diff options
| author | Stephen Hines <srhines@google.com> | 2015-03-23 12:10:34 -0700 |
|---|---|---|
| committer | Stephen Hines <srhines@google.com> | 2015-03-23 12:10:34 -0700 |
| commit | ebe69fe11e48d322045d5949c83283927a0d790b (patch) | |
| tree | c92f1907a6b8006628a4b01615f38264d29834ea /test/CodeGen/R600/llvm.AMDGPU.rsq.clamped.ll | |
| parent | b7d2e72b02a4cb8034f32f8247a2558d2434e121 (diff) | |
| download | external_llvm-ebe69fe11e48d322045d5949c83283927a0d790b.zip external_llvm-ebe69fe11e48d322045d5949c83283927a0d790b.tar.gz external_llvm-ebe69fe11e48d322045d5949c83283927a0d790b.tar.bz2 | |
Update aosp/master LLVM for rebase to r230699.
Change-Id: I2b5be30509658cb8266be782de0ab24f9099f9b9
Diffstat (limited to 'test/CodeGen/R600/llvm.AMDGPU.rsq.clamped.ll')
| -rw-r--r-- | test/CodeGen/R600/llvm.AMDGPU.rsq.clamped.ll | 11 |
1 files changed, 10 insertions, 1 deletions
diff --git a/test/CodeGen/R600/llvm.AMDGPU.rsq.clamped.ll b/test/CodeGen/R600/llvm.AMDGPU.rsq.clamped.ll index 6bf9f0c..eeff253 100644 --- a/test/CodeGen/R600/llvm.AMDGPU.rsq.clamped.ll +++ b/test/CodeGen/R600/llvm.AMDGPU.rsq.clamped.ll @@ -1,4 +1,5 @@ -; RUN: llc -march=r600 -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s +; RUN: llc -march=amdgcn -mcpu=SI -verify-machineinstrs < %s | FileCheck -check-prefix=SI -check-prefix=FUNC %s +; RUN: llc -march=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck -check-prefix=VI -check-prefix=FUNC %s ; RUN: llc -march=r600 -mcpu=cypress -verify-machineinstrs < %s | FileCheck -check-prefix=EG -check-prefix=FUNC %s @@ -6,7 +7,15 @@ declare float @llvm.AMDGPU.rsq.clamped.f32(float) nounwind readnone ; FUNC-LABEL: {{^}}rsq_clamped_f32: ; SI: v_rsq_clamp_f32_e32 + +; VI: v_rsq_f32_e32 [[RSQ:v[0-9]+]], {{s[0-9]+}} +; VI: v_min_f32_e32 [[MIN:v[0-9]+]], 0x7f7fffff, [[RSQ]] +; TODO: this constant should be folded: +; VI: v_mov_b32_e32 [[MINFLT:v[0-9]+]], 0xff7fffff +; VI: v_max_f32_e32 {{v[0-9]+}}, [[MIN]], [[MINFLT]] + ; EG: RECIPSQRT_CLAMPED + define void @rsq_clamped_f32(float addrspace(1)* %out, float %src) nounwind { %rsq_clamped = call float @llvm.AMDGPU.rsq.clamped.f32(float %src) nounwind readnone store float %rsq_clamped, float addrspace(1)* %out, align 4 |