Skip to content

Commit

Permalink
[X86] LowerVectorAllZero - lower to CMP(MOVMSK(NOT(X)),0) instead of …
Browse files Browse the repository at this point in the history 
…CMP(MOVMSK(X),65535)

In most cases the NOT will still be scalarized, but it allows us to perform the CMP(X,0) combines inside combineCMP()
  • Loading branch information
@RKSimon
RKSimon committed Mar 23, 2023
1 parent 16b6826 commit ec294d2
Show file tree
Hide file tree
Showing 4 changed files with 33 additions and 33 deletions.
3 changes: 2 additions & 1 deletion llvm/lib/Target/X86/X86ISelLowering.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -24224,9 +24224,10 @@ static SDValue LowerVectorAllZero(const SDLoc &DL, SDValue V, ISD::CondCode CC,
V = DAG.getBitcast(MVT::v16i8, MaskBits(V));
V = DAG.getNode(X86ISD::PCMPEQ, DL, MVT::v16i8, V,
getZeroVector(MVT::v16i8, Subtarget, DAG, DL));
V = DAG.getNOT(DL, V, MVT::v16i8);
V = DAG.getNode(X86ISD::MOVMSK, DL, MVT::i32, V);
return DAG.getNode(X86ISD::CMP, DL, MVT::i32, V,
DAG.getConstant(0xFFFF, DL, MVT::i32));
DAG.getConstant(0, DL, MVT::i32));
}

// Check whether an OR'd reduction tree is PTEST-able, or if we can fallback to
Expand Down
4 changes: 2 additions & 2 deletions llvm/test/CodeGen/X86/pr45378.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@ define i1 @parseHeaders(ptr %ptr) nounwind {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -45,7 +45,7 @@ define i1 @parseHeaders2_scalar_or(ptr %ptr) nounwind {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down
18 changes: 9 additions & 9 deletions llvm/test/CodeGen/X86/ptest.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -10,7 +10,7 @@ define i32 @veccond128(<4 x i32> %input) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: je .LBB0_2
; SSE2-NEXT: # %bb.1: # %if-true-block
; SSE2-NEXT: xorl %eax, %eax
Expand Down Expand Up @@ -57,7 +57,7 @@ define i32 @veccond256(<8 x i32> %input) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: je .LBB1_2
; SSE2-NEXT: # %bb.1: # %if-true-block
; SSE2-NEXT: xorl %eax, %eax
Expand Down Expand Up @@ -109,7 +109,7 @@ define i32 @veccond512(<16 x i32> %input) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: je .LBB2_2
; SSE2-NEXT: # %bb.1: # %if-true-block
; SSE2-NEXT: xorl %eax, %eax
Expand Down Expand Up @@ -176,7 +176,7 @@ define i32 @vectest128(<4 x i32> %input) {
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %ecx
; SSE2-NEXT: xorl %eax, %eax
; SSE2-NEXT: cmpl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -207,7 +207,7 @@ define i32 @vectest256(<8 x i32> %input) {
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %ecx
; SSE2-NEXT: xorl %eax, %eax
; SSE2-NEXT: cmpl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -242,7 +242,7 @@ define i32 @vectest512(<16 x i32> %input) {
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %ecx
; SSE2-NEXT: xorl %eax, %eax
; SSE2-NEXT: cmpl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -286,7 +286,7 @@ define i32 @vecsel128(<4 x i32> %input, i32 %a, i32 %b) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %ecx
; SSE2-NEXT: cmpl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: cmovel %esi, %eax
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -317,7 +317,7 @@ define i32 @vecsel256(<8 x i32> %input, i32 %a, i32 %b) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %ecx
; SSE2-NEXT: cmpl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: cmovel %esi, %eax
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -352,7 +352,7 @@ define i32 @vecsel512(<16 x i32> %input, i32 %a, i32 %b) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %ecx
; SSE2-NEXT: cmpl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %ecx # imm = 0xFFFF
; SSE2-NEXT: cmovel %esi, %eax
; SSE2-NEXT: retq
;
Expand Down
41 changes: 20 additions & 21 deletions llvm/test/CodeGen/X86/vector-reduce-or-cmp.ll
View file
Original file line number Diff line number Diff line change
Expand Up @@ -16,7 +16,7 @@ define i1 @test_v2i64(<2 x i64> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand All @@ -43,7 +43,7 @@ define i1 @test_v4i64(<4 x i64> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -74,7 +74,7 @@ define i1 @test_v8i64(<8 x i64> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -128,7 +128,7 @@ define i1 @test_v16i64(<16 x i64> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -207,7 +207,7 @@ define i1 @test_v4i32(<4 x i32> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand All @@ -234,7 +234,7 @@ define i1 @test_v8i32(<8 x i32> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -265,7 +265,7 @@ define i1 @test_v16i32(<16 x i32> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -319,7 +319,7 @@ define i1 @test_v32i32(<32 x i32> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -417,7 +417,7 @@ define i1 @test_v8i16(<8 x i16> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand All @@ -444,7 +444,7 @@ define i1 @test_v16i16(<16 x i16> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -475,7 +475,7 @@ define i1 @test_v32i16(<32 x i16> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -529,7 +529,7 @@ define i1 @test_v64i16(<64 x i16> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -646,7 +646,7 @@ define i1 @test_v16i8(<16 x i8> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand All @@ -673,7 +673,7 @@ define i1 @test_v32i8(<32 x i8> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -704,7 +704,7 @@ define i1 @test_v64i8(<64 x i8> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -758,7 +758,7 @@ define i1 @test_v128i8(<128 x i8> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -866,7 +866,7 @@ define i1 @mask_v8i32(<8 x i32> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -913,7 +913,7 @@ define i1 @trunc_v16i16(<16 x i16> %a0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: setne %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -964,8 +964,7 @@ define i1 @mask_v128i8(<128 x i8> %a0) {
; SSE2-NEXT: por %xmm1, %xmm0
; SSE2-NEXT: psllw $7, %xmm0
; SSE2-NEXT: pmovmskb %xmm0, %eax
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: testl %eax, %eax
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down Expand Up @@ -1026,7 +1025,7 @@ define zeroext i1 @PR44781(ptr %0) {
; SSE2-NEXT: pxor %xmm1, %xmm1
; SSE2-NEXT: pcmpeqb %xmm0, %xmm1
; SSE2-NEXT: pmovmskb %xmm1, %eax
; SSE2-NEXT: cmpl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: xorl $65535, %eax # imm = 0xFFFF
; SSE2-NEXT: sete %al
; SSE2-NEXT: retq
;
Expand Down

0 comments on commit ec294d2

Please sign in to comment.