maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[InstCombine] allow icmp (bit-manipulation-intrinsic(), C) folds for vectors 

llvm-svn: 276523
This commit is contained in:
Sanjay Patel 2016-07-23 13:06:49 +00:00
parent 488cb137a9
commit 1271bf9178
3 changed files with 20 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

29
llvm/lib/Transforms/InstCombine/InstCombineCompares.cpp
View File

@ -2355,37 +2355,35 @@ Instruction *InstCombiner::foldICmpEqualityWithConstant(ICmpInst &ICI,
return nullptr;
}
Instruction *InstCombiner::foldICmpIntrinsicWithConstant(ICmpInst &ICI,
Instruction *LHSI,
ConstantInt *RHS) {
IntrinsicInst *II = dyn_cast<IntrinsicInst>(LHSI);
if (!II || !ICI.isEquality())
Instruction *InstCombiner::foldICmpIntrinsicWithConstant(ICmpInst &ICI) {
IntrinsicInst *II = dyn_cast<IntrinsicInst>(ICI.getOperand(0));
const APInt *Op1C;
if (!II || !ICI.isEquality() || !match(ICI.getOperand(1), m_APInt(Op1C)))
return nullptr;
// Handle icmp {eq|ne} <intrinsic>, intcst.
const APInt &RHSV = RHS->getValue();
switch (II->getIntrinsicID()) {
case Intrinsic::bswap:
Worklist.Add(II);
ICI.setOperand(0, II->getArgOperand(0));
ICI.setOperand(1, Builder->getInt(RHSV.byteSwap()));
ICI.setOperand(1, Builder->getInt(Op1C->byteSwap()));
return &ICI;
case Intrinsic::ctlz:
case Intrinsic::cttz:
// ctz(A) == bitwidth(a) -> A == 0 and likewise for !=
if (RHSV == RHS->getType()->getBitWidth()) {
if (*Op1C == Op1C->getBitWidth()) {
Worklist.Add(II);
ICI.setOperand(0, II->getArgOperand(0));
ICI.setOperand(1, ConstantInt::get(RHS->getType(), 0));
ICI.setOperand(1, ConstantInt::getNullValue(II->getType()));
return &ICI;
}
break;
case Intrinsic::ctpop:
// popcount(A) == 0 -> A == 0 and likewise for !=
if (RHS->isZero()) {
if (*Op1C == 0) {
Worklist.Add(II);
ICI.setOperand(0, II->getArgOperand(0));
ICI.setOperand(1, RHS);
ICI.setOperand(1, ConstantInt::getNullValue(II->getType()));
return &ICI;
}
break;
@ -3641,6 +3639,10 @@ Instruction *InstCombiner::visitICmpInst(ICmpInst &I) {
// See if we are doing a comparison between a constant and an instruction that
// can be folded into the comparison.
// FIXME: Use m_APInt instead of dyn_cast<ConstantInt> to allow these
// transforms for vectors.
if (ConstantInt *CI = dyn_cast<ConstantInt>(Op1)) {
// Since the RHS is a ConstantInt (CI), if the left hand side is an
// instruction, see if that instruction also has constants so that the
@ -3650,11 +3652,12 @@ Instruction *InstCombiner::visitICmpInst(ICmpInst &I) {
return Res;
if (Instruction *Res = foldICmpEqualityWithConstant(I, LHSI, CI))
return Res;
if (Instruction *Res = foldICmpIntrinsicWithConstant(I, LHSI, CI))
return Res;
}
}
if (Instruction *Res = foldICmpIntrinsicWithConstant(I))
return Res;
// Handle icmp with constant (but not simple integer constant) RHS
if (Constant *RHSC = dyn_cast<Constant>(Op1)) {
if (Instruction *LHSI = dyn_cast<Instruction>(Op0))

3
llvm/lib/Transforms/InstCombine/InstCombineInternal.h
View File

@ -585,8 +585,7 @@ private:
ConstantInt *RHS);
Instruction *foldICmpEqualityWithConstant(ICmpInst &ICI, Instruction *LHS,
ConstantInt *RHS);
Instruction *foldICmpIntrinsicWithConstant(ICmpInst &ICI, Instruction *LHS,
ConstantInt *RHS);
Instruction *foldICmpIntrinsicWithConstant(ICmpInst &ICI);
Instruction *OptAndOp(Instruction *Op, ConstantInt *OpRHS,
ConstantInt *AndRHS, BinaryOperator &TheAnd);

11
llvm/test/Transforms/InstCombine/intrinsics.ll
View File

@ -316,12 +316,9 @@ entry:
; CHECK-NEXT: store volatile i1 %pop.cmp, i1* %c
}
; FIXME: Vectors should get the same folds as scalars.
define <2 x i1> @ctlz_cmp_vec(<2 x i32> %a) {
; CHECK-LABEL: @ctlz_cmp_vec(
; CHECK-NEXT: [[X:%.*]] = tail call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %a, i1 false) #0
; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i32> [[X]], <i32 32, i32 32>
; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i32> %a, zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%x = tail call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %a, i1 false) nounwind readnone
@ -331,8 +328,7 @@ define <2 x i1> @ctlz_cmp_vec(<2 x i32> %a) {
define <2 x i1> @cttz_cmp_vec(<2 x i32> %a) {
; CHECK-LABEL: @cttz_cmp_vec(
; CHECK-NEXT: [[X:%.*]] = tail call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %a, i1 false) #0
; CHECK-NEXT: [[CMP:%.*]] = icmp ne <2 x i32> [[X]], <i32 32, i32 32>
; CHECK-NEXT: [[CMP:%.*]] = icmp ne <2 x i32> %a, zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%x = tail call <2 x i32> @llvm.ctlz.v2i32(<2 x i32> %a, i1 false) nounwind readnone
@ -342,8 +338,7 @@ define <2 x i1> @cttz_cmp_vec(<2 x i32> %a) {
define <2 x i1> @ctpop_cmp_vec(<2 x i32> %a) {
; CHECK-LABEL: @ctpop_cmp_vec(
; CHECK-NEXT: [[X:%.*]] = tail call <2 x i32> @llvm.ctpop.v2i32(<2 x i32> %a) #0
; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i32> [[X]], zeroinitializer
; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i32> %a, zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%x = tail call <2 x i32> @llvm.ctpop.v2i32(<2 x i32> %a) nounwind readnone