maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[X86] Correct the implementation of the avx512 masked fmsubadd autoupgrade code to not leave the negate unconnected. 

This was causing us to generate fmaddsub instead of fmsubadd if
rounding control is not 4.
This commit is contained in:
Craig Topper 2020-02-07 15:04:22 -08:00
parent 598d9dd846
commit bb717d3f46
2 changed files with 3 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
llvm/lib/IR/AutoUpgrade.cpp
View File

@ -3259,8 +3259,7 @@ void llvm::UpgradeIntrinsicCall(CallInst *CI, Function *NewFn) {
Ops[2] = Builder.CreateFNeg(Ops[2]);
Rep = Builder.CreateCall(Intrinsic::getDeclaration(F->getParent(), IID),
{CI->getArgOperand(0), CI->getArgOperand(1),
CI->getArgOperand(2), CI->getArgOperand(4)});
Ops);
} else {
int NumElts = CI->getType()->getVectorNumElements();

4
llvm/test/CodeGen/X86/avx512-fma-intrinsics-upgrade.ll
View File

@ -735,7 +735,7 @@ define <8 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_512_rne(<8 x double>
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_512_rne:
; CHECK: ## %bb.0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmaddsub231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT: vfmsubadd231pd {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT: vmovapd %zmm2, %zmm0
; CHECK-NEXT: retq
%res = call <8 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 8)
@ -746,7 +746,7 @@ define <16 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_512_rne(<16 x float>
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_512_rne:
; CHECK: ## %bb.0:
; CHECK-NEXT: kmovw %edi, %k1
; CHECK-NEXT: vfmaddsub231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT: vfmsubadd231ps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT: vmovaps %zmm2, %zmm0
; CHECK-NEXT: retq
%res = call <16 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 8)