1 | /*===---- xmmintrin.h - SSE intrinsics -------------------------------------=== |
---|---|
2 | * |
3 | * Permission is hereby granted, free of charge, to any person obtaining a copy |
4 | * of this software and associated documentation files (the "Software"), to deal |
5 | * in the Software without restriction, including without limitation the rights |
6 | * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell |
7 | * copies of the Software, and to permit persons to whom the Software is |
8 | * furnished to do so, subject to the following conditions: |
9 | * |
10 | * The above copyright notice and this permission notice shall be included in |
11 | * all copies or substantial portions of the Software. |
12 | * |
13 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR |
14 | * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, |
15 | * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE |
16 | * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER |
17 | * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, |
18 | * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN |
19 | * THE SOFTWARE. |
20 | * |
21 | *===-----------------------------------------------------------------------=== |
22 | */ |
23 | |
24 | #ifndef __XMMINTRIN_H |
25 | #define __XMMINTRIN_H |
26 | |
27 | #include <mmintrin.h> |
28 | |
29 | typedef int __v4si __attribute__((__vector_size__(16))); |
30 | typedef float __v4sf __attribute__((__vector_size__(16))); |
31 | typedef float __m128 __attribute__((__vector_size__(16), __aligned__(16))); |
32 | |
33 | typedef float __m128_u __attribute__((__vector_size__(16), __aligned__(1))); |
34 | |
35 | /* Unsigned types */ |
36 | typedef unsigned int __v4su __attribute__((__vector_size__(16))); |
37 | |
38 | /* This header should only be included in a hosted environment as it depends on |
39 | * a standard library to provide allocation routines. */ |
40 | #if __STDC_HOSTED__ |
41 | #include <mm_malloc.h> |
42 | #endif |
43 | |
44 | /* Define the default attributes for the functions in this file. */ |
45 | #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("sse"), __min_vector_width__(128))) |
46 | #define __DEFAULT_FN_ATTRS_MMX __attribute__((__always_inline__, __nodebug__, __target__("mmx,sse"), __min_vector_width__(64))) |
47 | |
48 | /// Adds the 32-bit float values in the low-order bits of the operands. |
49 | /// |
50 | /// \headerfile <x86intrin.h> |
51 | /// |
52 | /// This intrinsic corresponds to the <c> VADDSS / ADDSS </c> instructions. |
53 | /// |
54 | /// \param __a |
55 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
56 | /// The lower 32 bits of this operand are used in the calculation. |
57 | /// \param __b |
58 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
59 | /// The lower 32 bits of this operand are used in the calculation. |
60 | /// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the sum |
61 | /// of the lower 32 bits of both operands. The upper 96 bits are copied from |
62 | /// the upper 96 bits of the first source operand. |
63 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
64 | _mm_add_ss(__m128 __a, __m128 __b) |
65 | { |
66 | __a[0] += __b[0]; |
67 | return __a; |
68 | } |
69 | |
70 | /// Adds two 128-bit vectors of [4 x float], and returns the results of |
71 | /// the addition. |
72 | /// |
73 | /// \headerfile <x86intrin.h> |
74 | /// |
75 | /// This intrinsic corresponds to the <c> VADDPS / ADDPS </c> instructions. |
76 | /// |
77 | /// \param __a |
78 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
79 | /// \param __b |
80 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
81 | /// \returns A 128-bit vector of [4 x float] containing the sums of both |
82 | /// operands. |
83 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
84 | _mm_add_ps(__m128 __a, __m128 __b) |
85 | { |
86 | return (__m128)((__v4sf)__a + (__v4sf)__b); |
87 | } |
88 | |
89 | /// Subtracts the 32-bit float value in the low-order bits of the second |
90 | /// operand from the corresponding value in the first operand. |
91 | /// |
92 | /// \headerfile <x86intrin.h> |
93 | /// |
94 | /// This intrinsic corresponds to the <c> VSUBSS / SUBSS </c> instructions. |
95 | /// |
96 | /// \param __a |
97 | /// A 128-bit vector of [4 x float] containing the minuend. The lower 32 bits |
98 | /// of this operand are used in the calculation. |
99 | /// \param __b |
100 | /// A 128-bit vector of [4 x float] containing the subtrahend. The lower 32 |
101 | /// bits of this operand are used in the calculation. |
102 | /// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the |
103 | /// difference of the lower 32 bits of both operands. The upper 96 bits are |
104 | /// copied from the upper 96 bits of the first source operand. |
105 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
106 | _mm_sub_ss(__m128 __a, __m128 __b) |
107 | { |
108 | __a[0] -= __b[0]; |
109 | return __a; |
110 | } |
111 | |
112 | /// Subtracts each of the values of the second operand from the first |
113 | /// operand, both of which are 128-bit vectors of [4 x float] and returns |
114 | /// the results of the subtraction. |
115 | /// |
116 | /// \headerfile <x86intrin.h> |
117 | /// |
118 | /// This intrinsic corresponds to the <c> VSUBPS / SUBPS </c> instructions. |
119 | /// |
120 | /// \param __a |
121 | /// A 128-bit vector of [4 x float] containing the minuend. |
122 | /// \param __b |
123 | /// A 128-bit vector of [4 x float] containing the subtrahend. |
124 | /// \returns A 128-bit vector of [4 x float] containing the differences between |
125 | /// both operands. |
126 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
127 | _mm_sub_ps(__m128 __a, __m128 __b) |
128 | { |
129 | return (__m128)((__v4sf)__a - (__v4sf)__b); |
130 | } |
131 | |
132 | /// Multiplies two 32-bit float values in the low-order bits of the |
133 | /// operands. |
134 | /// |
135 | /// \headerfile <x86intrin.h> |
136 | /// |
137 | /// This intrinsic corresponds to the <c> VMULSS / MULSS </c> instructions. |
138 | /// |
139 | /// \param __a |
140 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
141 | /// The lower 32 bits of this operand are used in the calculation. |
142 | /// \param __b |
143 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
144 | /// The lower 32 bits of this operand are used in the calculation. |
145 | /// \returns A 128-bit vector of [4 x float] containing the product of the lower |
146 | /// 32 bits of both operands. The upper 96 bits are copied from the upper 96 |
147 | /// bits of the first source operand. |
148 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
149 | _mm_mul_ss(__m128 __a, __m128 __b) |
150 | { |
151 | __a[0] *= __b[0]; |
152 | return __a; |
153 | } |
154 | |
155 | /// Multiplies two 128-bit vectors of [4 x float] and returns the |
156 | /// results of the multiplication. |
157 | /// |
158 | /// \headerfile <x86intrin.h> |
159 | /// |
160 | /// This intrinsic corresponds to the <c> VMULPS / MULPS </c> instructions. |
161 | /// |
162 | /// \param __a |
163 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
164 | /// \param __b |
165 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
166 | /// \returns A 128-bit vector of [4 x float] containing the products of both |
167 | /// operands. |
168 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
169 | _mm_mul_ps(__m128 __a, __m128 __b) |
170 | { |
171 | return (__m128)((__v4sf)__a * (__v4sf)__b); |
172 | } |
173 | |
174 | /// Divides the value in the low-order 32 bits of the first operand by |
175 | /// the corresponding value in the second operand. |
176 | /// |
177 | /// \headerfile <x86intrin.h> |
178 | /// |
179 | /// This intrinsic corresponds to the <c> VDIVSS / DIVSS </c> instructions. |
180 | /// |
181 | /// \param __a |
182 | /// A 128-bit vector of [4 x float] containing the dividend. The lower 32 |
183 | /// bits of this operand are used in the calculation. |
184 | /// \param __b |
185 | /// A 128-bit vector of [4 x float] containing the divisor. The lower 32 bits |
186 | /// of this operand are used in the calculation. |
187 | /// \returns A 128-bit vector of [4 x float] containing the quotients of the |
188 | /// lower 32 bits of both operands. The upper 96 bits are copied from the |
189 | /// upper 96 bits of the first source operand. |
190 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
191 | _mm_div_ss(__m128 __a, __m128 __b) |
192 | { |
193 | __a[0] /= __b[0]; |
194 | return __a; |
195 | } |
196 | |
197 | /// Divides two 128-bit vectors of [4 x float]. |
198 | /// |
199 | /// \headerfile <x86intrin.h> |
200 | /// |
201 | /// This intrinsic corresponds to the <c> VDIVPS / DIVPS </c> instructions. |
202 | /// |
203 | /// \param __a |
204 | /// A 128-bit vector of [4 x float] containing the dividend. |
205 | /// \param __b |
206 | /// A 128-bit vector of [4 x float] containing the divisor. |
207 | /// \returns A 128-bit vector of [4 x float] containing the quotients of both |
208 | /// operands. |
209 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
210 | _mm_div_ps(__m128 __a, __m128 __b) |
211 | { |
212 | return (__m128)((__v4sf)__a / (__v4sf)__b); |
213 | } |
214 | |
215 | /// Calculates the square root of the value stored in the low-order bits |
216 | /// of a 128-bit vector of [4 x float]. |
217 | /// |
218 | /// \headerfile <x86intrin.h> |
219 | /// |
220 | /// This intrinsic corresponds to the <c> VSQRTSS / SQRTSS </c> instructions. |
221 | /// |
222 | /// \param __a |
223 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
224 | /// used in the calculation. |
225 | /// \returns A 128-bit vector of [4 x float] containing the square root of the |
226 | /// value in the low-order bits of the operand. |
227 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
228 | _mm_sqrt_ss(__m128 __a) |
229 | { |
230 | return (__m128)__builtin_ia32_sqrtss((__v4sf)__a); |
231 | } |
232 | |
233 | /// Calculates the square roots of the values stored in a 128-bit vector |
234 | /// of [4 x float]. |
235 | /// |
236 | /// \headerfile <x86intrin.h> |
237 | /// |
238 | /// This intrinsic corresponds to the <c> VSQRTPS / SQRTPS </c> instructions. |
239 | /// |
240 | /// \param __a |
241 | /// A 128-bit vector of [4 x float]. |
242 | /// \returns A 128-bit vector of [4 x float] containing the square roots of the |
243 | /// values in the operand. |
244 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
245 | _mm_sqrt_ps(__m128 __a) |
246 | { |
247 | return __builtin_ia32_sqrtps((__v4sf)__a); |
248 | } |
249 | |
250 | /// Calculates the approximate reciprocal of the value stored in the |
251 | /// low-order bits of a 128-bit vector of [4 x float]. |
252 | /// |
253 | /// \headerfile <x86intrin.h> |
254 | /// |
255 | /// This intrinsic corresponds to the <c> VRCPSS / RCPSS </c> instructions. |
256 | /// |
257 | /// \param __a |
258 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
259 | /// used in the calculation. |
260 | /// \returns A 128-bit vector of [4 x float] containing the approximate |
261 | /// reciprocal of the value in the low-order bits of the operand. |
262 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
263 | _mm_rcp_ss(__m128 __a) |
264 | { |
265 | return (__m128)__builtin_ia32_rcpss((__v4sf)__a); |
266 | } |
267 | |
268 | /// Calculates the approximate reciprocals of the values stored in a |
269 | /// 128-bit vector of [4 x float]. |
270 | /// |
271 | /// \headerfile <x86intrin.h> |
272 | /// |
273 | /// This intrinsic corresponds to the <c> VRCPPS / RCPPS </c> instructions. |
274 | /// |
275 | /// \param __a |
276 | /// A 128-bit vector of [4 x float]. |
277 | /// \returns A 128-bit vector of [4 x float] containing the approximate |
278 | /// reciprocals of the values in the operand. |
279 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
280 | _mm_rcp_ps(__m128 __a) |
281 | { |
282 | return (__m128)__builtin_ia32_rcpps((__v4sf)__a); |
283 | } |
284 | |
285 | /// Calculates the approximate reciprocal of the square root of the value |
286 | /// stored in the low-order bits of a 128-bit vector of [4 x float]. |
287 | /// |
288 | /// \headerfile <x86intrin.h> |
289 | /// |
290 | /// This intrinsic corresponds to the <c> VRSQRTSS / RSQRTSS </c> instructions. |
291 | /// |
292 | /// \param __a |
293 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
294 | /// used in the calculation. |
295 | /// \returns A 128-bit vector of [4 x float] containing the approximate |
296 | /// reciprocal of the square root of the value in the low-order bits of the |
297 | /// operand. |
298 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
299 | _mm_rsqrt_ss(__m128 __a) |
300 | { |
301 | return __builtin_ia32_rsqrtss((__v4sf)__a); |
302 | } |
303 | |
304 | /// Calculates the approximate reciprocals of the square roots of the |
305 | /// values stored in a 128-bit vector of [4 x float]. |
306 | /// |
307 | /// \headerfile <x86intrin.h> |
308 | /// |
309 | /// This intrinsic corresponds to the <c> VRSQRTPS / RSQRTPS </c> instructions. |
310 | /// |
311 | /// \param __a |
312 | /// A 128-bit vector of [4 x float]. |
313 | /// \returns A 128-bit vector of [4 x float] containing the approximate |
314 | /// reciprocals of the square roots of the values in the operand. |
315 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
316 | _mm_rsqrt_ps(__m128 __a) |
317 | { |
318 | return __builtin_ia32_rsqrtps((__v4sf)__a); |
319 | } |
320 | |
321 | /// Compares two 32-bit float values in the low-order bits of both |
322 | /// operands and returns the lesser value in the low-order bits of the |
323 | /// vector of [4 x float]. |
324 | /// |
325 | /// \headerfile <x86intrin.h> |
326 | /// |
327 | /// This intrinsic corresponds to the <c> VMINSS / MINSS </c> instructions. |
328 | /// |
329 | /// \param __a |
330 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
331 | /// 32 bits of this operand are used in the comparison. |
332 | /// \param __b |
333 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
334 | /// 32 bits of this operand are used in the comparison. |
335 | /// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the |
336 | /// minimum value between both operands. The upper 96 bits are copied from |
337 | /// the upper 96 bits of the first source operand. |
338 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
339 | _mm_min_ss(__m128 __a, __m128 __b) |
340 | { |
341 | return __builtin_ia32_minss((__v4sf)__a, (__v4sf)__b); |
342 | } |
343 | |
344 | /// Compares two 128-bit vectors of [4 x float] and returns the lesser |
345 | /// of each pair of values. |
346 | /// |
347 | /// \headerfile <x86intrin.h> |
348 | /// |
349 | /// This intrinsic corresponds to the <c> VMINPS / MINPS </c> instructions. |
350 | /// |
351 | /// \param __a |
352 | /// A 128-bit vector of [4 x float] containing one of the operands. |
353 | /// \param __b |
354 | /// A 128-bit vector of [4 x float] containing one of the operands. |
355 | /// \returns A 128-bit vector of [4 x float] containing the minimum values |
356 | /// between both operands. |
357 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
358 | _mm_min_ps(__m128 __a, __m128 __b) |
359 | { |
360 | return __builtin_ia32_minps((__v4sf)__a, (__v4sf)__b); |
361 | } |
362 | |
363 | /// Compares two 32-bit float values in the low-order bits of both |
364 | /// operands and returns the greater value in the low-order bits of a 128-bit |
365 | /// vector of [4 x float]. |
366 | /// |
367 | /// \headerfile <x86intrin.h> |
368 | /// |
369 | /// This intrinsic corresponds to the <c> VMAXSS / MAXSS </c> instructions. |
370 | /// |
371 | /// \param __a |
372 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
373 | /// 32 bits of this operand are used in the comparison. |
374 | /// \param __b |
375 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
376 | /// 32 bits of this operand are used in the comparison. |
377 | /// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the |
378 | /// maximum value between both operands. The upper 96 bits are copied from |
