[X86][SSE] SimplifyDemandedBitsForTargetNode - PCMPGT(0,X) sign mask

For PCMPGT(0, X) patterns where we only demand the sign bit (e.g. BLENDV or MOVMSK) then we can use X directly.

Differential Revision: https://reviews.llvm.org/D57667

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353051 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 95827fbed5dec27c2fe378be3be77da962d5772e..18cb4d1a2cbecef0402e9a58b4f43491010f2336 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -33070,6 +33070,13 @@ bool X86TargetLowering::SimplifyDemandedBitsForTargetNode(
     }
     break;
   }
+  case X86ISD::PCMPGT:
+    // icmp sgt(0, R) == ashr(R, BitWidth-1).
+    // iff we only need the sign bit then we can use R directly.
+    if (OriginalDemandedBits.isSignMask() &&
+        ISD::isBuildVectorAllZeros(Op.getOperand(0).getNode()))
+      return TLO.CombineTo(Op, Op.getOperand(1));
+    break;
   case X86ISD::MOVMSK: {
     SDValue Src = Op.getOperand(0);
     MVT SrcVT = Src.getSimpleValueType();

diff --git a/test/CodeGen/X86/avx512-cvt-widen.ll b/test/CodeGen/X86/avx512-cvt-widen.ll
index 849a814e640dac1e014a959b49ce6eec56fc2ec1..01eba7912f940ba5322dfc5cd972edbb9ae49677 100644 (file)
--- a/test/CodeGen/X86/avx512-cvt-widen.ll
+++ b/test/CodeGen/X86/avx512-cvt-widen.ll
@@ -1909,8 +1909,6 @@ define <8 x float> @ubto8f32(<8 x i32> %a) {
 define <8 x double> @ubto8f64(<8 x i32> %a) {
 ; ALL-LABEL: ubto8f64:
 ; ALL:       # %bb.0:
-; ALL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; ALL-NEXT:    vpcmpgtd %ymm0, %ymm1, %ymm0
 ; ALL-NEXT:    vpsrld $31, %ymm0, %ymm0
 ; ALL-NEXT:    vcvtdq2pd %ymm0, %zmm0
 ; ALL-NEXT:    retq
@@ -1942,8 +1940,6 @@ define <4 x float> @ubto4f32(<4 x i32> %a) {
 define <4 x double> @ubto4f64(<4 x i32> %a) {
 ; ALL-LABEL: ubto4f64:
 ; ALL:       # %bb.0:
-; ALL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; ALL-NEXT:    vpcmpgtd %xmm0, %xmm1, %xmm0
 ; ALL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; ALL-NEXT:    vcvtdq2pd %xmm0, %ymm0
 ; ALL-NEXT:    retq

diff --git a/test/CodeGen/X86/avx512-cvt.ll b/test/CodeGen/X86/avx512-cvt.ll
index 264b1ea841fd3105f2b94ed84b306e5edbff2c52..170e5a30798c063bc574643ea7c6d8ff664ab4ab 100644 (file)
--- a/test/CodeGen/X86/avx512-cvt.ll
+++ b/test/CodeGen/X86/avx512-cvt.ll
@@ -1897,8 +1897,6 @@ define <8 x float> @ubto8f32(<8 x i32> %a) {
 define <8 x double> @ubto8f64(<8 x i32> %a) {
 ; ALL-LABEL: ubto8f64:
 ; ALL:       # %bb.0:
-; ALL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; ALL-NEXT:    vpcmpgtd %ymm0, %ymm1, %ymm0
 ; ALL-NEXT:    vpsrld $31, %ymm0, %ymm0
 ; ALL-NEXT:    vcvtdq2pd %ymm0, %zmm0
 ; ALL-NEXT:    retq
@@ -1930,8 +1928,6 @@ define <4 x float> @ubto4f32(<4 x i32> %a) {
 define <4 x double> @ubto4f64(<4 x i32> %a) {
 ; ALL-LABEL: ubto4f64:
 ; ALL:       # %bb.0:
-; ALL-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; ALL-NEXT:    vpcmpgtd %xmm0, %xmm1, %xmm0
 ; ALL-NEXT:    vpsrld $31, %xmm0, %xmm0
 ; ALL-NEXT:    vcvtdq2pd %xmm0, %ymm0
 ; ALL-NEXT:    retq

diff --git a/test/CodeGen/X86/bitcast-setcc-128.ll b/test/CodeGen/X86/bitcast-setcc-128.ll
index fb585974e5b740cdf972ec5a9054802a4687abf4..e9dea85e14dfca54c8a98906ea08bfe3c54716f1 100644 (file)
--- a/test/CodeGen/X86/bitcast-setcc-128.ll
+++ b/test/CodeGen/X86/bitcast-setcc-128.ll
@@ -683,9 +683,6 @@ define i64 @v16i8_widened_with_ones(<16 x i8> %a, <16 x i8> %b) {
 ; AVX1:       # %bb.0: # %entry
 ; AVX1-NEXT:    vpcmpeqb %xmm1, %xmm0, %xmm0
 ; AVX1-NEXT:    vpsllw $7, %xmm0, %xmm0
-; AVX1-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
 ; AVX1-NEXT:    orl $-65536, %ecx # imm = 0xFFFF0000
 ; AVX1-NEXT:    movabsq $-4294967296, %rax # imm = 0xFFFFFFFF00000000

