RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-19 02:52:53 +02:00
Code Issues Projects Releases Wiki Activity

[X86][AVX] Extend v2f64 BROADCAST(LOAD) -> BROADCAST_LOAD to v2i64/v4f32/v4i32

Browse Source 
Minor precursor fix for D66004, but helps the SSE41 tests as well as they run with -disable-peephole
This commit is contained in:
Simon Pilgrim 2020-08-01 12:28:10 +01:00
parent f0f5988a8b
commit d8abe46054
2 changed files with 7 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
lib/Target/X86/X86ISelLowering.cpp
View File

@ -36144,15 +36144,16 @@ static SDValue combineTargetShuffle(SDValue N, SelectionDAG &DAG,
}
// vbroadcast(vector load X) -> vbroadcast_load
if (SrcVT == MVT::v2f64 && Src.hasOneUse() &&
ISD::isNormalLoad(Src.getNode())) {
if ((SrcVT == MVT::v2f64 || SrcVT == MVT::v4f32 || SrcVT == MVT::v2i64 ||
SrcVT == MVT::v4i32) &&
Src.hasOneUse() && ISD::isNormalLoad(Src.getNode())) {
LoadSDNode *LN = cast<LoadSDNode>(Src);
// Unless the load is volatile or atomic.
if (LN->isSimple()) {
SDVTList Tys = DAG.getVTList(VT, MVT::Other);
SDValue Ops[] = { LN->getChain(), LN->getBasePtr() };
SDValue Ops[] = {LN->getChain(), LN->getBasePtr()};
SDValue BcastLd = DAG.getMemIntrinsicNode(
X86ISD::VBROADCAST_LOAD, DL, Tys, Ops, MVT::f64,
X86ISD::VBROADCAST_LOAD, DL, Tys, Ops, SrcVT.getScalarType(),
LN->getPointerInfo(), LN->getOriginalAlign(),
LN->getMemOperand()->getFlags());
DCI.CombineTo(N.getNode(), BcastLd);

6
test/CodeGen/X86/sse41.ll
View File

@ -654,8 +654,7 @@ define <4 x i32> @pinsrd_from_shufflevector_i32(<4 x i32> %a, <4 x i32>* nocaptu
; X86-AVX512-LABEL: pinsrd_from_shufflevector_i32:
; X86-AVX512: ## %bb.0: ## %entry
; X86-AVX512-NEXT: movl {{[0-9]+}}(%esp), %eax ## encoding: [0x8b,0x44,0x24,0x04]
; X86-AVX512-NEXT: vmovaps (%eax), %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0x08]
; X86-AVX512-NEXT: vbroadcastss %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x18,0xc9]
; X86-AVX512-NEXT: vbroadcastss (%eax), %xmm1 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x18,0x08]
; X86-AVX512-NEXT: vblendps $8, %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x0c,0xc1,0x08]
; X86-AVX512-NEXT: ## xmm0 = xmm0[0,1,2],xmm1[3]
; X86-AVX512-NEXT: retl ## encoding: [0xc3]
@ -678,8 +677,7 @@ define <4 x i32> @pinsrd_from_shufflevector_i32(<4 x i32> %a, <4 x i32>* nocaptu
;
; X64-AVX512-LABEL: pinsrd_from_shufflevector_i32:
; X64-AVX512: ## %bb.0: ## %entry
; X64-AVX512-NEXT: vmovaps (%rdi), %xmm1 ## EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0x0f]
; X64-AVX512-NEXT: vbroadcastss %xmm1, %xmm1 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x18,0xc9]
; X64-AVX512-NEXT: vbroadcastss (%rdi), %xmm1 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x18,0x0f]
; X64-AVX512-NEXT: vblendps $8, %xmm1, %xmm0, %xmm0 ## encoding: [0xc4,0xe3,0x79,0x0c,0xc1,0x08]
; X64-AVX512-NEXT: ## xmm0 = xmm0[0,1,2],xmm1[3]
; X64-AVX512-NEXT: retq ## encoding: [0xc3]