| 1 | ; Simple bit of IR to mimic CUDA's libdevice. We want to be |
|---|
| 2 | ; able to link with it and we need to make sure all __nvvm_reflect |
|---|
| 3 | ; calls are eliminated by the time PTX has been produced. |
|---|
| 4 | |
|---|
| 5 | target triple = "nvptx-unknown-cuda" |
|---|
| 6 | |
|---|
| 7 | declare i32 @__nvvm_reflect(i8*) |
|---|
| 8 | |
|---|
| 9 | @"$str" = private addrspace(1) constant [8 x i8] c"USE_MUL\00" |
|---|
| 10 | |
|---|
| 11 | define void @unused_subfunc(float %a) { |
|---|
| 12 | ret void |
|---|
| 13 | } |
|---|
| 14 | |
|---|
| 15 | define void @used_subfunc(float %a) { |
|---|
| 16 | ret void |
|---|
| 17 | } |
|---|
| 18 | |
|---|
| 19 | define float @_Z17device_mul_or_addff(float %a, float %b) { |
|---|
| 20 | %reflect = call i32 @__nvvm_reflect(i8* addrspacecast (i8 addrspace(1)* getelementptr inbounds ([8 x i8], [8 x i8] addrspace(1)* @"$str", i32 0, i32 0) to i8*)) |
|---|
| 21 | %cmp = icmp ne i32 %reflect, 0 |
|---|
| 22 | br i1 %cmp, label %use_mul, label %use_add |
|---|
| 23 | |
|---|
| 24 | use_mul: |
|---|
| 25 | %ret1 = fmul float %a, %b |
|---|
| 26 | br label %exit |
|---|
| 27 | |
|---|
| 28 | use_add: |
|---|
| 29 | %ret2 = fadd float %a, %b |
|---|
| 30 | br label %exit |
|---|
| 31 | |
|---|
| 32 | exit: |
|---|
| 33 | %ret = phi float [%ret1, %use_mul], [%ret2, %use_add] |
|---|
| 34 | |
|---|
| 35 | call void @used_subfunc(float %ret) |
|---|
| 36 | |
|---|
| 37 | ret float %ret |
|---|
| 38 | } |
|---|
| 39 | |
|---|