summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorSimon Pilgrim <llvm-dev@redking.me.uk>2019-09-08 12:44:22 +0000
committerSimon Pilgrim <llvm-dev@redking.me.uk>2019-09-08 12:44:22 +0000
commit178cd2cd3a69046c88ade1b31796f3f90b7e86a0 (patch)
treef25ccf75fb95b3389347e0d05136d911190e5c46
parent9d570020708d633e9979030a5787034887426a7c (diff)
downloadbcm5719-llvm-178cd2cd3a69046c88ade1b31796f3f90b7e86a0.tar.gz
bcm5719-llvm-178cd2cd3a69046c88ade1b31796f3f90b7e86a0.zip
[X86][SSE] Fix out of range shift introduced in D67070/rL371328
Use APInt to create the comparison mask instead. llvm-svn: 371330
-rw-r--r--llvm/lib/Target/X86/X86ISelLowering.cpp3
-rw-r--r--llvm/test/CodeGen/X86/vector-reduce-and-bool.ll8
2 files changed, 6 insertions, 5 deletions
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index f5c9971a3c5..5e5ec5c6c8b 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -35944,7 +35944,8 @@ static SDValue combineHorizontalPredicateResult(SDNode *Extract,
CondCode = ISD::CondCode::SETNE;
} else {
// all_of -> MOVMSK == ((1 << NumElts) - 1)
- CmpC = DAG.getConstant((1ULL << NumElts) - 1, DL, CmpVT);
+ CmpC = DAG.getConstant(APInt::getLowBitsSet(CmpVT.getSizeInBits(), NumElts),
+ DL, CmpVT);
CondCode = ISD::CondCode::SETEQ;
}
diff --git a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
index b9cdb2f0225..49add302bf4 100644
--- a/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
+++ b/llvm/test/CodeGen/X86/vector-reduce-and-bool.ll
@@ -866,7 +866,7 @@ define i1 @trunc_v64i8_v64i1(<64 x i8>) {
; AVX512BW-NEXT: vpsllw $7, %zmm0, %zmm0
; AVX512BW-NEXT: vpmovb2m %zmm0, %k0
; AVX512BW-NEXT: kortestq %k0, %k0
-; AVX512BW-NEXT: sete %al
+; AVX512BW-NEXT: setb %al
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
@@ -875,7 +875,7 @@ define i1 @trunc_v64i8_v64i1(<64 x i8>) {
; AVX512VL-NEXT: vpsllw $7, %zmm0, %zmm0
; AVX512VL-NEXT: vpmovb2m %zmm0, %k0
; AVX512VL-NEXT: kortestq %k0, %k0
-; AVX512VL-NEXT: sete %al
+; AVX512VL-NEXT: setb %al
; AVX512VL-NEXT: vzeroupper
; AVX512VL-NEXT: retq
%a = trunc <64 x i8> %0 to <64 x i1>
@@ -1744,7 +1744,7 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
; AVX512BW: # %bb.0:
; AVX512BW-NEXT: vptestnmb %zmm0, %zmm0, %k0
; AVX512BW-NEXT: kortestq %k0, %k0
-; AVX512BW-NEXT: sete %al
+; AVX512BW-NEXT: setb %al
; AVX512BW-NEXT: vzeroupper
; AVX512BW-NEXT: retq
;
@@ -1752,7 +1752,7 @@ define i1 @icmp_v64i8_v64i1(<64 x i8>) {
; AVX512VL: # %bb.0:
; AVX512VL-NEXT: vptestnmb %zmm0, %zmm0, %k0
; AVX512VL-NEXT: kortestq %k0, %k0
-; AVX512VL-NEXT: sete %al
+; AVX512VL-NEXT: setb %al
; AVX512VL-NEXT: vzeroupper
; AVX512VL-NEXT: retq
%a = icmp eq <64 x i8> %0, zeroinitializer
OpenPOWER on IntegriCloud