this, Ops, /*IsCompress*/true);
10231
10232  case X86::BI__builtin_ia32_gather3div2df:
10233  case X86::BI__builtin_ia32_gather3div2di:
10234  case X86::BI__builtin_ia32_gather3div4df:
10235  case X86::BI__builtin_ia32_gather3div4di:
10236  case X86::BI__builtin_ia32_gather3div4sf:
10237  case X86::BI__builtin_ia32_gather3div4si:
10238  case X86::BI__builtin_ia32_gather3div8sf:
10239  case X86::BI__builtin_ia32_gather3div8si:
10240  case X86::BI__builtin_ia32_gather3siv2df:
10241  case X86::BI__builtin_ia32_gather3siv2di:
10242  case X86::BI__builtin_ia32_gather3siv4df:
10243  case X86::BI__builtin_ia32_gather3siv4di:
10244  case X86::BI__builtin_ia32_gather3siv4sf:
10245  case X86::BI__builtin_ia32_gather3siv4si:
10246  case X86::BI__builtin_ia32_gather3siv8sf:
10247  case X86::BI__builtin_ia32_gather3siv8si:
10248  case X86::BI__builtin_ia32_gathersiv8df:
10249  case X86::BI__builtin_ia32_gathersiv16sf:
10250  case X86::BI__builtin_ia32_gatherdiv8df:
10251  case X86::BI__builtin_ia32_gatherdiv16sf:
10252  case X86::BI__builtin_ia32_gathersiv8di:
10253  case X86::BI__builtin_ia32_gathersiv16si:
10254  case X86::BI__builtin_ia32_gatherdiv8di:
10255  case X86::BI__builtin_ia32_gatherdiv16si: {
10256    Intrinsic::ID IID;
10257    switch (BuiltinID) {
10258    default: llvm_unreachable("Unexpected builtin");
10259    case X86::BI__builtin_ia32_gather3div2df:
10260      IID = Intrinsic::x86_avx512_mask_gather3div2_df;
10261      break;
10262    case X86::BI__builtin_ia32_gather3div2di:
10263      IID = Intrinsic::x86_avx512_mask_gather3div2_di;
10264      break;
10265    case X86::BI__builtin_ia32_gather3div4df:
10266      IID = Intrinsic::x86_avx512_mask_gather3div4_df;
10267      break;
10268    case X86::BI__builtin_ia32_gather3div4di:
10269      IID = Intrinsic::x86_avx512_mask_gather3div4_di;
10270      break;
10271    case X86::BI__builtin_ia32_gather3div4sf:
10272      IID = Intrinsic::x86_avx512_mask_gather3div4_sf;
10273      break;
10274    case X86::BI__builtin_ia32_gather3div4si:
10275      IID = Intrinsic::x86_avx512_mask_gather3div4_si;
10276      break;
10277    case X86::BI__builtin_ia32_gather3div8sf:
10278      IID = Intrinsic::x86_avx512_mask_gather3div8_sf;
10279      break;
10280    case X86::BI__builtin_ia32_gather3div8si:
10281      IID = Intrinsic::x86_avx512_mask_gather3div8_si;
10282      break;
10283    case X86::BI__builtin_ia32_gather3siv2df:
10284      IID = Intrinsic::x86_avx512_mask_gather3siv2_df;
10285      break;
10286    case X86::BI__builtin_ia32_gather3siv2di:
10287      IID = Intrinsic::x86_avx512_mask_gather3siv2_di;
10288      break;
10289    case X86::BI__builtin_ia32_gather3siv4df:
10290      IID = Intrinsic::x86_avx512_mask_gather3siv4_df;
10291      break;
10292    case X86::BI__builtin_ia32_gather3siv4di:
10293      IID = Intrinsic::x86_avx512_mask_gather3siv4_di;
10294      break;
10295    case X86::BI__builtin_ia32_gather3siv4sf:
10296      IID = Intrinsic::x86_avx512_mask_gather3siv4_sf;
10297      break;
10298    case X86::BI__builtin_ia32_gather3siv4si:
10299      IID = Intrinsic::x86_avx512_mask_gather3siv4_si;
10300      break;
10301    case X86::BI__builtin_ia32_gather3siv8sf:
10302      IID = Intrinsic::x86_avx512_mask_gather3siv8_sf;
10303      break;
10304    case X86::BI__builtin_ia32_gather3siv8si:
10305      IID = Intrinsic::x86_avx512_mask_gather3siv8_si;
10306      break;
10307    case X86::BI__builtin_ia32_gathersiv8df:
10308      IID = Intrinsic::x86_avx512_mask_gather_dpd_512;
10309      break;
10310    case X86::BI__builtin_ia32_gathersiv16sf:
10311      IID = Intrinsic::x86_avx512_mask_gather_dps_512;
10312      break;
10313    case X86::BI__builtin_ia32_gatherdiv8df:
10314      IID = Intrinsic::x86_avx512_mask_gather_qpd_512;
10315      break;
10316    case X86::BI__builtin_ia32_gatherdiv16sf:
10317      IID = Intrinsic::x86_avx512_mask_gather_qps_512;
10318      break;
10319    case X86::BI__builtin_ia32_gathersiv8di:
10320      IID = Intrinsic::x86_avx512_mask_gather_dpq_512;
10321      break;
10322    case X86::BI__builtin_ia32_gathersiv16si:
10323      IID = Intrinsic::x86_avx512_mask_gather_dpi_512;
10324      break;
10325    case X86::BI__builtin_ia32_gatherdiv8di:
10326      IID = Intrinsic::x86_avx512_mask_gather_qpq_512;
10327      break;
10328    case X86::BI__builtin_ia32_gatherdiv16si:
10329      IID = Intrinsic::x86_avx512_mask_gather_qpi_512;
10330      break;
10331    }
10332
10333    unsigned MinElts = std::min(Ops[0]->getType()->getVectorNumElements(),
10334                                Ops[2]->getType()->getVectorNumElements());
10335    Ops[3] = getMaskVecValue(*this, Ops[3], MinElts);
10336    Function *Intr = CGM.getIntrinsic(IID);
10337    return Builder.CreateCall(Intr, Ops);
10338  }
10339
10340  case X86::BI__builtin_ia32_scattersiv8df:
10341  case X86::BI__builtin_ia32_scattersiv16sf:
10342  case X86::BI__builtin_ia32_scatterdiv8df:
10343  case X86::BI__builtin_ia32_scatterdiv16sf:
10344  case X86::BI__builtin_ia32_scattersiv8di:
10345  case X86::BI__builtin_ia32_scattersiv16si:
10346  case X86::BI__builtin_ia32_scatterdiv8di:
10347  case X86::BI__builtin_ia32_scatterdiv16si:
10348  case X86::BI__builtin_ia32_scatterdiv2df:
10349  case X86::BI__builtin_ia32_scatterdiv2di:
10350  case X86::BI__builtin_ia32_scatterdiv4df:
10351  case X86::BI__builtin_ia32_scatterdiv4di:
10352  case X86::BI__builtin_ia32_scatterdiv4sf:
10353  case X86::BI__builtin_ia32_scatterdiv4si:
10354  case X86::BI__builtin_ia32_scatterdiv8sf:
10355  case X86::BI__builtin_ia32_scatterdiv8si:
10356  case X86::BI__builtin_ia32_scattersiv2df:
10357  case X86::BI__builtin_ia32_scattersiv2di:
10358  case X86::BI__builtin_ia32_scattersiv4df:
10359  case X86::BI__builtin_ia32_scattersiv4di:
10360  case X86::BI__builtin_ia32_scattersiv4sf:
10361  case X86::BI__builtin_ia32_scattersiv4si:
10362  case X86::BI__builtin_ia32_scattersiv8sf:
10363  case X86::BI__builtin_ia32_scattersiv8si: {
10364    Intrinsic::ID IID;
10365    switch (BuiltinID) {
10366    default: llvm_unreachable("Unexpected builtin");
10367    case X86::BI__builtin_ia32_scattersiv8df:
10368      IID = Intrinsic::x86_avx512_mask_scatter_dpd_512;
10369      break;
10370    case X86::BI__builtin_ia32_scattersiv16sf:
10371      IID = Intrinsic::x86_avx512_mask_scatter_dps_512;
10372      break;
10373    case X86::BI__builtin_ia32_scatterdiv8df:
10374      IID = Intrinsic::x86_avx512_mask_scatter_qpd_512;
10375      break;
10376    case X86::BI__builtin_ia32_scatterdiv16sf:
10377      IID = Intrinsic::x86_avx512_mask_scatter_qps_512;
10378      break;
10379    case X86::BI__builtin_ia32_scattersiv8di:
10380      IID = Intrinsic::x86_avx512_mask_scatter_dpq_512;
10381      break;
10382    case X86::BI__builtin_ia32_scattersiv16si:
10383      IID = Intrinsic::x86_avx512_mask_scatter_dpi_512;
10384      break;
10385    case X86::BI__builtin_ia32_scatterdiv8di:
10386      IID = Intrinsic::x86_avx512_mask_scatter_qpq_512;
10387      break;
10388    case X86::BI__builtin_ia32_scatterdiv16si:
10389      IID = Intrinsic::x86_avx512_mask_scatter_qpi_512;
10390      break;
10391    case X86::BI__builtin_ia32_scatterdiv2df:
10392      IID = Intrinsic::x86_avx512_mask_scatterdiv2_df;
10393      break;
10394    case X86::BI__builtin_ia32_scatterdiv2di:
10395      IID = Intrinsic::x86_avx512_mask_scatterdiv2_di;
10396      break;
10397    case X86::BI__builtin_ia32_scatterdiv4df:
10398      IID = Intrinsic::x86_avx512_mask_scatterdiv4_df;
10399      break;
10400    case X86::BI__builtin_ia32_scatterdiv4di:
10401      IID = Intrinsic::x86_avx512_mask_scatterdiv4_di;
10402      break;
10403    case X86::BI__builtin_ia32_scatterdiv4sf:
10404      IID = Intrinsic::x86_avx512_mask_scatterdiv4_sf;
10405      break;
10406    case X86::BI__builtin_ia32_scatterdiv4si:
10407      IID = Intrinsic::x86_avx512_mask_scatterdiv4_si;
10408      break;
10409    case X86::BI__builtin_ia32_scatterdiv8sf:
10410      IID = Intrinsic::x86_avx512_mask_scatterdiv8_sf;
10411      break;
10412    case X86::BI__builtin_ia32_scatterdiv8si:
10413      IID = Intrinsic::x86_avx512_mask_scatterdiv8_si;
10414      break;
10415    case X86::BI__builtin_ia32_scattersiv2df:
10416      IID = Intrinsic::x86_avx512_mask_scattersiv2_df;
10417      break;
10418    case X86::BI__builtin_ia32_scattersiv2di:
10419      IID = Intrinsic::x86_avx512_mask_scattersiv2_di;
10420      break;
10421    case X86::BI__builtin_ia32_scattersiv4df:
10422      IID = Intrinsic::x86_avx512_mask_scattersiv4_df;
10423      break;
10424    case X86::BI__builtin_ia32_scattersiv4di:
10425      IID = Intrinsic::x86_avx512_mask_scattersiv4_di;
10426      break;
10427    case X86::BI__builtin_ia32_scattersiv4sf:
10428      IID = Intrinsic::x86_avx512_mask_scattersiv4_sf;
10429      break;
10430    case X86::BI__builtin_ia32_scattersiv4si:
10431      IID = Intrinsic::x86_avx512_mask_scattersiv4_si;
10432      break;
10433    case X86::BI__builtin_ia32_scattersiv8sf:
10434      IID = Intrinsic::x86_avx512_mask_scattersiv8_sf;
10435      break;
10436    case X86::BI__builtin_ia32_scattersiv8si:
10437      IID = Intrinsic::x86_avx512_mask_scattersiv8_si;
10438      break;
10439    }
10440
10441    unsigned MinElts = std::min(Ops[2]->getType()->getVectorNumElements(),
10442                                Ops[3]->getType()->getVectorNumElements());
10443    Ops[1] = getMaskVecValue(*this, Ops[1], MinElts);
10444    Function *Intr = CGM.getIntrinsic(IID);
10445    return Builder.CreateCall(Intr, Ops);
10446  }
10447
10448  case X86::BI__builtin_ia32_storehps:
10449  case X86::BI__builtin_ia32_storelps: {
10450    llvm::Type *PtrTy = llvm::PointerType::getUnqual(Int64Ty);
10451    llvm::Type *VecTy = llvm::VectorType::get(Int64Ty, 2);
10452
10453    // cast val v2i64
10454    Ops[1] = Builder.CreateBitCast(Ops[1], VecTy, "cast");
10455
10456    // extract (0, 1)
10457    unsigned Index = BuiltinID == X86::BI__builtin_ia32_storelps ? 0 : 1;
10458    Ops[1] = Builder.CreateExtractElement(Ops[1], Index, "extract");
10459
10460    // cast pointer to i64 & store
10461    Ops[0] = Builder.CreateBitCast(Ops[0], PtrTy);
10462    return Builder.CreateDefaultAlignedStore(Ops[1], Ops[0]);
10463  }
10464  case X86::BI__builtin_ia32_vextractf128_pd256:
10465  case X86::BI__builtin_ia32_vextractf128_ps256:
10466  case X86::BI__builtin_ia32_vextractf128_si256:
10467  case X86::BI__builtin_ia32_extract128i256:
10468  case X86::BI__builtin_ia32_extractf64x4_mask:
10469  case X86::BI__builtin_ia32_extractf32x4_mask:
10470  case X86::BI__builtin_ia32_extracti64x4_mask:
10471  case X86::BI__builtin_ia32_extracti32x4_mask:
10472  case X86::BI__builtin_ia32_extractf32x8_mask:
10473  case X86::BI__builtin_ia32_extracti32x8_mask:
10474  case X86::BI__builtin_ia32_extractf32x4_256_mask:
10475  case X86::BI__builtin_ia32_extracti32x4_256_mask:
10476  case X86::BI__builtin_ia32_extractf64x2_256_mask:
10477  case X86::BI__builtin_ia32_extracti64x2_256_mask:
10478  case X86::BI__builtin_ia32_extractf64x2_512_mask:
10479  case X86::BI__builtin_ia32_extracti64x2_512_mask: {
10480    llvm::Type *DstTy = ConvertType(E->getType());
10481    unsigned NumElts = DstTy->getVectorNumElements();
10482    unsigned SrcNumElts = Ops[0]->getType()->getVectorNumElements();
10483    unsigned SubVectors = SrcNumElts / NumElts;
10484    unsigned Index = cast<ConstantInt>(Ops[1])->getZExtValue();
10485     (0) . __assert_fail ("llvm..isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 10485, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors");
10486    Index &= SubVectors - 1; // Remove any extra bits.
10487    Index *= NumElts;
10488
10489    uint32_t Indices[16];
10490    for (unsigned i = 0; i != NumElts; ++i)
10491      Indices[i] = i + Index;
10492
10493    Value *Res = Builder.CreateShuffleVector(Ops[0],
10494                                             UndefValue::get(Ops[0]->getType()),
10495                                             makeArrayRef(Indices, NumElts),
10496                                             "extract");
10497
10498    if (Ops.size() == 4)
10499      Res = EmitX86Select(*this, Ops[3], Res, Ops[2]);
10500
10501    return Res;
10502  }
10503  case X86::BI__builtin_ia32_vinsertf128_pd256:
10504  case X86::BI__builtin_ia32_vinsertf128_ps256:
10505  case X86::BI__builtin_ia32_vinsertf128_si256:
10506  case X86::BI__builtin_ia32_insert128i256:
10507  case X86::BI__builtin_ia32_insertf64x4:
10508  case X86::BI__builtin_ia32_insertf32x4:
10509  case X86::BI__builtin_ia32_inserti64x4:
10510  case X86::BI__builtin_ia32_inserti32x4:
10511  case X86::BI__builtin_ia32_insertf32x8:
10512  case X86::BI__builtin_ia32_inserti32x8:
10513  case X86::BI__builtin_ia32_insertf32x4_256:
10514  case X86::BI__builtin_ia32_inserti32x4_256:
10515  case X86::BI__builtin_ia32_insertf64x2_256:
10516  case X86::BI__builtin_ia32_inserti64x2_256:
10517  case X86::BI__builtin_ia32_insertf64x2_512:
10518  case X86::BI__builtin_ia32_inserti64x2_512: {
10519    unsigned DstNumElts = Ops[0]->getType()->getVectorNumElements();
10520    unsigned SrcNumElts = Ops[1]->getType()->getVectorNumElements();
10521    unsigned SubVectors = DstNumElts / SrcNumElts;
10522    unsigned Index = cast<ConstantInt>(Ops[2])->getZExtValue();
10523     (0) . __assert_fail ("llvm..isPowerOf2_32(SubVectors) && \"Expected power of 2 subvectors\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 10523, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(llvm::isPowerOf2_32(SubVectors) && "Expected power of 2 subvectors");
10524    Index &= SubVectors - 1; // Remove any extra bits.
10525    Index *= SrcNumElts;
10526
10527    uint32_t Indices[16];
10528    for (unsigned i = 0; i != DstNumElts; ++i)
10529      Indices[i] = (i >= SrcNumElts) ? SrcNumElts + (i % SrcNumElts) : i;
10530
10531    Value *Op1 = Builder.CreateShuffleVector(Ops[1],
10532                                             UndefValue::get(Ops[1]->getType()),
10533                                             makeArrayRef(Indices, DstNumElts),
10534                                             "widen");
10535
10536    for (unsigned i = 0; i != DstNumElts; ++i) {
10537      if (i >= Index && i < (Index + SrcNumElts))
10538        Indices[i] = (i - Index) + DstNumElts;
10539      else
10540        Indices[i] = i;
10541    }
10542
10543    return Builder.CreateShuffleVector(Ops[0], Op1,
10544                                       makeArrayRef(Indices, DstNumElts),
10545                                       "insert");
10546  }
10547  case X86::BI__builtin_ia32_pmovqd512_mask:
10548  case X86::BI__builtin_ia32_pmovwb512_mask: {
10549    Value *Res = Builder.CreateTrunc(Ops[0], Ops[1]->getType());
10550    return EmitX86Select(*this, Ops[2], Res, Ops[1]);
10551  }
10552  case X86::BI__builtin_ia32_pmovdb512_mask:
10553  case X86::BI__builtin_ia32_pmovdw512_mask:
10554  case X86::BI__builtin_ia32_pmovqw512_mask: {
10555    if (const auto *C = dyn_cast<Constant>(Ops[2]))
10556      if (C->isAllOnesValue())
10557        return Builder.CreateTrunc(Ops[0], Ops[1]->getType());
10558
10559    Intrinsic::ID IID;
10560    switch (BuiltinID) {
10561    default: llvm_unreachable("Unsupported intrinsic!");
10562    case X86::BI__builtin_ia32_pmovdb512_mask:
10563      IID = Intrinsic::x86_avx512_mask_pmov_db_512;
10564      break;
10565    case X86::BI__builtin_ia32_pmovdw512_mask:
10566      IID = Intrinsic::x86_avx512_mask_pmov_dw_512;
10567      break;
10568    case X86::BI__builtin_ia32_pmovqw512_mask:
10569      IID = Intrinsic::x86_avx512_mask_pmov_qw_512;
10570      break;
10571    }
10572
10573    Function *Intr = CGM.getIntrinsic(IID);
10574    return Builder.CreateCall(Intr, Ops);
10575  }
10576  case X86::BI__builtin_ia32_pblendw128:
10577  case X86::BI__builtin_ia32_blendpd:
10578  case X86::BI__builtin_ia32_blendps:
10579  case X86::BI__builtin_ia32_blendpd256:
10580  case X86::BI__builtin_ia32_blendps256:
10581  case X86::BI__builtin_ia32_pblendw256:
10582  case X86::BI__builtin_ia32_pblendd128:
10583  case X86::BI__builtin_ia32_pblendd256: {
10584    unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
10585    unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
10586
10587    uint32_t Indices[16];
10588    // If there are more than 8 elements, the immediate is used twice so make
10589    // sure we handle that.
10590    for (unsigned i = 0; i != NumElts; ++i)
10591      Indices[i] = ((Imm >> (i % 8)) & 0x1) ? NumElts + i : i;
10592
10593    return Builder.CreateShuffleVector(Ops[0], Ops[1],
10594                                       makeArrayRef(Indices, NumElts),
10595                                       "blend");
10596  }
10597  case X86::BI__builtin_ia32_pshuflw:
10598  case X86::BI__builtin_ia32_pshuflw256:
10599  case X86::BI__builtin_ia32_pshuflw512: {
10600    uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
10601    llvm::Type *Ty = Ops[0]->getType();
10602    unsigned NumElts = Ty->getVectorNumElements();
10603
10604    // Splat the 8-bits of immediate 4 times to help the loop wrap around.
10605    Imm = (Imm & 0xff) * 0x01010101;
10606
10607    uint32_t Indices[32];
10608    for (unsigned l = 0; l != NumElts; l += 8) {
10609      for (unsigned i = 0; i != 4; ++i) {
10610        Indices[l + i] = l + (Imm & 3);
10611        Imm >>= 2;
10612      }
10613      for (unsigned i = 4; i != 8; ++i)
10614        Indices[l + i] = l + i;
10615    }
10616
10617    return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
10618                                       makeArrayRef(Indices, NumElts),
10619                                       "pshuflw");
10620  }
10621  case X86::BI__builtin_ia32_pshufhw:
10622  case X86::BI__builtin_ia32_pshufhw256:
10623  case X86::BI__builtin_ia32_pshufhw512: {
10624    uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
10625    llvm::Type *Ty = Ops[0]->getType();
10626    unsigned NumElts = Ty->getVectorNumElements();
10627
10628    // Splat the 8-bits of immediate 4 times to help the loop wrap around.
10629    Imm = (Imm & 0xff) * 0x01010101;
10630
10631    uint32_t Indices[32];
10632    for (unsigned l = 0; l != NumElts; l += 8) {
10633      for (unsigned i = 0; i != 4; ++i)
10634        Indices[l + i] = l + i;
10635      for (unsigned i = 4; i != 8; ++i) {
10636        Indices[l + i] = l + 4 + (Imm & 3);
10637        Imm >>= 2;
10638      }
10639    }
10640
10641    return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
10642                                       makeArrayRef(Indices, NumElts),
10643                                       "pshufhw");
10644  }
10645  case X86::BI__builtin_ia32_pshufd:
10646  case X86::BI__builtin_ia32_pshufd256:
10647  case X86::BI__builtin_ia32_pshufd512:
10648  case X86::BI__builtin_ia32_vpermilpd:
10649  case X86::BI__builtin_ia32_vpermilps:
10650  case X86::BI__builtin_ia32_vpermilpd256:
10651  case X86::BI__builtin_ia32_vpermilps256:
10652  case X86::BI__builtin_ia32_vpermilpd512:
10653  case X86::BI__builtin_ia32_vpermilps512: {
10654    uint32_t Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
10655    llvm::Type *Ty = Ops[0]->getType();
10656    unsigned NumElts = Ty->getVectorNumElements();
10657    unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128;
10658    unsigned NumLaneElts = NumElts / NumLanes;
10659
10660    // Splat the 8-bits of immediate 4 times to help the loop wrap around.
10661    Imm = (Imm & 0xff) * 0x01010101;
10662
10663    uint32_t Indices[16];
10664    for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
10665      for (unsigned i = 0; i != NumLaneElts; ++i) {
10666        Indices[i + l] = (Imm % NumLaneElts) + l;
10667        Imm /= NumLaneElts;
10668      }
10669    }
10670
10671    return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
10672                                       makeArrayRef(Indices, NumElts),
10673                                       "permil");
10674  }
10675  case X86::BI__builtin_ia32_shufpd:
10676  case X86::BI__builtin_ia32_shufpd256:
10677  case X86::BI__builtin_ia32_shufpd512:
10678  case X86::BI__builtin_ia32_shufps:
10679  case X86::BI__builtin_ia32_shufps256:
10680  case X86::BI__builtin_ia32_shufps512: {
10681    uint32_t Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
10682    llvm::Type *Ty = Ops[0]->getType();
10683    unsigned NumElts = Ty->getVectorNumElements();
10684    unsigned NumLanes = Ty->getPrimitiveSizeInBits() / 128;
10685    unsigned NumLaneElts = NumElts / NumLanes;
10686
10687    // Splat the 8-bits of immediate 4 times to help the loop wrap around.
10688    Imm = (Imm & 0xff) * 0x01010101;
10689
10690    uint32_t Indices[16];
10691    for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
10692      for (unsigned i = 0; i != NumLaneElts; ++i) {
10693        unsigned Index = Imm % NumLaneElts;
10694        Imm /= NumLaneElts;
10695        if (i >= (NumLaneElts / 2))
10696          Index += NumElts;
10697        Indices[l + i] = l + Index;
10698      }
10699    }
10700
10701    return Builder.CreateShuffleVector(Ops[0], Ops[1],
10702                                       makeArrayRef(Indices, NumElts),
10703                                       "shufp");
10704  }
10705  case X86::BI__builtin_ia32_permdi256:
10706  case X86::BI__builtin_ia32_permdf256:
10707  case X86::BI__builtin_ia32_permdi512:
10708  case X86::BI__builtin_ia32_permdf512: {
10709    unsigned Imm = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
10710    llvm::Type *Ty = Ops[0]->getType();
10711    unsigned NumElts = Ty->getVectorNumElements();
10712
10713    // These intrinsics operate on 256-bit lanes of four 64-bit elements.
10714    uint32_t Indices[8];
10715    for (unsigned l = 0; l != NumElts; l += 4)
10716      for (unsigned i = 0; i != 4; ++i)
10717        Indices[l + i] = l + ((Imm >> (2 * i)) & 0x3);
10718
10719    return Builder.CreateShuffleVector(Ops[0], UndefValue::get(Ty),
10720                                       makeArrayRef(Indices, NumElts),
10721                                       "perm");
10722  }
10723  case X86::BI__builtin_ia32_palignr128:
10724  case X86::BI__builtin_ia32_palignr256:
10725  case X86::BI__builtin_ia32_palignr512: {
10726    unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff;
10727
10728    unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
10729    assert(NumElts % 16 == 0);
10730
10731    // If palignr is shifting the pair of vectors more than the size of two
10732    // lanes, emit zero.
