[llvm] cce6490 - [X86] combineSetCCMOVMSK - match all_of patterns with X86ISD::CMP as well as X86ISD::SUB
Simon Pilgrim via llvm-commits
llvm-commits at lists.llvm.org
Fri Jan 28 03:43:32 PST 2022
Author: Simon Pilgrim
Date: 2022-01-28T11:43:10Z
New Revision: cce6490eca5be1661178078e7cccdcdfbda622f4
URL: https://github.com/llvm/llvm-project/commit/cce6490eca5be1661178078e7cccdcdfbda622f4
DIFF: https://github.com/llvm/llvm-project/commit/cce6490eca5be1661178078e7cccdcdfbda622f4.diff
LOG: [X86] combineSetCCMOVMSK - match all_of patterns with X86ISD::CMP as well as X86ISD::SUB
Previous folds by combineSetCCMOVMSK might have converted these to CMP when changing the bitwidth, and the CMP->SUB fold might not have happened (or will happen)
Added:
Modified:
llvm/lib/Target/X86/X86ISelLowering.cpp
llvm/test/CodeGen/X86/movmsk-cmp.ll
llvm/test/CodeGen/X86/vector-compare-all_of.ll
llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 961c39832b62..ecd3831317f8 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -44467,8 +44467,8 @@ static SDValue combineSetCCMOVMSK(SDValue EFLAGS, X86::CondCode &CC,
unsigned NumEltBits = VecVT.getScalarSizeInBits();
bool IsAnyOf = CmpOpcode == X86ISD::CMP && CmpVal.isZero();
- bool IsAllOf = CmpOpcode == X86ISD::SUB && NumElts <= CmpBits &&
- CmpVal.isMask(NumElts);
+ bool IsAllOf = (CmpOpcode == X86ISD::SUB || CmpOpcode == X86ISD::CMP) &&
+ NumElts <= CmpBits && CmpVal.isMask(NumElts);
if (!IsAnyOf && !IsAllOf)
return SDValue();
diff --git a/llvm/test/CodeGen/X86/movmsk-cmp.ll b/llvm/test/CodeGen/X86/movmsk-cmp.ll
index 67ebbe979ee5..925f0a9eecd4 100644
--- a/llvm/test/CodeGen/X86/movmsk-cmp.ll
+++ b/llvm/test/CodeGen/X86/movmsk-cmp.ll
@@ -661,10 +661,8 @@ define i1 @allones_v16i32_sign(<16 x i32> %arg) {
; AVX2-NEXT: vpcmpgtd %ymm1, %ymm2, %ymm1
; AVX2-NEXT: vpcmpgtd %ymm0, %ymm2, %ymm0
; AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
-; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vpmovmskb %xmm0, %eax
-; AVX2-NEXT: cmpw $-1, %ax
+; AVX2-NEXT: vpmovmskb %ymm0, %eax
+; AVX2-NEXT: cmpl $-1, %eax
; AVX2-NEXT: sete %al
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
@@ -1047,12 +1045,8 @@ define i1 @allzeros_v8i32_not(<8 x i32> %a0) {
; AVX1-LABEL: allzeros_v8i32_not:
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vmovmskps %xmm0, %eax
-; AVX1-NEXT: cmpl $15, %eax
+; AVX1-NEXT: vpor %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vptest %xmm0, %xmm0
; AVX1-NEXT: setne %al
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
@@ -2025,10 +2019,8 @@ define i1 @allones_v16i32_and1(<16 x i32> %arg) {
; AVX2-NEXT: vpslld $31, %ymm0, %ymm0
; AVX2-NEXT: vpsrad $31, %ymm0, %ymm0
; AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
-; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vpmovmskb %xmm0, %eax
-; AVX2-NEXT: cmpw $-1, %ax
+; AVX2-NEXT: vpmovmskb %ymm0, %eax
+; AVX2-NEXT: cmpl $-1, %eax
; AVX2-NEXT: sete %al
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
@@ -3316,10 +3308,8 @@ define i1 @allones_v16i32_and4(<16 x i32> %arg) {
; AVX2-NEXT: vpslld $29, %ymm0, %ymm0
; AVX2-NEXT: vpsrad $31, %ymm0, %ymm0
; AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
-; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vpmovmskb %xmm0, %eax
-; AVX2-NEXT: cmpw $-1, %ax
+; AVX2-NEXT: vpmovmskb %ymm0, %eax
+; AVX2-NEXT: cmpl $-1, %eax
; AVX2-NEXT: sete %al
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
diff --git a/llvm/test/CodeGen/X86/vector-compare-all_of.ll b/llvm/test/CodeGen/X86/vector-compare-all_of.ll
index 0d2331b08ede..7f9fad7aab49 100644
--- a/llvm/test/CodeGen/X86/vector-compare-all_of.ll
+++ b/llvm/test/CodeGen/X86/vector-compare-all_of.ll
@@ -1236,12 +1236,11 @@ define i1 @bool_reduction_v8i32(<8 x i32> %x, <8 x i32> %y) {
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
; AVX1-NEXT: vpminud %xmm2, %xmm3, %xmm2
-; AVX1-NEXT: vpcmpeqd %xmm2, %xmm3, %xmm2
+; AVX1-NEXT: vpsubd %xmm2, %xmm3, %xmm2
; AVX1-NEXT: vpminud %xmm1, %xmm0, %xmm1
-; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vmovmskps %xmm0, %eax
-; AVX1-NEXT: cmpl $15, %eax
+; AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vpor %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vptest %xmm0, %xmm0
