diff options
Diffstat (limited to 'test/CodeGen/R600/rsq.ll')
| -rw-r--r-- | test/CodeGen/R600/rsq.ll | 40 |
1 files changed, 38 insertions, 2 deletions
diff --git a/test/CodeGen/R600/rsq.ll b/test/CodeGen/R600/rsq.ll index d792c9f..b8a23df 100644 --- a/test/CodeGen/R600/rsq.ll +++ b/test/CodeGen/R600/rsq.ll @@ -1,6 +1,7 @@ -; RUN: llc -march=r600 -mcpu=SI -mattr=-fp32-denormals -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=SI-UNSAFE -check-prefix=SI %s -; RUN: llc -march=r600 -mcpu=SI -mattr=-fp32-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=SI-SAFE -check-prefix=SI %s +; RUN: llc -march=amdgcn -mcpu=SI -mattr=-fp32-denormals -verify-machineinstrs -enable-unsafe-fp-math < %s | FileCheck -check-prefix=SI-UNSAFE -check-prefix=SI %s +; RUN: llc -march=amdgcn -mcpu=SI -mattr=-fp32-denormals -verify-machineinstrs < %s | FileCheck -check-prefix=SI-SAFE -check-prefix=SI %s +declare i32 @llvm.r600.read.tidig.x() nounwind readnone declare float @llvm.sqrt.f32(float) nounwind readnone declare double @llvm.sqrt.f64(double) nounwind readnone @@ -36,3 +37,38 @@ define void @rsq_f32_sgpr(float addrspace(1)* noalias %out, float %val) nounwind store float %div, float addrspace(1)* %out, align 4 ret void } + +; Recognize that this is rsqrt(a) * rcp(b) * c, +; not 1 / ( 1 / sqrt(a)) * rcp(b) * c. + +; SI-LABEL: @rsqrt_fmul +; SI-DAG: buffer_load_dword [[A:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64{{$}} +; SI-DAG: buffer_load_dword [[B:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:4 +; SI-DAG: buffer_load_dword [[C:v[0-9]+]], {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, 0 addr64 offset:8 + +; SI-UNSAFE-DAG: v_rsq_f32_e32 [[RSQA:v[0-9]+]], [[A]] +; SI-UNSAFE-DAG: v_rcp_f32_e32 [[RCPB:v[0-9]+]], [[B]] +; SI-UNSAFE-DAG: v_mul_f32_e32 [[TMP:v[0-9]+]], [[RCPB]], [[RSQA]] +; SI-UNSAFE: v_mul_f32_e32 [[RESULT:v[0-9]+]], [[TMP]], [[C]] +; SI-UNSAFE: buffer_store_dword [[RESULT]] + +; SI-SAFE-NOT: v_rsq_f32 + +; SI: s_endpgm +define void @rsqrt_fmul(float addrspace(1)* %out, float addrspace(1)* %in) { + %tid = call i32 @llvm.r600.read.tidig.x() nounwind readnone + %out.gep = getelementptr float addrspace(1)* %out, i32 %tid + %gep.0 = getelementptr float addrspace(1)* %in, i32 %tid + %gep.1 = getelementptr float addrspace(1)* %gep.0, i32 1 + %gep.2 = getelementptr float addrspace(1)* %gep.0, i32 2 + + %a = load float addrspace(1)* %gep.0 + %b = load float addrspace(1)* %gep.1 + %c = load float addrspace(1)* %gep.2 + + %x = call float @llvm.sqrt.f32(float %a) + %y = fmul float %x, %b + %z = fdiv float %c, %y + store float %z, float addrspace(1)* %out.gep + ret void +} |