10733    if (ShiftVal >= 32)
10734      return llvm::Constant::getNullValue(ConvertType(E->getType()));
10735
10736    // If palignr is shifting the pair of input vectors more than one lane,
10737    // but less than two lanes, convert to shifting in zeroes.
10738    if (ShiftVal > 16) {
10739      ShiftVal -= 16;
10740      Ops[1] = Ops[0];
10741      Ops[0] = llvm::Constant::getNullValue(Ops[0]->getType());
10742    }
10743
10744    uint32_t Indices[64];
10745    // 256-bit palignr operates on 128-bit lanes so we need to handle that
10746    for (unsigned l = 0; l != NumElts; l += 16) {
10747      for (unsigned i = 0; i != 16; ++i) {
10748        unsigned Idx = ShiftVal + i;
10749        if (Idx >= 16)
10750          Idx += NumElts - 16; // End of lane, switch operand.
10751        Indices[l + i] = Idx + l;
10752      }
10753    }
10754
10755    return Builder.CreateShuffleVector(Ops[1], Ops[0],
10756                                       makeArrayRef(Indices, NumElts),
10757                                       "palignr");
10758  }
10759  case X86::BI__builtin_ia32_alignd128:
10760  case X86::BI__builtin_ia32_alignd256:
10761  case X86::BI__builtin_ia32_alignd512:
10762  case X86::BI__builtin_ia32_alignq128:
10763  case X86::BI__builtin_ia32_alignq256:
10764  case X86::BI__builtin_ia32_alignq512: {
10765    unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
10766    unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0xff;
10767
10768    // Mask the shift amount to width of two vectors.
10769    ShiftVal &= (2 * NumElts) - 1;
10770
10771    uint32_t Indices[16];
10772    for (unsigned i = 0; i != NumElts; ++i)
10773      Indices[i] = i + ShiftVal;
10774
10775    return Builder.CreateShuffleVector(Ops[1], Ops[0],
10776                                       makeArrayRef(Indices, NumElts),
10777                                       "valign");
10778  }
10779  case X86::BI__builtin_ia32_shuf_f32x4_256:
10780  case X86::BI__builtin_ia32_shuf_f64x2_256:
10781  case X86::BI__builtin_ia32_shuf_i32x4_256:
10782  case X86::BI__builtin_ia32_shuf_i64x2_256:
10783  case X86::BI__builtin_ia32_shuf_f32x4:
10784  case X86::BI__builtin_ia32_shuf_f64x2:
10785  case X86::BI__builtin_ia32_shuf_i32x4:
10786  case X86::BI__builtin_ia32_shuf_i64x2: {
10787    unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
10788    llvm::Type *Ty = Ops[0]->getType();
10789    unsigned NumElts = Ty->getVectorNumElements();
10790    unsigned NumLanes = Ty->getPrimitiveSizeInBits() == 512 ? 4 : 2;
10791    unsigned NumLaneElts = NumElts / NumLanes;
10792
10793    uint32_t Indices[16];
10794    for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
10795      unsigned Index = (Imm % NumLanes) * NumLaneElts;
10796      Imm /= NumLanes; // Discard the bits we just used.
10797      if (l >= (NumElts / 2))
10798        Index += NumElts; // Switch to other source.
10799      for (unsigned i = 0; i != NumLaneElts; ++i) {
10800        Indices[l + i] = Index + i;
10801      }
10802    }
10803
10804    return Builder.CreateShuffleVector(Ops[0], Ops[1],
10805                                       makeArrayRef(Indices, NumElts),
10806                                       "shuf");
10807  }
10808
10809  case X86::BI__builtin_ia32_vperm2f128_pd256:
10810  case X86::BI__builtin_ia32_vperm2f128_ps256:
10811  case X86::BI__builtin_ia32_vperm2f128_si256:
10812  case X86::BI__builtin_ia32_permti256: {
10813    unsigned Imm = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
10814    unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
10815
10816    // This takes a very simple approach since there are two lanes and a
10817    // shuffle can have 2 inputs. So we reserve the first input for the first
10818    // lane and the second input for the second lane. This may result in
10819    // duplicate sources, but this can be dealt with in the backend.
10820
10821    Value *OutOps[2];
10822    uint32_t Indices[8];
10823    for (unsigned l = 0; l != 2; ++l) {
10824      // Determine the source for this lane.
10825      if (Imm & (1 << ((l * 4) + 3)))
10826        OutOps[l] = llvm::ConstantAggregateZero::get(Ops[0]->getType());
10827      else if (Imm & (1 << ((l * 4) + 1)))
10828        OutOps[l] = Ops[1];
10829      else
10830        OutOps[l] = Ops[0];
10831
10832      for (unsigned i = 0; i != NumElts/2; ++i) {
10833        // Start with ith element of the source for this lane.
10834        unsigned Idx = (l * NumElts) + i;
10835        // If bit 0 of the immediate half is set, switch to the high half of
10836        // the source.
10837        if (Imm & (1 << (l * 4)))
10838          Idx += NumElts/2;
10839        Indices[(l * (NumElts/2)) + i] = Idx;
10840      }
10841    }
10842
10843    return Builder.CreateShuffleVector(OutOps[0], OutOps[1],
10844                                       makeArrayRef(Indices, NumElts),
10845                                       "vperm");
10846  }
10847
10848  case X86::BI__builtin_ia32_pslldqi128_byteshift:
10849  case X86::BI__builtin_ia32_pslldqi256_byteshift:
10850  case X86::BI__builtin_ia32_pslldqi512_byteshift: {
10851    unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
10852    llvm::Type *ResultType = Ops[0]->getType();
10853    // Builtin type is vXi64 so multiply by 8 to get bytes.
10854    unsigned NumElts = ResultType->getVectorNumElements() * 8;
10855
10856    // If pslldq is shifting the vector more than 15 bytes, emit zero.
10857    if (ShiftVal >= 16)
10858      return llvm::Constant::getNullValue(ResultType);
10859
10860    uint32_t Indices[64];
10861    // 256/512-bit pslldq operates on 128-bit lanes so we need to handle that
10862    for (unsigned l = 0; l != NumElts; l += 16) {
10863      for (unsigned i = 0; i != 16; ++i) {
10864        unsigned Idx = NumElts + i - ShiftVal;
10865        if (Idx < NumElts) Idx -= NumElts - 16; // end of lane, switch operand.
10866        Indices[l + i] = Idx + l;
10867      }
10868    }
10869
10870    llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, NumElts);
10871    Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast");
10872    Value *Zero = llvm::Constant::getNullValue(VecTy);
10873    Value *SV = Builder.CreateShuffleVector(Zero, Cast,
10874                                            makeArrayRef(Indices, NumElts),
10875                                            "pslldq");
10876    return Builder.CreateBitCast(SV, Ops[0]->getType(), "cast");
10877  }
10878  case X86::BI__builtin_ia32_psrldqi128_byteshift:
10879  case X86::BI__builtin_ia32_psrldqi256_byteshift:
10880  case X86::BI__builtin_ia32_psrldqi512_byteshift: {
10881    unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
10882    llvm::Type *ResultType = Ops[0]->getType();
10883    // Builtin type is vXi64 so multiply by 8 to get bytes.
10884    unsigned NumElts = ResultType->getVectorNumElements() * 8;
10885
10886    // If psrldq is shifting the vector more than 15 bytes, emit zero.
10887    if (ShiftVal >= 16)
10888      return llvm::Constant::getNullValue(ResultType);
10889
10890    uint32_t Indices[64];
10891    // 256/512-bit psrldq operates on 128-bit lanes so we need to handle that
10892    for (unsigned l = 0; l != NumElts; l += 16) {
10893      for (unsigned i = 0; i != 16; ++i) {
10894        unsigned Idx = i + ShiftVal;
10895        if (Idx >= 16) Idx += NumElts - 16; // end of lane, switch operand.
10896        Indices[l + i] = Idx + l;
10897      }
10898    }
10899
10900    llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, NumElts);
10901    Value *Cast = Builder.CreateBitCast(Ops[0], VecTy, "cast");
10902    Value *Zero = llvm::Constant::getNullValue(VecTy);
10903    Value *SV = Builder.CreateShuffleVector(Cast, Zero,
10904                                            makeArrayRef(Indices, NumElts),
10905                                            "psrldq");
10906    return Builder.CreateBitCast(SV, ResultType, "cast");
10907  }
10908  case X86::BI__builtin_ia32_kshiftliqi:
10909  case X86::BI__builtin_ia32_kshiftlihi:
10910  case X86::BI__builtin_ia32_kshiftlisi:
10911  case X86::BI__builtin_ia32_kshiftlidi: {
10912    unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
10913    unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
10914
10915    if (ShiftVal >= NumElts)
10916      return llvm::Constant::getNullValue(Ops[0]->getType());
10917
10918    Value *In = getMaskVecValue(*this, Ops[0], NumElts);
10919
10920    uint32_t Indices[64];
10921    for (unsigned i = 0; i != NumElts; ++i)
10922      Indices[i] = NumElts + i - ShiftVal;
10923
10924    Value *Zero = llvm::Constant::getNullValue(In->getType());
10925    Value *SV = Builder.CreateShuffleVector(Zero, In,
10926                                            makeArrayRef(Indices, NumElts),
10927                                            "kshiftl");
10928    return Builder.CreateBitCast(SV, Ops[0]->getType());
10929  }
10930  case X86::BI__builtin_ia32_kshiftriqi:
10931  case X86::BI__builtin_ia32_kshiftrihi:
10932  case X86::BI__builtin_ia32_kshiftrisi:
10933  case X86::BI__builtin_ia32_kshiftridi: {
10934    unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() & 0xff;
10935    unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
10936
10937    if (ShiftVal >= NumElts)
10938      return llvm::Constant::getNullValue(Ops[0]->getType());
10939
10940    Value *In = getMaskVecValue(*this, Ops[0], NumElts);
10941
10942    uint32_t Indices[64];
10943    for (unsigned i = 0; i != NumElts; ++i)
10944      Indices[i] = i + ShiftVal;
10945
10946    Value *Zero = llvm::Constant::getNullValue(In->getType());
10947    Value *SV = Builder.CreateShuffleVector(In, Zero,
10948                                            makeArrayRef(Indices, NumElts),
10949                                            "kshiftr");
10950    return Builder.CreateBitCast(SV, Ops[0]->getType());
10951  }
10952  case X86::BI__builtin_ia32_movnti:
10953  case X86::BI__builtin_ia32_movnti64:
10954  case X86::BI__builtin_ia32_movntsd:
10955  case X86::BI__builtin_ia32_movntss: {
10956    llvm::MDNode *Node = llvm::MDNode::get(
10957        getLLVMContext(), llvm::ConstantAsMetadata::get(Builder.getInt32(1)));
10958
10959    Value *Ptr = Ops[0];
10960    Value *Src = Ops[1];
10961
10962    // Extract the 0'th element of the source vector.
10963    if (BuiltinID == X86::BI__builtin_ia32_movntsd ||
10964        BuiltinID == X86::BI__builtin_ia32_movntss)
10965      Src = Builder.CreateExtractElement(Src, (uint64_t)0, "extract");
10966
10967    // Convert the type of the pointer to a pointer to the stored type.
10968    Value *BC = Builder.CreateBitCast(
10969        Ptr, llvm::PointerType::getUnqual(Src->getType()), "cast");
10970
10971    // Unaligned nontemporal store of the scalar value.
10972    StoreInst *SI = Builder.CreateDefaultAlignedStore(Src, BC);
10973    SI->setMetadata(CGM.getModule().getMDKindID("nontemporal"), Node);
10974    SI->setAlignment(1);
10975    return SI;
10976  }
10977  // Rotate is a special case of funnel shift - 1st 2 args are the same.
10978  case X86::BI__builtin_ia32_vprotb:
10979  case X86::BI__builtin_ia32_vprotw:
10980  case X86::BI__builtin_ia32_vprotd:
10981  case X86::BI__builtin_ia32_vprotq:
10982  case X86::BI__builtin_ia32_vprotbi:
10983  case X86::BI__builtin_ia32_vprotwi:
10984  case X86::BI__builtin_ia32_vprotdi:
10985  case X86::BI__builtin_ia32_vprotqi:
10986  case X86::BI__builtin_ia32_prold128:
10987  case X86::BI__builtin_ia32_prold256:
10988  case X86::BI__builtin_ia32_prold512:
10989  case X86::BI__builtin_ia32_prolq128:
10990  case X86::BI__builtin_ia32_prolq256:
10991  case X86::BI__builtin_ia32_prolq512:
10992  case X86::BI__builtin_ia32_prolvd128:
10993  case X86::BI__builtin_ia32_prolvd256:
10994  case X86::BI__builtin_ia32_prolvd512:
10995  case X86::BI__builtin_ia32_prolvq128:
10996  case X86::BI__builtin_ia32_prolvq256:
10997  case X86::BI__builtin_ia32_prolvq512:
10998    return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], false);
10999  case X86::BI__builtin_ia32_prord128:
11000  case X86::BI__builtin_ia32_prord256:
11001  case X86::BI__builtin_ia32_prord512:
11002  case X86::BI__builtin_ia32_prorq128:
11003  case X86::BI__builtin_ia32_prorq256:
11004  case X86::BI__builtin_ia32_prorq512:
11005  case X86::BI__builtin_ia32_prorvd128:
11006  case X86::BI__builtin_ia32_prorvd256:
11007  case X86::BI__builtin_ia32_prorvd512:
11008  case X86::BI__builtin_ia32_prorvq128:
11009  case X86::BI__builtin_ia32_prorvq256:
11010  case X86::BI__builtin_ia32_prorvq512:
11011    return EmitX86FunnelShift(*this, Ops[0], Ops[0], Ops[1], true);
11012  case X86::BI__builtin_ia32_selectb_128:
11013  case X86::BI__builtin_ia32_selectb_256:
11014  case X86::BI__builtin_ia32_selectb_512:
11015  case X86::BI__builtin_ia32_selectw_128:
11016  case X86::BI__builtin_ia32_selectw_256:
11017  case X86::BI__builtin_ia32_selectw_512:
11018  case X86::BI__builtin_ia32_selectd_128:
11019  case X86::BI__builtin_ia32_selectd_256:
11020  case X86::BI__builtin_ia32_selectd_512:
11021  case X86::BI__builtin_ia32_selectq_128:
11022  case X86::BI__builtin_ia32_selectq_256:
11023  case X86::BI__builtin_ia32_selectq_512:
11024  case X86::BI__builtin_ia32_selectps_128:
11025  case X86::BI__builtin_ia32_selectps_256:
11026  case X86::BI__builtin_ia32_selectps_512:
11027  case X86::BI__builtin_ia32_selectpd_128:
11028  case X86::BI__builtin_ia32_selectpd_256:
11029  case X86::BI__builtin_ia32_selectpd_512:
11030    return EmitX86Select(*this, Ops[0], Ops[1], Ops[2]);
11031  case X86::BI__builtin_ia32_selectss_128:
11032  case X86::BI__builtin_ia32_selectsd_128: {
11033    Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0);
11034    Value *B = Builder.CreateExtractElement(Ops[2], (uint64_t)0);
11035    A = EmitX86ScalarSelect(*this, Ops[0], A, B);
11036    return Builder.CreateInsertElement(Ops[1], A, (uint64_t)0);
11037  }
11038  case X86::BI__builtin_ia32_cmpb128_mask:
11039  case X86::BI__builtin_ia32_cmpb256_mask:
11040  case X86::BI__builtin_ia32_cmpb512_mask:
11041  case X86::BI__builtin_ia32_cmpw128_mask:
11042  case X86::BI__builtin_ia32_cmpw256_mask:
11043  case X86::BI__builtin_ia32_cmpw512_mask:
11044  case X86::BI__builtin_ia32_cmpd128_mask:
11045  case X86::BI__builtin_ia32_cmpd256_mask:
11046  case X86::BI__builtin_ia32_cmpd512_mask:
11047  case X86::BI__builtin_ia32_cmpq128_mask:
11048  case X86::BI__builtin_ia32_cmpq256_mask:
11049  case X86::BI__builtin_ia32_cmpq512_mask: {
11050    unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
11051    return EmitX86MaskedCompare(*this, CC, true, Ops);
11052  }
11053  case X86::BI__builtin_ia32_ucmpb128_mask:
11054  case X86::BI__builtin_ia32_ucmpb256_mask:
11055  case X86::BI__builtin_ia32_ucmpb512_mask:
11056  case X86::BI__builtin_ia32_ucmpw128_mask:
11057  case X86::BI__builtin_ia32_ucmpw256_mask:
11058  case X86::BI__builtin_ia32_ucmpw512_mask:
11059  case X86::BI__builtin_ia32_ucmpd128_mask:
11060  case X86::BI__builtin_ia32_ucmpd256_mask:
11061  case X86::BI__builtin_ia32_ucmpd512_mask:
11062  case X86::BI__builtin_ia32_ucmpq128_mask:
11063  case X86::BI__builtin_ia32_ucmpq256_mask:
11064  case X86::BI__builtin_ia32_ucmpq512_mask: {
11065    unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x7;
11066    return EmitX86MaskedCompare(*this, CC, false, Ops);
11067  }
11068  case X86::BI__builtin_ia32_vpcomb:
11069  case X86::BI__builtin_ia32_vpcomw:
11070  case X86::BI__builtin_ia32_vpcomd:
11071  case X86::BI__builtin_ia32_vpcomq:
11072    return EmitX86vpcom(*this, Ops, true);
11073  case X86::BI__builtin_ia32_vpcomub:
11074  case X86::BI__builtin_ia32_vpcomuw:
11075  case X86::BI__builtin_ia32_vpcomud:
11076  case X86::BI__builtin_ia32_vpcomuq:
11077    return EmitX86vpcom(*this, Ops, false);
11078
11079  case X86::BI__builtin_ia32_kortestcqi:
11080  case X86::BI__builtin_ia32_kortestchi:
11081  case X86::BI__builtin_ia32_kortestcsi:
11082  case X86::BI__builtin_ia32_kortestcdi: {
11083    Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops);
11084    Value *C = llvm::Constant::getAllOnesValue(Ops[0]->getType());
11085    Value *Cmp = Builder.CreateICmpEQ(Or, C);
11086    return Builder.CreateZExt(Cmp, ConvertType(E->getType()));
11087  }
11088  case X86::BI__builtin_ia32_kortestzqi:
11089  case X86::BI__builtin_ia32_kortestzhi:
11090  case X86::BI__builtin_ia32_kortestzsi:
11091  case X86::BI__builtin_ia32_kortestzdi: {
11092    Value *Or = EmitX86MaskLogic(*this, Instruction::Or, Ops);
11093    Value *C = llvm::Constant::getNullValue(Ops[0]->getType());
11094    Value *Cmp = Builder.CreateICmpEQ(Or, C);
11095    return Builder.CreateZExt(Cmp, ConvertType(E->getType()));
11096  }
11097
11098  case X86::BI__builtin_ia32_ktestcqi:
11099  case X86::BI__builtin_ia32_ktestzqi:
11100  case X86::BI__builtin_ia32_ktestchi:
11101  case X86::BI__builtin_ia32_ktestzhi:
11102  case X86::BI__builtin_ia32_ktestcsi:
11103  case X86::BI__builtin_ia32_ktestzsi:
11104  case X86::BI__builtin_ia32_ktestcdi:
11105  case X86::BI__builtin_ia32_ktestzdi: {
11106    Intrinsic::ID IID;
11107    switch (BuiltinID) {
11108    default: llvm_unreachable("Unsupported intrinsic!");
11109    case X86::BI__builtin_ia32_ktestcqi:
11110      IID = Intrinsic::x86_avx512_ktestc_b;
11111      break;
11112    case X86::BI__builtin_ia32_ktestzqi:
11113      IID = Intrinsic::x86_avx512_ktestz_b;
11114      break;
11115    case X86::BI__builtin_ia32_ktestchi:
11116      IID = Intrinsic::x86_avx512_ktestc_w;
11117      break;
11118    case X86::BI__builtin_ia32_ktestzhi:
11119      IID = Intrinsic::x86_avx512_ktestz_w;
11120      break;
11121    case X86::BI__builtin_ia32_ktestcsi:
11122      IID = Intrinsic::x86_avx512_ktestc_d;
11123      break;
11124    case X86::BI__builtin_ia32_ktestzsi:
11125      IID = Intrinsic::x86_avx512_ktestz_d;
11126      break;
11127    case X86::BI__builtin_ia32_ktestcdi:
11128      IID = Intrinsic::x86_avx512_ktestc_q;
11129      break;
11130    case X86::BI__builtin_ia32_ktestzdi:
11131      IID = Intrinsic::x86_avx512_ktestz_q;
11132      break;
11133    }
11134
11135    unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
11136    Value *LHS = getMaskVecValue(*this, Ops[0], NumElts);
11137    Value *RHS = getMaskVecValue(*this, Ops[1], NumElts);
11138    Function *Intr = CGM.getIntrinsic(IID);
11139    return Builder.CreateCall(Intr, {LHS, RHS});
11140  }
11141
11142  case X86::BI__builtin_ia32_kaddqi:
11143  case X86::BI__builtin_ia32_kaddhi:
11144  case X86::BI__builtin_ia32_kaddsi:
11145  case X86::BI__builtin_ia32_kadddi: {
11146    Intrinsic::ID IID;
11147    switch (BuiltinID) {
11148    default: llvm_unreachable("Unsupported intrinsic!");
11149    case X86::BI__builtin_ia32_kaddqi:
11150      IID = Intrinsic::x86_avx512_kadd_b;
11151      break;
11152    case X86::BI__builtin_ia32_kaddhi:
11153      IID = Intrinsic::x86_avx512_kadd_w;
11154      break;
11155    case X86::BI__builtin_ia32_kaddsi:
11156      IID = Intrinsic::x86_avx512_kadd_d;
11157      break;
11158    case X86::BI__builtin_ia32_kadddi:
11159      IID = Intrinsic::x86_avx512_kadd_q;
11160      break;
11161    }
11162
11163    unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
11164    Value *LHS = getMaskVecValue(*this, Ops[0], NumElts);
11165    Value *RHS = getMaskVecValue(*this, Ops[1], NumElts);
11166    Function *Intr = CGM.getIntrinsic(IID);
11167    Value *Res = Builder.CreateCall(Intr, {LHS, RHS});
11168    return Builder.CreateBitCast(Res, Ops[0]->getType());
11169  }
11170  case X86::BI__builtin_ia32_kandqi:
11171  case X86::BI__builtin_ia32_kandhi:
11172  case X86::BI__builtin_ia32_kandsi:
11173  case X86::BI__builtin_ia32_kanddi:
11174    return EmitX86MaskLogic(*this, Instruction::And, Ops);
11175  case X86::BI__builtin_ia32_kandnqi:
11176  case X86::BI__builtin_ia32_kandnhi:
11177  case X86::BI__builtin_ia32_kandnsi:
11178  case X86::BI__builtin_ia32_kandndi:
11179    return EmitX86MaskLogic(*this, Instruction::And, Ops, true);
11180  case X86::BI__builtin_ia32_korqi:
11181  case X86::BI__builtin_ia32_korhi:
11182  case X86::BI__builtin_ia32_korsi:
11183  case X86::BI__builtin_ia32_kordi:
11184    return EmitX86MaskLogic(*this, Instruction::Or, Ops);
11185  case X86::BI__builtin_ia32_kxnorqi:
11186  case X86::BI__builtin_ia32_kxnorhi:
11187  case X86::BI__builtin_ia32_kxnorsi:
11188  case X86::BI__builtin_ia32_kxnordi:
11189    return EmitX86MaskLogic(*this, Instruction::Xor, Ops, true);
11190  case X86::BI__builtin_ia32_kxorqi:
11191  case X86::BI__builtin_ia32_kxorhi:
11192  case X86::BI__builtin_ia32_kxorsi:
11193  case X86::BI__builtin_ia32_kxordi:
11194    return EmitX86MaskLogic(*this, Instruction::Xor,  Ops);
11195  case X86::BI__builtin_ia32_knotqi:
11196  case X86::BI__builtin_ia32_knothi:
11197  case X86::BI__builtin_ia32_knotsi:
11198  case X86::BI__builtin_ia32_knotdi: {
11199    unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
11200    Value *Res = getMaskVecValue(*this, Ops[0], NumElts);
11201    return Builder.CreateBitCast(Builder.CreateNot(Res),
11202                                 Ops[0]->getType());
11203  }
11204  case X86::BI__builtin_ia32_kmovb:
11205  case X86::BI__builtin_ia32_kmovw:
11206  case X86::BI__builtin_ia32_kmovd:
11207  case X86::BI__builtin_ia32_kmovq: {
11208    // Bitcast to vXi1 type and then back to integer. This gets the mask
11209    // register type into the IR, but might be optimized out depending on
11210    // what's around it.
11211    unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
11212    Value *Res = getMaskVecValue(*this, Ops[0], NumElts);
11213    return Builder.CreateBitCast(Res, Ops[0]->getType());
11214  }
11215
11216  case X86::BI__builtin_ia32_kunpckdi:
11217  case X86::BI__builtin_ia32_kunpcksi:
11218  case X86::BI__builtin_ia32_kunpckhi: {
11219    unsigned NumElts = Ops[0]->getType()->getIntegerBitWidth();
11220    Value *LHS = getMaskVecValue(*this, Ops[0], NumElts);
11221    Value *RHS = getMaskVecValue(*this, Ops[1], NumElts);
11222    uint32_t Indices[64];
11223    for (unsigned i = 0; i != NumElts; ++i)
11224      Indices[i] = i;
11225
11226    // First extract half of each vector. This gives better codegen than
11227    // doing it in a single shuffle.