379 | /// the upper 96 bits of the first source operand. |
380 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
381 | _mm_max_ss(__m128 __a, __m128 __b) |
382 | { |
383 | return __builtin_ia32_maxss((__v4sf)__a, (__v4sf)__b); |
384 | } |
385 | |
386 | /// Compares two 128-bit vectors of [4 x float] and returns the greater |
387 | /// of each pair of values. |
388 | /// |
389 | /// \headerfile <x86intrin.h> |
390 | /// |
391 | /// This intrinsic corresponds to the <c> VMAXPS / MAXPS </c> instructions. |
392 | /// |
393 | /// \param __a |
394 | /// A 128-bit vector of [4 x float] containing one of the operands. |
395 | /// \param __b |
396 | /// A 128-bit vector of [4 x float] containing one of the operands. |
397 | /// \returns A 128-bit vector of [4 x float] containing the maximum values |
398 | /// between both operands. |
399 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
400 | _mm_max_ps(__m128 __a, __m128 __b) |
401 | { |
402 | return __builtin_ia32_maxps((__v4sf)__a, (__v4sf)__b); |
403 | } |
404 | |
405 | /// Performs a bitwise AND of two 128-bit vectors of [4 x float]. |
406 | /// |
407 | /// \headerfile <x86intrin.h> |
408 | /// |
409 | /// This intrinsic corresponds to the <c> VANDPS / ANDPS </c> instructions. |
410 | /// |
411 | /// \param __a |
412 | /// A 128-bit vector containing one of the source operands. |
413 | /// \param __b |
414 | /// A 128-bit vector containing one of the source operands. |
415 | /// \returns A 128-bit vector of [4 x float] containing the bitwise AND of the |
416 | /// values between both operands. |
417 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
418 | _mm_and_ps(__m128 __a, __m128 __b) |
419 | { |
420 | return (__m128)((__v4su)__a & (__v4su)__b); |
421 | } |
422 | |
423 | /// Performs a bitwise AND of two 128-bit vectors of [4 x float], using |
424 | /// the one's complement of the values contained in the first source |
425 | /// operand. |
426 | /// |
427 | /// \headerfile <x86intrin.h> |
428 | /// |
429 | /// This intrinsic corresponds to the <c> VANDNPS / ANDNPS </c> instructions. |
430 | /// |
431 | /// \param __a |
432 | /// A 128-bit vector of [4 x float] containing the first source operand. The |
433 | /// one's complement of this value is used in the bitwise AND. |
434 | /// \param __b |
435 | /// A 128-bit vector of [4 x float] containing the second source operand. |
436 | /// \returns A 128-bit vector of [4 x float] containing the bitwise AND of the |
437 | /// one's complement of the first operand and the values in the second |
438 | /// operand. |
439 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
440 | _mm_andnot_ps(__m128 __a, __m128 __b) |
441 | { |
442 | return (__m128)(~(__v4su)__a & (__v4su)__b); |
443 | } |
444 | |
445 | /// Performs a bitwise OR of two 128-bit vectors of [4 x float]. |
446 | /// |
447 | /// \headerfile <x86intrin.h> |
448 | /// |
449 | /// This intrinsic corresponds to the <c> VORPS / ORPS </c> instructions. |
450 | /// |
451 | /// \param __a |
452 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
453 | /// \param __b |
454 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
455 | /// \returns A 128-bit vector of [4 x float] containing the bitwise OR of the |
456 | /// values between both operands. |
457 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
458 | _mm_or_ps(__m128 __a, __m128 __b) |
459 | { |
460 | return (__m128)((__v4su)__a | (__v4su)__b); |
461 | } |
462 | |
463 | /// Performs a bitwise exclusive OR of two 128-bit vectors of |
464 | /// [4 x float]. |
465 | /// |
466 | /// \headerfile <x86intrin.h> |
467 | /// |
468 | /// This intrinsic corresponds to the <c> VXORPS / XORPS </c> instructions. |
469 | /// |
470 | /// \param __a |
471 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
472 | /// \param __b |
473 | /// A 128-bit vector of [4 x float] containing one of the source operands. |
474 | /// \returns A 128-bit vector of [4 x float] containing the bitwise exclusive OR |
475 | /// of the values between both operands. |
476 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
477 | _mm_xor_ps(__m128 __a, __m128 __b) |
478 | { |
479 | return (__m128)((__v4su)__a ^ (__v4su)__b); |
480 | } |
481 | |
482 | /// Compares two 32-bit float values in the low-order bits of both |
483 | /// operands for equality and returns the result of the comparison in the |
484 | /// low-order bits of a vector [4 x float]. |
485 | /// |
486 | /// \headerfile <x86intrin.h> |
487 | /// |
488 | /// This intrinsic corresponds to the <c> VCMPEQSS / CMPEQSS </c> instructions. |
489 | /// |
490 | /// \param __a |
491 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
492 | /// 32 bits of this operand are used in the comparison. |
493 | /// \param __b |
494 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
495 | /// 32 bits of this operand are used in the comparison. |
496 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
497 | /// in the low-order bits. |
498 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
499 | _mm_cmpeq_ss(__m128 __a, __m128 __b) |
500 | { |
501 | return (__m128)__builtin_ia32_cmpeqss((__v4sf)__a, (__v4sf)__b); |
502 | } |
503 | |
504 | /// Compares each of the corresponding 32-bit float values of the |
505 | /// 128-bit vectors of [4 x float] for equality. |
506 | /// |
507 | /// \headerfile <x86intrin.h> |
508 | /// |
509 | /// This intrinsic corresponds to the <c> VCMPEQPS / CMPEQPS </c> instructions. |
510 | /// |
511 | /// \param __a |
512 | /// A 128-bit vector of [4 x float]. |
513 | /// \param __b |
514 | /// A 128-bit vector of [4 x float]. |
515 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
516 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
517 | _mm_cmpeq_ps(__m128 __a, __m128 __b) |
518 | { |
519 | return (__m128)__builtin_ia32_cmpeqps((__v4sf)__a, (__v4sf)__b); |
520 | } |
521 | |
522 | /// Compares two 32-bit float values in the low-order bits of both |
523 | /// operands to determine if the value in the first operand is less than the |
524 | /// corresponding value in the second operand and returns the result of the |
525 | /// comparison in the low-order bits of a vector of [4 x float]. |
526 | /// |
527 | /// \headerfile <x86intrin.h> |
528 | /// |
529 | /// This intrinsic corresponds to the <c> VCMPLTSS / CMPLTSS </c> instructions. |
530 | /// |
531 | /// \param __a |
532 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
533 | /// 32 bits of this operand are used in the comparison. |
534 | /// \param __b |
535 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
536 | /// 32 bits of this operand are used in the comparison. |
537 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
538 | /// in the low-order bits. |
539 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
540 | _mm_cmplt_ss(__m128 __a, __m128 __b) |
541 | { |
542 | return (__m128)__builtin_ia32_cmpltss((__v4sf)__a, (__v4sf)__b); |
543 | } |
544 | |
545 | /// Compares each of the corresponding 32-bit float values of the |
546 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
547 | /// operand are less than those in the second operand. |
548 | /// |
549 | /// \headerfile <x86intrin.h> |
550 | /// |
551 | /// This intrinsic corresponds to the <c> VCMPLTPS / CMPLTPS </c> instructions. |
552 | /// |
553 | /// \param __a |
554 | /// A 128-bit vector of [4 x float]. |
555 | /// \param __b |
556 | /// A 128-bit vector of [4 x float]. |
557 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
558 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
559 | _mm_cmplt_ps(__m128 __a, __m128 __b) |
560 | { |
561 | return (__m128)__builtin_ia32_cmpltps((__v4sf)__a, (__v4sf)__b); |
562 | } |
563 | |
564 | /// Compares two 32-bit float values in the low-order bits of both |
565 | /// operands to determine if the value in the first operand is less than or |
566 | /// equal to the corresponding value in the second operand and returns the |
567 | /// result of the comparison in the low-order bits of a vector of |
568 | /// [4 x float]. |
569 | /// |
570 | /// \headerfile <x86intrin.h> |
571 | /// |
572 | /// This intrinsic corresponds to the <c> VCMPLESS / CMPLESS </c> instructions. |
573 | /// |
574 | /// \param __a |
575 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
576 | /// 32 bits of this operand are used in the comparison. |
577 | /// \param __b |
578 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
579 | /// 32 bits of this operand are used in the comparison. |
580 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
581 | /// in the low-order bits. |
582 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
583 | _mm_cmple_ss(__m128 __a, __m128 __b) |
584 | { |
585 | return (__m128)__builtin_ia32_cmpless((__v4sf)__a, (__v4sf)__b); |
586 | } |
587 | |
588 | /// Compares each of the corresponding 32-bit float values of the |
589 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
590 | /// operand are less than or equal to those in the second operand. |
591 | /// |
592 | /// \headerfile <x86intrin.h> |
593 | /// |
594 | /// This intrinsic corresponds to the <c> VCMPLEPS / CMPLEPS </c> instructions. |
595 | /// |
596 | /// \param __a |
597 | /// A 128-bit vector of [4 x float]. |
598 | /// \param __b |
599 | /// A 128-bit vector of [4 x float]. |
600 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
601 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
602 | _mm_cmple_ps(__m128 __a, __m128 __b) |
603 | { |
604 | return (__m128)__builtin_ia32_cmpleps((__v4sf)__a, (__v4sf)__b); |
605 | } |
606 | |
607 | /// Compares two 32-bit float values in the low-order bits of both |
608 | /// operands to determine if the value in the first operand is greater than |
609 | /// the corresponding value in the second operand and returns the result of |
610 | /// the comparison in the low-order bits of a vector of [4 x float]. |
611 | /// |
612 | /// \headerfile <x86intrin.h> |
613 | /// |
614 | /// This intrinsic corresponds to the <c> VCMPLTSS / CMPLTSS </c> instructions. |
615 | /// |
616 | /// \param __a |
617 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
618 | /// 32 bits of this operand are used in the comparison. |
619 | /// \param __b |
620 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
621 | /// 32 bits of this operand are used in the comparison. |
622 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
623 | /// in the low-order bits. |
624 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
625 | _mm_cmpgt_ss(__m128 __a, __m128 __b) |
626 | { |
627 | return (__m128)__builtin_shufflevector((__v4sf)__a, |
628 | (__v4sf)__builtin_ia32_cmpltss((__v4sf)__b, (__v4sf)__a), |
629 | 4, 1, 2, 3); |
630 | } |
631 | |
632 | /// Compares each of the corresponding 32-bit float values of the |
633 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
634 | /// operand are greater than those in the second operand. |
635 | /// |
636 | /// \headerfile <x86intrin.h> |
637 | /// |
638 | /// This intrinsic corresponds to the <c> VCMPLTPS / CMPLTPS </c> instructions. |
639 | /// |
640 | /// \param __a |
641 | /// A 128-bit vector of [4 x float]. |
642 | /// \param __b |
643 | /// A 128-bit vector of [4 x float]. |
644 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
645 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
646 | _mm_cmpgt_ps(__m128 __a, __m128 __b) |
647 | { |
648 | return (__m128)__builtin_ia32_cmpltps((__v4sf)__b, (__v4sf)__a); |
649 | } |
650 | |
651 | /// Compares two 32-bit float values in the low-order bits of both |
652 | /// operands to determine if the value in the first operand is greater than |
653 | /// or equal to the corresponding value in the second operand and returns |
654 | /// the result of the comparison in the low-order bits of a vector of |
655 | /// [4 x float]. |
656 | /// |
657 | /// \headerfile <x86intrin.h> |
658 | /// |
659 | /// This intrinsic corresponds to the <c> VCMPLESS / CMPLESS </c> instructions. |
660 | /// |
661 | /// \param __a |
662 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
663 | /// 32 bits of this operand are used in the comparison. |
664 | /// \param __b |
665 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
666 | /// 32 bits of this operand are used in the comparison. |
667 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
668 | /// in the low-order bits. |
669 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
670 | _mm_cmpge_ss(__m128 __a, __m128 __b) |
671 | { |
672 | return (__m128)__builtin_shufflevector((__v4sf)__a, |
673 | (__v4sf)__builtin_ia32_cmpless((__v4sf)__b, (__v4sf)__a), |
674 | 4, 1, 2, 3); |
675 | } |
676 | |
677 | /// Compares each of the corresponding 32-bit float values of the |
678 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
679 | /// operand are greater than or equal to those in the second operand. |
680 | /// |
681 | /// \headerfile <x86intrin.h> |
682 | /// |
683 | /// This intrinsic corresponds to the <c> VCMPLEPS / CMPLEPS </c> instructions. |
684 | /// |
685 | /// \param __a |
686 | /// A 128-bit vector of [4 x float]. |
687 | /// \param __b |
688 | /// A 128-bit vector of [4 x float]. |
689 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
690 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
691 | _mm_cmpge_ps(__m128 __a, __m128 __b) |
692 | { |
693 | return (__m128)__builtin_ia32_cmpleps((__v4sf)__b, (__v4sf)__a); |
694 | } |
695 | |
696 | /// Compares two 32-bit float values in the low-order bits of both |
697 | /// operands for inequality and returns the result of the comparison in the |
698 | /// low-order bits of a vector of [4 x float]. |
699 | /// |
700 | /// \headerfile <x86intrin.h> |
701 | /// |
702 | /// This intrinsic corresponds to the <c> VCMPNEQSS / CMPNEQSS </c> |
703 | /// instructions. |
704 | /// |
705 | /// \param __a |
706 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
707 | /// 32 bits of this operand are used in the comparison. |
708 | /// \param __b |
709 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
710 | /// 32 bits of this operand are used in the comparison. |
711 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
712 | /// in the low-order bits. |
713 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
714 | _mm_cmpneq_ss(__m128 __a, __m128 __b) |
715 | { |
716 | return (__m128)__builtin_ia32_cmpneqss((__v4sf)__a, (__v4sf)__b); |
717 | } |
718 | |
719 | /// Compares each of the corresponding 32-bit float values of the |
720 | /// 128-bit vectors of [4 x float] for inequality. |
721 | /// |
722 | /// \headerfile <x86intrin.h> |
723 | /// |
724 | /// This intrinsic corresponds to the <c> VCMPNEQPS / CMPNEQPS </c> |
725 | /// instructions. |
726 | /// |
727 | /// \param __a |
728 | /// A 128-bit vector of [4 x float]. |
729 | /// \param __b |
730 | /// A 128-bit vector of [4 x float]. |
731 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
732 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
733 | _mm_cmpneq_ps(__m128 __a, __m128 __b) |
734 | { |
735 | return (__m128)__builtin_ia32_cmpneqps((__v4sf)__a, (__v4sf)__b); |
736 | } |
737 | |
738 | /// Compares two 32-bit float values in the low-order bits of both |
739 | /// operands to determine if the value in the first operand is not less than |
740 | /// the corresponding value in the second operand and returns the result of |
741 | /// the comparison in the low-order bits of a vector of [4 x float]. |
742 | /// |
743 | /// \headerfile <x86intrin.h> |
744 | /// |
745 | /// This intrinsic corresponds to the <c> VCMPNLTSS / CMPNLTSS </c> |
746 | /// instructions. |
747 | /// |
748 | /// \param __a |
749 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
750 | /// 32 bits of this operand are used in the comparison. |
751 | /// \param __b |
752 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
753 | /// 32 bits of this operand are used in the comparison. |
754 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
755 | /// in the low-order bits. |
756 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
757 | _mm_cmpnlt_ss(__m128 __a, __m128 __b) |
758 | { |
