| // NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 6 |
| // REQUIRES: amdgpu-registered-target |
| |
| // RUN: %clang --target=%target -mcpu=%cpu %libclc_lib -cl-std=CL3.0 -O2 -fno-discard-value-names -emit-llvm -S -o - %s | FileCheck %s --check-prefix=%check_prefix |
| |
| // AMDGCN-LABEL: define hidden <4 x float> @test_float4( |
| // AMDGCN-SAME: <4 x float> noundef [[X:%.*]], <4 x float> noundef [[Y:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { |
| // AMDGCN-NEXT: [[ENTRY:.*:]] |
| // AMDGCN-NEXT: [[TMP0:%.*]] = extractelement <4 x float> [[X]], i64 1 |
| // AMDGCN-NEXT: [[TMP1:%.*]] = extractelement <4 x float> [[Y]], i64 2 |
| // AMDGCN-NEXT: [[TMP2:%.*]] = fmul contract float [[TMP0]], [[TMP1]] |
| // AMDGCN-NEXT: [[TMP3:%.*]] = extractelement <4 x float> [[X]], i64 2 |
| // AMDGCN-NEXT: [[TMP4:%.*]] = extractelement <4 x float> [[Y]], i64 1 |
| // AMDGCN-NEXT: [[TMP5:%.*]] = fmul contract float [[TMP3]], [[TMP4]] |
| // AMDGCN-NEXT: [[TMP6:%.*]] = fsub contract float [[TMP2]], [[TMP5]] |
| // AMDGCN-NEXT: [[TMP7:%.*]] = extractelement <4 x float> [[Y]], i64 0 |
| // AMDGCN-NEXT: [[TMP8:%.*]] = fmul contract float [[TMP3]], [[TMP7]] |
| // AMDGCN-NEXT: [[TMP9:%.*]] = extractelement <4 x float> [[X]], i64 0 |
| // AMDGCN-NEXT: [[TMP10:%.*]] = fmul contract float [[TMP9]], [[TMP1]] |
| // AMDGCN-NEXT: [[TMP11:%.*]] = fsub contract float [[TMP8]], [[TMP10]] |
| // AMDGCN-NEXT: [[TMP12:%.*]] = fmul contract float [[TMP9]], [[TMP4]] |
| // AMDGCN-NEXT: [[TMP13:%.*]] = fmul contract float [[TMP0]], [[TMP7]] |
| // AMDGCN-NEXT: [[TMP14:%.*]] = fsub contract float [[TMP12]], [[TMP13]] |
| // AMDGCN-NEXT: [[TMP15:%.*]] = insertelement <4 x float> <float poison, float poison, float poison, float 0.000000e+00>, float [[TMP6]], i64 0 |
| // AMDGCN-NEXT: [[TMP16:%.*]] = insertelement <4 x float> [[TMP15]], float [[TMP11]], i64 1 |
| // AMDGCN-NEXT: [[TMP17:%.*]] = insertelement <4 x float> [[TMP16]], float [[TMP14]], i64 2 |
| // AMDGCN-NEXT: ret <4 x float> [[TMP17]] |
| // |
| float4 test_float4(float4 x, float4 y) { |
| return cross(x, y); |
| } |