11228    LHS = Builder.CreateShuffleVector(LHS, LHS,
11229                                      makeArrayRef(Indices, NumElts / 2));
11230    RHS = Builder.CreateShuffleVector(RHS, RHS,
11231                                      makeArrayRef(Indices, NumElts / 2));
11232    // Concat the vectors.
11233    // NOTE: Operands are swapped to match the intrinsic definition.
11234    Value *Res = Builder.CreateShuffleVector(RHS, LHS,
11235                                             makeArrayRef(Indices, NumElts));
11236    return Builder.CreateBitCast(Res, Ops[0]->getType());
11237  }
11238
11239  case X86::BI__builtin_ia32_vplzcntd_128:
11240  case X86::BI__builtin_ia32_vplzcntd_256:
11241  case X86::BI__builtin_ia32_vplzcntd_512:
11242  case X86::BI__builtin_ia32_vplzcntq_128:
11243  case X86::BI__builtin_ia32_vplzcntq_256:
11244  case X86::BI__builtin_ia32_vplzcntq_512: {
11245    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, Ops[0]->getType());
11246    return Builder.CreateCall(F, {Ops[0],Builder.getInt1(false)});
11247  }
11248  case X86::BI__builtin_ia32_sqrtss:
11249  case X86::BI__builtin_ia32_sqrtsd: {
11250    Value *A = Builder.CreateExtractElement(Ops[0], (uint64_t)0);
11251    Function *F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType());
11252    A = Builder.CreateCall(F, {A});
11253    return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0);
11254  }
11255  case X86::BI__builtin_ia32_sqrtsd_round_mask:
11256  case X86::BI__builtin_ia32_sqrtss_round_mask: {
11257    unsigned CC = cast<llvm::ConstantInt>(Ops[4])->getZExtValue();
11258    // Support only if the rounding mode is 4 (AKA CUR_DIRECTION),
11259    // otherwise keep the intrinsic.
11260    if (CC != 4) {
11261      Intrinsic::ID IID = BuiltinID == X86::BI__builtin_ia32_sqrtsd_round_mask ?
11262                          Intrinsic::x86_avx512_mask_sqrt_sd :
11263                          Intrinsic::x86_avx512_mask_sqrt_ss;
11264      return Builder.CreateCall(CGM.getIntrinsic(IID), Ops);
11265    }
11266    Value *A = Builder.CreateExtractElement(Ops[1], (uint64_t)0);
11267    Function *F = CGM.getIntrinsic(Intrinsic::sqrt, A->getType());
11268    A = Builder.CreateCall(F, A);
11269    Value *Src = Builder.CreateExtractElement(Ops[2], (uint64_t)0);
11270    A = EmitX86ScalarSelect(*this, Ops[3], A, Src);
11271    return Builder.CreateInsertElement(Ops[0], A, (uint64_t)0);
11272  }
11273  case X86::BI__builtin_ia32_sqrtpd256:
11274  case X86::BI__builtin_ia32_sqrtpd:
11275  case X86::BI__builtin_ia32_sqrtps256:
11276  case X86::BI__builtin_ia32_sqrtps:
11277  case X86::BI__builtin_ia32_sqrtps512:
11278  case X86::BI__builtin_ia32_sqrtpd512: {
11279    if (Ops.size() == 2) {
11280      unsigned CC = cast<llvm::ConstantInt>(Ops[1])->getZExtValue();
11281      // Support only if the rounding mode is 4 (AKA CUR_DIRECTION),
11282      // otherwise keep the intrinsic.
11283      if (CC != 4) {
11284        Intrinsic::ID IID = BuiltinID == X86::BI__builtin_ia32_sqrtps512 ?
11285                            Intrinsic::x86_avx512_sqrt_ps_512 :
11286                            Intrinsic::x86_avx512_sqrt_pd_512;
11287        return Builder.CreateCall(CGM.getIntrinsic(IID), Ops);
11288      }
11289    }
11290    Function *F = CGM.getIntrinsic(Intrinsic::sqrt, Ops[0]->getType());
11291    return Builder.CreateCall(F, Ops[0]);
11292  }
11293  case X86::BI__builtin_ia32_pabsb128:
11294  case X86::BI__builtin_ia32_pabsw128:
11295  case X86::BI__builtin_ia32_pabsd128:
11296  case X86::BI__builtin_ia32_pabsb256:
11297  case X86::BI__builtin_ia32_pabsw256:
11298  case X86::BI__builtin_ia32_pabsd256:
11299  case X86::BI__builtin_ia32_pabsq128:
11300  case X86::BI__builtin_ia32_pabsq256:
11301  case X86::BI__builtin_ia32_pabsb512:
11302  case X86::BI__builtin_ia32_pabsw512:
11303  case X86::BI__builtin_ia32_pabsd512:
11304  case X86::BI__builtin_ia32_pabsq512:
11305    return EmitX86Abs(*this, Ops);
11306
11307  case X86::BI__builtin_ia32_pmaxsb128:
11308  case X86::BI__builtin_ia32_pmaxsw128:
11309  case X86::BI__builtin_ia32_pmaxsd128:
11310  case X86::BI__builtin_ia32_pmaxsq128:
11311  case X86::BI__builtin_ia32_pmaxsb256:
11312  case X86::BI__builtin_ia32_pmaxsw256:
11313  case X86::BI__builtin_ia32_pmaxsd256:
11314  case X86::BI__builtin_ia32_pmaxsq256:
11315  case X86::BI__builtin_ia32_pmaxsb512:
11316  case X86::BI__builtin_ia32_pmaxsw512:
11317  case X86::BI__builtin_ia32_pmaxsd512:
11318  case X86::BI__builtin_ia32_pmaxsq512:
11319    return EmitX86MinMax(*this, ICmpInst::ICMP_SGT, Ops);
11320  case X86::BI__builtin_ia32_pmaxub128:
11321  case X86::BI__builtin_ia32_pmaxuw128:
11322  case X86::BI__builtin_ia32_pmaxud128:
11323  case X86::BI__builtin_ia32_pmaxuq128:
11324  case X86::BI__builtin_ia32_pmaxub256:
11325  case X86::BI__builtin_ia32_pmaxuw256:
11326  case X86::BI__builtin_ia32_pmaxud256:
11327  case X86::BI__builtin_ia32_pmaxuq256:
11328  case X86::BI__builtin_ia32_pmaxub512:
11329  case X86::BI__builtin_ia32_pmaxuw512:
11330  case X86::BI__builtin_ia32_pmaxud512:
11331  case X86::BI__builtin_ia32_pmaxuq512:
11332    return EmitX86MinMax(*this, ICmpInst::ICMP_UGT, Ops);
11333  case X86::BI__builtin_ia32_pminsb128:
11334  case X86::BI__builtin_ia32_pminsw128:
11335  case X86::BI__builtin_ia32_pminsd128:
11336  case X86::BI__builtin_ia32_pminsq128:
11337  case X86::BI__builtin_ia32_pminsb256:
11338  case X86::BI__builtin_ia32_pminsw256:
11339  case X86::BI__builtin_ia32_pminsd256:
11340  case X86::BI__builtin_ia32_pminsq256:
11341  case X86::BI__builtin_ia32_pminsb512:
11342  case X86::BI__builtin_ia32_pminsw512:
11343  case X86::BI__builtin_ia32_pminsd512:
11344  case X86::BI__builtin_ia32_pminsq512:
11345    return EmitX86MinMax(*this, ICmpInst::ICMP_SLT, Ops);
11346  case X86::BI__builtin_ia32_pminub128:
11347  case X86::BI__builtin_ia32_pminuw128:
11348  case X86::BI__builtin_ia32_pminud128:
11349  case X86::BI__builtin_ia32_pminuq128:
11350  case X86::BI__builtin_ia32_pminub256:
11351  case X86::BI__builtin_ia32_pminuw256:
11352  case X86::BI__builtin_ia32_pminud256:
11353  case X86::BI__builtin_ia32_pminuq256:
11354  case X86::BI__builtin_ia32_pminub512:
11355  case X86::BI__builtin_ia32_pminuw512:
11356  case X86::BI__builtin_ia32_pminud512:
11357  case X86::BI__builtin_ia32_pminuq512:
11358    return EmitX86MinMax(*this, ICmpInst::ICMP_ULT, Ops);
11359
11360  case X86::BI__builtin_ia32_pmuludq128:
11361  case X86::BI__builtin_ia32_pmuludq256:
11362  case X86::BI__builtin_ia32_pmuludq512:
11363    return EmitX86Muldq(*this, /*IsSigned*/false, Ops);
11364
11365  case X86::BI__builtin_ia32_pmuldq128:
11366  case X86::BI__builtin_ia32_pmuldq256:
11367  case X86::BI__builtin_ia32_pmuldq512:
11368    return EmitX86Muldq(*this, /*IsSigned*/true, Ops);
11369
11370  case X86::BI__builtin_ia32_pternlogd512_mask:
11371  case X86::BI__builtin_ia32_pternlogq512_mask:
11372  case X86::BI__builtin_ia32_pternlogd128_mask:
11373  case X86::BI__builtin_ia32_pternlogd256_mask:
11374  case X86::BI__builtin_ia32_pternlogq128_mask:
11375  case X86::BI__builtin_ia32_pternlogq256_mask:
11376    return EmitX86Ternlog(*this, /*ZeroMask*/false, Ops);
11377
11378  case X86::BI__builtin_ia32_pternlogd512_maskz:
11379  case X86::BI__builtin_ia32_pternlogq512_maskz:
11380  case X86::BI__builtin_ia32_pternlogd128_maskz:
11381  case X86::BI__builtin_ia32_pternlogd256_maskz:
11382  case X86::BI__builtin_ia32_pternlogq128_maskz:
11383  case X86::BI__builtin_ia32_pternlogq256_maskz:
11384    return EmitX86Ternlog(*this, /*ZeroMask*/true, Ops);
11385
11386  case X86::BI__builtin_ia32_vpshldd128:
11387  case X86::BI__builtin_ia32_vpshldd256:
11388  case X86::BI__builtin_ia32_vpshldd512:
11389  case X86::BI__builtin_ia32_vpshldq128:
11390  case X86::BI__builtin_ia32_vpshldq256:
11391  case X86::BI__builtin_ia32_vpshldq512:
11392  case X86::BI__builtin_ia32_vpshldw128:
11393  case X86::BI__builtin_ia32_vpshldw256:
11394  case X86::BI__builtin_ia32_vpshldw512:
11395    return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false);
11396
11397  case X86::BI__builtin_ia32_vpshrdd128:
11398  case X86::BI__builtin_ia32_vpshrdd256:
11399  case X86::BI__builtin_ia32_vpshrdd512:
11400  case X86::BI__builtin_ia32_vpshrdq128:
11401  case X86::BI__builtin_ia32_vpshrdq256:
11402  case X86::BI__builtin_ia32_vpshrdq512:
11403  case X86::BI__builtin_ia32_vpshrdw128:
11404  case X86::BI__builtin_ia32_vpshrdw256:
11405  case X86::BI__builtin_ia32_vpshrdw512:
11406    // Ops 0 and 1 are swapped.
11407    return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true);
11408
11409  case X86::BI__builtin_ia32_vpshldvd128:
11410  case X86::BI__builtin_ia32_vpshldvd256:
11411  case X86::BI__builtin_ia32_vpshldvd512:
11412  case X86::BI__builtin_ia32_vpshldvq128:
11413  case X86::BI__builtin_ia32_vpshldvq256:
11414  case X86::BI__builtin_ia32_vpshldvq512:
11415  case X86::BI__builtin_ia32_vpshldvw128:
11416  case X86::BI__builtin_ia32_vpshldvw256:
11417  case X86::BI__builtin_ia32_vpshldvw512:
11418    return EmitX86FunnelShift(*this, Ops[0], Ops[1], Ops[2], false);
11419
11420  case X86::BI__builtin_ia32_vpshrdvd128:
11421  case X86::BI__builtin_ia32_vpshrdvd256:
11422  case X86::BI__builtin_ia32_vpshrdvd512:
11423  case X86::BI__builtin_ia32_vpshrdvq128:
11424  case X86::BI__builtin_ia32_vpshrdvq256:
11425  case X86::BI__builtin_ia32_vpshrdvq512:
11426  case X86::BI__builtin_ia32_vpshrdvw128:
11427  case X86::BI__builtin_ia32_vpshrdvw256:
11428  case X86::BI__builtin_ia32_vpshrdvw512:
11429    // Ops 0 and 1 are swapped.
11430    return EmitX86FunnelShift(*this, Ops[1], Ops[0], Ops[2], true);
11431
11432  // 3DNow!
11433  case X86::BI__builtin_ia32_pswapdsf:
11434  case X86::BI__builtin_ia32_pswapdsi: {
11435    llvm::Type *MMXTy = llvm::Type::getX86_MMXTy(getLLVMContext());
11436    Ops[0] = Builder.CreateBitCast(Ops[0], MMXTy, "cast");
11437    llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_3dnowa_pswapd);
11438    return Builder.CreateCall(F, Ops, "pswapd");
11439  }
11440  case X86::BI__builtin_ia32_rdrand16_step:
11441  case X86::BI__builtin_ia32_rdrand32_step:
11442  case X86::BI__builtin_ia32_rdrand64_step:
11443  case X86::BI__builtin_ia32_rdseed16_step:
11444  case X86::BI__builtin_ia32_rdseed32_step:
11445  case X86::BI__builtin_ia32_rdseed64_step: {
11446    Intrinsic::ID ID;
11447    switch (BuiltinID) {
11448    default: llvm_unreachable("Unsupported intrinsic!");
11449    case X86::BI__builtin_ia32_rdrand16_step:
11450      ID = Intrinsic::x86_rdrand_16;
11451      break;
11452    case X86::BI__builtin_ia32_rdrand32_step:
11453      ID = Intrinsic::x86_rdrand_32;
11454      break;
11455    case X86::BI__builtin_ia32_rdrand64_step:
11456      ID = Intrinsic::x86_rdrand_64;
11457      break;
11458    case X86::BI__builtin_ia32_rdseed16_step:
11459      ID = Intrinsic::x86_rdseed_16;
11460      break;
11461    case X86::BI__builtin_ia32_rdseed32_step:
11462      ID = Intrinsic::x86_rdseed_32;
11463      break;
11464    case X86::BI__builtin_ia32_rdseed64_step:
11465      ID = Intrinsic::x86_rdseed_64;
11466      break;
11467    }
11468
11469    Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID));
11470    Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 0),
11471                                      Ops[0]);
11472    return Builder.CreateExtractValue(Call, 1);
11473  }
11474  case X86::BI__builtin_ia32_addcarryx_u32:
11475  case X86::BI__builtin_ia32_addcarryx_u64:
11476  case X86::BI__builtin_ia32_subborrow_u32:
11477  case X86::BI__builtin_ia32_subborrow_u64: {
11478    Intrinsic::ID IID;
11479    switch (BuiltinID) {
11480    default: llvm_unreachable("Unsupported intrinsic!");
11481    case X86::BI__builtin_ia32_addcarryx_u32:
11482      IID = Intrinsic::x86_addcarry_32;
11483      break;
11484    case X86::BI__builtin_ia32_addcarryx_u64:
11485      IID = Intrinsic::x86_addcarry_64;
11486      break;
11487    case X86::BI__builtin_ia32_subborrow_u32:
11488      IID = Intrinsic::x86_subborrow_32;
11489      break;
11490    case X86::BI__builtin_ia32_subborrow_u64:
11491      IID = Intrinsic::x86_subborrow_64;
11492      break;
11493    }
11494
11495    Value *Call = Builder.CreateCall(CGM.getIntrinsic(IID),
11496                                     { Ops[0], Ops[1], Ops[2] });
11497    Builder.CreateDefaultAlignedStore(Builder.CreateExtractValue(Call, 1),
11498                                      Ops[3]);
11499    return Builder.CreateExtractValue(Call, 0);
11500  }
11501
11502  case X86::BI__builtin_ia32_fpclassps128_mask:
11503  case X86::BI__builtin_ia32_fpclassps256_mask:
11504  case X86::BI__builtin_ia32_fpclassps512_mask:
11505  case X86::BI__builtin_ia32_fpclasspd128_mask:
11506  case X86::BI__builtin_ia32_fpclasspd256_mask:
11507  case X86::BI__builtin_ia32_fpclasspd512_mask: {
11508    unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
11509    Value *MaskIn = Ops[2];
11510    Ops.erase(&Ops[2]);
11511
11512    Intrinsic::ID ID;
11513    switch (BuiltinID) {
11514    default: llvm_unreachable("Unsupported intrinsic!");
11515    case X86::BI__builtin_ia32_fpclassps128_mask:
11516      ID = Intrinsic::x86_avx512_fpclass_ps_128;
11517      break;
11518    case X86::BI__builtin_ia32_fpclassps256_mask:
11519      ID = Intrinsic::x86_avx512_fpclass_ps_256;
11520      break;
11521    case X86::BI__builtin_ia32_fpclassps512_mask:
11522      ID = Intrinsic::x86_avx512_fpclass_ps_512;
11523      break;
11524    case X86::BI__builtin_ia32_fpclasspd128_mask:
11525      ID = Intrinsic::x86_avx512_fpclass_pd_128;
11526      break;
11527    case X86::BI__builtin_ia32_fpclasspd256_mask:
11528      ID = Intrinsic::x86_avx512_fpclass_pd_256;
11529      break;
11530    case X86::BI__builtin_ia32_fpclasspd512_mask:
11531      ID = Intrinsic::x86_avx512_fpclass_pd_512;
11532      break;
11533    }
11534
11535    Value *Fpclass = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
11536    return EmitX86MaskedCompareResult(*this, Fpclass, NumElts, MaskIn);
11537  }
11538
11539  case X86::BI__builtin_ia32_vpmultishiftqb128:
11540  case X86::BI__builtin_ia32_vpmultishiftqb256:
11541  case X86::BI__builtin_ia32_vpmultishiftqb512: {
11542    Intrinsic::ID ID;
11543    switch (BuiltinID) {
11544    default: llvm_unreachable("Unsupported intrinsic!");
11545    case X86::BI__builtin_ia32_vpmultishiftqb128:
11546      ID = Intrinsic::x86_avx512_pmultishift_qb_128;
11547      break;
11548    case X86::BI__builtin_ia32_vpmultishiftqb256:
11549      ID = Intrinsic::x86_avx512_pmultishift_qb_256;
11550      break;
11551    case X86::BI__builtin_ia32_vpmultishiftqb512:
11552      ID = Intrinsic::x86_avx512_pmultishift_qb_512;
11553      break;
11554    }
11555
11556    return Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
11557  }
11558
11559  case X86::BI__builtin_ia32_vpshufbitqmb128_mask:
11560  case X86::BI__builtin_ia32_vpshufbitqmb256_mask:
11561  case X86::BI__builtin_ia32_vpshufbitqmb512_mask: {
11562    unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
11563    Value *MaskIn = Ops[2];
11564    Ops.erase(&Ops[2]);
11565
11566    Intrinsic::ID ID;
11567    switch (BuiltinID) {
11568    default: llvm_unreachable("Unsupported intrinsic!");
11569    case X86::BI__builtin_ia32_vpshufbitqmb128_mask:
11570      ID = Intrinsic::x86_avx512_vpshufbitqmb_128;
11571      break;
11572    case X86::BI__builtin_ia32_vpshufbitqmb256_mask:
11573      ID = Intrinsic::x86_avx512_vpshufbitqmb_256;
11574      break;
11575    case X86::BI__builtin_ia32_vpshufbitqmb512_mask:
11576      ID = Intrinsic::x86_avx512_vpshufbitqmb_512;
11577      break;
11578    }
11579
11580    Value *Shufbit = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
11581    return EmitX86MaskedCompareResult(*this, Shufbit, NumElts, MaskIn);
11582  }
11583
11584  // packed comparison intrinsics
11585  case X86::BI__builtin_ia32_cmpeqps:
11586  case X86::BI__builtin_ia32_cmpeqpd:
11587    return getVectorFCmpIR(CmpInst::FCMP_OEQ);
11588  case X86::BI__builtin_ia32_cmpltps:
11589  case X86::BI__builtin_ia32_cmpltpd:
11590    return getVectorFCmpIR(CmpInst::FCMP_OLT);
11591  case X86::BI__builtin_ia32_cmpleps:
11592  case X86::BI__builtin_ia32_cmplepd:
11593    return getVectorFCmpIR(CmpInst::FCMP_OLE);
11594  case X86::BI__builtin_ia32_cmpunordps:
11595  case X86::BI__builtin_ia32_cmpunordpd:
11596    return getVectorFCmpIR(CmpInst::FCMP_UNO);
11597  case X86::BI__builtin_ia32_cmpneqps:
11598  case X86::BI__builtin_ia32_cmpneqpd:
11599    return getVectorFCmpIR(CmpInst::FCMP_UNE);
11600  case X86::BI__builtin_ia32_cmpnltps:
11601  case X86::BI__builtin_ia32_cmpnltpd:
11602    return getVectorFCmpIR(CmpInst::FCMP_UGE);
11603  case X86::BI__builtin_ia32_cmpnleps:
11604  case X86::BI__builtin_ia32_cmpnlepd:
11605    return getVectorFCmpIR(CmpInst::FCMP_UGT);
11606  case X86::BI__builtin_ia32_cmpordps:
11607  case X86::BI__builtin_ia32_cmpordpd:
11608    return getVectorFCmpIR(CmpInst::FCMP_ORD);
11609  case X86::BI__builtin_ia32_cmpps:
11610  case X86::BI__builtin_ia32_cmpps256:
11611  case X86::BI__builtin_ia32_cmppd:
11612  case X86::BI__builtin_ia32_cmppd256:
11613  case X86::BI__builtin_ia32_cmpps128_mask:
11614  case X86::BI__builtin_ia32_cmpps256_mask:
11615  case X86::BI__builtin_ia32_cmpps512_mask:
11616  case X86::BI__builtin_ia32_cmppd128_mask:
11617  case X86::BI__builtin_ia32_cmppd256_mask:
11618  case X86::BI__builtin_ia32_cmppd512_mask: {
11619    // Lowering vector comparisons to fcmp instructions, while
11620    // ignoring signalling behaviour requested
11621    // ignoring rounding mode requested
11622    // This is is only possible as long as FENV_ACCESS is not implemented.
11623    // See also: https://reviews.llvm.org/D45616
11624
11625    // The third argument is the comparison condition, and integer in the
11626    // range [0, 31]
11627    unsigned CC = cast<llvm::ConstantInt>(Ops[2])->getZExtValue() & 0x1f;
11628
11629    // Lowering to IR fcmp instruction.
11630    // Ignoring requested signaling behaviour,
11631    // e.g. both _CMP_GT_OS & _CMP_GT_OQ are translated to FCMP_OGT.