759 | return (__m128)__builtin_ia32_cmpnltss((__v4sf)__a, (__v4sf)__b); |
760 | } |
761 | |
762 | /// Compares each of the corresponding 32-bit float values of the |
763 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
764 | /// operand are not less than those in the second operand. |
765 | /// |
766 | /// \headerfile <x86intrin.h> |
767 | /// |
768 | /// This intrinsic corresponds to the <c> VCMPNLTPS / CMPNLTPS </c> |
769 | /// instructions. |
770 | /// |
771 | /// \param __a |
772 | /// A 128-bit vector of [4 x float]. |
773 | /// \param __b |
774 | /// A 128-bit vector of [4 x float]. |
775 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
776 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
777 | _mm_cmpnlt_ps(__m128 __a, __m128 __b) |
778 | { |
779 | return (__m128)__builtin_ia32_cmpnltps((__v4sf)__a, (__v4sf)__b); |
780 | } |
781 | |
782 | /// Compares two 32-bit float values in the low-order bits of both |
783 | /// operands to determine if the value in the first operand is not less than |
784 | /// or equal to the corresponding value in the second operand and returns |
785 | /// the result of the comparison in the low-order bits of a vector of |
786 | /// [4 x float]. |
787 | /// |
788 | /// \headerfile <x86intrin.h> |
789 | /// |
790 | /// This intrinsic corresponds to the <c> VCMPNLESS / CMPNLESS </c> |
791 | /// instructions. |
792 | /// |
793 | /// \param __a |
794 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
795 | /// 32 bits of this operand are used in the comparison. |
796 | /// \param __b |
797 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
798 | /// 32 bits of this operand are used in the comparison. |
799 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
800 | /// in the low-order bits. |
801 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
802 | _mm_cmpnle_ss(__m128 __a, __m128 __b) |
803 | { |
804 | return (__m128)__builtin_ia32_cmpnless((__v4sf)__a, (__v4sf)__b); |
805 | } |
806 | |
807 | /// Compares each of the corresponding 32-bit float values of the |
808 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
809 | /// operand are not less than or equal to those in the second operand. |
810 | /// |
811 | /// \headerfile <x86intrin.h> |
812 | /// |
813 | /// This intrinsic corresponds to the <c> VCMPNLEPS / CMPNLEPS </c> |
814 | /// instructions. |
815 | /// |
816 | /// \param __a |
817 | /// A 128-bit vector of [4 x float]. |
818 | /// \param __b |
819 | /// A 128-bit vector of [4 x float]. |
820 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
821 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
822 | _mm_cmpnle_ps(__m128 __a, __m128 __b) |
823 | { |
824 | return (__m128)__builtin_ia32_cmpnleps((__v4sf)__a, (__v4sf)__b); |
825 | } |
826 | |
827 | /// Compares two 32-bit float values in the low-order bits of both |
828 | /// operands to determine if the value in the first operand is not greater |
829 | /// than the corresponding value in the second operand and returns the |
830 | /// result of the comparison in the low-order bits of a vector of |
831 | /// [4 x float]. |
832 | /// |
833 | /// \headerfile <x86intrin.h> |
834 | /// |
835 | /// This intrinsic corresponds to the <c> VCMPNLTSS / CMPNLTSS </c> |
836 | /// instructions. |
837 | /// |
838 | /// \param __a |
839 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
840 | /// 32 bits of this operand are used in the comparison. |
841 | /// \param __b |
842 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
843 | /// 32 bits of this operand are used in the comparison. |
844 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
845 | /// in the low-order bits. |
846 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
847 | _mm_cmpngt_ss(__m128 __a, __m128 __b) |
848 | { |
849 | return (__m128)__builtin_shufflevector((__v4sf)__a, |
850 | (__v4sf)__builtin_ia32_cmpnltss((__v4sf)__b, (__v4sf)__a), |
851 | 4, 1, 2, 3); |
852 | } |
853 | |
854 | /// Compares each of the corresponding 32-bit float values of the |
855 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
856 | /// operand are not greater than those in the second operand. |
857 | /// |
858 | /// \headerfile <x86intrin.h> |
859 | /// |
860 | /// This intrinsic corresponds to the <c> VCMPNLTPS / CMPNLTPS </c> |
861 | /// instructions. |
862 | /// |
863 | /// \param __a |
864 | /// A 128-bit vector of [4 x float]. |
865 | /// \param __b |
866 | /// A 128-bit vector of [4 x float]. |
867 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
868 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
869 | _mm_cmpngt_ps(__m128 __a, __m128 __b) |
870 | { |
871 | return (__m128)__builtin_ia32_cmpnltps((__v4sf)__b, (__v4sf)__a); |
872 | } |
873 | |
874 | /// Compares two 32-bit float values in the low-order bits of both |
875 | /// operands to determine if the value in the first operand is not greater |
876 | /// than or equal to the corresponding value in the second operand and |
877 | /// returns the result of the comparison in the low-order bits of a vector |
878 | /// of [4 x float]. |
879 | /// |
880 | /// \headerfile <x86intrin.h> |
881 | /// |
882 | /// This intrinsic corresponds to the <c> VCMPNLESS / CMPNLESS </c> |
883 | /// instructions. |
884 | /// |
885 | /// \param __a |
886 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
887 | /// 32 bits of this operand are used in the comparison. |
888 | /// \param __b |
889 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
890 | /// 32 bits of this operand are used in the comparison. |
891 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
892 | /// in the low-order bits. |
893 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
894 | _mm_cmpnge_ss(__m128 __a, __m128 __b) |
895 | { |
896 | return (__m128)__builtin_shufflevector((__v4sf)__a, |
897 | (__v4sf)__builtin_ia32_cmpnless((__v4sf)__b, (__v4sf)__a), |
898 | 4, 1, 2, 3); |
899 | } |
900 | |
901 | /// Compares each of the corresponding 32-bit float values of the |
902 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
903 | /// operand are not greater than or equal to those in the second operand. |
904 | /// |
905 | /// \headerfile <x86intrin.h> |
906 | /// |
907 | /// This intrinsic corresponds to the <c> VCMPNLEPS / CMPNLEPS </c> |
908 | /// instructions. |
909 | /// |
910 | /// \param __a |
911 | /// A 128-bit vector of [4 x float]. |
912 | /// \param __b |
913 | /// A 128-bit vector of [4 x float]. |
914 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
915 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
916 | _mm_cmpnge_ps(__m128 __a, __m128 __b) |
917 | { |
918 | return (__m128)__builtin_ia32_cmpnleps((__v4sf)__b, (__v4sf)__a); |
919 | } |
920 | |
921 | /// Compares two 32-bit float values in the low-order bits of both |
922 | /// operands to determine if the value in the first operand is ordered with |
923 | /// respect to the corresponding value in the second operand and returns the |
924 | /// result of the comparison in the low-order bits of a vector of |
925 | /// [4 x float]. |
926 | /// |
927 | /// \headerfile <x86intrin.h> |
928 | /// |
929 | /// This intrinsic corresponds to the <c> VCMPORDSS / CMPORDSS </c> |
930 | /// instructions. |
931 | /// |
932 | /// \param __a |
933 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
934 | /// 32 bits of this operand are used in the comparison. |
935 | /// \param __b |
936 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
937 | /// 32 bits of this operand are used in the comparison. |
938 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
939 | /// in the low-order bits. |
940 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
941 | _mm_cmpord_ss(__m128 __a, __m128 __b) |
942 | { |
943 | return (__m128)__builtin_ia32_cmpordss((__v4sf)__a, (__v4sf)__b); |
944 | } |
945 | |
946 | /// Compares each of the corresponding 32-bit float values of the |
947 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
948 | /// operand are ordered with respect to those in the second operand. |
949 | /// |
950 | /// \headerfile <x86intrin.h> |
951 | /// |
952 | /// This intrinsic corresponds to the <c> VCMPORDPS / CMPORDPS </c> |
953 | /// instructions. |
954 | /// |
955 | /// \param __a |
956 | /// A 128-bit vector of [4 x float]. |
957 | /// \param __b |
958 | /// A 128-bit vector of [4 x float]. |
959 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
960 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
961 | _mm_cmpord_ps(__m128 __a, __m128 __b) |
962 | { |
963 | return (__m128)__builtin_ia32_cmpordps((__v4sf)__a, (__v4sf)__b); |
964 | } |
965 | |
966 | /// Compares two 32-bit float values in the low-order bits of both |
967 | /// operands to determine if the value in the first operand is unordered |
968 | /// with respect to the corresponding value in the second operand and |
969 | /// returns the result of the comparison in the low-order bits of a vector |
970 | /// of [4 x float]. |
971 | /// |
972 | /// \headerfile <x86intrin.h> |
973 | /// |
974 | /// This intrinsic corresponds to the <c> VCMPUNORDSS / CMPUNORDSS </c> |
975 | /// instructions. |
976 | /// |
977 | /// \param __a |
978 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
979 | /// 32 bits of this operand are used in the comparison. |
980 | /// \param __b |
981 | /// A 128-bit vector of [4 x float] containing one of the operands. The lower |
982 | /// 32 bits of this operand are used in the comparison. |
983 | /// \returns A 128-bit vector of [4 x float] containing the comparison results |
984 | /// in the low-order bits. |
985 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
986 | _mm_cmpunord_ss(__m128 __a, __m128 __b) |
987 | { |
988 | return (__m128)__builtin_ia32_cmpunordss((__v4sf)__a, (__v4sf)__b); |
989 | } |
990 | |
991 | /// Compares each of the corresponding 32-bit float values of the |
992 | /// 128-bit vectors of [4 x float] to determine if the values in the first |
993 | /// operand are unordered with respect to those in the second operand. |
994 | /// |
995 | /// \headerfile <x86intrin.h> |
996 | /// |
997 | /// This intrinsic corresponds to the <c> VCMPUNORDPS / CMPUNORDPS </c> |
998 | /// instructions. |
999 | /// |
1000 | /// \param __a |
1001 | /// A 128-bit vector of [4 x float]. |
1002 | /// \param __b |
1003 | /// A 128-bit vector of [4 x float]. |
1004 | /// \returns A 128-bit vector of [4 x float] containing the comparison results. |
1005 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1006 | _mm_cmpunord_ps(__m128 __a, __m128 __b) |
1007 | { |
1008 | return (__m128)__builtin_ia32_cmpunordps((__v4sf)__a, (__v4sf)__b); |
1009 | } |
1010 | |
1011 | /// Compares two 32-bit float values in the low-order bits of both |
1012 | /// operands for equality and returns the result of the comparison. |
1013 | /// |
1014 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1015 | /// |
1016 | /// \headerfile <x86intrin.h> |
1017 | /// |
1018 | /// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> |
1019 | /// instructions. |
1020 | /// |
1021 | /// \param __a |
1022 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1023 | /// used in the comparison. |
1024 | /// \param __b |
1025 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1026 | /// used in the comparison. |
1027 | /// \returns An integer containing the comparison results. If either of the |
1028 | /// two lower 32-bit values is NaN, 0 is returned. |
1029 | static __inline__ int __DEFAULT_FN_ATTRS |
1030 | _mm_comieq_ss(__m128 __a, __m128 __b) |
1031 | { |
1032 | return __builtin_ia32_comieq((__v4sf)__a, (__v4sf)__b); |
1033 | } |
1034 | |
1035 | /// Compares two 32-bit float values in the low-order bits of both |
1036 | /// operands to determine if the first operand is less than the second |
1037 | /// operand and returns the result of the comparison. |
1038 | /// |
1039 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1040 | /// |
1041 | /// \headerfile <x86intrin.h> |
1042 | /// |
1043 | /// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> |
1044 | /// instructions. |
1045 | /// |
1046 | /// \param __a |
1047 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1048 | /// used in the comparison. |
1049 | /// \param __b |
1050 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1051 | /// used in the comparison. |
1052 | /// \returns An integer containing the comparison results. If either of the two |
1053 | /// lower 32-bit values is NaN, 0 is returned. |
1054 | static __inline__ int __DEFAULT_FN_ATTRS |
1055 | _mm_comilt_ss(__m128 __a, __m128 __b) |
1056 | { |
1057 | return __builtin_ia32_comilt((__v4sf)__a, (__v4sf)__b); |
1058 | } |
1059 | |
1060 | /// Compares two 32-bit float values in the low-order bits of both |
1061 | /// operands to determine if the first operand is less than or equal to the |
1062 | /// second operand and returns the result of the comparison. |
1063 | /// |
1064 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1065 | /// |
1066 | /// \headerfile <x86intrin.h> |
1067 | /// |
1068 | /// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> instructions. |
1069 | /// |
1070 | /// \param __a |
1071 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1072 | /// used in the comparison. |
1073 | /// \param __b |
1074 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1075 | /// used in the comparison. |
1076 | /// \returns An integer containing the comparison results. If either of the two |
1077 | /// lower 32-bit values is NaN, 0 is returned. |
1078 | static __inline__ int __DEFAULT_FN_ATTRS |
1079 | _mm_comile_ss(__m128 __a, __m128 __b) |
1080 | { |
1081 | return __builtin_ia32_comile((__v4sf)__a, (__v4sf)__b); |
1082 | } |
1083 | |
1084 | /// Compares two 32-bit float values in the low-order bits of both |
1085 | /// operands to determine if the first operand is greater than the second |
1086 | /// operand and returns the result of the comparison. |
1087 | /// |
1088 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1089 | /// |
1090 | /// \headerfile <x86intrin.h> |
1091 | /// |
1092 | /// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> instructions. |
1093 | /// |
1094 | /// \param __a |
1095 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1096 | /// used in the comparison. |
1097 | /// \param __b |
1098 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1099 | /// used in the comparison. |
1100 | /// \returns An integer containing the comparison results. If either of the |
1101 | /// two lower 32-bit values is NaN, 0 is returned. |
1102 | static __inline__ int __DEFAULT_FN_ATTRS |
1103 | _mm_comigt_ss(__m128 __a, __m128 __b) |
1104 | { |
1105 | return __builtin_ia32_comigt((__v4sf)__a, (__v4sf)__b); |
1106 | } |
1107 | |
1108 | /// Compares two 32-bit float values in the low-order bits of both |
1109 | /// operands to determine if the first operand is greater than or equal to |
1110 | /// the second operand and returns the result of the comparison. |
1111 | /// |
1112 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1113 | /// |
1114 | /// \headerfile <x86intrin.h> |
1115 | /// |
1116 | /// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> instructions. |
1117 | /// |
1118 | /// \param __a |
1119 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1120 | /// used in the comparison. |
1121 | /// \param __b |
1122 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1123 | /// used in the comparison. |
1124 | /// \returns An integer containing the comparison results. If either of the two |
1125 | /// lower 32-bit values is NaN, 0 is returned. |
1126 | static __inline__ int __DEFAULT_FN_ATTRS |
1127 | _mm_comige_ss(__m128 __a, __m128 __b) |
1128 | { |
1129 | return __builtin_ia32_comige((__v4sf)__a, (__v4sf)__b); |
1130 | } |
1131 | |
1132 | /// Compares two 32-bit float values in the low-order bits of both |
1133 | /// operands to determine if the first operand is not equal to the second |
1134 | /// operand and returns the result of the comparison. |
1135 | /// |
1136 | /// If either of the two lower 32-bit values is NaN, 1 is returned. |
1137 | /// |