; AVX1-NEXT: sete %al
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
diff --git a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
index d2011bed2d3a..30df18ad2f2b 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
@@ -1063,25 +1063,12 @@ define i1 @icmp0_v4i64_v4i1(<4 x i64>) {
; SSE41-NEXT: sete %al
; SSE41-NEXT: retq
;
-; AVX1-LABEL: icmp0_v4i64_v4i1:
-; AVX1: # %bb.0:
-; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpcmpeqq %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpcmpeqq %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vmovmskpd %xmm0, %eax
-; AVX1-NEXT: cmpl $3, %eax
-; AVX1-NEXT: sete %al
-; AVX1-NEXT: vzeroupper
-; AVX1-NEXT: retq
-;
-; AVX2-LABEL: icmp0_v4i64_v4i1:
-; AVX2: # %bb.0:
-; AVX2-NEXT: vptest %ymm0, %ymm0
-; AVX2-NEXT: sete %al
-; AVX2-NEXT: vzeroupper
-; AVX2-NEXT: retq
+; AVX-LABEL: icmp0_v4i64_v4i1:
+; AVX: # %bb.0:
+; AVX-NEXT: vptest %ymm0, %ymm0
+; AVX-NEXT: sete %al
+; AVX-NEXT: vzeroupper
+; AVX-NEXT: retq
;
; AVX512F-LABEL: icmp0_v4i64_v4i1:
; AVX512F: # %bb.0:
@@ -1129,25 +1116,12 @@ define i1 @icmp0_v8i32_v8i1(<8 x i32>) {
; SSE-NEXT: sete %al
; SSE-NEXT: retq
;
-; AVX1-LABEL: icmp0_v8i32_v8i1:
-; AVX1: # %bb.0:
-; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1
-; AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT: vpcmpeqd %xmm2, %xmm1, %xmm1
-; AVX1-NEXT: vpcmpeqd %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vmovmskps %xmm0, %eax
-; AVX1-NEXT: cmpl $15, %eax
-; AVX1-NEXT: sete %al
-; AVX1-NEXT: vzeroupper
-; AVX1-NEXT: retq
-;
-; AVX2-LABEL: icmp0_v8i32_v8i1:
-; AVX2: # %bb.0:
-; AVX2-NEXT: vptest %ymm0, %ymm0
-; AVX2-NEXT: sete %al
-; AVX2-NEXT: vzeroupper
-; AVX2-NEXT: retq
+; AVX-LABEL: icmp0_v8i32_v8i1:
+; AVX: # %bb.0:
+; AVX-NEXT: vptest %ymm0, %ymm0
+; AVX-NEXT: sete %al
+; AVX-NEXT: vzeroupper
+; AVX-NEXT: retq
;
; AVX512F-LABEL: icmp0_v8i32_v8i1:
; AVX512F: # %bb.0:
@@ -1439,10 +1413,8 @@ define i1 @icmp0_v16i32_v16i1(<16 x i32>) {
; AVX2-NEXT: vpcmpeqd %ymm2, %ymm1, %ymm1
; AVX2-NEXT: vpcmpeqd %ymm2, %ymm0, %ymm0
; AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
-; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vpmovmskb %xmm0, %eax
-; AVX2-NEXT: cmpw $-1, %ax
+; AVX2-NEXT: vpmovmskb %ymm0, %eax
+; AVX2-NEXT: cmpl $-1, %eax
; AVX2-NEXT: sete %al
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
@@ -1876,11 +1848,10 @@ define i1 @icmp_v4i64_v4i1(<4 x i64>, <4 x i64>) {
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT: vpcmpeqq %xmm2, %xmm3, %xmm2
-; AVX1-NEXT: vpcmpeqq %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vmovmskpd %xmm0, %eax
-; AVX1-NEXT: cmpl $3, %eax
+; AVX1-NEXT: vpsubq %xmm2, %xmm3, %xmm2
+; AVX1-NEXT: vpsubq %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vpor %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vptest %xmm0, %xmm0
; AVX1-NEXT: sete %al
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
@@ -1944,11 +1915,10 @@ define i1 @icmp_v8i32_v8i1(<8 x i32>, <8 x i32>) {
; AVX1: # %bb.0:
; AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2
; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3
-; AVX1-NEXT: vpcmpeqd %xmm2, %xmm3, %xmm2
-; AVX1-NEXT: vpcmpeqd %xmm1, %xmm0, %xmm0
-; AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0
-; AVX1-NEXT: vmovmskps %xmm0, %eax
-; AVX1-NEXT: cmpl $15, %eax
+; AVX1-NEXT: vpsubd %xmm2, %xmm3, %xmm2
+; AVX1-NEXT: vpsubd %xmm1, %xmm0, %xmm0
+; AVX1-NEXT: vpor %xmm2, %xmm0, %xmm0
+; AVX1-NEXT: vptest %xmm0, %xmm0
; AVX1-NEXT: sete %al
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
@@ -2280,10 +2250,8 @@ define i1 @icmp_v16i32_v16i1(<16 x i32>, <16 x i32>) {
; AVX2-NEXT: vpcmpeqd %ymm3, %ymm1, %ymm1
; AVX2-NEXT: vpcmpeqd %ymm2, %ymm0, %ymm0
; AVX2-NEXT: vpackssdw %ymm1, %ymm0, %ymm0
-; AVX2-NEXT: vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT: vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT: vpmovmskb %xmm0, %eax
-; AVX2-NEXT: cmpw $-1, %ax
+; AVX2-NEXT: vpmovmskb %ymm0, %eax
+; AVX2-NEXT: cmpl $-1, %eax
; AVX2-NEXT: sete %al
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
More information about the llvm-commits
mailing list