11632    FCmpInst::Predicate Pred;
11633    switch (CC) {
11634    case 0x00: Pred = FCmpInst::FCMP_OEQ;   break;
11635    case 0x01: Pred = FCmpInst::FCMP_OLT;   break;
11636    case 0x02: Pred = FCmpInst::FCMP_OLE;   break;
11637    case 0x03: Pred = FCmpInst::FCMP_UNO;   break;
11638    case 0x04: Pred = FCmpInst::FCMP_UNE;   break;
11639    case 0x05: Pred = FCmpInst::FCMP_UGE;   break;
11640    case 0x06: Pred = FCmpInst::FCMP_UGT;   break;
11641    case 0x07: Pred = FCmpInst::FCMP_ORD;   break;
11642    case 0x08: Pred = FCmpInst::FCMP_UEQ;   break;
11643    case 0x09: Pred = FCmpInst::FCMP_ULT;   break;
11644    case 0x0a: Pred = FCmpInst::FCMP_ULE;   break;
11645    case 0x0b: Pred = FCmpInst::FCMP_FALSE; break;
11646    case 0x0c: Pred = FCmpInst::FCMP_ONE;   break;
11647    case 0x0d: Pred = FCmpInst::FCMP_OGE;   break;
11648    case 0x0e: Pred = FCmpInst::FCMP_OGT;   break;
11649    case 0x0f: Pred = FCmpInst::FCMP_TRUE;  break;
11650    case 0x10: Pred = FCmpInst::FCMP_OEQ;   break;
11651    case 0x11: Pred = FCmpInst::FCMP_OLT;   break;
11652    case 0x12: Pred = FCmpInst::FCMP_OLE;   break;
11653    case 0x13: Pred = FCmpInst::FCMP_UNO;   break;
11654    case 0x14: Pred = FCmpInst::FCMP_UNE;   break;
11655    case 0x15: Pred = FCmpInst::FCMP_UGE;   break;
11656    case 0x16: Pred = FCmpInst::FCMP_UGT;   break;
11657    case 0x17: Pred = FCmpInst::FCMP_ORD;   break;
11658    case 0x18: Pred = FCmpInst::FCMP_UEQ;   break;
11659    case 0x19: Pred = FCmpInst::FCMP_ULT;   break;
11660    case 0x1a: Pred = FCmpInst::FCMP_ULE;   break;
11661    case 0x1b: Pred = FCmpInst::FCMP_FALSE; break;
11662    case 0x1c: Pred = FCmpInst::FCMP_ONE;   break;
11663    case 0x1d: Pred = FCmpInst::FCMP_OGE;   break;
11664    case 0x1e: Pred = FCmpInst::FCMP_OGT;   break;
11665    case 0x1f: Pred = FCmpInst::FCMP_TRUE;  break;
11666    default: llvm_unreachable("Unhandled CC");
11667    }
11668
11669    // Builtins without the _mask suffix return a vector of integers
11670    // of the same width as the input vectors
11671    switch (BuiltinID) {
11672    case X86::BI__builtin_ia32_cmpps512_mask:
11673    case X86::BI__builtin_ia32_cmppd512_mask:
11674    case X86::BI__builtin_ia32_cmpps128_mask:
11675    case X86::BI__builtin_ia32_cmpps256_mask:
11676    case X86::BI__builtin_ia32_cmppd128_mask:
11677    case X86::BI__builtin_ia32_cmppd256_mask: {
11678      unsigned NumElts = Ops[0]->getType()->getVectorNumElements();
11679      Value *Cmp = Builder.CreateFCmp(Pred, Ops[0], Ops[1]);
11680      return EmitX86MaskedCompareResult(*this, Cmp, NumElts, Ops[3]);
11681    }
11682    default:
11683      return getVectorFCmpIR(Pred);
11684    }
11685  }
11686
11687  // SSE scalar comparison intrinsics
11688  case X86::BI__builtin_ia32_cmpeqss:
11689    return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 0);
11690  case X86::BI__builtin_ia32_cmpltss:
11691    return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 1);
11692  case X86::BI__builtin_ia32_cmpless:
11693    return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 2);
11694  case X86::BI__builtin_ia32_cmpunordss:
11695    return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 3);
11696  case X86::BI__builtin_ia32_cmpneqss:
11697    return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 4);
11698  case X86::BI__builtin_ia32_cmpnltss:
11699    return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 5);
11700  case X86::BI__builtin_ia32_cmpnless:
11701    return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 6);
11702  case X86::BI__builtin_ia32_cmpordss:
11703    return getCmpIntrinsicCall(Intrinsic::x86_sse_cmp_ss, 7);
11704  case X86::BI__builtin_ia32_cmpeqsd:
11705    return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 0);
11706  case X86::BI__builtin_ia32_cmpltsd:
11707    return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 1);
11708  case X86::BI__builtin_ia32_cmplesd:
11709    return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 2);
11710  case X86::BI__builtin_ia32_cmpunordsd:
11711    return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 3);
11712  case X86::BI__builtin_ia32_cmpneqsd:
11713    return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 4);
11714  case X86::BI__builtin_ia32_cmpnltsd:
11715    return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 5);
11716  case X86::BI__builtin_ia32_cmpnlesd:
11717    return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 6);
11718  case X86::BI__builtin_ia32_cmpordsd:
11719    return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 7);
11720
11721  case X86::BI__emul:
11722  case X86::BI__emulu: {
11723    llvm::Type *Int64Ty = llvm::IntegerType::get(getLLVMContext(), 64);
11724    bool isSigned = (BuiltinID == X86::BI__emul);
11725    Value *LHS = Builder.CreateIntCast(Ops[0], Int64Ty, isSigned);
11726    Value *RHS = Builder.CreateIntCast(Ops[1], Int64Ty, isSigned);
11727    return Builder.CreateMul(LHS, RHS, "", !isSigned, isSigned);
11728  }
11729  case X86::BI__mulh:
11730  case X86::BI__umulh:
11731  case X86::BI_mul128:
11732  case X86::BI_umul128: {
11733    llvm::Type *ResType = ConvertType(E->getType());
11734    llvm::Type *Int128Ty = llvm::IntegerType::get(getLLVMContext(), 128);
11735
11736    bool IsSigned = (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI_mul128);
11737    Value *LHS = Builder.CreateIntCast(Ops[0], Int128Ty, IsSigned);
11738    Value *RHS = Builder.CreateIntCast(Ops[1], Int128Ty, IsSigned);
11739
11740    Value *MulResult, *HigherBits;
11741    if (IsSigned) {
11742      MulResult = Builder.CreateNSWMul(LHS, RHS);
11743      HigherBits = Builder.CreateAShr(MulResult, 64);
11744    } else {
11745      MulResult = Builder.CreateNUWMul(LHS, RHS);
11746      HigherBits = Builder.CreateLShr(MulResult, 64);
11747    }
11748    HigherBits = Builder.CreateIntCast(HigherBits, ResType, IsSigned);
11749
11750    if (BuiltinID == X86::BI__mulh || BuiltinID == X86::BI__umulh)
11751      return HigherBits;
11752
11753    Address HighBitsAddress = EmitPointerWithAlignment(E->getArg(2));
11754    Builder.CreateStore(HigherBits, HighBitsAddress);
11755    return Builder.CreateIntCast(MulResult, ResType, IsSigned);
11756  }
11757
11758  case X86::BI__faststorefence: {
11759    return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent,
11760                               llvm::SyncScope::System);
11761  }
11762  case X86::BI__shiftleft128:
11763  case X86::BI__shiftright128: {
11764    // FIXME: Once fshl/fshr no longer add an unneeded and and cmov, do this:
11765    // llvm::Function *F = CGM.getIntrinsic(
11766    //   BuiltinID == X86::BI__shiftleft128 ? Intrinsic::fshl : Intrinsic::fshr,
11767    //   Int64Ty);
11768    // Ops[2] = Builder.CreateZExt(Ops[2], Int64Ty);
11769    // return Builder.CreateCall(F, Ops);
11770    llvm::Type *Int128Ty = Builder.getInt128Ty();
11771    Value *HighPart128 =
11772        Builder.CreateShl(Builder.CreateZExt(Ops[1], Int128Ty), 64);
11773    Value *LowPart128 = Builder.CreateZExt(Ops[0], Int128Ty);
11774    Value *Val = Builder.CreateOr(HighPart128, LowPart128);
11775    Value *Amt = Builder.CreateAnd(Builder.CreateZExt(Ops[2], Int128Ty),
11776                                   llvm::ConstantInt::get(Int128Ty, 0x3f));
11777    Value *Res;
11778    if (BuiltinID == X86::BI__shiftleft128)
11779      Res = Builder.CreateLShr(Builder.CreateShl(Val, Amt), 64);
11780    else
11781      Res = Builder.CreateLShr(Val, Amt);
11782    return Builder.CreateTrunc(Res, Int64Ty);
11783  }
11784  case X86::BI_ReadWriteBarrier:
11785  case X86::BI_ReadBarrier:
11786  case X86::BI_WriteBarrier: {
11787    return Builder.CreateFence(llvm::AtomicOrdering::SequentiallyConsistent,
11788                               llvm::SyncScope::SingleThread);
11789  }
11790  case X86::BI_BitScanForward:
11791  case X86::BI_BitScanForward64:
11792    return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanForward, E);
11793  case X86::BI_BitScanReverse:
11794  case X86::BI_BitScanReverse64:
11795    return EmitMSVCBuiltinExpr(MSVCIntrin::_BitScanReverse, E);
11796
11797  case X86::BI_InterlockedAnd64:
11798    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedAnd, E);
11799  case X86::BI_InterlockedExchange64:
11800    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchange, E);
11801  case X86::BI_InterlockedExchangeAdd64:
11802    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeAdd, E);
11803  case X86::BI_InterlockedExchangeSub64:
11804    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedExchangeSub, E);
11805  case X86::BI_InterlockedOr64:
11806    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E);
11807  case X86::BI_InterlockedXor64:
11808    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E);
11809  case X86::BI_InterlockedDecrement64:
11810    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedDecrement, E);
11811  case X86::BI_InterlockedIncrement64:
11812    return EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedIncrement, E);
11813  case X86::BI_InterlockedCompareExchange128: {
11814    // InterlockedCompareExchange128 doesn't directly refer to 128bit ints,
11815    // instead it takes pointers to 64bit ints for Destination and
11816    // ComparandResult, and exchange is taken as two 64bit ints (high & low).
11817    // The previous value is written to ComparandResult, and success is
11818    // returned.
11819
11820    llvm::Type *Int128Ty = Builder.getInt128Ty();
11821    llvm::Type *Int128PtrTy = Int128Ty->getPointerTo();
11822
11823    Value *Destination =
11824        Builder.CreateBitCast(Ops[0], Int128PtrTy);
11825    Value *ExchangeHigh128 = Builder.CreateZExt(Ops[1], Int128Ty);
11826    Value *ExchangeLow128 = Builder.CreateZExt(Ops[2], Int128Ty);
11827    Address ComparandResult(Builder.CreateBitCast(Ops[3], Int128PtrTy),
11828                            getContext().toCharUnitsFromBits(128));
11829
11830    Value *Exchange = Builder.CreateOr(
11831        Builder.CreateShl(ExchangeHigh128, 64, "", false, false),
11832        ExchangeLow128);
11833
11834    Value *Comparand = Builder.CreateLoad(ComparandResult);
11835
11836    AtomicCmpXchgInst *CXI =
11837        Builder.CreateAtomicCmpXchg(Destination, Comparand, Exchange,
11838                                    AtomicOrdering::SequentiallyConsistent,
11839                                    AtomicOrdering::SequentiallyConsistent);
11840    CXI->setVolatile(true);
11841
11842    // Write the result back to the inout pointer.
11843    Builder.CreateStore(Builder.CreateExtractValue(CXI, 0), ComparandResult);
11844
11845    // Get the success boolean and zero extend it to i8.
11846    Value *Success = Builder.CreateExtractValue(CXI, 1);
11847    return Builder.CreateZExt(Success, ConvertType(E->getType()));
11848  }
11849
11850  case X86::BI_AddressOfReturnAddress: {
11851    Function *F = CGM.getIntrinsic(Intrinsic::addressofreturnaddress);
11852    return Builder.CreateCall(F);
11853  }
11854  case X86::BI__stosb: {
11855    // We treat __stosb as a volatile memset - it may not generate "rep stosb"
11856    // instruction, but it will create a memset that won't be optimized away.
11857    return Builder.CreateMemSet(Ops[0], Ops[1], Ops[2], 1, true);
11858  }
11859  case X86::BI__ud2:
11860    // llvm.trap makes a ud2a instruction on x86.
11861    return EmitTrapCall(Intrinsic::trap);
11862  case X86::BI__int2c: {
11863    // This syscall signals a driver assertion failure in x86 NT kernels.
11864    llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, false);
11865    llvm::InlineAsm *IA =
11866        llvm::InlineAsm::get(FTy, "int $$0x2c", "", /*SideEffects=*/true);
11867    llvm::AttributeList NoReturnAttr = llvm::AttributeList::get(
11868        getLLVMContext(), llvm::AttributeList::FunctionIndex,
11869        llvm::Attribute::NoReturn);
11870    llvm::CallInst *CI = Builder.CreateCall(IA);
11871    CI->setAttributes(NoReturnAttr);
11872    return CI;
11873  }
11874  case X86::BI__readfsbyte:
11875  case X86::BI__readfsword:
11876  case X86::BI__readfsdword:
11877  case X86::BI__readfsqword: {
11878    llvm::Type *IntTy = ConvertType(E->getType());
11879    Value *Ptr =
11880        Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 257));
11881    LoadInst *Load = Builder.CreateAlignedLoad(
11882        IntTy, Ptr, getContext().getTypeAlignInChars(E->getType()));
11883    Load->setVolatile(true);
11884    return Load;
11885  }
11886  case X86::BI__readgsbyte:
11887  case X86::BI__readgsword:
11888  case X86::BI__readgsdword:
11889  case X86::BI__readgsqword: {
11890    llvm::Type *IntTy = ConvertType(E->getType());
11891    Value *Ptr =
11892        Builder.CreateIntToPtr(Ops[0], llvm::PointerType::get(IntTy, 256));
11893    LoadInst *Load = Builder.CreateAlignedLoad(
11894        IntTy, Ptr, getContext().getTypeAlignInChars(E->getType()));
11895    Load->setVolatile(true);
11896    return Load;
11897  }
11898  case X86::BI__builtin_ia32_paddsb512:
11899  case X86::BI__builtin_ia32_paddsw512:
11900  case X86::BI__builtin_ia32_paddsb256:
11901  case X86::BI__builtin_ia32_paddsw256:
11902  case X86::BI__builtin_ia32_paddsb128:
11903  case X86::BI__builtin_ia32_paddsw128:
11904    return EmitX86AddSubSatExpr(*this, Ops, true, true);
11905  case X86::BI__builtin_ia32_paddusb512:
11906  case X86::BI__builtin_ia32_paddusw512:
11907  case X86::BI__builtin_ia32_paddusb256:
11908  case X86::BI__builtin_ia32_paddusw256:
11909  case X86::BI__builtin_ia32_paddusb128:
11910  case X86::BI__builtin_ia32_paddusw128:
11911    return EmitX86AddSubSatExpr(*this, Ops, false, true);
11912  case X86::BI__builtin_ia32_psubsb512:
11913  case X86::BI__builtin_ia32_psubsw512:
11914  case X86::BI__builtin_ia32_psubsb256:
11915  case X86::BI__builtin_ia32_psubsw256:
11916  case X86::BI__builtin_ia32_psubsb128:
11917  case X86::BI__builtin_ia32_psubsw128:
11918    return EmitX86AddSubSatExpr(*this, Ops, true, false);
11919  case X86::BI__builtin_ia32_psubusb512:
11920  case X86::BI__builtin_ia32_psubusw512:
11921  case X86::BI__builtin_ia32_psubusb256:
11922  case X86::BI__builtin_ia32_psubusw256:
11923  case X86::BI__builtin_ia32_psubusb128:
11924  case X86::BI__builtin_ia32_psubusw128:
11925    return EmitX86AddSubSatExpr(*this, Ops, false, false);
11926  }
11927}
11928
11929Value *CodeGenFunction::EmitPPCBuiltinExpr(unsigned BuiltinID,
11930                                           const CallExpr *E) {
11931  SmallVector<Value*, 4> Ops;
11932
11933  for (unsigned i = 0, e = E->getNumArgs(); i != e; i++)
11934    Ops.push_back(EmitScalarExpr(E->getArg(i)));
11935
11936  Intrinsic::ID ID = Intrinsic::not_intrinsic;
11937
11938  switch (BuiltinID) {
11939  default: return nullptr;
11940
11941  // __builtin_ppc_get_timebase is GCC 4.8+'s PowerPC-specific name for what we
11942  // call __builtin_readcyclecounter.
11943  case PPC::BI__builtin_ppc_get_timebase:
11944    return Builder.CreateCall(CGM.getIntrinsic(Intrinsic::readcyclecounter));
11945
11946  // vec_ld, vec_xl_be, vec_lvsl, vec_lvsr
11947  case PPC::BI__builtin_altivec_lvx:
11948  case PPC::BI__builtin_altivec_lvxl:
11949  case PPC::BI__builtin_altivec_lvebx:
11950  case PPC::BI__builtin_altivec_lvehx:
11951  case PPC::BI__builtin_altivec_lvewx:
11952  case PPC::BI__builtin_altivec_lvsl:
11953  case PPC::BI__builtin_altivec_lvsr:
11954  case PPC::BI__builtin_vsx_lxvd2x:
11955  case PPC::BI__builtin_vsx_lxvw4x:
11956  case PPC::BI__builtin_vsx_lxvd2x_be:
11957  case PPC::BI__builtin_vsx_lxvw4x_be:
11958  case PPC::BI__builtin_vsx_lxvl:
11959  case PPC::BI__builtin_vsx_lxvll:
11960  {
11961    if(BuiltinID == PPC::BI__builtin_vsx_lxvl ||
11962       BuiltinID == PPC::BI__builtin_vsx_lxvll){
11963      Ops[0] = Builder.CreateBitCast(Ops[0], Int8PtrTy);
11964    }else {
11965      Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy);
11966      Ops[0] = Builder.CreateGEP(Ops[1], Ops[0]);
11967      Ops.pop_back();
11968    }
11969
11970    switch (BuiltinID) {
11971    default: llvm_unreachable("Unsupported ld/lvsl/lvsr intrinsic!");
11972    case PPC::BI__builtin_altivec_lvx:
11973      ID = Intrinsic::ppc_altivec_lvx;
11974      break;
11975    case PPC::BI__builtin_altivec_lvxl:
11976      ID = Intrinsic::ppc_altivec_lvxl;
11977      break;
11978    case PPC::BI__builtin_altivec_lvebx:
11979      ID = Intrinsic::ppc_altivec_lvebx;
11980      break;
11981    case PPC::BI__builtin_altivec_lvehx:
11982      ID = Intrinsic::ppc_altivec_lvehx;
11983      break;
11984    case PPC::BI__builtin_altivec_lvewx:
11985      ID = Intrinsic::ppc_altivec_lvewx;
11986      break;
11987    case PPC::BI__builtin_altivec_lvsl:
11988      ID = Intrinsic::ppc_altivec_lvsl;
11989      break;
11990    case PPC::BI__builtin_altivec_lvsr:
11991      ID = Intrinsic::ppc_altivec_lvsr;
11992      break;
11993    case PPC::BI__builtin_vsx_lxvd2x:
11994      ID = Intrinsic::ppc_vsx_lxvd2x;
11995      break;
11996    case PPC::BI__builtin_vsx_lxvw4x:
11997      ID = Intrinsic::ppc_vsx_lxvw4x;
11998      break;
11999    case PPC::BI__builtin_vsx_lxvd2x_be:
12000      ID = Intrinsic::ppc_vsx_lxvd2x_be;
12001      break;
12002    case PPC::BI__builtin_vsx_lxvw4x_be:
12003      ID = Intrinsic::ppc_vsx_lxvw4x_be;
12004      break;
12005    case PPC::BI__builtin_vsx_lxvl:
12006      ID = Intrinsic::ppc_vsx_lxvl;
12007      break;
12008    case PPC::BI__builtin_vsx_lxvll:
12009      ID = Intrinsic::ppc_vsx_lxvll;
12010      break;
12011    }
12012    llvm::Function *F = CGM.getIntrinsic(ID);
12013    return Builder.CreateCall(F, Ops, "");
12014  }
12015
12016  // vec_st, vec_xst_be
12017  case PPC::BI__builtin_altivec_stvx:
12018  case PPC::BI__builtin_altivec_stvxl:
12019  case PPC::BI__builtin_altivec_stvebx:
12020  case PPC::BI__builtin_altivec_stvehx:
12021  case PPC::BI__builtin_altivec_stvewx:
12022  case PPC::BI__builtin_vsx_stxvd2x:
12023  case PPC::BI__builtin_vsx_stxvw4x:
12024  case PPC::BI__builtin_vsx_stxvd2x_be:
12025  case PPC::BI__builtin_vsx_stxvw4x_be:
12026  case PPC::BI__builtin_vsx_stxvl:
12027  case PPC::BI__builtin_vsx_stxvll:
12028  {
12029    if(BuiltinID == PPC::BI__builtin_vsx_stxvl ||
12030      BuiltinID == PPC::BI__builtin_vsx_stxvll ){
12031      Ops[1] = Builder.CreateBitCast(Ops[1], Int8PtrTy);
12032    }else {
12033      Ops[2] = Builder.CreateBitCast(Ops[2], Int8PtrTy);
12034      Ops[1] = Builder.CreateGEP(Ops[2], Ops[1]);
12035      Ops.pop_back();
12036    }
12037
12038    switch (BuiltinID) {
12039    default: llvm_unreachable("Unsupported st intrinsic!");
12040    case PPC::BI__builtin_altivec_stvx:
12041      ID = Intrinsic::ppc_altivec_stvx;
12042      break;
12043    case PPC::BI__builtin_altivec_stvxl:
12044      ID = Intrinsic::ppc_altivec_stvxl;
12045      break;
12046    case PPC::BI__builtin_altivec_stvebx:
12047      ID = Intrinsic::ppc_altivec_stvebx;
12048      break;
12049    case PPC::BI__builtin_altivec_stvehx:
12050      ID = Intrinsic::ppc_altivec_stvehx;
12051      break;
12052    case PPC::BI__builtin_altivec_stvewx:
12053      ID = Intrinsic::ppc_altivec_stvewx;
12054      break;
12055    case PPC::BI__builtin_vsx_stxvd2x:
12056      ID = Intrinsic::ppc_vsx_stxvd2x;
12057      break;
12058    case PPC::BI__builtin_vsx_stxvw4x:
12059      ID = Intrinsic::ppc_vsx_stxvw4x;
12060      break;
12061    case PPC::BI__builtin_vsx_stxvd2x_be:
12062      ID = Intrinsic::ppc_vsx_stxvd2x_be;
12063      break;
12064    case PPC::BI__builtin_vsx_stxvw4x_be:
12065      ID = Intrinsic::ppc_vsx_stxvw4x_be;
12066      break;
12067    case PPC::BI__builtin_vsx_stxvl:
12068      ID = Intrinsic::ppc_vsx_stxvl;
12069      break;
12070    case PPC::BI__builtin_vsx_stxvll:
12071      ID = Intrinsic::ppc_vsx_stxvll;
12072      break;
12073    }
12074    llvm::Function *F = CGM.getIntrinsic(ID);
12075    return Builder.CreateCall(F, Ops, "");
12076  }
12077  // Square root
12078  case PPC::BI__builtin_vsx_xvsqrtsp:
12079  case PPC::BI__builtin_vsx_xvsqrtdp: {
12080    llvm::Type *ResultType = ConvertType(E->getType());
12081    Value *X = EmitScalarExpr(E->getArg(0));
12082    ID = Intrinsic::sqrt;
12083    llvm::Function *F = CGM.getIntrinsic(ID, ResultType);
12084    return Builder.CreateCall(F, X);
12085  }
12086  // Count leading zeros
12087  case PPC::BI__builtin_altivec_vclzb:
12088  case PPC::BI__builtin_altivec_vclzh:
12089  case PPC::BI__builtin_altivec_vclzw:
12090  case PPC::BI__builtin_altivec_vclzd: {
12091    llvm::Type *ResultType = ConvertType(E->getType());
12092    Value *X = EmitScalarExpr(E->getArg(0));
12093    Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
12094    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType);
12095    return Builder.CreateCall(F, {X, Undef});
12096  }
12097  case PPC::BI__builtin_altivec_vctzb:
12098  case PPC::BI__builtin_altivec_vctzh:
12099  case PPC::BI__builtin_altivec_vctzw:
12100  case PPC::BI__builtin_altivec_vctzd: {
12101    llvm::Type *ResultType = ConvertType(E->getType());
12102    Value *X = EmitScalarExpr(E->getArg(0));
12103    Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
12104    Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType);
12105    return Builder.CreateCall(F, {X, Undef});
12106  }
12107  case PPC::BI__builtin_altivec_vpopcntb:
12108  case PPC::BI__builtin_altivec_vpopcnth:
12109  case PPC::BI__builtin_altivec_vpopcntw:
12110  case PPC::BI__builtin_altivec_vpopcntd: {
12111    llvm::Type *ResultType = ConvertType(E->getType());
12112    Value *X = EmitScalarExpr(E->getArg(0));
12113    llvm::Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
12114    return Builder.CreateCall(F, X);
12115  }
12116  // Copy sign
12117  case PPC::BI__builtin_vsx_xvcpsgnsp:
12118  case PPC::BI__builtin_vsx_xvcpsgndp: {
12119    llvm::Type *ResultType = ConvertType(E->getType());
12120    Value *X = EmitScalarExpr(E->getArg(0));
12121    Value *Y = EmitScalarExpr(E->getArg(1));
12122    ID = Intrinsic::copysign;
12123    llvm::Function *F = CGM.getIntrinsic(ID, ResultType);
12124    return Builder.CreateCall(F, {X, Y});
12125  }
12126  // Rounding/truncation
12127  case PPC::BI__builtin_vsx_xvrspip:
12128  case PPC::BI__builtin_vsx_xvrdpip:
12129  case PPC::BI__builtin_vsx_xvrdpim:
12130  case PPC::BI__builtin_vsx_xvrspim:
12131  case PPC::BI__builtin_vsx_xvrdpi:
12132  case PPC::BI__builtin_vsx_xvrspi:
12133  case PPC::BI__builtin_vsx_xvrdpic:
12134  case PPC::BI__builtin_vsx_xvrspic:
12135  case PPC::BI__builtin_vsx_xvrdpiz:
12136  case PPC::BI__builtin_vsx_xvrspiz: {
12137    llvm::Type *ResultType = ConvertType(E->getType());
12138    Value *X = EmitScalarExpr(E->getArg(0));
12139    if (BuiltinID == PPC::BI__builtin_vsx_xvrdpim ||
12140        BuiltinID == PPC::BI__builtin_vsx_xvrspim)
12141      ID = Intrinsic::floor;
12142    else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpi ||
12143             BuiltinID == PPC::BI__builtin_vsx_xvrspi)
12144      ID = Intrinsic::round;
12145    else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpic ||
12146             BuiltinID == PPC::BI__builtin_vsx_xvrspic)
12147      ID = Intrinsic::nearbyint;
12148    else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpip ||
12149             BuiltinID == PPC::BI__builtin_vsx_xvrspip)
12150      ID = Intrinsic::ceil;
12151    else if (BuiltinID == PPC::BI__builtin_vsx_xvrdpiz ||
12152             BuiltinID == PPC::BI__builtin_vsx_xvrspiz)
12153      ID = Intrinsic::trunc;
12154    llvm::Function *F = CGM.getIntrinsic(ID, ResultType);
12155    return Builder.CreateCall(F, X);
12156  }
12157
12158  // Absolute value
12159  case PPC::BI__builtin_vsx_xvabsdp:
12160  case PPC::BI__builtin_vsx_xvabssp: {
12161    llvm::Type *ResultType = ConvertType(E->getType());
12162    Value *X = EmitScalarExpr(E->getArg(0));
12163    llvm::Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
12164    return Builder.CreateCall(F, X);
12165  }
12166
12167  // FMA variations
12168  case PPC::BI__builtin_vsx_xvmaddadp:
12169  case PPC::BI__builtin_vsx_xvmaddasp:
12170  case PPC::BI__builtin_vsx_xvnmaddadp:
12171  case PPC::BI__builtin_vsx_xvnmaddasp:
12172  case PPC::BI__builtin_vsx_xvmsubadp:
12173  case PPC::BI__builtin_vsx_xvmsubasp:
12174  case PPC::BI__builtin_vsx_xvnmsubadp:
12175  case PPC::BI__builtin_vsx_xvnmsubasp: {
12176    llvm::Type *ResultType = ConvertType(E->getType());
12177    Value *X = EmitScalarExpr(E->getArg(0));
12178    Value *Y = EmitScalarExpr(E->getArg(1));
12179    Value *Z = EmitScalarExpr(E->getArg(2));
12180    Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
12181    llvm::Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
12182    switch (BuiltinID) {
12183      case PPC::BI__builtin_vsx_xvmaddadp:
12184      case PPC::BI__builtin_vsx_xvmaddasp:
12185        return Builder.CreateCall(F, {X, Y, Z});
12186      case PPC::BI__builtin_vsx_xvnmaddadp:
12187      case PPC::BI__builtin_vsx_xvnmaddasp:
12188        return Builder.CreateFSub(Zero,
12189                                  Builder.CreateCall(F, {X, Y, Z}), "sub");
12190      case PPC::BI__builtin_vsx_xvmsubadp:
12191      case PPC::BI__builtin_vsx_xvmsubasp:
12192        return Builder.CreateCall(F,
12193                                  {X, Y, Builder.CreateFSub(Zero, Z, "sub")});
12194      case PPC::BI__builtin_vsx_xvnmsubadp:
12195      case PPC::BI__builtin_vsx_xvnmsubasp:
12196        Value *FsubRes =
12197          Builder.CreateCall(F, {X, Y, Builder.CreateFSub(Zero, Z, "sub")});
12198        return Builder.CreateFSub(Zero, FsubRes, "sub");
12199    }
12200    llvm_unreachable("Unknown FMA operation");
12201    return nullptr; // Suppress no-return warning
12202  }
12203
12204  case PPC::BI__builtin_vsx_insertword: {
12205    llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxinsertw);
12206
12207    // Third argument is a compile time constant int. It must be clamped to
12208    // to the range [0, 12].