1138 | /// \headerfile <x86intrin.h> |
1139 | /// |
1140 | /// This intrinsic corresponds to the <c> VCOMISS / COMISS </c> instructions. |
1141 | /// |
1142 | /// \param __a |
1143 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1144 | /// used in the comparison. |
1145 | /// \param __b |
1146 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1147 | /// used in the comparison. |
1148 | /// \returns An integer containing the comparison results. If either of the |
1149 | /// two lower 32-bit values is NaN, 1 is returned. |
1150 | static __inline__ int __DEFAULT_FN_ATTRS |
1151 | _mm_comineq_ss(__m128 __a, __m128 __b) |
1152 | { |
1153 | return __builtin_ia32_comineq((__v4sf)__a, (__v4sf)__b); |
1154 | } |
1155 | |
1156 | /// Performs an unordered comparison of two 32-bit float values using |
1157 | /// the low-order bits of both operands to determine equality and returns |
1158 | /// the result of the comparison. |
1159 | /// |
1160 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1161 | /// |
1162 | /// \headerfile <x86intrin.h> |
1163 | /// |
1164 | /// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. |
1165 | /// |
1166 | /// \param __a |
1167 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1168 | /// used in the comparison. |
1169 | /// \param __b |
1170 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1171 | /// used in the comparison. |
1172 | /// \returns An integer containing the comparison results. If either of the two |
1173 | /// lower 32-bit values is NaN, 0 is returned. |
1174 | static __inline__ int __DEFAULT_FN_ATTRS |
1175 | _mm_ucomieq_ss(__m128 __a, __m128 __b) |
1176 | { |
1177 | return __builtin_ia32_ucomieq((__v4sf)__a, (__v4sf)__b); |
1178 | } |
1179 | |
1180 | /// Performs an unordered comparison of two 32-bit float values using |
1181 | /// the low-order bits of both operands to determine if the first operand is |
1182 | /// less than the second operand and returns the result of the comparison. |
1183 | /// |
1184 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1185 | /// |
1186 | /// \headerfile <x86intrin.h> |
1187 | /// |
1188 | /// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. |
1189 | /// |
1190 | /// \param __a |
1191 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1192 | /// used in the comparison. |
1193 | /// \param __b |
1194 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1195 | /// used in the comparison. |
1196 | /// \returns An integer containing the comparison results. If either of the two |
1197 | /// lower 32-bit values is NaN, 0 is returned. |
1198 | static __inline__ int __DEFAULT_FN_ATTRS |
1199 | _mm_ucomilt_ss(__m128 __a, __m128 __b) |
1200 | { |
1201 | return __builtin_ia32_ucomilt((__v4sf)__a, (__v4sf)__b); |
1202 | } |
1203 | |
1204 | /// Performs an unordered comparison of two 32-bit float values using |
1205 | /// the low-order bits of both operands to determine if the first operand is |
1206 | /// less than or equal to the second operand and returns the result of the |
1207 | /// comparison. |
1208 | /// |
1209 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1210 | /// |
1211 | /// \headerfile <x86intrin.h> |
1212 | /// |
1213 | /// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. |
1214 | /// |
1215 | /// \param __a |
1216 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1217 | /// used in the comparison. |
1218 | /// \param __b |
1219 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1220 | /// used in the comparison. |
1221 | /// \returns An integer containing the comparison results. If either of the two |
1222 | /// lower 32-bit values is NaN, 0 is returned. |
1223 | static __inline__ int __DEFAULT_FN_ATTRS |
1224 | _mm_ucomile_ss(__m128 __a, __m128 __b) |
1225 | { |
1226 | return __builtin_ia32_ucomile((__v4sf)__a, (__v4sf)__b); |
1227 | } |
1228 | |
1229 | /// Performs an unordered comparison of two 32-bit float values using |
1230 | /// the low-order bits of both operands to determine if the first operand is |
1231 | /// greater than the second operand and returns the result of the |
1232 | /// comparison. |
1233 | /// |
1234 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1235 | /// |
1236 | /// \headerfile <x86intrin.h> |
1237 | /// |
1238 | /// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. |
1239 | /// |
1240 | /// \param __a |
1241 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1242 | /// used in the comparison. |
1243 | /// \param __b |
1244 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1245 | /// used in the comparison. |
1246 | /// \returns An integer containing the comparison results. If either of the two |
1247 | /// lower 32-bit values is NaN, 0 is returned. |
1248 | static __inline__ int __DEFAULT_FN_ATTRS |
1249 | _mm_ucomigt_ss(__m128 __a, __m128 __b) |
1250 | { |
1251 | return __builtin_ia32_ucomigt((__v4sf)__a, (__v4sf)__b); |
1252 | } |
1253 | |
1254 | /// Performs an unordered comparison of two 32-bit float values using |
1255 | /// the low-order bits of both operands to determine if the first operand is |
1256 | /// greater than or equal to the second operand and returns the result of |
1257 | /// the comparison. |
1258 | /// |
1259 | /// If either of the two lower 32-bit values is NaN, 0 is returned. |
1260 | /// |
1261 | /// \headerfile <x86intrin.h> |
1262 | /// |
1263 | /// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. |
1264 | /// |
1265 | /// \param __a |
1266 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1267 | /// used in the comparison. |
1268 | /// \param __b |
1269 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1270 | /// used in the comparison. |
1271 | /// \returns An integer containing the comparison results. If either of the two |
1272 | /// lower 32-bit values is NaN, 0 is returned. |
1273 | static __inline__ int __DEFAULT_FN_ATTRS |
1274 | _mm_ucomige_ss(__m128 __a, __m128 __b) |
1275 | { |
1276 | return __builtin_ia32_ucomige((__v4sf)__a, (__v4sf)__b); |
1277 | } |
1278 | |
1279 | /// Performs an unordered comparison of two 32-bit float values using |
1280 | /// the low-order bits of both operands to determine inequality and returns |
1281 | /// the result of the comparison. |
1282 | /// |
1283 | /// If either of the two lower 32-bit values is NaN, 1 is returned. |
1284 | /// |
1285 | /// \headerfile <x86intrin.h> |
1286 | /// |
1287 | /// This intrinsic corresponds to the <c> VUCOMISS / UCOMISS </c> instructions. |
1288 | /// |
1289 | /// \param __a |
1290 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1291 | /// used in the comparison. |
1292 | /// \param __b |
1293 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1294 | /// used in the comparison. |
1295 | /// \returns An integer containing the comparison results. If either of the two |
1296 | /// lower 32-bit values is NaN, 1 is returned. |
1297 | static __inline__ int __DEFAULT_FN_ATTRS |
1298 | _mm_ucomineq_ss(__m128 __a, __m128 __b) |
1299 | { |
1300 | return __builtin_ia32_ucomineq((__v4sf)__a, (__v4sf)__b); |
1301 | } |
1302 | |
1303 | /// Converts a float value contained in the lower 32 bits of a vector of |
1304 | /// [4 x float] into a 32-bit integer. |
1305 | /// |
1306 | /// \headerfile <x86intrin.h> |
1307 | /// |
1308 | /// This intrinsic corresponds to the <c> VCVTSS2SI / CVTSS2SI </c> |
1309 | /// instructions. |
1310 | /// |
1311 | /// \param __a |
1312 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1313 | /// used in the conversion. |
1314 | /// \returns A 32-bit integer containing the converted value. |
1315 | static __inline__ int __DEFAULT_FN_ATTRS |
1316 | _mm_cvtss_si32(__m128 __a) |
1317 | { |
1318 | return __builtin_ia32_cvtss2si((__v4sf)__a); |
1319 | } |
1320 | |
1321 | /// Converts a float value contained in the lower 32 bits of a vector of |
1322 | /// [4 x float] into a 32-bit integer. |
1323 | /// |
1324 | /// \headerfile <x86intrin.h> |
1325 | /// |
1326 | /// This intrinsic corresponds to the <c> VCVTSS2SI / CVTSS2SI </c> |
1327 | /// instructions. |
1328 | /// |
1329 | /// \param __a |
1330 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1331 | /// used in the conversion. |
1332 | /// \returns A 32-bit integer containing the converted value. |
1333 | static __inline__ int __DEFAULT_FN_ATTRS |
1334 | _mm_cvt_ss2si(__m128 __a) |
1335 | { |
1336 | return _mm_cvtss_si32(__a); |
1337 | } |
1338 | |
1339 | #ifdef __x86_64__ |
1340 | |
1341 | /// Converts a float value contained in the lower 32 bits of a vector of |
1342 | /// [4 x float] into a 64-bit integer. |
1343 | /// |
1344 | /// \headerfile <x86intrin.h> |
1345 | /// |
1346 | /// This intrinsic corresponds to the <c> VCVTSS2SI / CVTSS2SI </c> |
1347 | /// instructions. |
1348 | /// |
1349 | /// \param __a |
1350 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1351 | /// used in the conversion. |
1352 | /// \returns A 64-bit integer containing the converted value. |
1353 | static __inline__ long long __DEFAULT_FN_ATTRS |
1354 | _mm_cvtss_si64(__m128 __a) |
1355 | { |
1356 | return __builtin_ia32_cvtss2si64((__v4sf)__a); |
1357 | } |
1358 | |
1359 | #endif |
1360 | |
1361 | /// Converts two low-order float values in a 128-bit vector of |
1362 | /// [4 x float] into a 64-bit vector of [2 x i32]. |
1363 | /// |
1364 | /// \headerfile <x86intrin.h> |
1365 | /// |
1366 | /// This intrinsic corresponds to the <c> CVTPS2PI </c> instruction. |
1367 | /// |
1368 | /// \param __a |
1369 | /// A 128-bit vector of [4 x float]. |
1370 | /// \returns A 64-bit integer vector containing the converted values. |
1371 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
1372 | _mm_cvtps_pi32(__m128 __a) |
1373 | { |
1374 | return (__m64)__builtin_ia32_cvtps2pi((__v4sf)__a); |
1375 | } |
1376 | |
1377 | /// Converts two low-order float values in a 128-bit vector of |
1378 | /// [4 x float] into a 64-bit vector of [2 x i32]. |
1379 | /// |
1380 | /// \headerfile <x86intrin.h> |
1381 | /// |
1382 | /// This intrinsic corresponds to the <c> CVTPS2PI </c> instruction. |
1383 | /// |
1384 | /// \param __a |
1385 | /// A 128-bit vector of [4 x float]. |
1386 | /// \returns A 64-bit integer vector containing the converted values. |
1387 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
1388 | _mm_cvt_ps2pi(__m128 __a) |
1389 | { |
1390 | return _mm_cvtps_pi32(__a); |
1391 | } |
1392 | |
1393 | /// Converts a float value contained in the lower 32 bits of a vector of |
1394 | /// [4 x float] into a 32-bit integer, truncating the result when it is |
1395 | /// inexact. |
1396 | /// |
1397 | /// \headerfile <x86intrin.h> |
1398 | /// |
1399 | /// This intrinsic corresponds to the <c> VCVTTSS2SI / CVTTSS2SI </c> |
1400 | /// instructions. |
1401 | /// |
1402 | /// \param __a |
1403 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1404 | /// used in the conversion. |
1405 | /// \returns A 32-bit integer containing the converted value. |
1406 | static __inline__ int __DEFAULT_FN_ATTRS |
1407 | _mm_cvttss_si32(__m128 __a) |
1408 | { |
1409 | return __builtin_ia32_cvttss2si((__v4sf)__a); |
1410 | } |
1411 | |
1412 | /// Converts a float value contained in the lower 32 bits of a vector of |
1413 | /// [4 x float] into a 32-bit integer, truncating the result when it is |
1414 | /// inexact. |
1415 | /// |
1416 | /// \headerfile <x86intrin.h> |
1417 | /// |
1418 | /// This intrinsic corresponds to the <c> VCVTTSS2SI / CVTTSS2SI </c> |
1419 | /// instructions. |
1420 | /// |
1421 | /// \param __a |
1422 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1423 | /// used in the conversion. |
1424 | /// \returns A 32-bit integer containing the converted value. |
1425 | static __inline__ int __DEFAULT_FN_ATTRS |
1426 | _mm_cvtt_ss2si(__m128 __a) |
1427 | { |
1428 | return _mm_cvttss_si32(__a); |
1429 | } |
1430 | |
1431 | #ifdef __x86_64__ |
1432 | /// Converts a float value contained in the lower 32 bits of a vector of |
1433 | /// [4 x float] into a 64-bit integer, truncating the result when it is |
1434 | /// inexact. |
1435 | /// |
1436 | /// \headerfile <x86intrin.h> |
1437 | /// |
1438 | /// This intrinsic corresponds to the <c> VCVTTSS2SI / CVTTSS2SI </c> |
1439 | /// instructions. |
1440 | /// |
1441 | /// \param __a |
1442 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1443 | /// used in the conversion. |
1444 | /// \returns A 64-bit integer containing the converted value. |
1445 | static __inline__ long long __DEFAULT_FN_ATTRS |
1446 | _mm_cvttss_si64(__m128 __a) |
1447 | { |
1448 | return __builtin_ia32_cvttss2si64((__v4sf)__a); |
1449 | } |
1450 | #endif |
1451 | |
1452 | /// Converts two low-order float values in a 128-bit vector of |
1453 | /// [4 x float] into a 64-bit vector of [2 x i32], truncating the result |
1454 | /// when it is inexact. |
1455 | /// |
1456 | /// \headerfile <x86intrin.h> |
1457 | /// |
1458 | /// This intrinsic corresponds to the <c> CVTTPS2PI / VTTPS2PI </c> |
1459 | /// instructions. |
1460 | /// |
1461 | /// \param __a |
1462 | /// A 128-bit vector of [4 x float]. |
1463 | /// \returns A 64-bit integer vector containing the converted values. |
1464 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
1465 | _mm_cvttps_pi32(__m128 __a) |
1466 | { |
1467 | return (__m64)__builtin_ia32_cvttps2pi((__v4sf)__a); |
1468 | } |
1469 | |
1470 | /// Converts two low-order float values in a 128-bit vector of [4 x |
1471 | /// float] into a 64-bit vector of [2 x i32], truncating the result when it |
1472 | /// is inexact. |
1473 | /// |
1474 | /// \headerfile <x86intrin.h> |
1475 | /// |
1476 | /// This intrinsic corresponds to the <c> CVTTPS2PI </c> instruction. |
1477 | /// |
1478 | /// \param __a |
1479 | /// A 128-bit vector of [4 x float]. |
1480 | /// \returns A 64-bit integer vector containing the converted values. |
1481 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
1482 | _mm_cvtt_ps2pi(__m128 __a) |
1483 | { |
1484 | return _mm_cvttps_pi32(__a); |
1485 | } |
1486 | |
1487 | /// Converts a 32-bit signed integer value into a floating point value |
1488 | /// and writes it to the lower 32 bits of the destination. The remaining |
1489 | /// higher order elements of the destination vector are copied from the |
1490 | /// corresponding elements in the first operand. |
1491 | /// |
1492 | /// \headerfile <x86intrin.h> |
1493 | /// |
1494 | /// This intrinsic corresponds to the <c> VCVTSI2SS / CVTSI2SS </c> instruction. |
1495 | /// |
1496 | /// \param __a |
1497 | /// A 128-bit vector of [4 x float]. |
1498 | /// \param __b |
1499 | /// A 32-bit signed integer operand containing the value to be converted. |
1500 | /// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the |
1501 | /// converted value of the second operand. The upper 96 bits are copied from |
1502 | /// the upper 96 bits of the first operand. |
1503 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1504 | _mm_cvtsi32_ss(__m128 __a, int __b) |
1505 | { |
1506 | __a[0] = __b; |
1507 | return __a; |
1508 | } |
1509 | |
1510 | /// Converts a 32-bit signed integer value into a floating point value |
1511 | /// and writes it to the lower 32 bits of the destination. The remaining |
1512 | /// higher order elements of the destination are copied from the |
1513 | /// corresponding elements in the first operand. |
1514 | /// |
1515 | /// \headerfile <x86intrin.h> |
1516 | /// |
1517 | /// This intrinsic corresponds to the <c> VCVTSI2SS / CVTSI2SS </c> instruction. |
1518 | /// |
1519 | /// \param __a |
1520 | /// A 128-bit vector of [4 x float]. |
1521 | /// \param __b |
1522 | /// A 32-bit signed integer operand containing the value to be converted. |
1523 | /// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the |
1524 | /// converted value of the second operand. The upper 96 bits are copied from |
1525 | /// the upper 96 bits of the first operand. |
1526 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1527 | _mm_cvt_si2ss(__m128 __a, int __b) |
1528 | { |
1529 | return _mm_cvtsi32_ss(__a, __b); |
1530 | } |
1531 | |
1532 | #ifdef __x86_64__ |
1533 | |
1534 | /// Converts a 64-bit signed integer value into a floating point value |