diff --git a/test/CodeGen/X86/bitcast-setcc-256.ll b/test/CodeGen/X86/bitcast-setcc-256.ll
index b0af971366c99f54d96b1daa56be45fce48a618e..7489c785f145836f03c7fe5fc3ff5da8cfe5492d 100644 (file)
--- a/test/CodeGen/X86/bitcast-setcc-256.ll
+++ b/test/CodeGen/X86/bitcast-setcc-256.ll
@@ -314,11 +314,8 @@ define void @bitcast_32i8_store(i32* %p, <32 x i8> %a0) {
 ;
 ; AVX1-LABEL: bitcast_32i8_store:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
-; AVX1-NEXT:    vpmovmskb %xmm2, %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
 ; AVX1-NEXT:    shll $16, %ecx
 ; AVX1-NEXT:    orl %eax, %ecx

diff --git a/test/CodeGen/X86/bitcast-setcc-512.ll b/test/CodeGen/X86/bitcast-setcc-512.ll
index 340c7ab3afd95aaa9ba4df02a70a89f355bbc7c1..b6a38aa132b86623b016e9cbb5548f960d446593 100644 (file)
--- a/test/CodeGen/X86/bitcast-setcc-512.ll
+++ b/test/CodeGen/X86/bitcast-setcc-512.ll
@@ -418,18 +418,13 @@ define void @bitcast_64i8_store(i64* %p, <64 x i8> %a0) {
 ;
 ; AVX1-LABEL: bitcast_64i8_store:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm3
-; AVX1-NEXT:    vpmovmskb %xmm3, %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
 ; AVX1-NEXT:    shll $16, %ecx
 ; AVX1-NEXT:    orl %eax, %ecx
-; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm2, %xmm0
-; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    vpmovmskb %xmm1, %eax
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %edx
 ; AVX1-NEXT:    shll $16, %edx
 ; AVX1-NEXT:    orl %eax, %edx

diff --git a/test/CodeGen/X86/movmsk-cmp.ll b/test/CodeGen/X86/movmsk-cmp.ll
index bc16d8c710eea668bcf108717e85cc36c85c6613..3ff101761dbd1926e0c24c66ee672a92ca5634d5 100644 (file)
--- a/test/CodeGen/X86/movmsk-cmp.ll
+++ b/test/CodeGen/X86/movmsk-cmp.ll
@@ -86,11 +86,8 @@ define i1 @allones_v32i8_sign(<32 x i8> %arg) {
 ;
 ; AVX1-LABEL: allones_v32i8_sign:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
-; AVX1-NEXT:    vpmovmskb %xmm2, %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
 ; AVX1-NEXT:    shll $16, %ecx
 ; AVX1-NEXT:    orl %eax, %ecx
@@ -140,11 +137,8 @@ define i1 @allzeros_v32i8_sign(<32 x i8> %arg) {
 ;
 ; AVX1-LABEL: allzeros_v32i8_sign:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
-; AVX1-NEXT:    vpmovmskb %xmm2, %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
 ; AVX1-NEXT:    shll $16, %ecx
 ; AVX1-NEXT:    orl %eax, %ecx
@@ -200,18 +194,13 @@ define i1 @allones_v64i8_sign(<64 x i8> %arg) {
 ;
 ; AVX1-LABEL: allones_v64i8_sign:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm3
-; AVX1-NEXT:    vpmovmskb %xmm3, %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
 ; AVX1-NEXT:    shll $16, %ecx
 ; AVX1-NEXT:    orl %eax, %ecx
-; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm2, %xmm0
-; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    vpmovmskb %xmm1, %eax
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %edx
 ; AVX1-NEXT:    shll $16, %edx
 ; AVX1-NEXT:    orl %eax, %edx
@@ -275,18 +264,13 @@ define i1 @allzeros_v64i8_sign(<64 x i8> %arg) {
 ;
 ; AVX1-LABEL: allzeros_v64i8_sign:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm2, %xmm2, %xmm2
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm3
-; AVX1-NEXT:    vpmovmskb %xmm3, %eax
+; AVX1-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
 ; AVX1-NEXT:    shll $16, %ecx
 ; AVX1-NEXT:    orl %eax, %ecx
-; AVX1-NEXT:    vpcmpgtb %xmm1, %xmm2, %xmm0
-; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    vpmovmskb %xmm1, %eax
 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm2, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %edx
 ; AVX1-NEXT:    shll $16, %edx
 ; AVX1-NEXT:    orl %eax, %edx
@@ -4756,11 +4740,8 @@ define i32 @movmskb256(<32 x i8> %x) {
 ;
 ; AVX1-LABEL: movmskb256:
 ; AVX1:       # %bb.0:
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm2
-; AVX1-NEXT:    vpmovmskb %xmm2, %ecx
+; AVX1-NEXT:    vpmovmskb %xmm0, %ecx
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
-; AVX1-NEXT:    vpcmpgtb %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpmovmskb %xmm0, %eax
 ; AVX1-NEXT:    shll $16, %eax
 ; AVX1-NEXT:    orl %ecx, %eax