12209    ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
12210     (0) . __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 12211, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(ArgCI &&
12211 (0) . __assert_fail ("ArgCI && \"Third arg to xxinsertw intrinsic must be constant integer\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 12211, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">           "Third arg to xxinsertw intrinsic must be constant integer");
12212    const int64_t MaxIndex = 12;
12213    int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex);
12214
12215    // The builtin semantics don't exactly match the xxinsertw instructions
12216    // semantics (which ppc_vsx_xxinsertw follows). The builtin extracts the
12217    // word from the first argument, and inserts it in the second argument. The
12218    // instruction extracts the word from its second input register and inserts
12219    // it into its first input register, so swap the first and second arguments.
12220    std::swap(Ops[0], Ops[1]);
12221
12222    // Need to cast the second argument from a vector of unsigned int to a
12223    // vector of long long.
12224    Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int64Ty, 2));
12225
12226    if (getTarget().isLittleEndian()) {
12227      // Create a shuffle mask of (1, 0)
12228      Constant *ShuffleElts[2] = { ConstantInt::get(Int32Ty, 1),
12229                                   ConstantInt::get(Int32Ty, 0)
12230                                 };
12231      Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts);
12232
12233      // Reverse the double words in the vector we will extract from.
12234      Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2));
12235      Ops[0] = Builder.CreateShuffleVector(Ops[0], Ops[0], ShuffleMask);
12236
12237      // Reverse the index.
12238      Index = MaxIndex - Index;
12239    }
12240
12241    // Intrinsic expects the first arg to be a vector of int.
12242    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4));
12243    Ops[2] = ConstantInt::getSigned(Int32Ty, Index);
12244    return Builder.CreateCall(F, Ops);
12245  }
12246
12247  case PPC::BI__builtin_vsx_extractuword: {
12248    llvm::Function *F = CGM.getIntrinsic(Intrinsic::ppc_vsx_xxextractuw);
12249
12250    // Intrinsic expects the first argument to be a vector of doublewords.
12251    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2));
12252
12253    // The second argument is a compile time constant int that needs to
12254    // be clamped to the range [0, 12].
12255    ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[1]);
12256     (0) . __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 12257, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(ArgCI &&
12257 (0) . __assert_fail ("ArgCI && \"Second Arg to xxextractuw intrinsic must be a constant integer!\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 12257, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">           "Second Arg to xxextractuw intrinsic must be a constant integer!");
12258    const int64_t MaxIndex = 12;
12259    int64_t Index = clamp(ArgCI->getSExtValue(), 0, MaxIndex);
12260
12261    if (getTarget().isLittleEndian()) {
12262      // Reverse the index.
12263      Index = MaxIndex - Index;
12264      Ops[1] = ConstantInt::getSigned(Int32Ty, Index);
12265
12266      // Emit the call, then reverse the double words of the results vector.
12267      Value *Call = Builder.CreateCall(F, Ops);
12268
12269      // Create a shuffle mask of (1, 0)
12270      Constant *ShuffleElts[2] = { ConstantInt::get(Int32Ty, 1),
12271                                   ConstantInt::get(Int32Ty, 0)
12272                                 };
12273      Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts);
12274
12275      Value *ShuffleCall = Builder.CreateShuffleVector(Call, Call, ShuffleMask);
12276      return ShuffleCall;
12277    } else {
12278      Ops[1] = ConstantInt::getSigned(Int32Ty, Index);
12279      return Builder.CreateCall(F, Ops);
12280    }
12281  }
12282
12283  case PPC::BI__builtin_vsx_xxpermdi: {
12284    ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
12285     (0) . __assert_fail ("ArgCI && \"Third arg must be constant integer!\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 12285, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(ArgCI && "Third arg must be constant integer!");
12286
12287    unsigned Index = ArgCI->getZExtValue();
12288    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int64Ty, 2));
12289    Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int64Ty, 2));
12290
12291    // Account for endianness by treating this as just a shuffle. So we use the
12292    // same indices for both LE and BE in order to produce expected results in
12293    // both cases.
12294    unsigned ElemIdx0 = (Index & 2) >> 1;
12295    unsigned ElemIdx1 = 2 + (Index & 1);
12296
12297    Constant *ShuffleElts[2] = {ConstantInt::get(Int32Ty, ElemIdx0),
12298                                ConstantInt::get(Int32Ty, ElemIdx1)};
12299    Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts);
12300
12301    Value *ShuffleCall =
12302        Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleMask);
12303    QualType BIRetType = E->getType();
12304    auto RetTy = ConvertType(BIRetType);
12305    return Builder.CreateBitCast(ShuffleCall, RetTy);
12306  }
12307
12308  case PPC::BI__builtin_vsx_xxsldwi: {
12309    ConstantInt *ArgCI = dyn_cast<ConstantInt>(Ops[2]);
12310     (0) . __assert_fail ("ArgCI && \"Third argument must be a compile time constant\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 12310, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(ArgCI && "Third argument must be a compile time constant");
12311    unsigned Index = ArgCI->getZExtValue() & 0x3;
12312    Ops[0] = Builder.CreateBitCast(Ops[0], llvm::VectorType::get(Int32Ty, 4));
12313    Ops[1] = Builder.CreateBitCast(Ops[1], llvm::VectorType::get(Int32Ty, 4));
12314
12315    // Create a shuffle mask
12316    unsigned ElemIdx0;
12317    unsigned ElemIdx1;
12318    unsigned ElemIdx2;
12319    unsigned ElemIdx3;
12320    if (getTarget().isLittleEndian()) {
12321      // Little endian element N comes from element 8+N-Index of the
12322      // concatenated wide vector (of course, using modulo arithmetic on
12323      // the total number of elements).
12324      ElemIdx0 = (8 - Index) % 8;
12325      ElemIdx1 = (9 - Index) % 8;
12326      ElemIdx2 = (10 - Index) % 8;
12327      ElemIdx3 = (11 - Index) % 8;
12328    } else {
12329      // Big endian ElemIdx<N> = Index + N
12330      ElemIdx0 = Index;
12331      ElemIdx1 = Index + 1;
12332      ElemIdx2 = Index + 2;
12333      ElemIdx3 = Index + 3;
12334    }
12335
12336    Constant *ShuffleElts[4] = {ConstantInt::get(Int32Ty, ElemIdx0),
12337                                ConstantInt::get(Int32Ty, ElemIdx1),
12338                                ConstantInt::get(Int32Ty, ElemIdx2),
12339                                ConstantInt::get(Int32Ty, ElemIdx3)};
12340
12341    Constant *ShuffleMask = llvm::ConstantVector::get(ShuffleElts);
12342    Value *ShuffleCall =
12343        Builder.CreateShuffleVector(Ops[0], Ops[1], ShuffleMask);
12344    QualType BIRetType = E->getType();
12345    auto RetTy = ConvertType(BIRetType);
12346    return Builder.CreateBitCast(ShuffleCall, RetTy);
12347  }
12348
12349  case PPC::BI__builtin_pack_vector_int128: {
12350    bool isLittleEndian = getTarget().isLittleEndian();
12351    Value *UndefValue =
12352        llvm::UndefValue::get(llvm::VectorType::get(Ops[0]->getType(), 2));
12353    Value *Res = Builder.CreateInsertElement(
12354        UndefValue, Ops[0], (uint64_t)(isLittleEndian ? 1 : 0));
12355    Res = Builder.CreateInsertElement(Res, Ops[1],
12356                                      (uint64_t)(isLittleEndian ? 0 : 1));
12357    return Builder.CreateBitCast(Res, ConvertType(E->getType()));
12358  }
12359
12360  case PPC::BI__builtin_unpack_vector_int128: {
12361    ConstantInt *Index = cast<ConstantInt>(Ops[1]);
12362    Value *Unpacked = Builder.CreateBitCast(
12363        Ops[0], llvm::VectorType::get(ConvertType(E->getType()), 2));
12364
12365    if (getTarget().isLittleEndian())
12366      Index = ConstantInt::get(Index->getType(), 1 - Index->getZExtValue());
12367
12368    return Builder.CreateExtractElement(Unpacked, Index);
12369  }
12370  }
12371}
12372
12373Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
12374                                              const CallExpr *E) {
12375  switch (BuiltinID) {
12376  case AMDGPU::BI__builtin_amdgcn_div_scale:
12377  case AMDGPU::BI__builtin_amdgcn_div_scalef: {
12378    // Translate from the intrinsics's struct return to the builtin's out
12379    // argument.
12380
12381    Address FlagOutPtr = EmitPointerWithAlignment(E->getArg(3));
12382
12383    llvm::Value *X = EmitScalarExpr(E->getArg(0));
12384    llvm::Value *Y = EmitScalarExpr(E->getArg(1));
12385    llvm::Value *Z = EmitScalarExpr(E->getArg(2));
12386
12387    llvm::Function *Callee = CGM.getIntrinsic(Intrinsic::amdgcn_div_scale,
12388                                           X->getType());
12389
12390    llvm::Value *Tmp = Builder.CreateCall(Callee, {X, Y, Z});
12391
12392    llvm::Value *Result = Builder.CreateExtractValue(Tmp, 0);
12393    llvm::Value *Flag = Builder.CreateExtractValue(Tmp, 1);
12394
12395    llvm::Type *RealFlagType
12396      = FlagOutPtr.getPointer()->getType()->getPointerElementType();
12397
12398    llvm::Value *FlagExt = Builder.CreateZExt(Flag, RealFlagType);
12399    Builder.CreateStore(FlagExt, FlagOutPtr);
12400    return Result;
12401  }
12402  case AMDGPU::BI__builtin_amdgcn_div_fmas:
12403  case AMDGPU::BI__builtin_amdgcn_div_fmasf: {
12404    llvm::Value *Src0 = EmitScalarExpr(E->getArg(0));
12405    llvm::Value *Src1 = EmitScalarExpr(E->getArg(1));
12406    llvm::Value *Src2 = EmitScalarExpr(E->getArg(2));
12407    llvm::Value *Src3 = EmitScalarExpr(E->getArg(3));
12408
12409    llvm::Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_div_fmas,
12410                                      Src0->getType());
12411    llvm::Value *Src3ToBool = Builder.CreateIsNotNull(Src3);
12412    return Builder.CreateCall(F, {Src0, Src1, Src2, Src3ToBool});
12413  }
12414
12415  case AMDGPU::BI__builtin_amdgcn_ds_swizzle:
12416    return emitBinaryBuiltin(*this, E, Intrinsic::amdgcn_ds_swizzle);
12417  case AMDGPU::BI__builtin_amdgcn_mov_dpp:
12418  case AMDGPU::BI__builtin_amdgcn_update_dpp: {
12419    llvm::SmallVector<llvm::Value *, 6> Args;
12420    for (unsigned I = 0; I != E->getNumArgs(); ++I)
12421      Args.push_back(EmitScalarExpr(E->getArg(I)));
12422    assert(Args.size() == 5 || Args.size() == 6);
12423    if (Args.size() == 5)
12424      Args.insert(Args.begin(), llvm::UndefValue::get(Args[0]->getType()));
12425    Function *F =
12426        CGM.getIntrinsic(Intrinsic::amdgcn_update_dpp, Args[0]->getType());
12427    return Builder.CreateCall(F, Args);
12428  }
12429  case AMDGPU::BI__builtin_amdgcn_div_fixup:
12430  case AMDGPU::BI__builtin_amdgcn_div_fixupf:
12431  case AMDGPU::BI__builtin_amdgcn_div_fixuph:
12432    return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_div_fixup);
12433  case AMDGPU::BI__builtin_amdgcn_trig_preop:
12434  case AMDGPU::BI__builtin_amdgcn_trig_preopf:
12435    return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_trig_preop);
12436  case AMDGPU::BI__builtin_amdgcn_rcp:
12437  case AMDGPU::BI__builtin_amdgcn_rcpf:
12438  case AMDGPU::BI__builtin_amdgcn_rcph:
12439    return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rcp);
12440  case AMDGPU::BI__builtin_amdgcn_rsq:
12441  case AMDGPU::BI__builtin_amdgcn_rsqf:
12442  case AMDGPU::BI__builtin_amdgcn_rsqh:
12443    return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq);
12444  case AMDGPU::BI__builtin_amdgcn_rsq_clamp:
12445  case AMDGPU::BI__builtin_amdgcn_rsq_clampf:
12446    return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_rsq_clamp);
12447  case AMDGPU::BI__builtin_amdgcn_sinf:
12448  case AMDGPU::BI__builtin_amdgcn_sinh:
12449    return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_sin);
12450  case AMDGPU::BI__builtin_amdgcn_cosf:
12451  case AMDGPU::BI__builtin_amdgcn_cosh:
12452    return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_cos);
12453  case AMDGPU::BI__builtin_amdgcn_log_clampf:
12454    return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_log_clamp);
12455  case AMDGPU::BI__builtin_amdgcn_ldexp:
12456  case AMDGPU::BI__builtin_amdgcn_ldexpf:
12457  case AMDGPU::BI__builtin_amdgcn_ldexph:
12458    return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_ldexp);
12459  case AMDGPU::BI__builtin_amdgcn_frexp_mant:
12460  case AMDGPU::BI__builtin_amdgcn_frexp_mantf:
12461  case AMDGPU::BI__builtin_amdgcn_frexp_manth:
12462    return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_frexp_mant);
12463  case AMDGPU::BI__builtin_amdgcn_frexp_exp:
12464  case AMDGPU::BI__builtin_amdgcn_frexp_expf: {
12465    Value *Src0 = EmitScalarExpr(E->getArg(0));
12466    Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp,
12467                                { Builder.getInt32Ty(), Src0->getType() });
12468    return Builder.CreateCall(F, Src0);
12469  }
12470  case AMDGPU::BI__builtin_amdgcn_frexp_exph: {
12471    Value *Src0 = EmitScalarExpr(E->getArg(0));
12472    Function *F = CGM.getIntrinsic(Intrinsic::amdgcn_frexp_exp,
12473                                { Builder.getInt16Ty(), Src0->getType() });
12474    return Builder.CreateCall(F, Src0);
12475  }
12476  case AMDGPU::BI__builtin_amdgcn_fract:
12477  case AMDGPU::BI__builtin_amdgcn_fractf:
12478  case AMDGPU::BI__builtin_amdgcn_fracth:
12479    return emitUnaryBuiltin(*this, E, Intrinsic::amdgcn_fract);
12480  case AMDGPU::BI__builtin_amdgcn_lerp:
12481    return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_lerp);
12482  case AMDGPU::BI__builtin_amdgcn_uicmp:
12483  case AMDGPU::BI__builtin_amdgcn_uicmpl:
12484  case AMDGPU::BI__builtin_amdgcn_sicmp:
12485  case AMDGPU::BI__builtin_amdgcn_sicmpl:
12486    return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_icmp);
12487  case AMDGPU::BI__builtin_amdgcn_fcmp:
12488  case AMDGPU::BI__builtin_amdgcn_fcmpf:
12489    return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_fcmp);
12490  case AMDGPU::BI__builtin_amdgcn_class:
12491  case AMDGPU::BI__builtin_amdgcn_classf:
12492  case AMDGPU::BI__builtin_amdgcn_classh:
12493    return emitFPIntBuiltin(*this, E, Intrinsic::amdgcn_class);
12494  case AMDGPU::BI__builtin_amdgcn_fmed3f:
12495  case AMDGPU::BI__builtin_amdgcn_fmed3h:
12496    return emitTernaryBuiltin(*this, E, Intrinsic::amdgcn_fmed3);
12497  case AMDGPU::BI__builtin_amdgcn_ds_append:
12498  case AMDGPU::BI__builtin_amdgcn_ds_consume: {
12499    Intrinsic::ID Intrin = BuiltinID == AMDGPU::BI__builtin_amdgcn_ds_append ?
12500      Intrinsic::amdgcn_ds_append : Intrinsic::amdgcn_ds_consume;
12501    Value *Src0 = EmitScalarExpr(E->getArg(0));
12502    Function *F = CGM.getIntrinsic(Intrin, { Src0->getType() });
12503    return Builder.CreateCall(F, { Src0, Builder.getFalse() });
12504  }
12505  case AMDGPU::BI__builtin_amdgcn_read_exec: {
12506    CallInst *CI = cast<CallInst>(
12507      EmitSpecialRegisterBuiltin(*this, E, Int64Ty, Int64Ty, true, "exec"));
12508    CI->setConvergent();
12509    return CI;
12510  }
12511  case AMDGPU::BI__builtin_amdgcn_read_exec_lo:
12512  case AMDGPU::BI__builtin_amdgcn_read_exec_hi: {
12513    StringRef RegName = BuiltinID == AMDGPU::BI__builtin_amdgcn_read_exec_lo ?
12514      "exec_lo" : "exec_hi";
12515    CallInst *CI = cast<CallInst>(
12516      EmitSpecialRegisterBuiltin(*this, E, Int32Ty, Int32Ty, true, RegName));
12517    CI->setConvergent();
12518    return CI;
12519  }
12520  // amdgcn workitem
12521  case AMDGPU::BI__builtin_amdgcn_workitem_id_x:
12522    return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_x, 0, 1024);
12523  case AMDGPU::BI__builtin_amdgcn_workitem_id_y:
12524    return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_y, 0, 1024);
12525  case AMDGPU::BI__builtin_amdgcn_workitem_id_z:
12526    return emitRangedBuiltin(*this, Intrinsic::amdgcn_workitem_id_z, 0, 1024);
12527
12528  // r600 intrinsics
12529  case AMDGPU::BI__builtin_r600_recipsqrt_ieee:
12530  case AMDGPU::BI__builtin_r600_recipsqrt_ieeef:
12531    return emitUnaryBuiltin(*this, E, Intrinsic::r600_recipsqrt_ieee);
12532  case AMDGPU::BI__builtin_r600_read_tidig_x:
12533    return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_x, 0, 1024);
12534  case AMDGPU::BI__builtin_r600_read_tidig_y:
12535    return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_y, 0, 1024);
12536  case AMDGPU::BI__builtin_r600_read_tidig_z:
12537    return emitRangedBuiltin(*this, Intrinsic::r600_read_tidig_z, 0, 1024);
12538  default:
12539    return nullptr;
12540  }
12541}
12542
12543/// Handle a SystemZ function in which the final argument is a pointer
12544/// to an int that receives the post-instruction CC value.  At the LLVM level
12545/// this is represented as a function that returns a {result, cc} pair.
12546static Value *EmitSystemZIntrinsicWithCC(CodeGenFunction &CGF,
12547                                         unsigned IntrinsicID,
12548                                         const CallExpr *E) {
12549  unsigned NumArgs = E->getNumArgs() - 1;
12550  SmallVector<Value *, 8> Args(NumArgs);
12551  for (unsigned I = 0; I < NumArgs; ++I)
12552    Args[I] = CGF.EmitScalarExpr(E->getArg(I));
12553  Address CCPtr = CGF.EmitPointerWithAlignment(E->getArg(NumArgs));
12554  Function *F = CGF.CGM.getIntrinsic(IntrinsicID);
12555  Value *Call = CGF.Builder.CreateCall(F, Args);
12556  Value *CC = CGF.Builder.CreateExtractValue(Call, 1);
12557  CGF.Builder.CreateStore(CC, CCPtr);
12558  return CGF.Builder.CreateExtractValue(Call, 0);
12559}
12560
12561Value *CodeGenFunction::EmitSystemZBuiltinExpr(unsigned BuiltinID,
12562                                               const CallExpr *E) {
12563  switch (BuiltinID) {
12564  case SystemZ::BI__builtin_tbegin: {
12565    Value *TDB = EmitScalarExpr(E->getArg(0));
12566    Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c);
12567    Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin);
12568    return Builder.CreateCall(F, {TDB, Control});
12569  }
12570  case SystemZ::BI__builtin_tbegin_nofloat: {
12571    Value *TDB = EmitScalarExpr(E->getArg(0));
12572    Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c);
12573    Function *F = CGM.getIntrinsic(Intrinsic::s390_tbegin_nofloat);
12574    return Builder.CreateCall(F, {TDB, Control});
12575  }
12576  case SystemZ::BI__builtin_tbeginc: {
12577    Value *TDB = llvm::ConstantPointerNull::get(Int8PtrTy);
12578    Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff08);
12579    Function *F = CGM.getIntrinsic(Intrinsic::s390_tbeginc);
12580    return Builder.CreateCall(F, {TDB, Control});
12581  }
12582  case SystemZ::BI__builtin_tabort: {
12583    Value *Data = EmitScalarExpr(E->getArg(0));
12584    Function *F = CGM.getIntrinsic(Intrinsic::s390_tabort);
12585    return Builder.CreateCall(F, Builder.CreateSExt(Data, Int64Ty, "tabort"));
12586  }
12587  case SystemZ::BI__builtin_non_tx_store: {
12588    Value *Address = EmitScalarExpr(E->getArg(0));
12589    Value *Data = EmitScalarExpr(E->getArg(1));
12590    Function *F = CGM.getIntrinsic(Intrinsic::s390_ntstg);
12591    return Builder.CreateCall(F, {Data, Address});
12592  }
12593
12594  // Vector builtins.  Note that most vector builtins are mapped automatically
12595  // to target-specific LLVM intrinsics.  The ones handled specially here can
12596  // be represented via standard LLVM IR, which is preferable to enable common
12597  // LLVM optimizations.
12598
12599  case SystemZ::BI__builtin_s390_vpopctb:
12600  case SystemZ::BI__builtin_s390_vpopcth:
12601  case SystemZ::BI__builtin_s390_vpopctf:
12602  case SystemZ::BI__builtin_s390_vpopctg: {
12603    llvm::Type *ResultType = ConvertType(E->getType());
12604    Value *X = EmitScalarExpr(E->getArg(0));
12605    Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
12606    return Builder.CreateCall(F, X);
12607  }
12608
12609  case SystemZ::BI__builtin_s390_vclzb:
12610  case SystemZ::BI__builtin_s390_vclzh:
12611  case SystemZ::BI__builtin_s390_vclzf:
12612  case SystemZ::BI__builtin_s390_vclzg: {
12613    llvm::Type *ResultType = ConvertType(E->getType());
12614    Value *X = EmitScalarExpr(E->getArg(0));
12615    Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
12616    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType);
12617    return Builder.CreateCall(F, {X, Undef});
12618  }
12619
12620  case SystemZ::BI__builtin_s390_vctzb:
12621  case SystemZ::BI__builtin_s390_vctzh:
12622  case SystemZ::BI__builtin_s390_vctzf:
12623  case SystemZ::BI__builtin_s390_vctzg: {
12624    llvm::Type *ResultType = ConvertType(E->getType());
12625    Value *X = EmitScalarExpr(E->getArg(0));
12626    Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
12627    Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType);
12628    return Builder.CreateCall(F, {X, Undef});
12629  }
12630
12631  case SystemZ::BI__builtin_s390_vfsqsb:
12632  case SystemZ::BI__builtin_s390_vfsqdb: {
12633    llvm::Type *ResultType = ConvertType(E->getType());
12634    Value *X = EmitScalarExpr(E->getArg(0));
12635    Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType);
12636    return Builder.CreateCall(F, X);
12637  }
12638  case SystemZ::BI__builtin_s390_vfmasb:
12639  case SystemZ::BI__builtin_s390_vfmadb: {
12640    llvm::Type *ResultType = ConvertType(E->getType());
12641    Value *X = EmitScalarExpr(E->getArg(0));
12642    Value *Y = EmitScalarExpr(E->getArg(1));
12643    Value *Z = EmitScalarExpr(E->getArg(2));
12644    Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
12645    return Builder.CreateCall(F, {X, Y, Z});
12646  }
12647  case SystemZ::BI__builtin_s390_vfmssb:
12648  case SystemZ::BI__builtin_s390_vfmsdb: {
12649    llvm::Type *ResultType = ConvertType(E->getType());
12650    Value *X = EmitScalarExpr(E->getArg(0));
12651    Value *Y = EmitScalarExpr(E->getArg(1));
12652    Value *Z = EmitScalarExpr(E->getArg(2));
12653    Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
12654    Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
12655    return Builder.CreateCall(F, {X, Y, Builder.CreateFSub(Zero, Z, "sub")});
12656  }
12657  case SystemZ::BI__builtin_s390_vfnmasb:
12658  case SystemZ::BI__builtin_s390_vfnmadb: {
12659    llvm::Type *ResultType = ConvertType(E->getType());
12660    Value *X = EmitScalarExpr(E->getArg(0));
12661    Value *Y = EmitScalarExpr(E->getArg(1));
12662    Value *Z = EmitScalarExpr(E->getArg(2));
12663    Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
12664    Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
12665    return Builder.CreateFSub(Zero, Builder.CreateCall(F, {X, Y, Z}), "sub");
12666  }
12667  case SystemZ::BI__builtin_s390_vfnmssb:
12668  case SystemZ::BI__builtin_s390_vfnmsdb: {
12669    llvm::Type *ResultType = ConvertType(E->getType());
12670    Value *X = EmitScalarExpr(E->getArg(0));
12671    Value *Y = EmitScalarExpr(E->getArg(1));
12672    Value *Z = EmitScalarExpr(E->getArg(2));
12673    Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
12674    Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
12675    Value *NegZ = Builder.CreateFSub(Zero, Z, "sub");
12676    return Builder.CreateFSub(Zero, Builder.CreateCall(F, {X, Y, NegZ}));
12677  }
12678  case SystemZ::BI__builtin_s390_vflpsb:
12679  case SystemZ::BI__builtin_s390_vflpdb: {
12680    llvm::Type *ResultType = ConvertType(E->getType());
12681    Value *X = EmitScalarExpr(E->getArg(0));
12682    Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
12683    return Builder.CreateCall(F, X);
12684  }
12685  case SystemZ::BI__builtin_s390_vflnsb:
12686  case SystemZ::BI__builtin_s390_vflndb: {
12687    llvm::Type *ResultType = ConvertType(E->getType());
12688    Value *X = EmitScalarExpr(E->getArg(0));
12689    Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
12690    Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
12691    return Builder.CreateFSub(Zero, Builder.CreateCall(F, X), "sub");
12692  }
12693  case SystemZ::BI__builtin_s390_vfisb:
12694  case SystemZ::BI__builtin_s390_vfidb: {
12695    llvm::Type *ResultType = ConvertType(E->getType());
12696    Value *X = EmitScalarExpr(E->getArg(0));
12697    // Constant-fold the M4 and M5 mask arguments.
