1 // RUN: %clang_cc1 -fcuda-is-device \
2 // RUN: -triple nvptx-nvidia-cuda -emit-llvm -o - %s | \
3 // RUN: FileCheck -check-prefixes=NOFTZ,PTXNOFTZ %s
4
5 // RUN: %clang_cc1 -fcuda-is-device -fdenormal-fp-math-f32=ieee \
6 // RUN: -triple nvptx-nvidia-cuda -emit-llvm -o - %s | \
7 // RUN: FileCheck -check-prefixes=NOFTZ,PTXNOFTZ %s
8
9 // RUN: %clang_cc1 -fcuda-is-device -fdenormal-fp-math-f32=preserve-sign \
10 // RUN: -triple nvptx-nvidia-cuda -emit-llvm -o - %s | \
11 // RUN: FileCheck -check-prefixes=FTZ,PTXFTZ %s
12
13 // RUN: %clang_cc1 -fcuda-is-device -x hip \
14 // RUN: -triple amdgcn-amd-amdhsa -target-cpu gfx900 -emit-llvm -o - %s | \
15 // RUN: FileCheck -check-prefix=NOFTZ %s
16
17 // RUN: %clang_cc1 -fcuda-is-device -x hip \
18 // RUN: -triple amdgcn-amd-amdhsa -target-cpu gfx900 -fdenormal-fp-math-f32=ieee -emit-llvm -o - %s | \
19 // RUN: FileCheck -check-prefix=NOFTZ %s
20
21 // RUN: %clang_cc1 -fcuda-is-device -x hip -fdenormal-fp-math-f32=preserve-sign \
22 // RUN: -triple amdgcn-amd-amdhsa -target-cpu gfx900 -emit-llvm -o - %s | \
23 // RUN: FileCheck -check-prefix=FTZ %s
24
25 #include "Inputs/cuda.h"
26
27 // Checks that device function calls get emitted with the "denormal-fp-math-f32"
28 // attribute set when we compile CUDA device code with
29 // -fdenormal-fp-math-f32. Further, check that we reflect the presence or
30 // absence of -fgpu-flush-denormals-to-zero in a module flag.
31
32 // AMDGCN targets always have f64/f16 denormals enabled.
33 //
34 // AMDGCN targets without fast FMAF (e.g. gfx803) always have f32 denormal
35 // flushing by default.
36 //
37 // For AMDGCN target with fast FMAF (e.g. gfx900), it has ieee denormals by
38 // default and preserve-sign when there with the option
39 // -fgpu-flush-denormals-to-zero.
40
41 // CHECK-LABEL: define void @foo() #0
foo()42 extern "C" __device__ void foo() {}
43
44 // FTZ: attributes #0 = {{.*}} "denormal-fp-math-f32"="preserve-sign,preserve-sign"
45 // NOFTZ-NOT: "denormal-fp-math-f32"
46
47 // PTXFTZ:!llvm.module.flags = !{{{.*}}[[MODFLAG:![0-9]+]]}
48 // PTXFTZ:[[MODFLAG]] = !{i32 4, !"nvvm-reflect-ftz", i32 1}
49
50 // PTXNOFTZ:!llvm.module.flags = !{{{.*}}[[MODFLAG:![0-9]+]]}
51 // PTXNOFTZ:[[MODFLAG]] = !{i32 4, !"nvvm-reflect-ftz", i32 0}
52