RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2025-01-31 20:51:52 +01:00
Code Issues Projects Releases Wiki Activity

[InstCombine] allow icmp-xor folds for vectors (PR33138)

Browse Source 
This fixes the first part of:
https://bugs.llvm.org/show_bug.cgi?id=33138

More work is needed for the bitcasted variant.

llvm-svn: 303660
This commit is contained in:
Sanjay Patel 2017-05-23 17:29:58 +00:00
parent 4530cf2da1
commit 5e73758477
2 changed files with 13 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
lib/Transforms/InstCombine/InstCombineCompares.cpp
View File

@ -3030,18 +3030,21 @@ Instruction *InstCombiner::foldICmpBinOp(ICmpInst &I) {
break;
case Instruction::Add:
case Instruction::Sub:
case Instruction::Xor:
case Instruction::Xor: {
if (I.isEquality()) // a+x icmp eq/ne b+x --> a icmp b
return new ICmpInst(Pred, BO0->getOperand(0), BO1->getOperand(0));
// icmp u/s (a ^ signmask), (b ^ signmask) --> icmp s/u a, b
if (ConstantInt *CI = dyn_cast<ConstantInt>(BO0->getOperand(1))) {
if (CI->getValue().isSignMask()) {
const APInt *C;
if (match(BO0->getOperand(1), m_APInt(C))) {
// icmp u/s (a ^ signmask), (b ^ signmask) --> icmp s/u a, b
if (C->isSignMask()) {
ICmpInst::Predicate NewPred =
I.isSigned() ? I.getUnsignedPredicate() : I.getSignedPredicate();
return new ICmpInst(NewPred, BO0->getOperand(0), BO1->getOperand(0));
}
if (BO0->getOpcode() == Instruction::Xor && CI->isMaxValue(true)) {
// icmp u/s (a ^ maxsignval), (b ^ maxsignval) --> icmp s/u' a, b
if (BO0->getOpcode() == Instruction::Xor && C->isMaxSignedValue()) {
ICmpInst::Predicate NewPred =
I.isSigned() ? I.getUnsignedPredicate() : I.getSignedPredicate();
NewPred = I.getSwappedPredicate(NewPred);
@ -3049,6 +3052,7 @@ Instruction *InstCombiner::foldICmpBinOp(ICmpInst &I) {
}
}
break;
}
case Instruction::Mul:
if (!I.isEquality())
break;

16
test/Transforms/InstCombine/icmp-xor-signbit.ll
View File

@ -18,9 +18,7 @@ define i1 @slt_to_ult(i8 %x, i8 %y) {
define <2 x i1> @slt_to_ult_splat(<2 x i8> %x, <2 x i8> %y) {
; CHECK-LABEL: @slt_to_ult_splat(
; CHECK-NEXT: [[A:%.*]] = xor <2 x i8> %x, <i8 -128, i8 -128>
; CHECK-NEXT: [[B:%.*]] = xor <2 x i8> %y, <i8 -128, i8 -128>
; CHECK-NEXT: [[CMP:%.*]] = icmp slt <2 x i8> [[A]], [[B]]
; CHECK-NEXT: [[CMP:%.*]] = icmp ult <2 x i8> %x, %y
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%a = xor <2 x i8> %x, <i8 128, i8 128>
@ -44,9 +42,7 @@ define i1 @ult_to_slt(i8 %x, i8 %y) {
define <2 x i1> @ult_to_slt_splat(<2 x i8> %x, <2 x i8> %y) {
; CHECK-LABEL: @ult_to_slt_splat(
; CHECK-NEXT: [[A:%.*]] = xor <2 x i8> %x, <i8 -128, i8 -128>
; CHECK-NEXT: [[B:%.*]] = xor <2 x i8> %y, <i8 -128, i8 -128>
; CHECK-NEXT: [[CMP:%.*]] = icmp ult <2 x i8> [[A]], [[B]]
; CHECK-NEXT: [[CMP:%.*]] = icmp slt <2 x i8> %x, %y
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%a = xor <2 x i8> %x, <i8 128, i8 128>
@ -70,9 +66,7 @@ define i1 @slt_to_ugt(i8 %x, i8 %y) {
define <2 x i1> @slt_to_ugt_splat(<2 x i8> %x, <2 x i8> %y) {
; CHECK-LABEL: @slt_to_ugt_splat(
; CHECK-NEXT: [[A:%.*]] = xor <2 x i8> %x, <i8 127, i8 127>
; CHECK-NEXT: [[B:%.*]] = xor <2 x i8> %y, <i8 127, i8 127>
; CHECK-NEXT: [[CMP:%.*]] = icmp slt <2 x i8> [[A]], [[B]]
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt <2 x i8> %x, %y
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%a = xor <2 x i8> %x, <i8 127, i8 127>
@ -96,9 +90,7 @@ define i1 @ult_to_sgt(i8 %x, i8 %y) {
define <2 x i1> @ult_to_sgt_splat(<2 x i8> %x, <2 x i8> %y) {
; CHECK-LABEL: @ult_to_sgt_splat(
; CHECK-NEXT: [[A:%.*]] = xor <2 x i8> %x, <i8 127, i8 127>
; CHECK-NEXT: [[B:%.*]] = xor <2 x i8> %y, <i8 127, i8 127>
; CHECK-NEXT: [[CMP:%.*]] = icmp ult <2 x i8> [[A]], [[B]]
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <2 x i8> %x, %y
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%a = xor <2 x i8> %x, <i8 127, i8 127>