RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-12-11 21:45:16 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[X86][MMX] Fixed i32 extraction on 32-bit targets

Browse Source 
MMX extraction often ends up as extract_i32(bitcast_v2i32(extract_i64(bitcast_v1i64(x86mmx v), 0)), 0) which fails to simplify on 32-bit targets as i64 isn't legal

llvm-svn: 296782
This commit is contained in:
Simon Pilgrim 2017-03-02 18:56:06 +00:00
parent e42e2cd56a
commit c4977030a6
2 changed files with 16 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
lib/Target/X86/X86ISelLowering.cpp
View File

@ -29027,6 +29027,16 @@ static SDValue combineExtractVectorElt(SDNode *N, SelectionDAG &DAG,
EVT VT = N->getValueType(0);
SDLoc dl(InputVector);
// Detect mmx extraction of all bits as a i64. It works better as a bitcast.
if (InputVector.getOpcode() == ISD::BITCAST && InputVector.hasOneUse() &&
VT == MVT::i64 && SrcVT == MVT::v1i64 && isNullConstant(EltIdx)) {
SDValue MMXSrc = InputVector.getOperand(0);
// The bitcast source is a direct mmx result.
if (MMXSrc.getValueType() == MVT::x86mmx)
return DAG.getBitcast(VT, InputVector);
}
// Detect mmx to i32 conversion through a v2i32 elt extract.
if (InputVector.getOpcode() == ISD::BITCAST && InputVector.hasOneUse() &&
VT == MVT::i32 && SrcVT == MVT::v2i32 && isNullConstant(EltIdx)) {

27
test/CodeGen/X86/vec_extract-mmx.ll
View File

@ -8,15 +8,14 @@ define i32 @test0(<1 x i64>* %v4) nounwind {
; X32-NEXT: pushl %ebp
; X32-NEXT: movl %esp, %ebp
; X32-NEXT: andl $-8, %esp
; X32-NEXT: subl $24, %esp
; X32-NEXT: subl $8, %esp
; X32-NEXT: movl 8(%ebp), %eax
; X32-NEXT: movl (%eax), %ecx
; X32-NEXT: movl 4(%eax), %eax
; X32-NEXT: movl %eax, {{[0-9]+}}(%esp)
; X32-NEXT: movl %ecx, (%esp)
; X32-NEXT: pshufw $238, (%esp), %mm0 # mm0 = mem[2,3,2,3]
; X32-NEXT: movq %mm0, {{[0-9]+}}(%esp)
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-NEXT: movd %mm0, %eax
; X32-NEXT: addl $32, %eax
; X32-NEXT: movl %ebp, %esp
; X32-NEXT: popl %ebp
@ -45,18 +44,11 @@ entry:
define i32 @test1(i32* nocapture readonly %ptr) nounwind {
; X32-LABEL: test1:
; X32: # BB#0: # %entry
; X32-NEXT: pushl %ebp
; X32-NEXT: movl %esp, %ebp
; X32-NEXT: andl $-8, %esp
; X32-NEXT: subl $16, %esp
; X32-NEXT: movl 8(%ebp), %eax
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-NEXT: movd (%eax), %mm0
; X32-NEXT: pshufw $232, %mm0, %mm0 # mm0 = mm0[0,2,2,3]
; X32-NEXT: movq %mm0, (%esp)
; X32-NEXT: movl (%esp), %eax
; X32-NEXT: movd %mm0, %eax
; X32-NEXT: emms
; X32-NEXT: movl %ebp, %esp
; X32-NEXT: popl %ebp
; X32-NEXT: retl
;
; X64-LABEL: test1:
@ -87,17 +79,10 @@ entry:
define i32 @test2(i32* nocapture readonly %ptr) nounwind {
; X32-LABEL: test2:
; X32: # BB#0: # %entry
; X32-NEXT: pushl %ebp
; X32-NEXT: movl %esp, %ebp
; X32-NEXT: andl $-8, %esp
; X32-NEXT: subl $16, %esp
; X32-NEXT: movl 8(%ebp), %eax
; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
; X32-NEXT: pshufw $232, (%eax), %mm0 # mm0 = mem[0,2,2,3]
; X32-NEXT: movq %mm0, (%esp)
; X32-NEXT: movl (%esp), %eax
; X32-NEXT: movd %mm0, %eax
; X32-NEXT: emms
; X32-NEXT: movl %ebp, %esp
; X32-NEXT: popl %ebp
; X32-NEXT: retl
;
; X64-LABEL: test2: