34 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=fma | FileCheck %s3 4; If fast-math-flags are propagated correctly, the mul1 expression5; should be recognized as a factor in the last fsub, so we should6; see a mul and add, not a mul and fma:7; a * b - (-a * b) ---> (a * b) + (a * b)8 9define float @fmf_should_not_break_cse(float %a, float %b) {10; CHECK-LABEL: fmf_should_not_break_cse:11; CHECK: # %bb.0:12; CHECK-NEXT: vmulss %xmm1, %xmm0, %xmm013; CHECK-NEXT: vaddss %xmm0, %xmm0, %xmm014; CHECK-NEXT: retq15 %mul1 = fmul fast float %a, %b16 %nega = fsub fast float 0.0, %a17 %mul2 = fmul fast float %nega, %b18 %abx2 = fsub fast float %mul1, %mul219 ret float %abx220}21 22define <4 x float> @fmf_should_not_break_cse_vector(<4 x float> %a, <4 x float> %b) {23; CHECK-LABEL: fmf_should_not_break_cse_vector:24; CHECK: # %bb.0:25; CHECK-NEXT: vmulps %xmm1, %xmm0, %xmm026; CHECK-NEXT: vaddps %xmm0, %xmm0, %xmm027; CHECK-NEXT: retq28 %mul1 = fmul fast <4 x float> %a, %b29 %nega = fsub fast <4 x float> <float 0.0, float 0.0, float 0.0, float 0.0>, %a30 %mul2 = fmul fast <4 x float> %nega, %b31 %abx2 = fsub fast <4 x float> %mul1, %mul232 ret <4 x float> %abx233}34