1535 | /// and writes it to the lower 32 bits of the destination. The remaining |
1536 | /// higher order elements of the destination are copied from the |
1537 | /// corresponding elements in the first operand. |
1538 | /// |
1539 | /// \headerfile <x86intrin.h> |
1540 | /// |
1541 | /// This intrinsic corresponds to the <c> VCVTSI2SS / CVTSI2SS </c> instruction. |
1542 | /// |
1543 | /// \param __a |
1544 | /// A 128-bit vector of [4 x float]. |
1545 | /// \param __b |
1546 | /// A 64-bit signed integer operand containing the value to be converted. |
1547 | /// \returns A 128-bit vector of [4 x float] whose lower 32 bits contain the |
1548 | /// converted value of the second operand. The upper 96 bits are copied from |
1549 | /// the upper 96 bits of the first operand. |
1550 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1551 | _mm_cvtsi64_ss(__m128 __a, long long __b) |
1552 | { |
1553 | __a[0] = __b; |
1554 | return __a; |
1555 | } |
1556 | |
1557 | #endif |
1558 | |
1559 | /// Converts two elements of a 64-bit vector of [2 x i32] into two |
1560 | /// floating point values and writes them to the lower 64-bits of the |
1561 | /// destination. The remaining higher order elements of the destination are |
1562 | /// copied from the corresponding elements in the first operand. |
1563 | /// |
1564 | /// \headerfile <x86intrin.h> |
1565 | /// |
1566 | /// This intrinsic corresponds to the <c> CVTPI2PS </c> instruction. |
1567 | /// |
1568 | /// \param __a |
1569 | /// A 128-bit vector of [4 x float]. |
1570 | /// \param __b |
1571 | /// A 64-bit vector of [2 x i32]. The elements in this vector are converted |
1572 | /// and written to the corresponding low-order elements in the destination. |
1573 | /// \returns A 128-bit vector of [4 x float] whose lower 64 bits contain the |
1574 | /// converted value of the second operand. The upper 64 bits are copied from |
1575 | /// the upper 64 bits of the first operand. |
1576 | static __inline__ __m128 __DEFAULT_FN_ATTRS_MMX |
1577 | _mm_cvtpi32_ps(__m128 __a, __m64 __b) |
1578 | { |
1579 | return __builtin_ia32_cvtpi2ps((__v4sf)__a, (__v2si)__b); |
1580 | } |
1581 | |
1582 | /// Converts two elements of a 64-bit vector of [2 x i32] into two |
1583 | /// floating point values and writes them to the lower 64-bits of the |
1584 | /// destination. The remaining higher order elements of the destination are |
1585 | /// copied from the corresponding elements in the first operand. |
1586 | /// |
1587 | /// \headerfile <x86intrin.h> |
1588 | /// |
1589 | /// This intrinsic corresponds to the <c> CVTPI2PS </c> instruction. |
1590 | /// |
1591 | /// \param __a |
1592 | /// A 128-bit vector of [4 x float]. |
1593 | /// \param __b |
1594 | /// A 64-bit vector of [2 x i32]. The elements in this vector are converted |
1595 | /// and written to the corresponding low-order elements in the destination. |
1596 | /// \returns A 128-bit vector of [4 x float] whose lower 64 bits contain the |
1597 | /// converted value from the second operand. The upper 64 bits are copied |
1598 | /// from the upper 64 bits of the first operand. |
1599 | static __inline__ __m128 __DEFAULT_FN_ATTRS_MMX |
1600 | _mm_cvt_pi2ps(__m128 __a, __m64 __b) |
1601 | { |
1602 | return _mm_cvtpi32_ps(__a, __b); |
1603 | } |
1604 | |
1605 | /// Extracts a float value contained in the lower 32 bits of a vector of |
1606 | /// [4 x float]. |
1607 | /// |
1608 | /// \headerfile <x86intrin.h> |
1609 | /// |
1610 | /// This intrinsic has no corresponding instruction. |
1611 | /// |
1612 | /// \param __a |
1613 | /// A 128-bit vector of [4 x float]. The lower 32 bits of this operand are |
1614 | /// used in the extraction. |
1615 | /// \returns A 32-bit float containing the extracted value. |
1616 | static __inline__ float __DEFAULT_FN_ATTRS |
1617 | _mm_cvtss_f32(__m128 __a) |
1618 | { |
1619 | return __a[0]; |
1620 | } |
1621 | |
1622 | /// Loads two packed float values from the address \a __p into the |
1623 | /// high-order bits of a 128-bit vector of [4 x float]. The low-order bits |
1624 | /// are copied from the low-order bits of the first operand. |
1625 | /// |
1626 | /// \headerfile <x86intrin.h> |
1627 | /// |
1628 | /// This intrinsic corresponds to the <c> VMOVHPD / MOVHPD </c> instruction. |
1629 | /// |
1630 | /// \param __a |
1631 | /// A 128-bit vector of [4 x float]. Bits [63:0] are written to bits [63:0] |
1632 | /// of the destination. |
1633 | /// \param __p |
1634 | /// A pointer to two packed float values. Bits [63:0] are written to bits |
1635 | /// [127:64] of the destination. |
1636 | /// \returns A 128-bit vector of [4 x float] containing the moved values. |
1637 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1638 | _mm_loadh_pi(__m128 __a, const __m64 *__p) |
1639 | { |
1640 | typedef float __mm_loadh_pi_v2f32 __attribute__((__vector_size__(8))); |
1641 | struct __mm_loadh_pi_struct { |
1642 | __mm_loadh_pi_v2f32 __u; |
1643 | } __attribute__((__packed__, __may_alias__)); |
1644 | __mm_loadh_pi_v2f32 __b = ((struct __mm_loadh_pi_struct*)__p)->__u; |
1645 | __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1); |
1646 | return __builtin_shufflevector(__a, __bb, 0, 1, 4, 5); |
1647 | } |
1648 | |
1649 | /// Loads two packed float values from the address \a __p into the |
1650 | /// low-order bits of a 128-bit vector of [4 x float]. The high-order bits |
1651 | /// are copied from the high-order bits of the first operand. |
1652 | /// |
1653 | /// \headerfile <x86intrin.h> |
1654 | /// |
1655 | /// This intrinsic corresponds to the <c> VMOVLPD / MOVLPD </c> instruction. |
1656 | /// |
1657 | /// \param __a |
1658 | /// A 128-bit vector of [4 x float]. Bits [127:64] are written to bits |
1659 | /// [127:64] of the destination. |
1660 | /// \param __p |
1661 | /// A pointer to two packed float values. Bits [63:0] are written to bits |
1662 | /// [63:0] of the destination. |
1663 | /// \returns A 128-bit vector of [4 x float] containing the moved values. |
1664 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1665 | _mm_loadl_pi(__m128 __a, const __m64 *__p) |
1666 | { |
1667 | typedef float __mm_loadl_pi_v2f32 __attribute__((__vector_size__(8))); |
1668 | struct __mm_loadl_pi_struct { |
1669 | __mm_loadl_pi_v2f32 __u; |
1670 | } __attribute__((__packed__, __may_alias__)); |
1671 | __mm_loadl_pi_v2f32 __b = ((struct __mm_loadl_pi_struct*)__p)->__u; |
1672 | __m128 __bb = __builtin_shufflevector(__b, __b, 0, 1, 0, 1); |
1673 | return __builtin_shufflevector(__a, __bb, 4, 5, 2, 3); |
1674 | } |
1675 | |
1676 | /// Constructs a 128-bit floating-point vector of [4 x float]. The lower |
1677 | /// 32 bits of the vector are initialized with the single-precision |
1678 | /// floating-point value loaded from a specified memory location. The upper |
1679 | /// 96 bits are set to zero. |
1680 | /// |
1681 | /// \headerfile <x86intrin.h> |
1682 | /// |
1683 | /// This intrinsic corresponds to the <c> VMOVSS / MOVSS </c> instruction. |
1684 | /// |
1685 | /// \param __p |
1686 | /// A pointer to a 32-bit memory location containing a single-precision |
1687 | /// floating-point value. |
1688 | /// \returns An initialized 128-bit floating-point vector of [4 x float]. The |
1689 | /// lower 32 bits contain the value loaded from the memory location. The |
1690 | /// upper 96 bits are set to zero. |
1691 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1692 | _mm_load_ss(const float *__p) |
1693 | { |
1694 | struct __mm_load_ss_struct { |
1695 | float __u; |
1696 | } __attribute__((__packed__, __may_alias__)); |
1697 | float __u = ((struct __mm_load_ss_struct*)__p)->__u; |
1698 | return __extension__ (__m128){ __u, 0, 0, 0 }; |
1699 | } |
1700 | |
1701 | /// Loads a 32-bit float value and duplicates it to all four vector |
1702 | /// elements of a 128-bit vector of [4 x float]. |
1703 | /// |
1704 | /// \headerfile <x86intrin.h> |
1705 | /// |
1706 | /// This intrinsic corresponds to the <c> VBROADCASTSS / MOVSS + shuffling </c> |
1707 | /// instruction. |
1708 | /// |
1709 | /// \param __p |
1710 | /// A pointer to a float value to be loaded and duplicated. |
1711 | /// \returns A 128-bit vector of [4 x float] containing the loaded and |
1712 | /// duplicated values. |
1713 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1714 | _mm_load1_ps(const float *__p) |
1715 | { |
1716 | struct __mm_load1_ps_struct { |
1717 | float __u; |
1718 | } __attribute__((__packed__, __may_alias__)); |
1719 | float __u = ((struct __mm_load1_ps_struct*)__p)->__u; |
1720 | return __extension__ (__m128){ __u, __u, __u, __u }; |
1721 | } |
1722 | |
1723 | #define _mm_load_ps1(p) _mm_load1_ps(p) |
1724 | |
1725 | /// Loads a 128-bit floating-point vector of [4 x float] from an aligned |
1726 | /// memory location. |
1727 | /// |
1728 | /// \headerfile <x86intrin.h> |
1729 | /// |
1730 | /// This intrinsic corresponds to the <c> VMOVAPS / MOVAPS </c> instruction. |
1731 | /// |
1732 | /// \param __p |
1733 | /// A pointer to a 128-bit memory location. The address of the memory |
1734 | /// location has to be 128-bit aligned. |
1735 | /// \returns A 128-bit vector of [4 x float] containing the loaded values. |
1736 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1737 | _mm_load_ps(const float *__p) |
1738 | { |
1739 | return *(__m128*)__p; |
1740 | } |
1741 | |
1742 | /// Loads a 128-bit floating-point vector of [4 x float] from an |
1743 | /// unaligned memory location. |
1744 | /// |
1745 | /// \headerfile <x86intrin.h> |
1746 | /// |
1747 | /// This intrinsic corresponds to the <c> VMOVUPS / MOVUPS </c> instruction. |
1748 | /// |
1749 | /// \param __p |
1750 | /// A pointer to a 128-bit memory location. The address of the memory |
1751 | /// location does not have to be aligned. |
1752 | /// \returns A 128-bit vector of [4 x float] containing the loaded values. |
1753 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1754 | _mm_loadu_ps(const float *__p) |
1755 | { |
1756 | struct __loadu_ps { |
1757 | __m128_u __v; |
1758 | } __attribute__((__packed__, __may_alias__)); |
1759 | return ((struct __loadu_ps*)__p)->__v; |
1760 | } |
1761 | |
1762 | /// Loads four packed float values, in reverse order, from an aligned |
1763 | /// memory location to 32-bit elements in a 128-bit vector of [4 x float]. |
1764 | /// |
1765 | /// \headerfile <x86intrin.h> |
1766 | /// |
1767 | /// This intrinsic corresponds to the <c> VMOVAPS / MOVAPS + shuffling </c> |
1768 | /// instruction. |
1769 | /// |
1770 | /// \param __p |
1771 | /// A pointer to a 128-bit memory location. The address of the memory |
1772 | /// location has to be 128-bit aligned. |
1773 | /// \returns A 128-bit vector of [4 x float] containing the moved values, loaded |
1774 | /// in reverse order. |
1775 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1776 | _mm_loadr_ps(const float *__p) |
1777 | { |
1778 | __m128 __a = _mm_load_ps(__p); |
1779 | return __builtin_shufflevector((__v4sf)__a, (__v4sf)__a, 3, 2, 1, 0); |
1780 | } |
1781 | |
1782 | /// Create a 128-bit vector of [4 x float] with undefined values. |
1783 | /// |
1784 | /// \headerfile <x86intrin.h> |
1785 | /// |
1786 | /// This intrinsic has no corresponding instruction. |
1787 | /// |
1788 | /// \returns A 128-bit vector of [4 x float] containing undefined values. |
1789 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1790 | _mm_undefined_ps(void) |
1791 | { |
1792 | return (__m128)__builtin_ia32_undef128(); |
1793 | } |
1794 | |
1795 | /// Constructs a 128-bit floating-point vector of [4 x float]. The lower |
1796 | /// 32 bits of the vector are initialized with the specified single-precision |
1797 | /// floating-point value. The upper 96 bits are set to zero. |
1798 | /// |
1799 | /// \headerfile <x86intrin.h> |
1800 | /// |
1801 | /// This intrinsic corresponds to the <c> VMOVSS / MOVSS </c> instruction. |
1802 | /// |
1803 | /// \param __w |
1804 | /// A single-precision floating-point value used to initialize the lower 32 |
1805 | /// bits of the result. |
1806 | /// \returns An initialized 128-bit floating-point vector of [4 x float]. The |
1807 | /// lower 32 bits contain the value provided in the source operand. The |
1808 | /// upper 96 bits are set to zero. |
1809 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1810 | _mm_set_ss(float __w) |
1811 | { |
1812 | return __extension__ (__m128){ __w, 0, 0, 0 }; |
1813 | } |
1814 | |
1815 | /// Constructs a 128-bit floating-point vector of [4 x float], with each |
1816 | /// of the four single-precision floating-point vector elements set to the |
1817 | /// specified single-precision floating-point value. |
1818 | /// |
1819 | /// \headerfile <x86intrin.h> |
1820 | /// |
1821 | /// This intrinsic corresponds to the <c> VPERMILPS / PERMILPS </c> instruction. |
1822 | /// |
1823 | /// \param __w |
1824 | /// A single-precision floating-point value used to initialize each vector |
1825 | /// element of the result. |
1826 | /// \returns An initialized 128-bit floating-point vector of [4 x float]. |
1827 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1828 | _mm_set1_ps(float __w) |
1829 | { |
1830 | return __extension__ (__m128){ __w, __w, __w, __w }; |
1831 | } |
1832 | |
1833 | /* Microsoft specific. */ |
1834 | /// Constructs a 128-bit floating-point vector of [4 x float], with each |
1835 | /// of the four single-precision floating-point vector elements set to the |
1836 | /// specified single-precision floating-point value. |
1837 | /// |
1838 | /// \headerfile <x86intrin.h> |
1839 | /// |
1840 | /// This intrinsic corresponds to the <c> VPERMILPS / PERMILPS </c> instruction. |
1841 | /// |
1842 | /// \param __w |
1843 | /// A single-precision floating-point value used to initialize each vector |
1844 | /// element of the result. |
1845 | /// \returns An initialized 128-bit floating-point vector of [4 x float]. |
1846 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1847 | _mm_set_ps1(float __w) |
1848 | { |
1849 | return _mm_set1_ps(__w); |
1850 | } |
1851 | |
1852 | /// Constructs a 128-bit floating-point vector of [4 x float] |
1853 | /// initialized with the specified single-precision floating-point values. |
1854 | /// |
1855 | /// \headerfile <x86intrin.h> |
1856 | /// |
1857 | /// This intrinsic is a utility function and does not correspond to a specific |
1858 | /// instruction. |
1859 | /// |
1860 | /// \param __z |
1861 | /// A single-precision floating-point value used to initialize bits [127:96] |
1862 | /// of the result. |
1863 | /// \param __y |
1864 | /// A single-precision floating-point value used to initialize bits [95:64] |
1865 | /// of the result. |
1866 | /// \param __x |
1867 | /// A single-precision floating-point value used to initialize bits [63:32] |
1868 | /// of the result. |
1869 | /// \param __w |
1870 | /// A single-precision floating-point value used to initialize bits [31:0] |
1871 | /// of the result. |
1872 | /// \returns An initialized 128-bit floating-point vector of [4 x float]. |
1873 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1874 | _mm_set_ps(float __z, float __y, float __x, float __w) |
1875 | { |
1876 | return __extension__ (__m128){ __w, __x, __y, __z }; |
1877 | } |
1878 | |
1879 | /// Constructs a 128-bit floating-point vector of [4 x float], |
1880 | /// initialized in reverse order with the specified 32-bit single-precision |
1881 | /// float-point values. |
1882 | /// |
1883 | /// \headerfile <x86intrin.h> |
1884 | /// |
1885 | /// This intrinsic is a utility function and does not correspond to a specific |
1886 | /// instruction. |
1887 | /// |
1888 | /// \param __z |
1889 | /// A single-precision floating-point value used to initialize bits [31:0] |
1890 | /// of the result. |
1891 | /// \param __y |
1892 | /// A single-precision floating-point value used to initialize bits [63:32] |
1893 | /// of the result. |
1894 | /// \param __x |
1895 | /// A single-precision floating-point value used to initialize bits [95:64] |
1896 | /// of the result. |
1897 | /// \param __w |
