RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-02-25 23:06:41 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[X86] Teach materializeVectorConstant to not call getZeroVector/getOnesVector on the types we already have isel patterns for.

Browse Source 
llvm-svn: 371343
This commit is contained in:
Craig Topper 2019-09-08 19:24:29 +00:00
parent 6736ed2107
commit 09af94b030
2 changed files with 4 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
lib/Target/X86/X86ISelLowering.cpp
View File

@ -9137,7 +9137,8 @@ static SDValue materializeVectorConstant(SDValue Op, SelectionDAG &DAG,
if (ISD::isBuildVectorAllZeros(Op.getNode())) {
// Canonicalize this to <4 x i32> to 1) ensure the zero vectors are CSE'd
// and 2) ensure that i64 scalars are eliminated on x86-32 hosts.
if (VT == MVT::v4i32 || VT == MVT::v8i32 || VT == MVT::v16i32)
if (VT.isFloatingPoint() ||
VT == MVT::v4i32 || VT == MVT::v8i32 || VT == MVT::v16i32)
return Op;
return getZeroVector(VT, Subtarget, DAG, DL);
@ -9147,8 +9148,7 @@ static SDValue materializeVectorConstant(SDValue Op, SelectionDAG &DAG,
// vectors or broken into v4i32 operations on 256-bit vectors. AVX2 can use
// vpcmpeqd on 256-bit vectors.
if (Subtarget.hasSSE2() && ISD::isBuildVectorAllOnes(Op.getNode())) {
if (VT == MVT::v4i32 || VT == MVT::v16i32 ||
(VT == MVT::v8i32 && Subtarget.hasInt256()))
if (VT == MVT::v4i32 || VT == MVT::v8i32 || VT == MVT::v16i32)
return Op;
return getOnesVector(VT, DAG, DL);

2
test/CodeGen/X86/fold-load-vec.ll
View File

@ -12,7 +12,7 @@ define void @sample_test(<4 x float>* %source, <2 x float>* %dest) nounwind {
; CHECK-NEXT: movq %rsi, {{[0-9]+}}(%rsp)
; CHECK-NEXT: xorps %xmm0, %xmm0
; CHECK-NEXT: movlps %xmm0, (%rsp)
; CHECK-NEXT: insertps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[2,3]
; CHECK-NEXT: unpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT: movlps %xmm0, (%rsp)
; CHECK-NEXT: movlps %xmm0, (%rsi)
; CHECK-NEXT: movq {{[0-9]+}}(%rsp), %rax