RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-02-13 08:54:59 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[TargetLowering] SimplifyDemandedVectorElts - support MUL and ANY_EXTEND_VECTOR_INREG

Browse Source 
Also fold ANY_EXTEND_VECTOR_INREG -> BITCAST if we only need the bottom element.

Fixes temporary regression introduced in rL363693.

llvm-svn: 363694
This commit is contained in:
Simon Pilgrim 2019-06-18 15:49:35 +00:00
parent 4133ecc78b
commit e16283f391
2 changed files with 9 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
lib/CodeGen/SelectionDAG/TargetLowering.cpp
View File

@ -2156,6 +2156,7 @@ bool TargetLowering::SimplifyDemandedVectorElts(
}
break;
}
case ISD::ANY_EXTEND_VECTOR_INREG:
case ISD::SIGN_EXTEND_VECTOR_INREG:
case ISD::ZERO_EXTEND_VECTOR_INREG: {
APInt SrcUndef, SrcZero;
@ -2168,6 +2169,13 @@ bool TargetLowering::SimplifyDemandedVectorElts(
KnownZero = SrcZero.zextOrTrunc(NumElts);
KnownUndef = SrcUndef.zextOrTrunc(NumElts);
if (Op.getOpcode() == ISD::ANY_EXTEND_VECTOR_INREG &&
Op.getValueSizeInBits() == Src.getValueSizeInBits() &&
DemandedSrcElts == 1 && TLO.DAG.getDataLayout().isLittleEndian()) {
// aext - if we just need the bottom element then we can bitcast.
return TLO.CombineTo(Op, TLO.DAG.getBitcast(VT, Src));
}
if (Op.getOpcode() == ISD::ZERO_EXTEND_VECTOR_INREG) {
// zext(undef) upper bits are guaranteed to be zero.
if (DemandedElts.isSubsetOf(KnownUndef))
@ -2201,6 +2209,7 @@ bool TargetLowering::SimplifyDemandedVectorElts(
KnownUndef = getKnownUndefForVectorBinop(Op, TLO.DAG, UndefLHS, UndefRHS);
break;
}
case ISD::MUL:
case ISD::AND: {
APInt SrcUndef, SrcZero;
if (SimplifyDemandedVectorElts(Op.getOperand(1), DemandedElts, SrcUndef,

1
test/CodeGen/X86/vector-reduce-mul-widen.ll
View File

@ -1837,7 +1837,6 @@ define i8 @test_v16i8(<16 x i8> %a0) {
; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpackuswb %xmm0, %xmm0, %xmm1
; AVX2-NEXT: vpsrlw $8, %xmm1, %xmm1
; AVX2-NEXT: vpmovzxbw {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero,xmm1[2],zero,xmm1[3],zero,xmm1[4],zero,xmm1[5],zero,xmm1[6],zero,xmm1[7],zero
; AVX2-NEXT: vpmullw %xmm1, %xmm0, %xmm0
; AVX2-NEXT: vpextrb $0, %xmm0, %eax
; AVX2-NEXT: # kill: def $al killed $al killed $eax