From d57faf858c7cd7dbf406051fc822a2a66c4cd120 Mon Sep 17 00:00:00 2001
From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Wed, 4 Oct 2017 18:00:42 +0000
Subject: [PATCH] [X86][AVX] Improve (i8 bitcast (v8i1 x)) handling for
 v8i64/v8f64 512-bit vector compare results.

AVX1/AVX2 targets were missing a chance to use vmovmskps for v8f32/v8i32 results for bool vector bitcasts


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@314921 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Target/X86/X86ISelLowering.cpp    | 11 +++++------
 test/CodeGen/X86/bitcast-setcc-512.ll | 20 ++++++--------------
 2 files changed, 11 insertions(+), 20 deletions(-)

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index 5df5a21f752..9aae58b5555 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -29505,9 +29505,8 @@ static SDValue combineBitcastvxi1(SelectionDAG &DAG, SDValue BitCast,
     FPCastVT = MVT::v4f32;
     // For cases such as (i4 bitcast (v4i1 setcc v4i64 v1, v2))
     // sign-extend to a 256-bit operation to avoid truncation.
-    if (N0->getOpcode() == ISD::SETCC &&
-        N0->getOperand(0)->getValueType(0).is256BitVector() &&
-        Subtarget.hasAVX()) {
+    if (N0->getOpcode() == ISD::SETCC && Subtarget.hasAVX() &&
+        N0->getOperand(0)->getValueType(0).is256BitVector()) {
       SExtVT = MVT::v4i64;
       FPCastVT = MVT::v4f64;
     }
@@ -29519,9 +29518,9 @@ static SDValue combineBitcastvxi1(SelectionDAG &DAG, SDValue BitCast,
     // If the setcc operand is 128-bit, prefer sign-extending to 128-bit over
     // 256-bit because the shuffle is cheaper than sign extending the result of
     // the compare.
-    if (N0->getOpcode() == ISD::SETCC &&
-        N0->getOperand(0)->getValueType(0).is256BitVector() &&
-        Subtarget.hasAVX()) {
+    if (N0->getOpcode() == ISD::SETCC && Subtarget.hasAVX() &&
+        (N0->getOperand(0)->getValueType(0).is256BitVector() ||
+         N0->getOperand(0)->getValueType(0).is512BitVector())) {
       SExtVT = MVT::v8i32;
       FPCastVT = MVT::v8f32;
     }
diff --git a/test/CodeGen/X86/bitcast-setcc-512.ll b/test/CodeGen/X86/bitcast-setcc-512.ll
index 4ab96c3945d..32044a83910 100644
--- a/test/CodeGen/X86/bitcast-setcc-512.ll
+++ b/test/CodeGen/X86/bitcast-setcc-512.ll
@@ -1074,9 +1074,8 @@ define i8 @v8i64(<8 x i64> %a, <8 x i64> %b) {
 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm3
 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpacksswb %xmm3, %xmm0, %xmm0
-; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
-; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
@@ -1087,10 +1086,7 @@ define i8 @v8i64(<8 x i64> %a, <8 x i64> %b) {
 ; AVX2-NEXT:    vpcmpgtq %ymm2, %ymm0, %ymm0
 ; AVX2-NEXT:    vpacksswb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
-; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
 ; AVX2-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
@@ -1138,9 +1134,8 @@ define i8 @v8f64(<8 x double> %a, <8 x double> %b) {
 ; AVX1-NEXT:    vcmpltpd %ymm0, %ymm2, %ymm0
 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
 ; AVX1-NEXT:    vpacksswb %xmm2, %xmm0, %xmm0
-; AVX1-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX1-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
-; AVX1-NEXT:    vpmovmskb %xmm0, %eax
+; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
+; AVX1-NEXT:    vmovmskps %ymm0, %eax
 ; AVX1-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
@@ -1151,10 +1146,7 @@ define i8 @v8f64(<8 x double> %a, <8 x double> %b) {
 ; AVX2-NEXT:    vcmpltpd %ymm0, %ymm2, %ymm0
 ; AVX2-NEXT:    vpacksswb %ymm1, %ymm0, %ymm0
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
-; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
-; AVX2-NEXT:    vpacksswb %xmm1, %xmm0, %xmm0
-; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,2,4,6,8,10,12,14,u,u,u,u,u,u,u,u]
-; AVX2-NEXT:    vpmovmskb %xmm0, %eax
+; AVX2-NEXT:    vmovmskps %ymm0, %eax
 ; AVX2-NEXT:    # kill: %AL<def> %AL<kill> %EAX<kill>
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
-- 
2.40.0