12698    llvm::APSInt M4, M5;
12699    bool IsConstM4 = E->getArg(1)->isIntegerConstantExpr(M4, getContext());
12700    bool IsConstM5 = E->getArg(2)->isIntegerConstantExpr(M5, getContext());
12701     (0) . __assert_fail ("IsConstM4 && IsConstM5 && \"Constant arg isn't actually constant?\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 12701, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(IsConstM4 && IsConstM5 && "Constant arg isn't actually constant?");
12702    (void)IsConstM4; (void)IsConstM5;
12703    // Check whether this instance can be represented via a LLVM standard
12704    // intrinsic.  We only support some combinations of M4 and M5.
12705    Intrinsic::ID ID = Intrinsic::not_intrinsic;
12706    switch (M4.getZExtValue()) {
12707    default: break;
12708    case 0:  // IEEE-inexact exception allowed
12709      switch (M5.getZExtValue()) {
12710      default: break;
12711      case 0: ID = Intrinsic::rint; break;
12712      }
12713      break;
12714    case 4:  // IEEE-inexact exception suppressed
12715      switch (M5.getZExtValue()) {
12716      default: break;
12717      case 0: ID = Intrinsic::nearbyint; break;
12718      case 1: ID = Intrinsic::round; break;
12719      case 5: ID = Intrinsic::trunc; break;
12720      case 6: ID = Intrinsic::ceil; break;
12721      case 7: ID = Intrinsic::floor; break;
12722      }
12723      break;
12724    }
12725    if (ID != Intrinsic::not_intrinsic) {
12726      Function *F = CGM.getIntrinsic(ID, ResultType);
12727      return Builder.CreateCall(F, X);
12728    }
12729    switch (BuiltinID) {
12730      case SystemZ::BI__builtin_s390_vfisb: ID = Intrinsic::s390_vfisb; break;
12731      case SystemZ::BI__builtin_s390_vfidb: ID = Intrinsic::s390_vfidb; break;
12732      default: llvm_unreachable("Unknown BuiltinID");
12733    }
12734    Function *F = CGM.getIntrinsic(ID);
12735    Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
12736    Value *M5Value = llvm::ConstantInt::get(getLLVMContext(), M5);
12737    return Builder.CreateCall(F, {X, M4Value, M5Value});
12738  }
12739  case SystemZ::BI__builtin_s390_vfmaxsb:
12740  case SystemZ::BI__builtin_s390_vfmaxdb: {
12741    llvm::Type *ResultType = ConvertType(E->getType());
12742    Value *X = EmitScalarExpr(E->getArg(0));
12743    Value *Y = EmitScalarExpr(E->getArg(1));
12744    // Constant-fold the M4 mask argument.
12745    llvm::APSInt M4;
12746    bool IsConstM4 = E->getArg(2)->isIntegerConstantExpr(M4, getContext());
12747     (0) . __assert_fail ("IsConstM4 && \"Constant arg isn't actually constant?\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 12747, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(IsConstM4 && "Constant arg isn't actually constant?");
12748    (void)IsConstM4;
12749    // Check whether this instance can be represented via a LLVM standard
12750    // intrinsic.  We only support some values of M4.
12751    Intrinsic::ID ID = Intrinsic::not_intrinsic;
12752    switch (M4.getZExtValue()) {
12753    default: break;
12754    case 4: ID = Intrinsic::maxnum; break;
12755    }
12756    if (ID != Intrinsic::not_intrinsic) {
12757      Function *F = CGM.getIntrinsic(ID, ResultType);
12758      return Builder.CreateCall(F, {X, Y});
12759    }
12760    switch (BuiltinID) {
12761      case SystemZ::BI__builtin_s390_vfmaxsb: ID = Intrinsic::s390_vfmaxsb; break;
12762      case SystemZ::BI__builtin_s390_vfmaxdb: ID = Intrinsic::s390_vfmaxdb; break;
12763      default: llvm_unreachable("Unknown BuiltinID");
12764    }
12765    Function *F = CGM.getIntrinsic(ID);
12766    Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
12767    return Builder.CreateCall(F, {X, Y, M4Value});
12768  }
12769  case SystemZ::BI__builtin_s390_vfminsb:
12770  case SystemZ::BI__builtin_s390_vfmindb: {
12771    llvm::Type *ResultType = ConvertType(E->getType());
12772    Value *X = EmitScalarExpr(E->getArg(0));
12773    Value *Y = EmitScalarExpr(E->getArg(1));
12774    // Constant-fold the M4 mask argument.
12775    llvm::APSInt M4;
12776    bool IsConstM4 = E->getArg(2)->isIntegerConstantExpr(M4, getContext());
12777     (0) . __assert_fail ("IsConstM4 && \"Constant arg isn't actually constant?\"", "/home/seafit/code_projects/clang_source/clang/lib/CodeGen/CGBuiltin.cpp", 12777, __PRETTY_FUNCTION__))" file_link="../../../include/assert.h.html#88" macro="true">assert(IsConstM4 && "Constant arg isn't actually constant?");
12778    (void)IsConstM4;
12779    // Check whether this instance can be represented via a LLVM standard
12780    // intrinsic.  We only support some values of M4.
12781    Intrinsic::ID ID = Intrinsic::not_intrinsic;
12782    switch (M4.getZExtValue()) {
12783    default: break;
12784    case 4: ID = Intrinsic::minnum; break;
12785    }
12786    if (ID != Intrinsic::not_intrinsic) {
12787      Function *F = CGM.getIntrinsic(ID, ResultType);
12788      return Builder.CreateCall(F, {X, Y});
12789    }
12790    switch (BuiltinID) {
12791      case SystemZ::BI__builtin_s390_vfminsb: ID = Intrinsic::s390_vfminsb; break;
12792      case SystemZ::BI__builtin_s390_vfmindb: ID = Intrinsic::s390_vfmindb; break;
12793      default: llvm_unreachable("Unknown BuiltinID");
12794    }
12795    Function *F = CGM.getIntrinsic(ID);
12796    Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
12797    return Builder.CreateCall(F, {X, Y, M4Value});
12798  }
12799
12800  // Vector intrinsics that output the post-instruction CC value.
12801
12802#define INTRINSIC_WITH_CC(NAME) \
12803    case SystemZ::BI__builtin_##NAME: \
12804      return EmitSystemZIntrinsicWithCC(*this, Intrinsic::NAME, E)
12805
12806  INTRINSIC_WITH_CC(s390_vpkshs);
12807  INTRINSIC_WITH_CC(s390_vpksfs);
12808  INTRINSIC_WITH_CC(s390_vpksgs);
12809
12810  INTRINSIC_WITH_CC(s390_vpklshs);
12811  INTRINSIC_WITH_CC(s390_vpklsfs);
12812  INTRINSIC_WITH_CC(s390_vpklsgs);
12813
12814  INTRINSIC_WITH_CC(s390_vceqbs);
12815  INTRINSIC_WITH_CC(s390_vceqhs);
12816  INTRINSIC_WITH_CC(s390_vceqfs);
12817  INTRINSIC_WITH_CC(s390_vceqgs);
12818
12819  INTRINSIC_WITH_CC(s390_vchbs);
12820  INTRINSIC_WITH_CC(s390_vchhs);
12821  INTRINSIC_WITH_CC(s390_vchfs);
12822  INTRINSIC_WITH_CC(s390_vchgs);
12823
12824  INTRINSIC_WITH_CC(s390_vchlbs);
12825  INTRINSIC_WITH_CC(s390_vchlhs);
12826  INTRINSIC_WITH_CC(s390_vchlfs);
12827  INTRINSIC_WITH_CC(s390_vchlgs);
12828
12829  INTRINSIC_WITH_CC(s390_vfaebs);
12830  INTRINSIC_WITH_CC(s390_vfaehs);
12831  INTRINSIC_WITH_CC(s390_vfaefs);
12832
12833  INTRINSIC_WITH_CC(s390_vfaezbs);
12834  INTRINSIC_WITH_CC(s390_vfaezhs);
12835  INTRINSIC_WITH_CC(s390_vfaezfs);
12836
12837  INTRINSIC_WITH_CC(s390_vfeebs);
12838  INTRINSIC_WITH_CC(s390_vfeehs);
12839  INTRINSIC_WITH_CC(s390_vfeefs);
12840
12841  INTRINSIC_WITH_CC(s390_vfeezbs);
12842  INTRINSIC_WITH_CC(s390_vfeezhs);
12843  INTRINSIC_WITH_CC(s390_vfeezfs);
12844
12845  INTRINSIC_WITH_CC(s390_vfenebs);
12846  INTRINSIC_WITH_CC(s390_vfenehs);
12847  INTRINSIC_WITH_CC(s390_vfenefs);
12848
12849  INTRINSIC_WITH_CC(s390_vfenezbs);
12850  INTRINSIC_WITH_CC(s390_vfenezhs);
12851  INTRINSIC_WITH_CC(s390_vfenezfs);
12852
12853  INTRINSIC_WITH_CC(s390_vistrbs);
12854  INTRINSIC_WITH_CC(s390_vistrhs);
12855  INTRINSIC_WITH_CC(s390_vistrfs);
12856
12857  INTRINSIC_WITH_CC(s390_vstrcbs);
12858  INTRINSIC_WITH_CC(s390_vstrchs);
12859  INTRINSIC_WITH_CC(s390_vstrcfs);
12860
12861  INTRINSIC_WITH_CC(s390_vstrczbs);
12862  INTRINSIC_WITH_CC(s390_vstrczhs);
12863  INTRINSIC_WITH_CC(s390_vstrczfs);
12864
12865  INTRINSIC_WITH_CC(s390_vfcesbs);
12866  INTRINSIC_WITH_CC(s390_vfcedbs);
12867  INTRINSIC_WITH_CC(s390_vfchsbs);
12868  INTRINSIC_WITH_CC(s390_vfchdbs);
12869  INTRINSIC_WITH_CC(s390_vfchesbs);
12870  INTRINSIC_WITH_CC(s390_vfchedbs);
12871
12872  INTRINSIC_WITH_CC(s390_vftcisb);
12873  INTRINSIC_WITH_CC(s390_vftcidb);
12874
12875#undef INTRINSIC_WITH_CC
12876
12877  default:
12878    return nullptr;
12879  }
12880}
12881
12882Value *CodeGenFunction::EmitNVPTXBuiltinExpr(unsigned BuiltinID,
12883                                             const CallExpr *E) {
12884  auto MakeLdg = [&](unsigned IntrinsicID) {
12885    Value *Ptr = EmitScalarExpr(E->getArg(0));
12886    clang::CharUnits Align =
12887        getNaturalPointeeTypeAlignment(E->getArg(0)->getType());
12888    return Builder.CreateCall(
12889        CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(),
12890                                       Ptr->getType()}),
12891        {Ptr, ConstantInt::get(Builder.getInt32Ty(), Align.getQuantity())});
12892  };
12893  auto MakeScopedAtomic = [&](unsigned IntrinsicID) {
12894    Value *Ptr = EmitScalarExpr(E->getArg(0));
12895    return Builder.CreateCall(
12896        CGM.getIntrinsic(IntrinsicID, {Ptr->getType()->getPointerElementType(),
12897                                       Ptr->getType()}),
12898        {Ptr, EmitScalarExpr(E->getArg(1))});
12899  };
12900  switch (BuiltinID) {
12901  case NVPTX::BI__nvvm_atom_add_gen_i:
12902  case NVPTX::BI__nvvm_atom_add_gen_l:
12903  case NVPTX::BI__nvvm_atom_add_gen_ll:
12904    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Add, E);
12905
12906  case NVPTX::BI__nvvm_atom_sub_gen_i:
12907  case NVPTX::BI__nvvm_atom_sub_gen_l:
12908  case NVPTX::BI__nvvm_atom_sub_gen_ll:
12909    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Sub, E);
12910
12911  case NVPTX::BI__nvvm_atom_and_gen_i:
12912  case NVPTX::BI__nvvm_atom_and_gen_l:
12913  case NVPTX::BI__nvvm_atom_and_gen_ll:
12914    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::And, E);
12915
12916  case NVPTX::BI__nvvm_atom_or_gen_i:
12917  case NVPTX::BI__nvvm_atom_or_gen_l:
12918  case NVPTX::BI__nvvm_atom_or_gen_ll:
12919    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Or, E);
12920
12921  case NVPTX::BI__nvvm_atom_xor_gen_i:
12922  case NVPTX::BI__nvvm_atom_xor_gen_l:
12923  case NVPTX::BI__nvvm_atom_xor_gen_ll:
12924    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xor, E);
12925
12926  case NVPTX::BI__nvvm_atom_xchg_gen_i:
12927  case NVPTX::BI__nvvm_atom_xchg_gen_l:
12928  case NVPTX::BI__nvvm_atom_xchg_gen_ll:
12929    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Xchg, E);
12930
12931  case NVPTX::BI__nvvm_atom_max_gen_i:
12932  case NVPTX::BI__nvvm_atom_max_gen_l:
12933  case NVPTX::BI__nvvm_atom_max_gen_ll:
12934    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Max, E);
12935
12936  case NVPTX::BI__nvvm_atom_max_gen_ui:
12937  case NVPTX::BI__nvvm_atom_max_gen_ul:
12938  case NVPTX::BI__nvvm_atom_max_gen_ull:
12939    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMax, E);
12940
12941  case NVPTX::BI__nvvm_atom_min_gen_i:
12942  case NVPTX::BI__nvvm_atom_min_gen_l:
12943  case NVPTX::BI__nvvm_atom_min_gen_ll:
12944    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::Min, E);
12945
12946  case NVPTX::BI__nvvm_atom_min_gen_ui:
12947  case NVPTX::BI__nvvm_atom_min_gen_ul:
12948  case NVPTX::BI__nvvm_atom_min_gen_ull:
12949    return MakeBinaryAtomicValue(*this, llvm::AtomicRMWInst::UMin, E);
12950
12951  case NVPTX::BI__nvvm_atom_cas_gen_i:
12952  case NVPTX::BI__nvvm_atom_cas_gen_l:
12953  case NVPTX::BI__nvvm_atom_cas_gen_ll:
12954    // __nvvm_atom_cas_gen_* should return the old value rather than the
12955    // success flag.
12956    return MakeAtomicCmpXchgValue(*this, E, /*ReturnBool=*/false);
12957
12958  case NVPTX::BI__nvvm_atom_add_gen_f: {
12959    Value *Ptr = EmitScalarExpr(E->getArg(0));
12960    Value *Val = EmitScalarExpr(E->getArg(1));
12961    // atomicrmw only deals with integer arguments so we need to use
12962    // LLVM's nvvm_atomic_load_add_f32 intrinsic for that.
12963    Function *FnALAF32 =
12964        CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_add_f32, Ptr->getType());
12965    return Builder.CreateCall(FnALAF32, {Ptr, Val});
12966  }
12967
12968  case NVPTX::BI__nvvm_atom_add_gen_d: {
12969    Value *Ptr = EmitScalarExpr(E->getArg(0));
12970    Value *Val = EmitScalarExpr(E->getArg(1));
12971    // atomicrmw only deals with integer arguments, so we need to use
12972    // LLVM's nvvm_atomic_load_add_f64 intrinsic.
12973    Function *FnALAF64 =
12974        CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_add_f64, Ptr->getType());
12975    return Builder.CreateCall(FnALAF64, {Ptr, Val});
12976  }
12977
12978  case NVPTX::BI__nvvm_atom_inc_gen_ui: {
12979    Value *Ptr = EmitScalarExpr(E->getArg(0));
12980    Value *Val = EmitScalarExpr(E->getArg(1));
12981    Function *FnALI32 =
12982        CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_inc_32, Ptr->getType());
12983    return Builder.CreateCall(FnALI32, {Ptr, Val});
12984  }
12985
12986  case NVPTX::BI__nvvm_atom_dec_gen_ui: {
12987    Value *Ptr = EmitScalarExpr(E->getArg(0));
12988    Value *Val = EmitScalarExpr(E->getArg(1));
12989    Function *FnALD32 =
12990        CGM.getIntrinsic(Intrinsic::nvvm_atomic_load_dec_32, Ptr->getType());
12991    return Builder.CreateCall(FnALD32, {Ptr, Val});
12992  }
12993
12994  case NVPTX::BI__nvvm_ldg_c:
12995  case NVPTX::BI__nvvm_ldg_c2:
12996  case NVPTX::BI__nvvm_ldg_c4:
12997  case NVPTX::BI__nvvm_ldg_s:
12998  case NVPTX::BI__nvvm_ldg_s2:
12999  case NVPTX::BI__nvvm_ldg_s4:
13000  case NVPTX::BI__nvvm_ldg_i:
13001  case NVPTX::BI__nvvm_ldg_i2:
13002  case NVPTX::BI__nvvm_ldg_i4:
13003  case NVPTX::BI__nvvm_ldg_l:
13004  case NVPTX::BI__nvvm_ldg_ll:
13005  case NVPTX::BI__nvvm_ldg_ll2:
13006  case NVPTX::BI__nvvm_ldg_uc:
13007  case NVPTX::BI__nvvm_ldg_uc2:
13008  case NVPTX::BI__nvvm_ldg_uc4:
13009  case NVPTX::BI__nvvm_ldg_us:
13010  case NVPTX::BI__nvvm_ldg_us2:
13011  case NVPTX::BI__nvvm_ldg_us4:
13012  case NVPTX::BI__nvvm_ldg_ui:
13013  case NVPTX::BI__nvvm_ldg_ui2:
13014  case NVPTX::BI__nvvm_ldg_ui4:
13015  case NVPTX::BI__nvvm_ldg_ul:
13016  case NVPTX::BI__nvvm_ldg_ull:
13017  case NVPTX::BI__nvvm_ldg_ull2:
13018    // PTX Interoperability section 2.2: "For a vector with an even number of
13019    // elements, its alignment is set to number of elements times the alignment
13020    // of its member: n*alignof(t)."