1898 | /// A single-precision floating-point value used to initialize bits [127:96] |
1899 | /// of the result. |
1900 | /// \returns An initialized 128-bit floating-point vector of [4 x float]. |
1901 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1902 | _mm_setr_ps(float __z, float __y, float __x, float __w) |
1903 | { |
1904 | return __extension__ (__m128){ __z, __y, __x, __w }; |
1905 | } |
1906 | |
1907 | /// Constructs a 128-bit floating-point vector of [4 x float] initialized |
1908 | /// to zero. |
1909 | /// |
1910 | /// \headerfile <x86intrin.h> |
1911 | /// |
1912 | /// This intrinsic corresponds to the <c> VXORPS / XORPS </c> instruction. |
1913 | /// |
1914 | /// \returns An initialized 128-bit floating-point vector of [4 x float] with |
1915 | /// all elements set to zero. |
1916 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
1917 | _mm_setzero_ps(void) |
1918 | { |
1919 | return __extension__ (__m128){ 0, 0, 0, 0 }; |
1920 | } |
1921 | |
1922 | /// Stores the upper 64 bits of a 128-bit vector of [4 x float] to a |
1923 | /// memory location. |
1924 | /// |
1925 | /// \headerfile <x86intrin.h> |
1926 | /// |
1927 | /// This intrinsic corresponds to the <c> VPEXTRQ / PEXTRQ </c> instruction. |
1928 | /// |
1929 | /// \param __p |
1930 | /// A pointer to a 64-bit memory location. |
1931 | /// \param __a |
1932 | /// A 128-bit vector of [4 x float] containing the values to be stored. |
1933 | static __inline__ void __DEFAULT_FN_ATTRS |
1934 | _mm_storeh_pi(__m64 *__p, __m128 __a) |
1935 | { |
1936 | __builtin_ia32_storehps((__v2si *)__p, (__v4sf)__a); |
1937 | } |
1938 | |
1939 | /// Stores the lower 64 bits of a 128-bit vector of [4 x float] to a |
1940 | /// memory location. |
1941 | /// |
1942 | /// \headerfile <x86intrin.h> |
1943 | /// |
1944 | /// This intrinsic corresponds to the <c> VMOVLPS / MOVLPS </c> instruction. |
1945 | /// |
1946 | /// \param __p |
1947 | /// A pointer to a memory location that will receive the float values. |
1948 | /// \param __a |
1949 | /// A 128-bit vector of [4 x float] containing the values to be stored. |
1950 | static __inline__ void __DEFAULT_FN_ATTRS |
1951 | _mm_storel_pi(__m64 *__p, __m128 __a) |
1952 | { |
1953 | __builtin_ia32_storelps((__v2si *)__p, (__v4sf)__a); |
1954 | } |
1955 | |
1956 | /// Stores the lower 32 bits of a 128-bit vector of [4 x float] to a |
1957 | /// memory location. |
1958 | /// |
1959 | /// \headerfile <x86intrin.h> |
1960 | /// |
1961 | /// This intrinsic corresponds to the <c> VMOVSS / MOVSS </c> instruction. |
1962 | /// |
1963 | /// \param __p |
1964 | /// A pointer to a 32-bit memory location. |
1965 | /// \param __a |
1966 | /// A 128-bit vector of [4 x float] containing the value to be stored. |
1967 | static __inline__ void __DEFAULT_FN_ATTRS |
1968 | _mm_store_ss(float *__p, __m128 __a) |
1969 | { |
1970 | struct __mm_store_ss_struct { |
1971 | float __u; |
1972 | } __attribute__((__packed__, __may_alias__)); |
1973 | ((struct __mm_store_ss_struct*)__p)->__u = __a[0]; |
1974 | } |
1975 | |
1976 | /// Stores a 128-bit vector of [4 x float] to an unaligned memory |
1977 | /// location. |
1978 | /// |
1979 | /// \headerfile <x86intrin.h> |
1980 | /// |
1981 | /// This intrinsic corresponds to the <c> VMOVUPS / MOVUPS </c> instruction. |
1982 | /// |
1983 | /// \param __p |
1984 | /// A pointer to a 128-bit memory location. The address of the memory |
1985 | /// location does not have to be aligned. |
1986 | /// \param __a |
1987 | /// A 128-bit vector of [4 x float] containing the values to be stored. |
1988 | static __inline__ void __DEFAULT_FN_ATTRS |
1989 | _mm_storeu_ps(float *__p, __m128 __a) |
1990 | { |
1991 | struct __storeu_ps { |
1992 | __m128_u __v; |
1993 | } __attribute__((__packed__, __may_alias__)); |
1994 | ((struct __storeu_ps*)__p)->__v = __a; |
1995 | } |
1996 | |
1997 | /// Stores a 128-bit vector of [4 x float] into an aligned memory |
1998 | /// location. |
1999 | /// |
2000 | /// \headerfile <x86intrin.h> |
2001 | /// |
2002 | /// This intrinsic corresponds to the <c> VMOVAPS / MOVAPS </c> instruction. |
2003 | /// |
2004 | /// \param __p |
2005 | /// A pointer to a 128-bit memory location. The address of the memory |
2006 | /// location has to be 16-byte aligned. |
2007 | /// \param __a |
2008 | /// A 128-bit vector of [4 x float] containing the values to be stored. |
2009 | static __inline__ void __DEFAULT_FN_ATTRS |
2010 | _mm_store_ps(float *__p, __m128 __a) |
2011 | { |
2012 | *(__m128*)__p = __a; |
2013 | } |
2014 | |
2015 | /// Stores the lower 32 bits of a 128-bit vector of [4 x float] into |
2016 | /// four contiguous elements in an aligned memory location. |
2017 | /// |
2018 | /// \headerfile <x86intrin.h> |
2019 | /// |
2020 | /// This intrinsic corresponds to <c> VMOVAPS / MOVAPS + shuffling </c> |
2021 | /// instruction. |
2022 | /// |
2023 | /// \param __p |
2024 | /// A pointer to a 128-bit memory location. |
2025 | /// \param __a |
2026 | /// A 128-bit vector of [4 x float] whose lower 32 bits are stored to each |
2027 | /// of the four contiguous elements pointed by \a __p. |
2028 | static __inline__ void __DEFAULT_FN_ATTRS |
2029 | _mm_store1_ps(float *__p, __m128 __a) |
2030 | { |
2031 | __a = __builtin_shufflevector((__v4sf)__a, (__v4sf)__a, 0, 0, 0, 0); |
2032 | _mm_store_ps(__p, __a); |
2033 | } |
2034 | |
2035 | /// Stores the lower 32 bits of a 128-bit vector of [4 x float] into |
2036 | /// four contiguous elements in an aligned memory location. |
2037 | /// |
2038 | /// \headerfile <x86intrin.h> |
2039 | /// |
2040 | /// This intrinsic corresponds to <c> VMOVAPS / MOVAPS + shuffling </c> |
2041 | /// instruction. |
2042 | /// |
2043 | /// \param __p |
2044 | /// A pointer to a 128-bit memory location. |
2045 | /// \param __a |
2046 | /// A 128-bit vector of [4 x float] whose lower 32 bits are stored to each |
2047 | /// of the four contiguous elements pointed by \a __p. |
2048 | static __inline__ void __DEFAULT_FN_ATTRS |
2049 | _mm_store_ps1(float *__p, __m128 __a) |
2050 | { |
2051 | _mm_store1_ps(__p, __a); |
2052 | } |
2053 | |
2054 | /// Stores float values from a 128-bit vector of [4 x float] to an |
2055 | /// aligned memory location in reverse order. |
2056 | /// |
2057 | /// \headerfile <x86intrin.h> |
2058 | /// |
2059 | /// This intrinsic corresponds to the <c> VMOVAPS / MOVAPS + shuffling </c> |
2060 | /// instruction. |
2061 | /// |
2062 | /// \param __p |
2063 | /// A pointer to a 128-bit memory location. The address of the memory |
2064 | /// location has to be 128-bit aligned. |
2065 | /// \param __a |
2066 | /// A 128-bit vector of [4 x float] containing the values to be stored. |
2067 | static __inline__ void __DEFAULT_FN_ATTRS |
2068 | _mm_storer_ps(float *__p, __m128 __a) |
2069 | { |
2070 | __a = __builtin_shufflevector((__v4sf)__a, (__v4sf)__a, 3, 2, 1, 0); |
2071 | _mm_store_ps(__p, __a); |
2072 | } |
2073 | |
2074 | #define _MM_HINT_ET0 7 |
2075 | #define _MM_HINT_ET1 6 |
2076 | #define _MM_HINT_T0 3 |
2077 | #define _MM_HINT_T1 2 |
2078 | #define _MM_HINT_T2 1 |
2079 | #define _MM_HINT_NTA 0 |
2080 | |
2081 | #ifndef _MSC_VER |
2082 | /* FIXME: We have to #define this because "sel" must be a constant integer, and |
2083 | Sema doesn't do any form of constant propagation yet. */ |
2084 | |
2085 | /// Loads one cache line of data from the specified address to a location |
2086 | /// closer to the processor. |
2087 | /// |
2088 | /// \headerfile <x86intrin.h> |
2089 | /// |
2090 | /// \code |
2091 | /// void _mm_prefetch(const void * a, const int sel); |
2092 | /// \endcode |
2093 | /// |
2094 | /// This intrinsic corresponds to the <c> PREFETCHNTA </c> instruction. |
2095 | /// |
2096 | /// \param a |
2097 | /// A pointer to a memory location containing a cache line of data. |
2098 | /// \param sel |
2099 | /// A predefined integer constant specifying the type of prefetch |
2100 | /// operation: \n |
2101 | /// _MM_HINT_NTA: Move data using the non-temporal access (NTA) hint. The |
2102 | /// PREFETCHNTA instruction will be generated. \n |
2103 | /// _MM_HINT_T0: Move data using the T0 hint. The PREFETCHT0 instruction will |
2104 | /// be generated. \n |
2105 | /// _MM_HINT_T1: Move data using the T1 hint. The PREFETCHT1 instruction will |
2106 | /// be generated. \n |
2107 | /// _MM_HINT_T2: Move data using the T2 hint. The PREFETCHT2 instruction will |
2108 | /// be generated. |
2109 | #define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), \ |
2110 | ((sel) >> 2) & 1, (sel) & 0x3)) |
2111 | #endif |
2112 | |
2113 | /// Stores a 64-bit integer in the specified aligned memory location. To |
2114 | /// minimize caching, the data is flagged as non-temporal (unlikely to be |
2115 | /// used again soon). |
2116 | /// |
2117 | /// \headerfile <x86intrin.h> |
2118 | /// |
2119 | /// This intrinsic corresponds to the <c> MOVNTQ </c> instruction. |
2120 | /// |
2121 | /// \param __p |
2122 | /// A pointer to an aligned memory location used to store the register value. |
2123 | /// \param __a |
2124 | /// A 64-bit integer containing the value to be stored. |
2125 | static __inline__ void __DEFAULT_FN_ATTRS_MMX |
2126 | _mm_stream_pi(__m64 *__p, __m64 __a) |
2127 | { |
2128 | __builtin_ia32_movntq(__p, __a); |
2129 | } |
2130 | |
2131 | /// Moves packed float values from a 128-bit vector of [4 x float] to a |
2132 | /// 128-bit aligned memory location. To minimize caching, the data is flagged |
2133 | /// as non-temporal (unlikely to be used again soon). |
2134 | /// |
2135 | /// \headerfile <x86intrin.h> |
2136 | /// |
2137 | /// This intrinsic corresponds to the <c> VMOVNTPS / MOVNTPS </c> instruction. |
2138 | /// |
2139 | /// \param __p |
2140 | /// A pointer to a 128-bit aligned memory location that will receive the |
2141 | /// single-precision floating-point values. |
2142 | /// \param __a |
2143 | /// A 128-bit vector of [4 x float] containing the values to be moved. |
2144 | static __inline__ void __DEFAULT_FN_ATTRS |
2145 | _mm_stream_ps(float *__p, __m128 __a) |
2146 | { |
2147 | __builtin_nontemporal_store((__v4sf)__a, (__v4sf*)__p); |
2148 | } |
2149 | |
2150 | #if defined(__cplusplus) |
2151 | extern "C" { |
2152 | #endif |
2153 | |
2154 | /// Forces strong memory ordering (serialization) between store |
2155 | /// instructions preceding this instruction and store instructions following |
2156 | /// this instruction, ensuring the system completes all previous stores |
2157 | /// before executing subsequent stores. |
2158 | /// |
2159 | /// \headerfile <x86intrin.h> |
2160 | /// |
2161 | /// This intrinsic corresponds to the <c> SFENCE </c> instruction. |
2162 | /// |
2163 | void _mm_sfence(void); |
2164 | |
2165 | #if defined(__cplusplus) |
2166 | } // extern "C" |
2167 | #endif |
2168 | |
2169 | /// Extracts 16-bit element from a 64-bit vector of [4 x i16] and |
2170 | /// returns it, as specified by the immediate integer operand. |
2171 | /// |
2172 | /// \headerfile <x86intrin.h> |
2173 | /// |
2174 | /// \code |
2175 | /// int _mm_extract_pi16(__m64 a, int n); |
2176 | /// \endcode |
2177 | /// |
2178 | /// This intrinsic corresponds to the <c> VPEXTRW / PEXTRW </c> instruction. |
2179 | /// |
2180 | /// \param a |
2181 | /// A 64-bit vector of [4 x i16]. |
2182 | /// \param n |
2183 | /// An immediate integer operand that determines which bits are extracted: \n |
2184 | /// 0: Bits [15:0] are copied to the destination. \n |
2185 | /// 1: Bits [31:16] are copied to the destination. \n |
2186 | /// 2: Bits [47:32] are copied to the destination. \n |
2187 | /// 3: Bits [63:48] are copied to the destination. |
2188 | /// \returns A 16-bit integer containing the extracted 16 bits of packed data. |
2189 | #define _mm_extract_pi16(a, n) \ |
2190 | (int)__builtin_ia32_vec_ext_v4hi((__m64)a, (int)n) |
2191 | |
2192 | /// Copies data from the 64-bit vector of [4 x i16] to the destination, |
2193 | /// and inserts the lower 16-bits of an integer operand at the 16-bit offset |
2194 | /// specified by the immediate operand \a n. |
2195 | /// |
2196 | /// \headerfile <x86intrin.h> |
2197 | /// |
2198 | /// \code |
2199 | /// __m64 _mm_insert_pi16(__m64 a, int d, int n); |
2200 | /// \endcode |
2201 | /// |
2202 | /// This intrinsic corresponds to the <c> PINSRW </c> instruction. |
2203 | /// |
2204 | /// \param a |
2205 | /// A 64-bit vector of [4 x i16]. |
2206 | /// \param d |
2207 | /// An integer. The lower 16-bit value from this operand is written to the |
2208 | /// destination at the offset specified by operand \a n. |
2209 | /// \param n |
2210 | /// An immediate integer operant that determines which the bits to be used |
2211 | /// in the destination. \n |
2212 | /// 0: Bits [15:0] are copied to the destination. \n |
2213 | /// 1: Bits [31:16] are copied to the destination. \n |
2214 | /// 2: Bits [47:32] are copied to the destination. \n |
2215 | /// 3: Bits [63:48] are copied to the destination. \n |
2216 | /// The remaining bits in the destination are copied from the corresponding |
2217 | /// bits in operand \a a. |
2218 | /// \returns A 64-bit integer vector containing the copied packed data from the |
2219 | /// operands. |
2220 | #define _mm_insert_pi16(a, d, n) \ |
2221 | (__m64)__builtin_ia32_vec_set_v4hi((__m64)a, (int)d, (int)n) |
2222 | |
2223 | /// Compares each of the corresponding packed 16-bit integer values of |
2224 | /// the 64-bit integer vectors, and writes the greater value to the |
2225 | /// corresponding bits in the destination. |
2226 | /// |
2227 | /// \headerfile <x86intrin.h> |
2228 | /// |
2229 | /// This intrinsic corresponds to the <c> PMAXSW </c> instruction. |
2230 | /// |
2231 | /// \param __a |
2232 | /// A 64-bit integer vector containing one of the source operands. |
2233 | /// \param __b |
2234 | /// A 64-bit integer vector containing one of the source operands. |
2235 | /// \returns A 64-bit integer vector containing the comparison results. |
2236 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2237 | _mm_max_pi16(__m64 __a, __m64 __b) |
2238 | { |
2239 | return (__m64)__builtin_ia32_pmaxsw((__v4hi)__a, (__v4hi)__b); |
2240 | } |
2241 | |
2242 | /// Compares each of the corresponding packed 8-bit unsigned integer |
2243 | /// values of the 64-bit integer vectors, and writes the greater value to the |
2244 | /// corresponding bits in the destination. |
2245 | /// |
2246 | /// \headerfile <x86intrin.h> |
2247 | /// |
2248 | /// This intrinsic corresponds to the <c> PMAXUB </c> instruction. |
2249 | /// |
2250 | /// \param __a |
2251 | /// A 64-bit integer vector containing one of the source operands. |
2252 | /// \param __b |
2253 | /// A 64-bit integer vector containing one of the source operands. |
2254 | /// \returns A 64-bit integer vector containing the comparison results. |
2255 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2256 | _mm_max_pu8(__m64 __a, __m64 __b) |
2257 | { |
2258 | return (__m64)__builtin_ia32_pmaxub((__v8qi)__a, (__v8qi)__b); |
2259 | } |
2260 | |
2261 | /// Compares each of the corresponding packed 16-bit integer values of |
2262 | /// the 64-bit integer vectors, and writes the lesser value to the |
2263 | /// corresponding bits in the destination. |
2264 | /// |
2265 | /// \headerfile <x86intrin.h> |
2266 | /// |
2267 | /// This intrinsic corresponds to the <c> PMINSW </c> instruction. |
2268 | /// |
2269 | /// \param __a |
2270 | /// A 64-bit integer vector containing one of the source operands. |
2271 | /// \param __b |
2272 | /// A 64-bit integer vector containing one of the source operands. |
2273 | /// \returns A 64-bit integer vector containing the comparison results. |
2274 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2275 | _mm_min_pi16(__m64 __a, __m64 __b) |
2276 | { |
2277 | return (__m64)__builtin_ia32_pminsw((__v4hi)__a, (__v4hi)__b); |
2278 | } |
2279 | |
2280 | /// Compares each of the corresponding packed 8-bit unsigned integer |
2281 | /// values of the 64-bit integer vectors, and writes the lesser value to the |
2282 | /// corresponding bits in the destination. |
2283 | /// |
2284 | /// \headerfile <x86intrin.h> |
2285 | /// |
2286 | /// This intrinsic corresponds to the <c> PMINUB </c> instruction. |
2287 | /// |
2288 | /// \param __a |
2289 | /// A 64-bit integer vector containing one of the source operands. |
2290 | /// \param __b |
2291 | /// A 64-bit integer vector containing one of the source operands. |
2292 | /// \returns A 64-bit integer vector containing the comparison results. |
2293 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2294 | _mm_min_pu8(__m64 __a, __m64 __b) |
2295 | { |
2296 | return (__m64)__builtin_ia32_pminub((__v8qi)__a, (__v8qi)__b); |
2297 | } |
2298 | |
2299 | /// Takes the most significant bit from each 8-bit element in a 64-bit |
2300 | /// integer vector to create an 8-bit mask value. Zero-extends the value to |
2301 | /// 32-bit integer and writes it to the destination. |
2302 | /// |
2303 | /// \headerfile <x86intrin.h> |
2304 | /// |
2305 | /// This intrinsic corresponds to the <c> PMOVMSKB </c> instruction. |
2306 | /// |
2307 | /// \param __a |
2308 | /// A 64-bit integer vector containing the values with bits to be extracted. |
2309 | /// \returns The most significant bit from each 8-bit element in \a __a, |
2310 | /// written to bits [7:0]. |
2311 | static __inline__ int __DEFAULT_FN_ATTRS_MMX |
2312 | _mm_movemask_pi8(__m64 __a) |
2313 | { |
2314 | return __builtin_ia32_pmovmskb((__v8qi)__a); |
2315 | } |
2316 | |
2317 | /// Multiplies packed 16-bit unsigned integer values and writes the |
2318 | /// high-order 16 bits of each 32-bit product to the corresponding bits in |
2319 | /// the destination. |
2320 | /// |
2321 | /// \headerfile <x86intrin.h> |
2322 | /// |
2323 | /// This intrinsic corresponds to the <c> PMULHUW </c> instruction. |
2324 | /// |
2325 | /// \param __a |
2326 | /// A 64-bit integer vector containing one of the source operands. |
2327 | /// \param __b |
2328 | /// A 64-bit integer vector containing one of the source operands. |
2329 | /// \returns A 64-bit integer vector containing the products of both operands. |
2330 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2331 | _mm_mulhi_pu16(__m64 __a, __m64 __b) |
2332 | { |
2333 | return (__m64)__builtin_ia32_pmulhuw((__v4hi)__a, (__v4hi)__b); |
2334 | } |
2335 | |
2336 | /// Shuffles the 4 16-bit integers from a 64-bit integer vector to the |
2337 | /// destination, as specified by the immediate value operand. |
2338 | /// |
2339 | /// \headerfile <x86intrin.h> |
2340 | /// |
2341 | /// \code |
2342 | /// __m64 _mm_shuffle_pi16(__m64 a, const int n); |
2343 | /// \endcode |
2344 | /// |
2345 | /// This intrinsic corresponds to the <c> PSHUFW </c> instruction. |
2346 | /// |
2347 | /// \param a |
2348 | /// A 64-bit integer vector containing the values to be shuffled. |
2349 | /// \param n |
2350 | /// An immediate value containing an 8-bit value specifying which elements to |
2351 | /// copy from \a a. The destinations within the 64-bit destination are |
2352 | /// assigned values as follows: \n |
2353 | /// Bits [1:0] are used to assign values to bits [15:0] in the |
2354 | /// destination. \n |
2355 | /// Bits [3:2] are used to assign values to bits [31:16] in the |
2356 | /// destination. \n |
2357 | /// Bits [5:4] are used to assign values to bits [47:32] in the |
2358 | /// destination. \n |
2359 | /// Bits [7:6] are used to assign values to bits [63:48] in the |
2360 | /// destination. \n |
2361 | /// Bit value assignments: \n |
2362 | /// 00: assigned from bits [15:0] of \a a. \n |
2363 | /// 01: assigned from bits [31:16] of \a a. \n |
2364 | /// 10: assigned from bits [47:32] of \a a. \n |
2365 | /// 11: assigned from bits [63:48] of \a a. |
2366 | /// \returns A 64-bit integer vector containing the shuffled values. |
2367 | #define _mm_shuffle_pi16(a, n) \ |
2368 | (__m64)__builtin_ia32_pshufw((__v4hi)(__m64)(a), (n)) |
2369 | |
2370 | /// Conditionally copies the values from each 8-bit element in the first |
2371 | /// 64-bit integer vector operand to the specified memory location, as |
2372 | /// specified by the most significant bit in the corresponding element in the |
2373 | /// second 64-bit integer vector operand. |
2374 | /// |
2375 | /// To minimize caching, the data is flagged as non-temporal |
2376 | /// (unlikely to be used again soon). |
2377 | /// |
2378 | /// \headerfile <x86intrin.h> |
2379 | /// |
2380 | /// This intrinsic corresponds to the <c> MASKMOVQ </c> instruction. |
2381 | /// |
2382 | /// \param __d |
2383 | /// A 64-bit integer vector containing the values with elements to be copied. |
2384 | /// \param __n |
2385 | /// A 64-bit integer vector operand. The most significant bit from each 8-bit |
2386 | /// element determines whether the corresponding element in operand \a __d |
2387 | /// is copied. If the most significant bit of a given element is 1, the |
2388 | /// corresponding element in operand \a __d is copied. |
2389 | /// \param __p |
2390 | /// A pointer to a 64-bit memory location that will receive the conditionally |
2391 | /// copied integer values. The address of the memory location does not have |
2392 | /// to be aligned. |
2393 | static __inline__ void __DEFAULT_FN_ATTRS_MMX |
2394 | _mm_maskmove_si64(__m64 __d, __m64 __n, char *__p) |
2395 | { |
2396 | __builtin_ia32_maskmovq((__v8qi)__d, (__v8qi)__n, __p); |
2397 | } |
2398 | |
2399 | /// Computes the rounded averages of the packed unsigned 8-bit integer |
2400 | /// values and writes the averages to the corresponding bits in the |
2401 | /// destination. |
2402 | /// |
2403 | /// \headerfile <x86intrin.h> |
2404 | /// |
2405 | /// This intrinsic corresponds to the <c> PAVGB </c> instruction. |
2406 | /// |
2407 | /// \param __a |
2408 | /// A 64-bit integer vector containing one of the source operands. |
2409 | /// \param __b |
2410 | /// A 64-bit integer vector containing one of the source operands. |
2411 | /// \returns A 64-bit integer vector containing the averages of both operands. |
2412 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2413 | _mm_avg_pu8(__m64 __a, __m64 __b) |
2414 | { |
2415 | return (__m64)__builtin_ia32_pavgb((__v8qi)__a, (__v8qi)__b); |
2416 | } |
2417 | |
2418 | /// Computes the rounded averages of the packed unsigned 16-bit integer |
2419 | /// values and writes the averages to the corresponding bits in the |
2420 | /// destination. |
2421 | /// |
2422 | /// \headerfile <x86intrin.h> |
2423 | /// |
2424 | /// This intrinsic corresponds to the <c> PAVGW </c> instruction. |
2425 | /// |
2426 | /// \param __a |
2427 | /// A 64-bit integer vector containing one of the source operands. |
2428 | /// \param __b |
2429 | /// A 64-bit integer vector containing one of the source operands. |
2430 | /// \returns A 64-bit integer vector containing the averages of both operands. |
2431 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2432 | _mm_avg_pu16(__m64 __a, __m64 __b) |
2433 | { |
2434 | return (__m64)__builtin_ia32_pavgw((__v4hi)__a, (__v4hi)__b); |
2435 | } |
2436 | |
2437 | /// Subtracts the corresponding 8-bit unsigned integer values of the two |
2438 | /// 64-bit vector operands and computes the absolute value for each of the |
2439 | /// difference. Then sum of the 8 absolute differences is written to the |
2440 | /// bits [15:0] of the destination; the remaining bits [63:16] are cleared. |
2441 | /// |
2442 | /// \headerfile <x86intrin.h> |
2443 | /// |
2444 | /// This intrinsic corresponds to the <c> PSADBW </c> instruction. |
2445 | /// |
2446 | /// \param __a |
2447 | /// A 64-bit integer vector containing one of the source operands. |
2448 | /// \param __b |
2449 | /// A 64-bit integer vector containing one of the source operands. |
2450 | /// \returns A 64-bit integer vector whose lower 16 bits contain the sums of the |
2451 | /// sets of absolute differences between both operands. The upper bits are |
2452 | /// cleared. |
2453 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2454 | _mm_sad_pu8(__m64 __a, __m64 __b) |
2455 | { |
2456 | return (__m64)__builtin_ia32_psadbw((__v8qi)__a, (__v8qi)__b); |
2457 | } |
2458 | |
2459 | #if defined(__cplusplus) |
2460 | extern "C" { |
2461 | #endif |
2462 | |
2463 | /// Returns the contents of the MXCSR register as a 32-bit unsigned |
2464 | /// integer value. |
2465 | /// |
2466 | /// There are several groups of macros associated with this |
2467 | /// intrinsic, including: |
2468 | /// <ul> |
2469 | /// <li> |
2470 | /// For checking exception states: _MM_EXCEPT_INVALID, _MM_EXCEPT_DIV_ZERO, |
2471 | /// _MM_EXCEPT_DENORM, _MM_EXCEPT_OVERFLOW, _MM_EXCEPT_UNDERFLOW, |
2472 | /// _MM_EXCEPT_INEXACT. There is a convenience wrapper |
2473 | /// _MM_GET_EXCEPTION_STATE(). |
2474 | /// </li> |
2475 | /// <li> |
2476 | /// For checking exception masks: _MM_MASK_UNDERFLOW, _MM_MASK_OVERFLOW, |
2477 | /// _MM_MASK_INVALID, _MM_MASK_DENORM, _MM_MASK_DIV_ZERO, _MM_MASK_INEXACT. |
2478 | /// There is a convenience wrapper _MM_GET_EXCEPTION_MASK(). |
2479 | /// </li> |
2480 | /// <li> |
2481 | /// For checking rounding modes: _MM_ROUND_NEAREST, _MM_ROUND_DOWN, |
2482 | /// _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO. There is a convenience wrapper |
2483 | /// _MM_GET_ROUNDING_MODE(). |
2484 | /// </li> |
2485 | /// <li> |
2486 | /// For checking flush-to-zero mode: _MM_FLUSH_ZERO_ON, _MM_FLUSH_ZERO_OFF. |
2487 | /// There is a convenience wrapper _MM_GET_FLUSH_ZERO_MODE(). |
2488 | /// </li> |
2489 | /// <li> |
2490 | /// For checking denormals-are-zero mode: _MM_DENORMALS_ZERO_ON, |
2491 | /// _MM_DENORMALS_ZERO_OFF. There is a convenience wrapper |
2492 | /// _MM_GET_DENORMALS_ZERO_MODE(). |
2493 | /// </li> |
2494 | /// </ul> |
2495 | /// |
2496 | /// For example, the following expression checks if an overflow exception has |
2497 | /// occurred: |
2498 | /// \code |
2499 | /// ( _mm_getcsr() & _MM_EXCEPT_OVERFLOW ) |
2500 | /// \endcode |
2501 | /// |
2502 | /// The following expression gets the current rounding mode: |
2503 | /// \code |
2504 | /// _MM_GET_ROUNDING_MODE() |
2505 | /// \endcode |
2506 | /// |
2507 | /// \headerfile <x86intrin.h> |
2508 | /// |
2509 | /// This intrinsic corresponds to the <c> VSTMXCSR / STMXCSR </c> instruction. |
2510 | /// |
2511 | /// \returns A 32-bit unsigned integer containing the contents of the MXCSR |
2512 | /// register. |
2513 | unsigned int _mm_getcsr(void); |
2514 | |
2515 | /// Sets the MXCSR register with the 32-bit unsigned integer value. |
2516 | /// |
2517 | /// There are several groups of macros associated with this intrinsic, |
2518 | /// including: |
2519 | /// <ul> |
2520 | /// <li> |
2521 | /// For setting exception states: _MM_EXCEPT_INVALID, _MM_EXCEPT_DIV_ZERO, |
2522 | /// _MM_EXCEPT_DENORM, _MM_EXCEPT_OVERFLOW, _MM_EXCEPT_UNDERFLOW, |
2523 | /// _MM_EXCEPT_INEXACT. There is a convenience wrapper |
2524 | /// _MM_SET_EXCEPTION_STATE(x) where x is one of these macros. |
2525 | /// </li> |
2526 | /// <li> |
2527 | /// For setting exception masks: _MM_MASK_UNDERFLOW, _MM_MASK_OVERFLOW, |
2528 | /// _MM_MASK_INVALID, _MM_MASK_DENORM, _MM_MASK_DIV_ZERO, _MM_MASK_INEXACT. |
2529 | /// There is a convenience wrapper _MM_SET_EXCEPTION_MASK(x) where x is one |
2530 | /// of these macros. |
2531 | /// </li> |
2532 | /// <li> |
2533 | /// For setting rounding modes: _MM_ROUND_NEAREST, _MM_ROUND_DOWN, |
2534 | /// _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO. There is a convenience wrapper |
2535 | /// _MM_SET_ROUNDING_MODE(x) where x is one of these macros. |
2536 | /// </li> |
2537 | /// <li> |
2538 | /// For setting flush-to-zero mode: _MM_FLUSH_ZERO_ON, _MM_FLUSH_ZERO_OFF. |
2539 | /// There is a convenience wrapper _MM_SET_FLUSH_ZERO_MODE(x) where x is |
2540 | /// one of these macros. |
2541 | /// </li> |
2542 | /// <li> |
2543 | /// For setting denormals-are-zero mode: _MM_DENORMALS_ZERO_ON, |
2544 | /// _MM_DENORMALS_ZERO_OFF. There is a convenience wrapper |
2545 | /// _MM_SET_DENORMALS_ZERO_MODE(x) where x is one of these macros. |
2546 | /// </li> |
2547 | /// </ul> |
2548 | /// |
2549 | /// For example, the following expression causes subsequent floating-point |
2550 | /// operations to round up: |
2551 | /// _mm_setcsr(_mm_getcsr() | _MM_ROUND_UP) |
2552 | /// |
2553 | /// The following example sets the DAZ and FTZ flags: |
2554 | /// \code |
2555 | /// void setFlags() { |
2556 | /// _MM_SET_FLUSH_ZERO_MODE(_MM_FLUSH_ZERO_ON); |
2557 | /// _MM_SET_DENORMALS_ZERO_MODE(_MM_DENORMALS_ZERO_ON); |
2558 | /// } |
2559 | /// \endcode |
2560 | /// |
2561 | /// \headerfile <x86intrin.h> |
2562 | /// |
2563 | /// This intrinsic corresponds to the <c> VLDMXCSR / LDMXCSR </c> instruction. |
2564 | /// |
2565 | /// \param __i |
2566 | /// A 32-bit unsigned integer value to be written to the MXCSR register. |
2567 | void _mm_setcsr(unsigned int __i); |
2568 | |
2569 | #if defined(__cplusplus) |
2570 | } // extern "C" |
2571 | #endif |
2572 | |
2573 | /// Selects 4 float values from the 128-bit operands of [4 x float], as |
2574 | /// specified by the immediate value operand. |
2575 | /// |
2576 | /// \headerfile <x86intrin.h> |
2577 | /// |
2578 | /// \code |
2579 | /// __m128 _mm_shuffle_ps(__m128 a, __m128 b, const int mask); |
2580 | /// \endcode |
2581 | /// |
2582 | /// This intrinsic corresponds to the <c> VSHUFPS / SHUFPS </c> instruction. |
2583 | /// |
2584 | /// \param a |
2585 | /// A 128-bit vector of [4 x float]. |
2586 | /// \param b |
2587 | /// A 128-bit vector of [4 x float]. |
2588 | /// \param mask |
2589 | /// An immediate value containing an 8-bit value specifying which elements to |
2590 | /// copy from \a a and \a b. \n |
2591 | /// Bits [3:0] specify the values copied from operand \a a. \n |
2592 | /// Bits [7:4] specify the values copied from operand \a b. \n |
2593 | /// The destinations within the 128-bit destination are assigned values as |
2594 | /// follows: \n |
2595 | /// Bits [1:0] are used to assign values to bits [31:0] in the |
2596 | /// destination. \n |
2597 | /// Bits [3:2] are used to assign values to bits [63:32] in the |
2598 | /// destination. \n |
2599 | /// Bits [5:4] are used to assign values to bits [95:64] in the |
2600 | /// destination. \n |
2601 | /// Bits [7:6] are used to assign values to bits [127:96] in the |
2602 | /// destination. \n |
2603 | /// Bit value assignments: \n |
2604 | /// 00: Bits [31:0] copied from the specified operand. \n |
2605 | /// 01: Bits [63:32] copied from the specified operand. \n |
2606 | /// 10: Bits [95:64] copied from the specified operand. \n |
2607 | /// 11: Bits [127:96] copied from the specified operand. |
2608 | /// \returns A 128-bit vector of [4 x float] containing the shuffled values. |
2609 | #define _mm_shuffle_ps(a, b, mask) \ |
2610 | (__m128)__builtin_ia32_shufps((__v4sf)(__m128)(a), (__v4sf)(__m128)(b), \ |
2611 | (int)(mask)) |
2612 | |
2613 | /// Unpacks the high-order (index 2,3) values from two 128-bit vectors of |
2614 | /// [4 x float] and interleaves them into a 128-bit vector of [4 x float]. |
2615 | /// |
2616 | /// \headerfile <x86intrin.h> |
2617 | /// |
2618 | /// This intrinsic corresponds to the <c> VUNPCKHPS / UNPCKHPS </c> instruction. |
2619 | /// |
2620 | /// \param __a |
2621 | /// A 128-bit vector of [4 x float]. \n |
2622 | /// Bits [95:64] are written to bits [31:0] of the destination. \n |
2623 | /// Bits [127:96] are written to bits [95:64] of the destination. |
2624 | /// \param __b |
2625 | /// A 128-bit vector of [4 x float]. |
2626 | /// Bits [95:64] are written to bits [63:32] of the destination. \n |
2627 | /// Bits [127:96] are written to bits [127:96] of the destination. |
2628 | /// \returns A 128-bit vector of [4 x float] containing the interleaved values. |
2629 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
2630 | _mm_unpackhi_ps(__m128 __a, __m128 __b) |
2631 | { |
2632 | return __builtin_shufflevector((__v4sf)__a, (__v4sf)__b, 2, 6, 3, 7); |
2633 | } |
2634 | |
2635 | /// Unpacks the low-order (index 0,1) values from two 128-bit vectors of |
2636 | /// [4 x float] and interleaves them into a 128-bit vector of [4 x float]. |