13021    return MakeLdg(Intrinsic::nvvm_ldg_global_i);
13022  case NVPTX::BI__nvvm_ldg_f:
13023  case NVPTX::BI__nvvm_ldg_f2:
13024  case NVPTX::BI__nvvm_ldg_f4:
13025  case NVPTX::BI__nvvm_ldg_d:
13026  case NVPTX::BI__nvvm_ldg_d2:
13027    return MakeLdg(Intrinsic::nvvm_ldg_global_f);
13028
13029  case NVPTX::BI__nvvm_atom_cta_add_gen_i:
13030  case NVPTX::BI__nvvm_atom_cta_add_gen_l:
13031  case NVPTX::BI__nvvm_atom_cta_add_gen_ll:
13032    return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_cta);
13033  case NVPTX::BI__nvvm_atom_sys_add_gen_i:
13034  case NVPTX::BI__nvvm_atom_sys_add_gen_l:
13035  case NVPTX::BI__nvvm_atom_sys_add_gen_ll:
13036    return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_i_sys);
13037  case NVPTX::BI__nvvm_atom_cta_add_gen_f:
13038  case NVPTX::BI__nvvm_atom_cta_add_gen_d:
13039    return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_cta);
13040  case NVPTX::BI__nvvm_atom_sys_add_gen_f:
13041  case NVPTX::BI__nvvm_atom_sys_add_gen_d:
13042    return MakeScopedAtomic(Intrinsic::nvvm_atomic_add_gen_f_sys);
13043  case NVPTX::BI__nvvm_atom_cta_xchg_gen_i:
13044  case NVPTX::BI__nvvm_atom_cta_xchg_gen_l:
13045  case NVPTX::BI__nvvm_atom_cta_xchg_gen_ll:
13046    return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_cta);
13047  case NVPTX::BI__nvvm_atom_sys_xchg_gen_i:
13048  case NVPTX::BI__nvvm_atom_sys_xchg_gen_l:
13049  case NVPTX::BI__nvvm_atom_sys_xchg_gen_ll:
13050    return MakeScopedAtomic(Intrinsic::nvvm_atomic_exch_gen_i_sys);
13051  case NVPTX::BI__nvvm_atom_cta_max_gen_i:
13052  case NVPTX::BI__nvvm_atom_cta_max_gen_ui:
13053  case NVPTX::BI__nvvm_atom_cta_max_gen_l:
13054  case NVPTX::BI__nvvm_atom_cta_max_gen_ul:
13055  case NVPTX::BI__nvvm_atom_cta_max_gen_ll:
13056  case NVPTX::BI__nvvm_atom_cta_max_gen_ull:
13057    return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_cta);
13058  case NVPTX::BI__nvvm_atom_sys_max_gen_i:
13059  case NVPTX::BI__nvvm_atom_sys_max_gen_ui:
13060  case NVPTX::BI__nvvm_atom_sys_max_gen_l:
13061  case NVPTX::BI__nvvm_atom_sys_max_gen_ul:
13062  case NVPTX::BI__nvvm_atom_sys_max_gen_ll:
13063  case NVPTX::BI__nvvm_atom_sys_max_gen_ull:
13064    return MakeScopedAtomic(Intrinsic::nvvm_atomic_max_gen_i_sys);
13065  case NVPTX::BI__nvvm_atom_cta_min_gen_i:
13066  case NVPTX::BI__nvvm_atom_cta_min_gen_ui:
13067  case NVPTX::BI__nvvm_atom_cta_min_gen_l:
13068  case NVPTX::BI__nvvm_atom_cta_min_gen_ul:
13069  case NVPTX::BI__nvvm_atom_cta_min_gen_ll:
13070  case NVPTX::BI__nvvm_atom_cta_min_gen_ull:
13071    return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_cta);
13072  case NVPTX::BI__nvvm_atom_sys_min_gen_i:
13073  case NVPTX::BI__nvvm_atom_sys_min_gen_ui:
13074  case NVPTX::BI__nvvm_atom_sys_min_gen_l:
13075  case NVPTX::BI__nvvm_atom_sys_min_gen_ul:
13076  case NVPTX::BI__nvvm_atom_sys_min_gen_ll:
13077  case NVPTX::BI__nvvm_atom_sys_min_gen_ull:
13078    return MakeScopedAtomic(Intrinsic::nvvm_atomic_min_gen_i_sys);
13079  case NVPTX::BI__nvvm_atom_cta_inc_gen_ui:
13080    return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_cta);
13081  case NVPTX::BI__nvvm_atom_cta_dec_gen_ui:
13082    return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_cta);
13083  case NVPTX::BI__nvvm_atom_sys_inc_gen_ui:
13084    return MakeScopedAtomic(Intrinsic::nvvm_atomic_inc_gen_i_sys);
13085  case NVPTX::BI__nvvm_atom_sys_dec_gen_ui:
13086    return MakeScopedAtomic(Intrinsic::nvvm_atomic_dec_gen_i_sys);
13087  case NVPTX::BI__nvvm_atom_cta_and_gen_i:
13088  case NVPTX::BI__nvvm_atom_cta_and_gen_l:
13089  case NVPTX::BI__nvvm_atom_cta_and_gen_ll:
13090    return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_cta);
13091  case NVPTX::BI__nvvm_atom_sys_and_gen_i:
13092  case NVPTX::BI__nvvm_atom_sys_and_gen_l:
13093  case NVPTX::BI__nvvm_atom_sys_and_gen_ll:
13094    return MakeScopedAtomic(Intrinsic::nvvm_atomic_and_gen_i_sys);
13095  case NVPTX::BI__nvvm_atom_cta_or_gen_i:
13096  case NVPTX::BI__nvvm_atom_cta_or_gen_l:
13097  case NVPTX::BI__nvvm_atom_cta_or_gen_ll:
13098    return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_cta);
13099  case NVPTX::BI__nvvm_atom_sys_or_gen_i:
13100  case NVPTX::BI__nvvm_atom_sys_or_gen_l:
13101  case NVPTX::BI__nvvm_atom_sys_or_gen_ll:
13102    return MakeScopedAtomic(Intrinsic::nvvm_atomic_or_gen_i_sys);
13103  case NVPTX::BI__nvvm_atom_cta_xor_gen_i:
13104  case NVPTX::BI__nvvm_atom_cta_xor_gen_l:
13105  case NVPTX::BI__nvvm_atom_cta_xor_gen_ll:
13106    return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_cta);
13107  case NVPTX::BI__nvvm_atom_sys_xor_gen_i:
13108  case NVPTX::BI__nvvm_atom_sys_xor_gen_l:
13109  case NVPTX::BI__nvvm_atom_sys_xor_gen_ll:
13110    return MakeScopedAtomic(Intrinsic::nvvm_atomic_xor_gen_i_sys);
13111  case NVPTX::BI__nvvm_atom_cta_cas_gen_i:
13112  case NVPTX::BI__nvvm_atom_cta_cas_gen_l:
13113  case NVPTX::BI__nvvm_atom_cta_cas_gen_ll: {
13114    Value *Ptr = EmitScalarExpr(E->getArg(0));
13115    return Builder.CreateCall(
13116        CGM.getIntrinsic(
13117            Intrinsic::nvvm_atomic_cas_gen_i_cta,
13118            {Ptr->getType()->getPointerElementType(), Ptr->getType()}),
13119        {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))});
13120  }
13121  case NVPTX::BI__nvvm_atom_sys_cas_gen_i:
13122  case NVPTX::BI__nvvm_atom_sys_cas_gen_l:
13123  case NVPTX::BI__nvvm_atom_sys_cas_gen_ll: {
13124    Value *Ptr = EmitScalarExpr(E->getArg(0));
13125    return Builder.CreateCall(
13126        CGM.getIntrinsic(
13127            Intrinsic::nvvm_atomic_cas_gen_i_sys,
13128            {Ptr->getType()->getPointerElementType(), Ptr->getType()}),
13129        {Ptr, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2))});
13130  }
13131  case NVPTX::BI__nvvm_match_all_sync_i32p:
13132  case NVPTX::BI__nvvm_match_all_sync_i64p: {
13133    Value *Mask = EmitScalarExpr(E->getArg(0));
13134    Value *Val = EmitScalarExpr(E->getArg(1));
13135    Address PredOutPtr = EmitPointerWithAlignment(E->getArg(2));
13136    Value *ResultPair = Builder.CreateCall(
13137        CGM.getIntrinsic(BuiltinID == NVPTX::BI__nvvm_match_all_sync_i32p
13138                             ? Intrinsic::nvvm_match_all_sync_i32p
13139                             : Intrinsic::nvvm_match_all_sync_i64p),
13140        {Mask, Val});
13141    Value *Pred = Builder.CreateZExt(Builder.CreateExtractValue(ResultPair, 1),
13142                                     PredOutPtr.getElementType());
13143    Builder.CreateStore(Pred, PredOutPtr);
13144    return Builder.CreateExtractValue(ResultPair, 0);
13145  }
13146  case NVPTX::BI__hmma_m16n16k16_ld_a:
13147  case NVPTX::BI__hmma_m16n16k16_ld_b:
13148  case NVPTX::BI__hmma_m16n16k16_ld_c_f16:
13149  case NVPTX::BI__hmma_m16n16k16_ld_c_f32:
13150  case NVPTX::BI__hmma_m32n8k16_ld_a:
13151  case NVPTX::BI__hmma_m32n8k16_ld_b:
13152  case NVPTX::BI__hmma_m32n8k16_ld_c_f16:
13153  case NVPTX::BI__hmma_m32n8k16_ld_c_f32:
13154  case NVPTX::BI__hmma_m8n32k16_ld_a:
13155  case NVPTX::BI__hmma_m8n32k16_ld_b:
13156  case NVPTX::BI__hmma_m8n32k16_ld_c_f16:
13157  case NVPTX::BI__hmma_m8n32k16_ld_c_f32: {
13158    Address Dst = EmitPointerWithAlignment(E->getArg(0));
13159    Value *Src = EmitScalarExpr(E->getArg(1));
13160    Value *Ldm = EmitScalarExpr(E->getArg(2));
13161    llvm::APSInt isColMajorArg;
13162    if (!E->getArg(3)->isIntegerConstantExpr(isColMajorArg, getContext()))
13163      return nullptr;
13164    bool isColMajor = isColMajorArg.getSExtValue();
13165    unsigned IID;
13166    unsigned NumResults;
13167    switch (BuiltinID) {
13168    case NVPTX::BI__hmma_m16n16k16_ld_a:
13169      IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_load_a_f16_col_stride
13170                       : Intrinsic::nvvm_wmma_m16n16k16_load_a_f16_row_stride;
13171      NumResults = 8;
13172      break;
13173    case NVPTX::BI__hmma_m16n16k16_ld_b:
13174      IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_load_b_f16_col_stride
13175                       : Intrinsic::nvvm_wmma_m16n16k16_load_b_f16_row_stride;
13176      NumResults = 8;
13177      break;
13178    case NVPTX::BI__hmma_m16n16k16_ld_c_f16:
13179      IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_load_c_f16_col_stride
13180                       : Intrinsic::nvvm_wmma_m16n16k16_load_c_f16_row_stride;
13181      NumResults = 4;
13182      break;
13183    case NVPTX::BI__hmma_m16n16k16_ld_c_f32:
13184      IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_load_c_f32_col_stride
13185                       : Intrinsic::nvvm_wmma_m16n16k16_load_c_f32_row_stride;
13186      NumResults = 8;
13187      break;
13188    case NVPTX::BI__hmma_m32n8k16_ld_a:
13189      IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_load_a_f16_col_stride
13190                       : Intrinsic::nvvm_wmma_m32n8k16_load_a_f16_row_stride;
13191      NumResults = 8;
13192      break;
13193    case NVPTX::BI__hmma_m32n8k16_ld_b:
13194      IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_load_b_f16_col_stride
13195                       : Intrinsic::nvvm_wmma_m32n8k16_load_b_f16_row_stride;
13196      NumResults = 8;
13197      break;
13198    case NVPTX::BI__hmma_m32n8k16_ld_c_f16:
13199      IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_load_c_f16_col_stride
13200                       : Intrinsic::nvvm_wmma_m32n8k16_load_c_f16_row_stride;
13201      NumResults = 4;
13202      break;
13203    case NVPTX::BI__hmma_m32n8k16_ld_c_f32:
13204      IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_load_c_f32_col_stride
13205                       : Intrinsic::nvvm_wmma_m32n8k16_load_c_f32_row_stride;
13206      NumResults = 8;
13207      break;
13208    case NVPTX::BI__hmma_m8n32k16_ld_a:
13209      IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_load_a_f16_col_stride
13210                       : Intrinsic::nvvm_wmma_m8n32k16_load_a_f16_row_stride;
13211      NumResults = 8;
13212      break;
13213    case NVPTX::BI__hmma_m8n32k16_ld_b:
13214      IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_load_b_f16_col_stride
13215                       : Intrinsic::nvvm_wmma_m8n32k16_load_b_f16_row_stride;
13216      NumResults = 8;
13217      break;
13218    case NVPTX::BI__hmma_m8n32k16_ld_c_f16:
13219      IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_load_c_f16_col_stride
13220                       : Intrinsic::nvvm_wmma_m8n32k16_load_c_f16_row_stride;
13221      NumResults = 4;
13222      break;
13223    case NVPTX::BI__hmma_m8n32k16_ld_c_f32:
13224      IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_load_c_f32_col_stride
13225                       : Intrinsic::nvvm_wmma_m8n32k16_load_c_f32_row_stride;
13226      NumResults = 8;
13227      break;
13228    default:
13229      llvm_unreachable("Unexpected builtin ID.");
13230    }
13231    Value *Result =
13232        Builder.CreateCall(CGM.getIntrinsic(IID, Src->getType()), {Src, Ldm});
13233
13234    // Save returned values.
13235    for (unsigned i = 0; i < NumResults; ++i) {
13236      Builder.CreateAlignedStore(
13237          Builder.CreateBitCast(Builder.CreateExtractValue(Result, i),
13238                                Dst.getElementType()),
13239          Builder.CreateGEP(Dst.getPointer(), llvm::ConstantInt::get(IntTy, i)),
13240          CharUnits::fromQuantity(4));
13241    }
13242    return Result;
13243  }
13244
13245  case NVPTX::BI__hmma_m16n16k16_st_c_f16:
13246  case NVPTX::BI__hmma_m16n16k16_st_c_f32:
13247  case NVPTX::BI__hmma_m32n8k16_st_c_f16:
13248  case NVPTX::BI__hmma_m32n8k16_st_c_f32:
13249  case NVPTX::BI__hmma_m8n32k16_st_c_f16:
13250  case NVPTX::BI__hmma_m8n32k16_st_c_f32: {
13251    Value *Dst = EmitScalarExpr(E->getArg(0));
13252    Address Src = EmitPointerWithAlignment(E->getArg(1));
13253    Value *Ldm = EmitScalarExpr(E->getArg(2));
13254    llvm::APSInt isColMajorArg;
13255    if (!E->getArg(3)->isIntegerConstantExpr(isColMajorArg, getContext()))
13256      return nullptr;
13257    bool isColMajor = isColMajorArg.getSExtValue();
13258    unsigned IID;
13259    unsigned NumResults = 8;
13260    // PTX Instructions (and LLVM intrinsics) are defined for slice _d_, yet
13261    // for some reason nvcc builtins use _c_.
13262    switch (BuiltinID) {
13263    case NVPTX::BI__hmma_m16n16k16_st_c_f16:
13264      IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_store_d_f16_col_stride
13265                       : Intrinsic::nvvm_wmma_m16n16k16_store_d_f16_row_stride;
13266      NumResults = 4;
13267      break;
13268    case NVPTX::BI__hmma_m16n16k16_st_c_f32:
13269      IID = isColMajor ? Intrinsic::nvvm_wmma_m16n16k16_store_d_f32_col_stride
13270                       : Intrinsic::nvvm_wmma_m16n16k16_store_d_f32_row_stride;
13271      break;
13272    case NVPTX::BI__hmma_m32n8k16_st_c_f16:
13273      IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_store_d_f16_col_stride
13274                       : Intrinsic::nvvm_wmma_m32n8k16_store_d_f16_row_stride;
13275      NumResults = 4;
13276      break;
13277    case NVPTX::BI__hmma_m32n8k16_st_c_f32:
13278      IID = isColMajor ? Intrinsic::nvvm_wmma_m32n8k16_store_d_f32_col_stride
13279                       : Intrinsic::nvvm_wmma_m32n8k16_store_d_f32_row_stride;
13280      break;
13281    case NVPTX::BI__hmma_m8n32k16_st_c_f16:
13282      IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_store_d_f16_col_stride
13283                       : Intrinsic::nvvm_wmma_m8n32k16_store_d_f16_row_stride;
13284      NumResults = 4;
13285      break;
13286    case NVPTX::BI__hmma_m8n32k16_st_c_f32:
13287      IID = isColMajor ? Intrinsic::nvvm_wmma_m8n32k16_store_d_f32_col_stride
13288                       : Intrinsic::nvvm_wmma_m8n32k16_store_d_f32_row_stride;
13289      break;
13290    default:
13291      llvm_unreachable("Unexpected builtin ID.");
13292    }
13293    Function *Intrinsic = CGM.getIntrinsic(IID, Dst->getType());
13294    llvm::Type *ParamType = Intrinsic->getFunctionType()->getParamType(1);
13295    SmallVector<Value *, 10> Values = {Dst};
13296    for (unsigned i = 0; i < NumResults; ++i) {
13297      Value *V = Builder.CreateAlignedLoad(
13298          Builder.CreateGEP(Src.getPointer(), llvm::ConstantInt::get(IntTy, i)),
13299          CharUnits::fromQuantity(4));
13300      Values.push_back(Builder.CreateBitCast(V, ParamType));
13301    }
13302    Values.push_back(Ldm);
13303    Value *Result = Builder.CreateCall(Intrinsic, Values);
13304    return Result;
13305  }
13306
13307  // BI__hmma_m16n16k16_mma_<Dtype><CType>(d, a, b, c, layout, satf) -->
13308  // Intrinsic::nvvm_wmma_m16n16k16_mma_sync<layout A,B><DType><CType><Satf>
13309  case NVPTX::BI__hmma_m16n16k16_mma_f16f16:
13310  case NVPTX::BI__hmma_m16n16k16_mma_f32f16:
13311  case NVPTX::BI__hmma_m16n16k16_mma_f32f32:
13312  case NVPTX::BI__hmma_m16n16k16_mma_f16f32:
13313  case NVPTX::BI__hmma_m32n8k16_mma_f16f16:
13314  case NVPTX::BI__hmma_m32n8k16_mma_f32f16:
13315  case NVPTX::BI__hmma_m32n8k16_mma_f32f32:
13316  case NVPTX::BI__hmma_m32n8k16_mma_f16f32:
13317  case NVPTX::BI__hmma_m8n32k16_mma_f16f16:
13318  case NVPTX::BI__hmma_m8n32k16_mma_f32f16:
13319  case NVPTX::BI__hmma_m8n32k16_mma_f32f32:
13320  case NVPTX::BI__hmma_m8n32k16_mma_f16f32: {
13321    Address Dst = EmitPointerWithAlignment(E->getArg(0));
13322    Address SrcA = EmitPointerWithAlignment(E->getArg(1));
13323    Address SrcB = EmitPointerWithAlignment(E->getArg(2));
13324    Address SrcC = EmitPointerWithAlignment(E->getArg(3));
13325    llvm::APSInt LayoutArg;
13326    if (!E->getArg(4)->isIntegerConstantExpr(LayoutArg, getContext()))
13327      return nullptr;
13328    int Layout = LayoutArg.getSExtValue();
13329    if (Layout < 0 || Layout > 3)
13330      return nullptr;
13331    llvm::APSInt SatfArg;
13332    if (!E->getArg(5)->isIntegerConstantExpr(SatfArg, getContext()))
13333      return nullptr;
13334    bool Satf = SatfArg.getSExtValue();
13335
13336    // clang-format off
13337#define MMA_VARIANTS(geom, type) {{                                 \
13338      Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type,             \
13339      Intrinsic::nvvm_wmma_##geom##_mma_row_row_##type##_satfinite, \
13340      Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type,             \
13341      Intrinsic::nvvm_wmma_##geom##_mma_row_col_##type##_satfinite, \
13342      Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type,             \
13343      Intrinsic::nvvm_wmma_##geom##_mma_col_row_##type##_satfinite, \
13344      Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type,             \
13345      Intrinsic::nvvm_wmma_##geom##_mma_col_col_##type##_satfinite  \
13346    }}
13347    // clang-format on
13348
13349    auto getMMAIntrinsic = [Layout, Satf](std::array<unsigned, 8> Variants) {
13350      unsigned Index = Layout * 2 + Satf;
13351      assert(Index < 8);
13352      return Variants[Index];
13353    };
13354    unsigned IID;
13355    unsigned NumEltsC;
13356    unsigned NumEltsD;
13357    switch (BuiltinID) {
13358    case NVPTX::BI__hmma_m16n16k16_mma_f16f16:
13359      IID = getMMAIntrinsic(MMA_VARIANTS(m16n16k16, f16_f16));
13360      NumEltsC = 4;
13361      NumEltsD = 4;
13362      break;
13363    case NVPTX::BI__hmma_m16n16k16_mma_f32f16:
13364      IID = getMMAIntrinsic(MMA_VARIANTS(m16n16k16, f32_f16));
13365      NumEltsC = 4;
13366      NumEltsD = 8;
13367      break;
13368    case NVPTX::BI__hmma_m16n16k16_mma_f16f32:
13369      IID = getMMAIntrinsic(MMA_VARIANTS(m16n16k16, f16_f32));
13370      NumEltsC = 8;
13371      NumEltsD = 4;
13372      break;
13373    case NVPTX::BI__hmma_m16n16k16_mma_f32f32:
13374      IID = getMMAIntrinsic(MMA_VARIANTS(m16n16k16, f32_f32));
13375      NumEltsC = 8;
13376      NumEltsD = 8;
13377      break;
13378    case NVPTX::BI__hmma_m32n8k16_mma_f16f16:
13379      IID = getMMAIntrinsic(MMA_VARIANTS(m32n8k16, f16_f16));
13380      NumEltsC = 4;
13381      NumEltsD = 4;
13382      break;
13383    case NVPTX::BI__hmma_m32n8k16_mma_f32f16:
13384      IID = getMMAIntrinsic(MMA_VARIANTS(m32n8k16, f32_f16));
13385      NumEltsC = 4;
13386      NumEltsD = 8;
13387      break;
13388    case NVPTX::BI__hmma_m32n8k16_mma_f16f32:
13389      IID = getMMAIntrinsic(MMA_VARIANTS(m32n8k16, f16_f32));
13390      NumEltsC = 8;
13391      NumEltsD = 4;
13392      break;
13393    case NVPTX::BI__hmma_m32n8k16_mma_f32f32:
13394      IID = getMMAIntrinsic(MMA_VARIANTS(m32n8k16, f32_f32));
13395      NumEltsC = 8;
13396      NumEltsD = 8;
13397      break;
13398    case NVPTX::BI__hmma_m8n32k16_mma_f16f16:
13399      IID = getMMAIntrinsic(MMA_VARIANTS(m8n32k16, f16_f16));
13400      NumEltsC = 4;
13401      NumEltsD = 4;
13402      break;
13403    case NVPTX::BI__hmma_m8n32k16_mma_f32f16:
13404      IID = getMMAIntrinsic(MMA_VARIANTS(m8n32k16, f32_f16));
13405      NumEltsC = 4;
13406      NumEltsD = 8;
13407      break;
13408    case NVPTX::BI__hmma_m8n32k16_mma_f16f32:
13409      IID = getMMAIntrinsic(MMA_VARIANTS(m8n32k16, f16_f32));
13410      NumEltsC = 8;
13411      NumEltsD = 4;
13412      break;
13413    case NVPTX::BI__hmma_m8n32k16_mma_f32f32:
13414      IID = getMMAIntrinsic(MMA_VARIANTS(m8n32k16, f32_f32));
13415      NumEltsC = 8;
13416      NumEltsD = 8;
13417      break;
13418    default:
13419      llvm_unreachable("Unexpected builtin ID.");
13420    }
13421#undef MMA_VARIANTS
13422
13423    SmallVector<Value *, 24> Values;
13424    Function *Intrinsic = CGM.getIntrinsic(IID);
13425    llvm::Type *ABType = Intrinsic->getFunctionType()->getParamType(0);
13426    // Load A
13427    for (unsigned i = 0; i < 8; ++i) {
13428      Value *V = Builder.CreateAlignedLoad(
13429          Builder.CreateGEP(SrcA.getPointer(),
13430                            llvm::ConstantInt::get(IntTy, i)),
13431          CharUnits::fromQuantity(4));
13432      Values.push_back(Builder.CreateBitCast(V, ABType));
13433    }
13434    // Load B
13435    for (unsigned i = 0; i < 8; ++i) {
13436      Value *V = Builder.CreateAlignedLoad(
13437          Builder.CreateGEP(SrcB.getPointer(),
13438                            llvm::ConstantInt::get(IntTy, i)),
13439          CharUnits::fromQuantity(4));
13440      Values.push_back(Builder.CreateBitCast(V, ABType));
13441    }
13442    // Load C
13443    llvm::Type *CType = Intrinsic->getFunctionType()->getParamType(16);
13444    for (unsigned i = 0; i < NumEltsC; ++i) {
13445      Value *V = Builder.CreateAlignedLoad(
13446          Builder.CreateGEP(SrcC.getPointer(),
13447                            llvm::ConstantInt::get(IntTy, i)),
13448          CharUnits::fromQuantity(4));
13449      Values.push_back(Builder.CreateBitCast(V, CType));
13450    }
13451    Value *Result = Builder.CreateCall(Intrinsic, Values);
13452    llvm::Type *DType = Dst.getElementType();
13453    for (unsigned i = 0; i < NumEltsD; ++i)
13454      Builder.CreateAlignedStore(
13455          Builder.CreateBitCast(Builder.CreateExtractValue(Result, i), DType),
13456          Builder.CreateGEP(Dst.getPointer(), llvm::ConstantInt::get(IntTy, i)),
13457          CharUnits::fromQuantity(4));
13458    return Result;
13459  }
13460  default:
13461    return nullptr;
13462  }
13463}
13464
13465Value *CodeGenFunction::EmitWebAssemblyBuiltinExpr(unsigned BuiltinID,
13466                                                   const CallExpr *E) {
13467  switch (BuiltinID) {
13468  case WebAssembly::BI__builtin_wasm_memory_size: {
13469    llvm::Type *ResultType = ConvertType(E->getType());
13470    Value *I = EmitScalarExpr(E->getArg(0));
13471    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_size, ResultType);
13472    return Builder.CreateCall(Callee, I);
13473  }
13474  case WebAssembly::BI__builtin_wasm_memory_grow: {
13475    llvm::Type *ResultType = ConvertType(E->getType());
13476    Value *Args[] = {
13477      EmitScalarExpr(E->getArg(0)),
13478      EmitScalarExpr(E->getArg(1))
13479    };
13480    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_grow, ResultType);
13481    return Builder.CreateCall(Callee, Args);
13482  }
13483  case WebAssembly::BI__builtin_wasm_memory_init: {
13484    llvm::APSInt SegConst;
13485    if (!E->getArg(0)->isIntegerConstantExpr(SegConst, getContext()))
13486      llvm_unreachable("Constant arg isn't actually constant?");
13487    llvm::APSInt MemConst;
13488    if (!E->getArg(1)->isIntegerConstantExpr(MemConst, getContext()))
13489      llvm_unreachable("Constant arg isn't actually constant?");
13490    if (!MemConst.isNullValue())
13491      ErrorUnsupported(E, "non-zero memory index");
13492    Value *Args[] = {llvm::ConstantInt::get(getLLVMContext(), SegConst),
13493                     llvm::ConstantInt::get(getLLVMContext(), MemConst),
13494                     EmitScalarExpr(E->getArg(2)), EmitScalarExpr(E->getArg(3)),
13495                     EmitScalarExpr(E->getArg(4))};
13496    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_memory_init);
13497    return Builder.CreateCall(Callee, Args);
13498  }
13499  case WebAssembly::BI__builtin_wasm_data_drop: {
13500    llvm::APSInt SegConst;
13501    if (!E->getArg(0)->isIntegerConstantExpr(SegConst, getContext()))
13502      llvm_unreachable("Constant arg isn't actually constant?");
13503    Value *Arg = llvm::ConstantInt::get(getLLVMContext(), SegConst);
13504    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_data_drop);
13505    return Builder.CreateCall(Callee, {Arg});
13506  }
13507  case WebAssembly::BI__builtin_wasm_throw: {
13508    Value *Tag = EmitScalarExpr(E->getArg(0));
13509    Value *Obj = EmitScalarExpr(E->getArg(1));
13510    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_throw);
13511    return Builder.CreateCall(Callee, {Tag, Obj});
13512  }
13513  case WebAssembly::BI__builtin_wasm_rethrow_in_catch: {
13514    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_rethrow_in_catch);
13515    return Builder.CreateCall(Callee);
13516  }
13517  case WebAssembly::BI__builtin_wasm_atomic_wait_i32: {
13518    Value *Addr = EmitScalarExpr(E->getArg(0));
13519    Value *Expected = EmitScalarExpr(E->getArg(1));