2637 | /// |
2638 | /// \headerfile <x86intrin.h> |
2639 | /// |
2640 | /// This intrinsic corresponds to the <c> VUNPCKLPS / UNPCKLPS </c> instruction. |
2641 | /// |
2642 | /// \param __a |
2643 | /// A 128-bit vector of [4 x float]. \n |
2644 | /// Bits [31:0] are written to bits [31:0] of the destination. \n |
2645 | /// Bits [63:32] are written to bits [95:64] of the destination. |
2646 | /// \param __b |
2647 | /// A 128-bit vector of [4 x float]. \n |
2648 | /// Bits [31:0] are written to bits [63:32] of the destination. \n |
2649 | /// Bits [63:32] are written to bits [127:96] of the destination. |
2650 | /// \returns A 128-bit vector of [4 x float] containing the interleaved values. |
2651 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
2652 | _mm_unpacklo_ps(__m128 __a, __m128 __b) |
2653 | { |
2654 | return __builtin_shufflevector((__v4sf)__a, (__v4sf)__b, 0, 4, 1, 5); |
2655 | } |
2656 | |
2657 | /// Constructs a 128-bit floating-point vector of [4 x float]. The lower |
2658 | /// 32 bits are set to the lower 32 bits of the second parameter. The upper |
2659 | /// 96 bits are set to the upper 96 bits of the first parameter. |
2660 | /// |
2661 | /// \headerfile <x86intrin.h> |
2662 | /// |
2663 | /// This intrinsic corresponds to the <c> VBLENDPS / BLENDPS / MOVSS </c> |
2664 | /// instruction. |
2665 | /// |
2666 | /// \param __a |
2667 | /// A 128-bit floating-point vector of [4 x float]. The upper 96 bits are |
2668 | /// written to the upper 96 bits of the result. |
2669 | /// \param __b |
2670 | /// A 128-bit floating-point vector of [4 x float]. The lower 32 bits are |
2671 | /// written to the lower 32 bits of the result. |
2672 | /// \returns A 128-bit floating-point vector of [4 x float]. |
2673 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
2674 | _mm_move_ss(__m128 __a, __m128 __b) |
2675 | { |
2676 | __a[0] = __b[0]; |
2677 | return __a; |
2678 | } |
2679 | |
2680 | /// Constructs a 128-bit floating-point vector of [4 x float]. The lower |
2681 | /// 64 bits are set to the upper 64 bits of the second parameter. The upper |
2682 | /// 64 bits are set to the upper 64 bits of the first parameter. |
2683 | /// |
2684 | /// \headerfile <x86intrin.h> |
2685 | /// |
2686 | /// This intrinsic corresponds to the <c> VUNPCKHPD / UNPCKHPD </c> instruction. |
2687 | /// |
2688 | /// \param __a |
2689 | /// A 128-bit floating-point vector of [4 x float]. The upper 64 bits are |
2690 | /// written to the upper 64 bits of the result. |
2691 | /// \param __b |
2692 | /// A 128-bit floating-point vector of [4 x float]. The upper 64 bits are |
2693 | /// written to the lower 64 bits of the result. |
2694 | /// \returns A 128-bit floating-point vector of [4 x float]. |
2695 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
2696 | _mm_movehl_ps(__m128 __a, __m128 __b) |
2697 | { |
2698 | return __builtin_shufflevector((__v4sf)__a, (__v4sf)__b, 6, 7, 2, 3); |
2699 | } |
2700 | |
2701 | /// Constructs a 128-bit floating-point vector of [4 x float]. The lower |
2702 | /// 64 bits are set to the lower 64 bits of the first parameter. The upper |
2703 | /// 64 bits are set to the lower 64 bits of the second parameter. |
2704 | /// |
2705 | /// \headerfile <x86intrin.h> |
2706 | /// |
2707 | /// This intrinsic corresponds to the <c> VUNPCKLPD / UNPCKLPD </c> instruction. |
2708 | /// |
2709 | /// \param __a |
2710 | /// A 128-bit floating-point vector of [4 x float]. The lower 64 bits are |
2711 | /// written to the lower 64 bits of the result. |
2712 | /// \param __b |
2713 | /// A 128-bit floating-point vector of [4 x float]. The lower 64 bits are |
2714 | /// written to the upper 64 bits of the result. |
2715 | /// \returns A 128-bit floating-point vector of [4 x float]. |
2716 | static __inline__ __m128 __DEFAULT_FN_ATTRS |
2717 | _mm_movelh_ps(__m128 __a, __m128 __b) |
2718 | { |
2719 | return __builtin_shufflevector((__v4sf)__a, (__v4sf)__b, 0, 1, 4, 5); |
2720 | } |
2721 | |
2722 | /// Converts a 64-bit vector of [4 x i16] into a 128-bit vector of [4 x |
2723 | /// float]. |
2724 | /// |
2725 | /// \headerfile <x86intrin.h> |
2726 | /// |
2727 | /// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. |
2728 | /// |
2729 | /// \param __a |
2730 | /// A 64-bit vector of [4 x i16]. The elements of the destination are copied |
2731 | /// from the corresponding elements in this operand. |
2732 | /// \returns A 128-bit vector of [4 x float] containing the copied and converted |
2733 | /// values from the operand. |
2734 | static __inline__ __m128 __DEFAULT_FN_ATTRS_MMX |
2735 | _mm_cvtpi16_ps(__m64 __a) |
2736 | { |
2737 | __m64 __b, __c; |
2738 | __m128 __r; |
2739 | |
2740 | __b = _mm_setzero_si64(); |
2741 | __b = _mm_cmpgt_pi16(__b, __a); |
2742 | __c = _mm_unpackhi_pi16(__a, __b); |
2743 | __r = _mm_setzero_ps(); |
2744 | __r = _mm_cvtpi32_ps(__r, __c); |
2745 | __r = _mm_movelh_ps(__r, __r); |
2746 | __c = _mm_unpacklo_pi16(__a, __b); |
2747 | __r = _mm_cvtpi32_ps(__r, __c); |
2748 | |
2749 | return __r; |
2750 | } |
2751 | |
2752 | /// Converts a 64-bit vector of 16-bit unsigned integer values into a |
2753 | /// 128-bit vector of [4 x float]. |
2754 | /// |
2755 | /// \headerfile <x86intrin.h> |
2756 | /// |
2757 | /// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. |
2758 | /// |
2759 | /// \param __a |
2760 | /// A 64-bit vector of 16-bit unsigned integer values. The elements of the |
2761 | /// destination are copied from the corresponding elements in this operand. |
2762 | /// \returns A 128-bit vector of [4 x float] containing the copied and converted |
2763 | /// values from the operand. |
2764 | static __inline__ __m128 __DEFAULT_FN_ATTRS_MMX |
2765 | _mm_cvtpu16_ps(__m64 __a) |
2766 | { |
2767 | __m64 __b, __c; |
2768 | __m128 __r; |
2769 | |
2770 | __b = _mm_setzero_si64(); |
2771 | __c = _mm_unpackhi_pi16(__a, __b); |
2772 | __r = _mm_setzero_ps(); |
2773 | __r = _mm_cvtpi32_ps(__r, __c); |
2774 | __r = _mm_movelh_ps(__r, __r); |
2775 | __c = _mm_unpacklo_pi16(__a, __b); |
2776 | __r = _mm_cvtpi32_ps(__r, __c); |
2777 | |
2778 | return __r; |
2779 | } |
2780 | |
2781 | /// Converts the lower four 8-bit values from a 64-bit vector of [8 x i8] |
2782 | /// into a 128-bit vector of [4 x float]. |
2783 | /// |
2784 | /// \headerfile <x86intrin.h> |
2785 | /// |
2786 | /// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. |
2787 | /// |
2788 | /// \param __a |
2789 | /// A 64-bit vector of [8 x i8]. The elements of the destination are copied |
2790 | /// from the corresponding lower 4 elements in this operand. |
2791 | /// \returns A 128-bit vector of [4 x float] containing the copied and converted |
2792 | /// values from the operand. |
2793 | static __inline__ __m128 __DEFAULT_FN_ATTRS_MMX |
2794 | _mm_cvtpi8_ps(__m64 __a) |
2795 | { |
2796 | __m64 __b; |
2797 | |
2798 | __b = _mm_setzero_si64(); |
2799 | __b = _mm_cmpgt_pi8(__b, __a); |
2800 | __b = _mm_unpacklo_pi8(__a, __b); |
2801 | |
2802 | return _mm_cvtpi16_ps(__b); |
2803 | } |
2804 | |
2805 | /// Converts the lower four unsigned 8-bit integer values from a 64-bit |
2806 | /// vector of [8 x u8] into a 128-bit vector of [4 x float]. |
2807 | /// |
2808 | /// \headerfile <x86intrin.h> |
2809 | /// |
2810 | /// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. |
2811 | /// |
2812 | /// \param __a |
2813 | /// A 64-bit vector of unsigned 8-bit integer values. The elements of the |
2814 | /// destination are copied from the corresponding lower 4 elements in this |
2815 | /// operand. |
2816 | /// \returns A 128-bit vector of [4 x float] containing the copied and converted |
2817 | /// values from the source operand. |
2818 | static __inline__ __m128 __DEFAULT_FN_ATTRS_MMX |
2819 | _mm_cvtpu8_ps(__m64 __a) |
2820 | { |
2821 | __m64 __b; |
2822 | |
2823 | __b = _mm_setzero_si64(); |
2824 | __b = _mm_unpacklo_pi8(__a, __b); |
2825 | |
2826 | return _mm_cvtpi16_ps(__b); |
2827 | } |
2828 | |
2829 | /// Converts the two 32-bit signed integer values from each 64-bit vector |
2830 | /// operand of [2 x i32] into a 128-bit vector of [4 x float]. |
2831 | /// |
2832 | /// \headerfile <x86intrin.h> |
2833 | /// |
2834 | /// This intrinsic corresponds to the <c> CVTPI2PS + COMPOSITE </c> instruction. |
2835 | /// |
2836 | /// \param __a |
2837 | /// A 64-bit vector of [2 x i32]. The lower elements of the destination are |
2838 | /// copied from the elements in this operand. |
2839 | /// \param __b |
2840 | /// A 64-bit vector of [2 x i32]. The upper elements of the destination are |
2841 | /// copied from the elements in this operand. |
2842 | /// \returns A 128-bit vector of [4 x float] whose lower 64 bits contain the |
2843 | /// copied and converted values from the first operand. The upper 64 bits |
2844 | /// contain the copied and converted values from the second operand. |
2845 | static __inline__ __m128 __DEFAULT_FN_ATTRS_MMX |
2846 | _mm_cvtpi32x2_ps(__m64 __a, __m64 __b) |
2847 | { |
2848 | __m128 __c; |
2849 | |
2850 | __c = _mm_setzero_ps(); |
2851 | __c = _mm_cvtpi32_ps(__c, __b); |
2852 | __c = _mm_movelh_ps(__c, __c); |
2853 | |
2854 | return _mm_cvtpi32_ps(__c, __a); |
2855 | } |
2856 | |
2857 | /// Converts each single-precision floating-point element of a 128-bit |
2858 | /// floating-point vector of [4 x float] into a 16-bit signed integer, and |
2859 | /// packs the results into a 64-bit integer vector of [4 x i16]. |
2860 | /// |
2861 | /// If the floating-point element is NaN or infinity, or if the |
2862 | /// floating-point element is greater than 0x7FFFFFFF or less than -0x8000, |
2863 | /// it is converted to 0x8000. Otherwise if the floating-point element is |
2864 | /// greater than 0x7FFF, it is converted to 0x7FFF. |
2865 | /// |
2866 | /// \headerfile <x86intrin.h> |
2867 | /// |
2868 | /// This intrinsic corresponds to the <c> CVTPS2PI + COMPOSITE </c> instruction. |
2869 | /// |
2870 | /// \param __a |
2871 | /// A 128-bit floating-point vector of [4 x float]. |
2872 | /// \returns A 64-bit integer vector of [4 x i16] containing the converted |
2873 | /// values. |
2874 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2875 | _mm_cvtps_pi16(__m128 __a) |
2876 | { |
2877 | __m64 __b, __c; |
2878 | |
2879 | __b = _mm_cvtps_pi32(__a); |
2880 | __a = _mm_movehl_ps(__a, __a); |
2881 | __c = _mm_cvtps_pi32(__a); |
2882 | |
2883 | return _mm_packs_pi32(__b, __c); |
2884 | } |
2885 | |
2886 | /// Converts each single-precision floating-point element of a 128-bit |
2887 | /// floating-point vector of [4 x float] into an 8-bit signed integer, and |
2888 | /// packs the results into the lower 32 bits of a 64-bit integer vector of |
2889 | /// [8 x i8]. The upper 32 bits of the vector are set to 0. |
2890 | /// |
2891 | /// If the floating-point element is NaN or infinity, or if the |
2892 | /// floating-point element is greater than 0x7FFFFFFF or less than -0x80, it |
2893 | /// is converted to 0x80. Otherwise if the floating-point element is greater |
2894 | /// than 0x7F, it is converted to 0x7F. |
2895 | /// |
2896 | /// \headerfile <x86intrin.h> |
2897 | /// |
2898 | /// This intrinsic corresponds to the <c> CVTPS2PI + COMPOSITE </c> instruction. |
2899 | /// |
2900 | /// \param __a |
2901 | /// 128-bit floating-point vector of [4 x float]. |
2902 | /// \returns A 64-bit integer vector of [8 x i8]. The lower 32 bits contain the |
2903 | /// converted values and the uppper 32 bits are set to zero. |
2904 | static __inline__ __m64 __DEFAULT_FN_ATTRS_MMX |
2905 | _mm_cvtps_pi8(__m128 __a) |
2906 | { |
2907 | __m64 __b, __c; |
2908 | |
2909 | __b = _mm_cvtps_pi16(__a); |
2910 | __c = _mm_setzero_si64(); |
2911 | |
2912 | return _mm_packs_pi16(__b, __c); |
2913 | } |
2914 | |
2915 | /// Extracts the sign bits from each single-precision floating-point |
2916 | /// element of a 128-bit floating-point vector of [4 x float] and returns the |
2917 | /// sign bits in bits [0:3] of the result. Bits [31:4] of the result are set |
2918 | /// to zero. |
2919 | /// |
2920 | /// \headerfile <x86intrin.h> |
2921 | /// |
2922 | /// This intrinsic corresponds to the <c> VMOVMSKPS / MOVMSKPS </c> instruction. |
2923 | /// |
2924 | /// \param __a |
2925 | /// A 128-bit floating-point vector of [4 x float]. |
2926 | /// \returns A 32-bit integer value. Bits [3:0] contain the sign bits from each |
2927 | /// single-precision floating-point element of the parameter. Bits [31:4] are |
2928 | /// set to zero. |
2929 | static __inline__ int __DEFAULT_FN_ATTRS |
2930 | _mm_movemask_ps(__m128 __a) |
2931 | { |
2932 | return __builtin_ia32_movmskps((__v4sf)__a); |
2933 | } |
2934 | |
2935 | |
2936 | #define _MM_ALIGN16 __attribute__((aligned(16))) |
2937 | |
2938 | #define _MM_SHUFFLE(z, y, x, w) (((z) << 6) | ((y) << 4) | ((x) << 2) | (w)) |
2939 | |
2940 | #define _MM_EXCEPT_INVALID (0x0001) |
2941 | #define _MM_EXCEPT_DENORM (0x0002) |
2942 | #define _MM_EXCEPT_DIV_ZERO (0x0004) |
2943 | #define _MM_EXCEPT_OVERFLOW (0x0008) |
2944 | #define _MM_EXCEPT_UNDERFLOW (0x0010) |
2945 | #define _MM_EXCEPT_INEXACT (0x0020) |
2946 | #define _MM_EXCEPT_MASK (0x003f) |
2947 | |
2948 | #define _MM_MASK_INVALID (0x0080) |
2949 | #define _MM_MASK_DENORM (0x0100) |
2950 | #define _MM_MASK_DIV_ZERO (0x0200) |
2951 | #define _MM_MASK_OVERFLOW (0x0400) |
2952 | #define _MM_MASK_UNDERFLOW (0x0800) |
2953 | #define _MM_MASK_INEXACT (0x1000) |
2954 | #define _MM_MASK_MASK (0x1f80) |
2955 | |
2956 | #define _MM_ROUND_NEAREST (0x0000) |
2957 | #define _MM_ROUND_DOWN (0x2000) |
2958 | #define _MM_ROUND_UP (0x4000) |
2959 | #define _MM_ROUND_TOWARD_ZERO (0x6000) |
2960 | #define _MM_ROUND_MASK (0x6000) |
2961 | |
2962 | #define _MM_FLUSH_ZERO_MASK (0x8000) |
2963 | #define _MM_FLUSH_ZERO_ON (0x8000) |
2964 | #define _MM_FLUSH_ZERO_OFF (0x0000) |
2965 | |
2966 | #define _MM_GET_EXCEPTION_MASK() (_mm_getcsr() & _MM_MASK_MASK) |
2967 | #define _MM_GET_EXCEPTION_STATE() (_mm_getcsr() & _MM_EXCEPT_MASK) |
2968 | #define _MM_GET_FLUSH_ZERO_MODE() (_mm_getcsr() & _MM_FLUSH_ZERO_MASK) |
2969 | #define _MM_GET_ROUNDING_MODE() (_mm_getcsr() & _MM_ROUND_MASK) |
2970 | |
2971 | #define _MM_SET_EXCEPTION_MASK(x) (_mm_setcsr((_mm_getcsr() & ~_MM_MASK_MASK) | (x))) |
2972 | #define _MM_SET_EXCEPTION_STATE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_EXCEPT_MASK) | (x))) |
2973 | #define _MM_SET_FLUSH_ZERO_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_FLUSH_ZERO_MASK) | (x))) |
2974 | #define _MM_SET_ROUNDING_MODE(x) (_mm_setcsr((_mm_getcsr() & ~_MM_ROUND_MASK) | (x))) |
2975 | |
2976 | #define _MM_TRANSPOSE4_PS(row0, row1, row2, row3) \ |
2977 | do { \ |
2978 | __m128 tmp3, tmp2, tmp1, tmp0; \ |
2979 | tmp0 = _mm_unpacklo_ps((row0), (row1)); \ |
2980 | tmp2 = _mm_unpacklo_ps((row2), (row3)); \ |
2981 | tmp1 = _mm_unpackhi_ps((row0), (row1)); \ |
2982 | tmp3 = _mm_unpackhi_ps((row2), (row3)); \ |
2983 | (row0) = _mm_movelh_ps(tmp0, tmp2); \ |
2984 | (row1) = _mm_movehl_ps(tmp2, tmp0); \ |
2985 | (row2) = _mm_movelh_ps(tmp1, tmp3); \ |
2986 | (row3) = _mm_movehl_ps(tmp3, tmp1); \ |
2987 | } while (0) |
2988 | |
2989 | /* Aliases for compatibility. */ |
2990 | #define _m_pextrw _mm_extract_pi16 |
2991 | #define _m_pinsrw _mm_insert_pi16 |
2992 | #define _m_pmaxsw _mm_max_pi16 |
2993 | #define _m_pmaxub _mm_max_pu8 |
2994 | #define _m_pminsw _mm_min_pi16 |
2995 | #define _m_pminub _mm_min_pu8 |
2996 | #define _m_pmovmskb _mm_movemask_pi8 |
2997 | #define _m_pmulhuw _mm_mulhi_pu16 |
2998 | #define _m_pshufw _mm_shuffle_pi16 |
2999 | #define _m_maskmovq _mm_maskmove_si64 |
3000 | #define _m_pavgb _mm_avg_pu8 |
3001 | #define _m_pavgw _mm_avg_pu16 |
3002 | #define _m_psadbw _mm_sad_pu8 |
3003 | #define _m_ _mm_ |
3004 | #define _m_ _mm_ |
3005 | |
3006 | #undef __DEFAULT_FN_ATTRS |
3007 | #undef __DEFAULT_FN_ATTRS_MMX |
3008 | |
3009 | /* Ugly hack for backwards-compatibility (compatible with gcc) */ |
3010 | #if defined(__SSE2__) && !__building_module(_Builtin_intrinsics) |
3011 | #include <emmintrin.h> |
3012 | #endif |
3013 | |
3014 | #endif /* __XMMINTRIN_H */ |
3015 |