13520    Value *Timeout = EmitScalarExpr(E->getArg(2));
13521    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_atomic_wait_i32);
13522    return Builder.CreateCall(Callee, {Addr, Expected, Timeout});
13523  }
13524  case WebAssembly::BI__builtin_wasm_atomic_wait_i64: {
13525    Value *Addr = EmitScalarExpr(E->getArg(0));
13526    Value *Expected = EmitScalarExpr(E->getArg(1));
13527    Value *Timeout = EmitScalarExpr(E->getArg(2));
13528    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_atomic_wait_i64);
13529    return Builder.CreateCall(Callee, {Addr, Expected, Timeout});
13530  }
13531  case WebAssembly::BI__builtin_wasm_atomic_notify: {
13532    Value *Addr = EmitScalarExpr(E->getArg(0));
13533    Value *Count = EmitScalarExpr(E->getArg(1));
13534    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_atomic_notify);
13535    return Builder.CreateCall(Callee, {Addr, Count});
13536  }
13537  case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f32:
13538  case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32_f64:
13539  case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f32:
13540  case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64_f64:
13541  case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i32x4_f32x4:
13542  case WebAssembly::BI__builtin_wasm_trunc_saturate_s_i64x2_f64x2: {
13543    Value *Src = EmitScalarExpr(E->getArg(0));
13544    llvm::Type *ResT = ConvertType(E->getType());
13545    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_trunc_saturate_signed,
13546                                     {ResT, Src->getType()});
13547    return Builder.CreateCall(Callee, {Src});
13548  }
13549  case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f32:
13550  case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32_f64:
13551  case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f32:
13552  case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64_f64:
13553  case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i32x4_f32x4:
13554  case WebAssembly::BI__builtin_wasm_trunc_saturate_u_i64x2_f64x2: {
13555    Value *Src = EmitScalarExpr(E->getArg(0));
13556    llvm::Type *ResT = ConvertType(E->getType());
13557    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_trunc_saturate_unsigned,
13558                                     {ResT, Src->getType()});
13559    return Builder.CreateCall(Callee, {Src});
13560  }
13561  case WebAssembly::BI__builtin_wasm_min_f32:
13562  case WebAssembly::BI__builtin_wasm_min_f64:
13563  case WebAssembly::BI__builtin_wasm_min_f32x4:
13564  case WebAssembly::BI__builtin_wasm_min_f64x2: {
13565    Value *LHS = EmitScalarExpr(E->getArg(0));
13566    Value *RHS = EmitScalarExpr(E->getArg(1));
13567    Function *Callee = CGM.getIntrinsic(Intrinsic::minimum,
13568                                     ConvertType(E->getType()));
13569    return Builder.CreateCall(Callee, {LHS, RHS});
13570  }
13571  case WebAssembly::BI__builtin_wasm_max_f32:
13572  case WebAssembly::BI__builtin_wasm_max_f64:
13573  case WebAssembly::BI__builtin_wasm_max_f32x4:
13574  case WebAssembly::BI__builtin_wasm_max_f64x2: {
13575    Value *LHS = EmitScalarExpr(E->getArg(0));
13576    Value *RHS = EmitScalarExpr(E->getArg(1));
13577    Function *Callee = CGM.getIntrinsic(Intrinsic::maximum,
13578                                     ConvertType(E->getType()));
13579    return Builder.CreateCall(Callee, {LHS, RHS});
13580  }
13581  case WebAssembly::BI__builtin_wasm_extract_lane_s_i8x16:
13582  case WebAssembly::BI__builtin_wasm_extract_lane_u_i8x16:
13583  case WebAssembly::BI__builtin_wasm_extract_lane_s_i16x8:
13584  case WebAssembly::BI__builtin_wasm_extract_lane_u_i16x8:
13585  case WebAssembly::BI__builtin_wasm_extract_lane_i32x4:
13586  case WebAssembly::BI__builtin_wasm_extract_lane_i64x2:
13587  case WebAssembly::BI__builtin_wasm_extract_lane_f32x4:
13588  case WebAssembly::BI__builtin_wasm_extract_lane_f64x2: {
13589    llvm::APSInt LaneConst;
13590    if (!E->getArg(1)->isIntegerConstantExpr(LaneConst, getContext()))
13591      llvm_unreachable("Constant arg isn't actually constant?");
13592    Value *Vec = EmitScalarExpr(E->getArg(0));
13593    Value *Lane = llvm::ConstantInt::get(getLLVMContext(), LaneConst);
13594    Value *Extract = Builder.CreateExtractElement(Vec, Lane);
13595    switch (BuiltinID) {
13596    case WebAssembly::BI__builtin_wasm_extract_lane_s_i8x16:
13597    case WebAssembly::BI__builtin_wasm_extract_lane_s_i16x8:
13598      return Builder.CreateSExt(Extract, ConvertType(E->getType()));
13599    case WebAssembly::BI__builtin_wasm_extract_lane_u_i8x16:
13600    case WebAssembly::BI__builtin_wasm_extract_lane_u_i16x8:
13601      return Builder.CreateZExt(Extract, ConvertType(E->getType()));
13602    case WebAssembly::BI__builtin_wasm_extract_lane_i32x4:
13603    case WebAssembly::BI__builtin_wasm_extract_lane_i64x2:
13604    case WebAssembly::BI__builtin_wasm_extract_lane_f32x4:
13605    case WebAssembly::BI__builtin_wasm_extract_lane_f64x2:
13606      return Extract;
13607    default:
13608      llvm_unreachable("unexpected builtin ID");
13609    }
13610  }
13611  case WebAssembly::BI__builtin_wasm_replace_lane_i8x16:
13612  case WebAssembly::BI__builtin_wasm_replace_lane_i16x8:
13613  case WebAssembly::BI__builtin_wasm_replace_lane_i32x4:
13614  case WebAssembly::BI__builtin_wasm_replace_lane_i64x2:
13615  case WebAssembly::BI__builtin_wasm_replace_lane_f32x4:
13616  case WebAssembly::BI__builtin_wasm_replace_lane_f64x2: {
13617    llvm::APSInt LaneConst;
13618    if (!E->getArg(1)->isIntegerConstantExpr(LaneConst, getContext()))
13619      llvm_unreachable("Constant arg isn't actually constant?");
13620    Value *Vec = EmitScalarExpr(E->getArg(0));
13621    Value *Lane = llvm::ConstantInt::get(getLLVMContext(), LaneConst);
13622    Value *Val = EmitScalarExpr(E->getArg(2));
13623    switch (BuiltinID) {
13624    case WebAssembly::BI__builtin_wasm_replace_lane_i8x16:
13625    case WebAssembly::BI__builtin_wasm_replace_lane_i16x8: {
13626      llvm::Type *ElemType = ConvertType(E->getType())->getVectorElementType();
13627      Value *Trunc = Builder.CreateTrunc(Val, ElemType);
13628      return Builder.CreateInsertElement(Vec, Trunc, Lane);
13629    }
13630    case WebAssembly::BI__builtin_wasm_replace_lane_i32x4:
13631    case WebAssembly::BI__builtin_wasm_replace_lane_i64x2:
13632    case WebAssembly::BI__builtin_wasm_replace_lane_f32x4:
13633    case WebAssembly::BI__builtin_wasm_replace_lane_f64x2:
13634      return Builder.CreateInsertElement(Vec, Val, Lane);
13635    default:
13636      llvm_unreachable("unexpected builtin ID");
13637    }
13638  }
13639  case WebAssembly::BI__builtin_wasm_add_saturate_s_i8x16:
13640  case WebAssembly::BI__builtin_wasm_add_saturate_u_i8x16:
13641  case WebAssembly::BI__builtin_wasm_add_saturate_s_i16x8:
13642  case WebAssembly::BI__builtin_wasm_add_saturate_u_i16x8:
13643  case WebAssembly::BI__builtin_wasm_sub_saturate_s_i8x16:
13644  case WebAssembly::BI__builtin_wasm_sub_saturate_u_i8x16:
13645  case WebAssembly::BI__builtin_wasm_sub_saturate_s_i16x8:
13646  case WebAssembly::BI__builtin_wasm_sub_saturate_u_i16x8: {
13647    unsigned IntNo;
13648    switch (BuiltinID) {
13649    case WebAssembly::BI__builtin_wasm_add_saturate_s_i8x16:
13650    case WebAssembly::BI__builtin_wasm_add_saturate_s_i16x8:
13651      IntNo = Intrinsic::sadd_sat;
13652      break;
13653    case WebAssembly::BI__builtin_wasm_add_saturate_u_i8x16:
13654    case WebAssembly::BI__builtin_wasm_add_saturate_u_i16x8:
13655      IntNo = Intrinsic::uadd_sat;
13656      break;
13657    case WebAssembly::BI__builtin_wasm_sub_saturate_s_i8x16:
13658    case WebAssembly::BI__builtin_wasm_sub_saturate_s_i16x8:
13659      IntNo = Intrinsic::wasm_sub_saturate_signed;
13660      break;
13661    case WebAssembly::BI__builtin_wasm_sub_saturate_u_i8x16:
13662    case WebAssembly::BI__builtin_wasm_sub_saturate_u_i16x8:
13663      IntNo = Intrinsic::wasm_sub_saturate_unsigned;
13664      break;
13665    default:
13666      llvm_unreachable("unexpected builtin ID");
13667    }
13668    Value *LHS = EmitScalarExpr(E->getArg(0));
13669    Value *RHS = EmitScalarExpr(E->getArg(1));
13670    Function *Callee = CGM.getIntrinsic(IntNo, ConvertType(E->getType()));
13671    return Builder.CreateCall(Callee, {LHS, RHS});
13672  }
13673  case WebAssembly::BI__builtin_wasm_bitselect: {
13674    Value *V1 = EmitScalarExpr(E->getArg(0));
13675    Value *V2 = EmitScalarExpr(E->getArg(1));
13676    Value *C = EmitScalarExpr(E->getArg(2));
13677    Function *Callee = CGM.getIntrinsic(Intrinsic::wasm_bitselect,
13678                                     ConvertType(E->getType()));
13679    return Builder.CreateCall(Callee, {V1, V2, C});
13680  }
13681  case WebAssembly::BI__builtin_wasm_any_true_i8x16:
13682  case WebAssembly::BI__builtin_wasm_any_true_i16x8:
13683  case WebAssembly::BI__builtin_wasm_any_true_i32x4:
13684  case WebAssembly::BI__builtin_wasm_any_true_i64x2:
13685  case WebAssembly::BI__builtin_wasm_all_true_i8x16:
13686  case WebAssembly::BI__builtin_wasm_all_true_i16x8:
13687  case WebAssembly::BI__builtin_wasm_all_true_i32x4:
13688  case WebAssembly::BI__builtin_wasm_all_true_i64x2: {
13689    unsigned IntNo;
13690    switch (BuiltinID) {
13691    case WebAssembly::BI__builtin_wasm_any_true_i8x16:
13692    case WebAssembly::BI__builtin_wasm_any_true_i16x8:
13693    case WebAssembly::BI__builtin_wasm_any_true_i32x4:
13694    case WebAssembly::BI__builtin_wasm_any_true_i64x2:
13695      IntNo = Intrinsic::wasm_anytrue;
13696      break;
13697    case WebAssembly::BI__builtin_wasm_all_true_i8x16:
13698    case WebAssembly::BI__builtin_wasm_all_true_i16x8:
13699    case WebAssembly::BI__builtin_wasm_all_true_i32x4:
13700    case WebAssembly::BI__builtin_wasm_all_true_i64x2:
13701      IntNo = Intrinsic::wasm_alltrue;
13702      break;
13703    default:
13704      llvm_unreachable("unexpected builtin ID");
13705    }
13706    Value *Vec = EmitScalarExpr(E->getArg(0));
13707    Function *Callee = CGM.getIntrinsic(IntNo, Vec->getType());
13708    return Builder.CreateCall(Callee, {Vec});
13709  }
13710  case WebAssembly::BI__builtin_wasm_abs_f32x4:
13711  case WebAssembly::BI__builtin_wasm_abs_f64x2: {
13712    Value *Vec = EmitScalarExpr(E->getArg(0));
13713    Function *Callee = CGM.getIntrinsic(Intrinsic::fabs, Vec->getType());
13714    return Builder.CreateCall(Callee, {Vec});
13715  }
13716  case WebAssembly::BI__builtin_wasm_sqrt_f32x4:
13717  case WebAssembly::BI__builtin_wasm_sqrt_f64x2: {
13718    Value *Vec = EmitScalarExpr(E->getArg(0));
13719    Function *Callee = CGM.getIntrinsic(Intrinsic::sqrt, Vec->getType());
13720    return Builder.CreateCall(Callee, {Vec});
13721  }
13722
13723  default:
13724    return nullptr;
13725  }
13726}
13727
13728Value *CodeGenFunction::EmitHexagonBuiltinExpr(unsigned BuiltinID,
13729                                               const CallExpr *E) {
13730  SmallVector<llvm::Value *, 4> Ops;
13731  Intrinsic::ID ID = Intrinsic::not_intrinsic;
13732
13733  auto MakeCircLd = [&](unsigned IntID, bool HasImm) {
13734    // The base pointer is passed by address, so it needs to be loaded.
13735    Address BP = EmitPointerWithAlignment(E->getArg(0));
13736    BP = Address(Builder.CreateBitCast(BP.getPointer(), Int8PtrPtrTy),
13737                 BP.getAlignment());
13738    llvm::Value *Base = Builder.CreateLoad(BP);
13739    // Operands are Base, Increment, Modifier, Start.
13740    if (HasImm)
13741      Ops = { Base, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)),
13742              EmitScalarExpr(E->getArg(3)) };
13743    else
13744      Ops = { Base, EmitScalarExpr(E->getArg(1)),
13745              EmitScalarExpr(E->getArg(2)) };
13746
13747    llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops);
13748    llvm::Value *NewBase = Builder.CreateExtractValue(Result, 1);
13749    llvm::Value *LV = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)),
13750                                            NewBase->getType()->getPointerTo());
13751    Address Dest = EmitPointerWithAlignment(E->getArg(0));
13752    // The intrinsic generates two results. The new value for the base pointer
13753    // needs to be stored.
13754    Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment());
13755    return Builder.CreateExtractValue(Result, 0);
13756  };
13757
13758  auto MakeCircSt = [&](unsigned IntID, bool HasImm) {
13759    // The base pointer is passed by address, so it needs to be loaded.
13760    Address BP = EmitPointerWithAlignment(E->getArg(0));
13761    BP = Address(Builder.CreateBitCast(BP.getPointer(), Int8PtrPtrTy),
13762                 BP.getAlignment());
13763    llvm::Value *Base = Builder.CreateLoad(BP);
13764    // Operands are Base, Increment, Modifier, Value, Start.
13765    if (HasImm)
13766      Ops = { Base, EmitScalarExpr(E->getArg(1)), EmitScalarExpr(E->getArg(2)),
13767              EmitScalarExpr(E->getArg(3)), EmitScalarExpr(E->getArg(4)) };
13768    else
13769      Ops = { Base, EmitScalarExpr(E->getArg(1)),
13770              EmitScalarExpr(E->getArg(2)), EmitScalarExpr(E->getArg(3)) };
13771
13772    llvm::Value *NewBase = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops);
13773    llvm::Value *LV = Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)),
13774                                            NewBase->getType()->getPointerTo());
13775    Address Dest = EmitPointerWithAlignment(E->getArg(0));
13776    // The intrinsic generates one result, which is the new value for the base
13777    // pointer. It needs to be stored.
13778    return Builder.CreateAlignedStore(NewBase, LV, Dest.getAlignment());
13779  };
13780
13781  // Handle the conversion of bit-reverse load intrinsics to bit code.
13782  // The intrinsic call after this function only reads from memory and the
13783  // write to memory is dealt by the store instruction.
13784  auto MakeBrevLd = [&](unsigned IntID, llvm::Type *DestTy) {
13785    // The intrinsic generates one result, which is the new value for the base
13786    // pointer. It needs to be returned. The result of the load instruction is
13787    // passed to intrinsic by address, so the value needs to be stored.
13788    llvm::Value *BaseAddress =
13789        Builder.CreateBitCast(EmitScalarExpr(E->getArg(0)), Int8PtrTy);
13790
13791    // Expressions like &(*pt++) will be incremented per evaluation.
13792    // EmitPointerWithAlignment and EmitScalarExpr evaluates the expression
13793    // per call.
13794    Address DestAddr = EmitPointerWithAlignment(E->getArg(1));
13795    DestAddr = Address(Builder.CreateBitCast(DestAddr.getPointer(), Int8PtrTy),
13796                       DestAddr.getAlignment());
13797    llvm::Value *DestAddress = DestAddr.getPointer();
13798
13799    // Operands are Base, Dest, Modifier.
13800    // The intrinsic format in LLVM IR is defined as
13801    // { ValueType, i8* } (i8*, i32).
13802    Ops = {BaseAddress, EmitScalarExpr(E->getArg(2))};
13803
13804    llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(IntID), Ops);
13805    // The value needs to be stored as the variable is passed by reference.
13806    llvm::Value *DestVal = Builder.CreateExtractValue(Result, 0);
13807
13808    // The store needs to be truncated to fit the destination type.
13809    // While i32 and i64 are natively supported on Hexagon, i8 and i16 needs
13810    // to be handled with stores of respective destination type.
13811    DestVal = Builder.CreateTrunc(DestVal, DestTy);
13812
13813    llvm::Value *DestForStore =
13814        Builder.CreateBitCast(DestAddress, DestVal->getType()->getPointerTo());
13815    Builder.CreateAlignedStore(DestVal, DestForStore, DestAddr.getAlignment());
13816    // The updated value of the base pointer is returned.
13817    return Builder.CreateExtractValue(Result, 1);
13818  };
13819
13820  switch (BuiltinID) {
13821  case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry:
13822  case Hexagon::BI__builtin_HEXAGON_V6_vaddcarry_128B: {
13823    Address Dest = EmitPointerWithAlignment(E->getArg(2));
13824    unsigned Size;
13825    if (BuiltinID == Hexagon::BI__builtin_HEXAGON_V6_vaddcarry) {
13826      Size = 512;
13827      ID = Intrinsic::hexagon_V6_vaddcarry;
13828    } else {
13829      Size = 1024;
13830      ID = Intrinsic::hexagon_V6_vaddcarry_128B;
13831    }
13832    Dest = Builder.CreateBitCast(Dest,
13833        llvm::VectorType::get(Builder.getInt1Ty(), Size)->getPointerTo(0));
13834    LoadInst *QLd = Builder.CreateLoad(Dest);
13835    Ops = { EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), QLd };
13836    llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
13837    llvm::Value *Vprd = Builder.CreateExtractValue(Result, 1);
13838    llvm::Value *Base = Builder.CreateBitCast(EmitScalarExpr(E->getArg(2)),
13839                                              Vprd->getType()->getPointerTo(0));
13840    Builder.CreateAlignedStore(Vprd, Base, Dest.getAlignment());
13841    return Builder.CreateExtractValue(Result, 0);
13842  }
13843  case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry:
13844  case Hexagon::BI__builtin_HEXAGON_V6_vsubcarry_128B: {
13845    Address Dest = EmitPointerWithAlignment(E->getArg(2));
13846    unsigned Size;
13847    if (BuiltinID == Hexagon::BI__builtin_HEXAGON_V6_vsubcarry) {
13848      Size = 512;
13849      ID = Intrinsic::hexagon_V6_vsubcarry;
13850    } else {
13851      Size = 1024;
13852      ID = Intrinsic::hexagon_V6_vsubcarry_128B;
13853    }
13854    Dest = Builder.CreateBitCast(Dest,
13855        llvm::VectorType::get(Builder.getInt1Ty(), Size)->getPointerTo(0));
13856    LoadInst *QLd = Builder.CreateLoad(Dest);
13857    Ops = { EmitScalarExpr(E->getArg(0)), EmitScalarExpr(E->getArg(1)), QLd };
13858    llvm::Value *Result = Builder.CreateCall(CGM.getIntrinsic(ID), Ops);
13859    llvm::Value *Vprd = Builder.CreateExtractValue(Result, 1);
13860    llvm::Value *Base = Builder.CreateBitCast(EmitScalarExpr(E->getArg(2)),
13861                                              Vprd->getType()->getPointerTo(0));
13862    Builder.CreateAlignedStore(Vprd, Base, Dest.getAlignment());
13863    return Builder.CreateExtractValue(Result, 0);
13864  }
13865  case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pci:
13866    return MakeCircLd(Intrinsic::hexagon_L2_loadrub_pci, /*HasImm*/true);
13867  case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pci:
13868    return MakeCircLd(Intrinsic::hexagon_L2_loadrb_pci,  /*HasImm*/true);
13869  case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pci:
13870    return MakeCircLd(Intrinsic::hexagon_L2_loadruh_pci, /*HasImm*/true);
13871  case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pci:
13872    return MakeCircLd(Intrinsic::hexagon_L2_loadrh_pci,  /*HasImm*/true);
13873  case Hexagon::BI__builtin_HEXAGON_L2_loadri_pci:
13874    return MakeCircLd(Intrinsic::hexagon_L2_loadri_pci,  /*HasImm*/true);
13875  case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pci:
13876    return MakeCircLd(Intrinsic::hexagon_L2_loadrd_pci,  /*HasImm*/true);
13877  case Hexagon::BI__builtin_HEXAGON_L2_loadrub_pcr:
13878    return MakeCircLd(Intrinsic::hexagon_L2_loadrub_pcr, /*HasImm*/false);
13879  case Hexagon::BI__builtin_HEXAGON_L2_loadrb_pcr:
13880    return MakeCircLd(Intrinsic::hexagon_L2_loadrb_pcr,  /*HasImm*/false);
13881  case Hexagon::BI__builtin_HEXAGON_L2_loadruh_pcr:
13882    return MakeCircLd(Intrinsic::hexagon_L2_loadruh_pcr, /*HasImm*/false);
13883  case Hexagon::BI__builtin_HEXAGON_L2_loadrh_pcr:
13884    return MakeCircLd(Intrinsic::hexagon_L2_loadrh_pcr,  /*HasImm*/false);
13885  case Hexagon::BI__builtin_HEXAGON_L2_loadri_pcr:
13886    return MakeCircLd(Intrinsic::hexagon_L2_loadri_pcr,  /*HasImm*/false);
13887  case Hexagon::BI__builtin_HEXAGON_L2_loadrd_pcr:
13888    return MakeCircLd(Intrinsic::hexagon_L2_loadrd_pcr,  /*HasImm*/false);
13889  case Hexagon::BI__builtin_HEXAGON_S2_storerb_pci:
13890    return MakeCircSt(Intrinsic::hexagon_S2_storerb_pci, /*HasImm*/true);
13891  case Hexagon::BI__builtin_HEXAGON_S2_storerh_pci:
13892    return MakeCircSt(Intrinsic::hexagon_S2_storerh_pci, /*HasImm*/true);
13893  case Hexagon::BI__builtin_HEXAGON_S2_storerf_pci:
13894    return MakeCircSt(Intrinsic::hexagon_S2_storerf_pci, /*HasImm*/true);
13895  case Hexagon::BI__builtin_HEXAGON_S2_storeri_pci:
13896    return MakeCircSt(Intrinsic::hexagon_S2_storeri_pci, /*HasImm*/true);
13897  case Hexagon::BI__builtin_HEXAGON_S2_storerd_pci:
13898    return MakeCircSt(Intrinsic::hexagon_S2_storerd_pci, /*HasImm*/true);
13899  case Hexagon::BI__builtin_HEXAGON_S2_storerb_pcr:
13900    return MakeCircSt(Intrinsic::hexagon_S2_storerb_pcr, /*HasImm*/false);
13901  case Hexagon::BI__builtin_HEXAGON_S2_storerh_pcr:
13902    return MakeCircSt(Intrinsic::hexagon_S2_storerh_pcr, /*HasImm*/false);
13903  case Hexagon::BI__builtin_HEXAGON_S2_storerf_pcr:
13904    return MakeCircSt(Intrinsic::hexagon_S2_storerf_pcr, /*HasImm*/false);
13905  case Hexagon::BI__builtin_HEXAGON_S2_storeri_pcr:
13906    return MakeCircSt(Intrinsic::hexagon_S2_storeri_pcr, /*HasImm*/false);
13907  case Hexagon::BI__builtin_HEXAGON_S2_storerd_pcr:
13908    return MakeCircSt(Intrinsic::hexagon_S2_storerd_pcr, /*HasImm*/false);
13909  case Hexagon::BI__builtin_brev_ldub:
13910    return MakeBrevLd(Intrinsic::hexagon_L2_loadrub_pbr, Int8Ty);
13911  case Hexagon::BI__builtin_brev_ldb:
13912    return MakeBrevLd(Intrinsic::hexagon_L2_loadrb_pbr, Int8Ty);
13913  case Hexagon::BI__builtin_brev_lduh:
13914    return MakeBrevLd(Intrinsic::hexagon_L2_loadruh_pbr, Int16Ty);
13915  case Hexagon::BI__builtin_brev_ldh:
13916    return MakeBrevLd(Intrinsic::hexagon_L2_loadrh_pbr, Int16Ty);
13917  case Hexagon::BI__builtin_brev_ldw:
13918    return MakeBrevLd(Intrinsic::hexagon_L2_loadri_pbr, Int32Ty);
13919  case Hexagon::BI__builtin_brev_ldd:
13920    return MakeBrevLd(Intrinsic::hexagon_L2_loadrd_pbr, Int64Ty);
13921  default:
13922    break;
13923  } // switch
13924
13925  return nullptr;
13926}
13927
clang::CodeGen::CodeGenModule::getBuiltinLibFunction
clang::CodeGen::CodeGenFunction::EmitVAStartEnd
clang::CodeGen::CodeGenFunction::evaluateOrEmitBuiltinObjectSize
clang::CodeGen::CodeGenFunction::emitBuiltinObjectSize
clang::CodeGen::CodeGenFunction::MSVCIntrin
clang::CodeGen::CodeGenFunction::EmitMSVCBuiltinExpr
clang::CodeGen::CodeGenFunction::EmitCheckedArgForBuiltin
clang::CodeGen::CodeGenFunction::generateBuiltinOSLogHelperFunction
clang::CodeGen::CodeGenFunction::emitBuiltinOSLogFormat
clang::CodeGen::CodeGenFunction::emitRotate
clang::CodeGen::CodeGenFunction::EmitBuiltinExpr
clang::CodeGen::CodeGenFunction::EmitTargetBuiltinExpr
clang::CodeGen::CodeGenFunction::EmitNeonSplat
clang::CodeGen::CodeGenFunction::EmitNeonCall
clang::CodeGen::CodeGenFunction::EmitNeonShiftVector
clang::CodeGen::CodeGenFunction::EmitNeonRShiftImm
clang::CodeGen::CodeGenFunction::LookupNeonLLVMIntrinsic
clang::CodeGen::CodeGenFunction::EmitCommonNeonBuiltinExpr
clang::CodeGen::CodeGenFunction::EmitAArch64CompareBuiltinExpr
clang::CodeGen::CodeGenFunction::GetValueForARMHint
clang::CodeGen::CodeGenFunction::EmitARMBuiltinExpr
clang::CodeGen::CodeGenFunction::vectorWrapScalar16
clang::CodeGen::CodeGenFunction::EmitAArch64BuiltinExpr
clang::CodeGen::CodeGenFunction::BuildVector
clang::CodeGen::CodeGenFunction::EmitX86CpuIs
clang::CodeGen::CodeGenFunction::EmitX86CpuIs
clang::CodeGen::CodeGenFunction::EmitX86CpuSupports
clang::CodeGen::CodeGenFunction::GetX86CpuSupportsMask
clang::CodeGen::CodeGenFunction::EmitX86CpuSupports
clang::CodeGen::CodeGenFunction::EmitX86CpuSupports
clang::CodeGen::CodeGenFunction::EmitX86CpuInit
clang::CodeGen::CodeGenFunction::EmitX86BuiltinExpr
clang::CodeGen::CodeGenFunction::EmitPPCBuiltinExpr
clang::CodeGen::CodeGenFunction::EmitAMDGPUBuiltinExpr
clang::CodeGen::CodeGenFunction::EmitSystemZBuiltinExpr
clang::CodeGen::CodeGenFunction::EmitNVPTXBuiltinExpr
clang::CodeGen::CodeGenFunction::EmitWebAssemblyBuiltinExpr
clang::CodeGen::CodeGenFunction::EmitHexagonBuiltinExpr
CodeThinker, Enjoy your reading code