RPCSX/llvm
1
0
Fork 0
mirror of https://github.com/RPCSX/llvm.git synced 2024-11-24 12:19:53 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

[InstCombine] use m_APInt to allow icmp (and X, Y), C folds for splat constant vectors

Browse Source 
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@279937 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Sanjay Patel 2016-08-28 18:18:00 +00:00
parent 337ddd9188
commit 2c15995350
5 changed files with 41 additions and 49 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

64
lib/Transforms/InstCombine/InstCombineCompares.cpp
View File

@ -1610,48 +1610,50 @@ Instruction *InstCombiner::foldICmpAndConstant(ICmpInst &Cmp,
if (Instruction *I = foldICmpAndConstConst(Cmp, And, C))
return I;
// FIXME: This check restricts all folds under here to scalar types.
ConstantInt *RHS = dyn_cast<ConstantInt>(Cmp.getOperand(1));
if (!RHS)
return nullptr;
// TODO: These all require that Y is constant too, so refactor with the above.
// Try to optimize things like "A[i]&42 == 0" to index computations.
if (LoadInst *LI = dyn_cast<LoadInst>(And->getOperand(0))) {
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(LI->getOperand(0)))
if (GlobalVariable *GV = dyn_cast<GlobalVariable>(GEP->getOperand(0)))
// Try to optimize things like "A[i] & 42 == 0" to index computations.
Value *X = And->getOperand(0);
Value *Y = And->getOperand(1);
if (auto *LI = dyn_cast<LoadInst>(X))
if (auto *GEP = dyn_cast<GetElementPtrInst>(LI->getOperand(0)))
if (auto *GV = dyn_cast<GlobalVariable>(GEP->getOperand(0)))
if (GV->isConstant() && GV->hasDefinitiveInitializer() &&
!LI->isVolatile() && isa<ConstantInt>(And->getOperand(1))) {
ConstantInt *C = cast<ConstantInt>(And->getOperand(1));
if (Instruction *Res = foldCmpLoadFromIndexedGlobal(GEP, GV, Cmp, C))
!LI->isVolatile() && isa<ConstantInt>(Y)) {
ConstantInt *C2 = cast<ConstantInt>(Y);
if (Instruction *Res = foldCmpLoadFromIndexedGlobal(GEP, GV, Cmp, C2))
return Res;
}
}
if (!Cmp.isEquality())
return nullptr;
// X & -C == -C -> X > u ~C
// X & -C != -C -> X <= u ~C
// iff C is a power of 2
if (Cmp.isEquality() && RHS == And->getOperand(1) && (-(*C)).isPowerOf2())
return new ICmpInst(Cmp.getPredicate() == ICmpInst::ICMP_EQ
? ICmpInst::ICMP_UGT
: ICmpInst::ICMP_ULE,
And->getOperand(0), SubOne(RHS));
if (Cmp.getOperand(1) == Y && (-(*C)).isPowerOf2()) {
auto NewPred = Cmp.getPredicate() == CmpInst::ICMP_EQ ? CmpInst::ICMP_UGT
: CmpInst::ICMP_ULE;
return new ICmpInst(NewPred, X, SubOne(cast<Constant>(Cmp.getOperand(1))));
}
// (icmp eq (and %A, C), 0) -> (icmp sgt (trunc %A), -1)
// iff C is a power of 2
if (Cmp.isEquality() && And->hasOneUse() && match(RHS, m_Zero())) {
if (auto *CI = dyn_cast<ConstantInt>(And->getOperand(1))) {
const APInt &AI = CI->getValue();
int32_t ExactLogBase2 = AI.exactLogBase2();
if (ExactLogBase2 != -1 && DL.isLegalInteger(ExactLogBase2 + 1)) {
Type *NTy = IntegerType::get(Cmp.getContext(), ExactLogBase2 + 1);
Value *Trunc = Builder->CreateTrunc(And->getOperand(0), NTy);
return new ICmpInst(Cmp.getPredicate() == ICmpInst::ICMP_EQ
? ICmpInst::ICMP_SGE
: ICmpInst::ICMP_SLT,
Trunc, Constant::getNullValue(NTy));
}
// (X & C2) == 0 -> (trunc X) >= 0
// (X & C2) != 0 -> (trunc X) < 0
// iff C2 is a power of 2 and it masks the sign bit of a legal integer type.
const APInt *C2;
if (And->hasOneUse() && *C == 0 && match(Y, m_APInt(C2))) {
int32_t ExactLogBase2 = C2->exactLogBase2();
if (ExactLogBase2 != -1 && DL.isLegalInteger(ExactLogBase2 + 1)) {
Type *NTy = IntegerType::get(Cmp.getContext(), ExactLogBase2 + 1);
if (And->getType()->isVectorTy())
NTy = VectorType::get(NTy, And->getType()->getVectorNumElements());
Value *Trunc = Builder->CreateTrunc(X, NTy);
auto NewPred = Cmp.getPredicate() == CmpInst::ICMP_EQ ? CmpInst::ICMP_SGE
: CmpInst::ICMP_SLT;
return new ICmpInst(NewPred, Trunc, Constant::getNullValue(NTy));
}
}
return nullptr;
}

10
test/Transforms/InstCombine/and-compare.ll
View File

@ -42,11 +42,10 @@ define i1 @test2(i64 %A) {
ret i1 %cmp
}
; FIXME: Vectors should fold the same way.
define <2 x i1> @test2vec(<2 x i64> %A) {
; CHECK-LABEL: @test2vec(
; CHECK-NEXT: [[AND:%.*]] = and <2 x i64> %A, <i64 128, i64 128>
; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i64> [[AND]], zeroinitializer
; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i64> %A to <2 x i8>
; CHECK-NEXT: [[CMP:%.*]] = icmp sgt <2 x i8> [[TMP1]], <i8 -1, i8 -1>
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%and = and <2 x i64> %A, <i64 128, i64 128>
@ -65,11 +64,10 @@ define i1 @test3(i64 %A) {
ret i1 %cmp
}
; FIXME: Vectors should fold the same way.
define <2 x i1> @test3vec(<2 x i64> %A) {
; CHECK-LABEL: @test3vec(
; CHECK-NEXT: [[AND:%.*]] = and <2 x i64> %A, <i64 128, i64 128>
; CHECK-NEXT: [[CMP:%.*]] = icmp ne <2 x i64> [[AND]], zeroinitializer
; CHECK-NEXT: [[TMP1:%.*]] = trunc <2 x i64> %A to <2 x i8>
; CHECK-NEXT: [[CMP:%.*]] = icmp slt <2 x i8> [[TMP1]], zeroinitializer
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%and = and <2 x i64> %A, <i64 128, i64 128>

4
test/Transforms/InstCombine/apint-shift.ll
View File

@ -280,11 +280,9 @@ define i1 @test19a(i39 %A) {
ret i1 %C
}
; FIXME: Vectors should fold too.
define <2 x i1> @test19a_vec(<2 x i39> %A) {
; CHECK-LABEL: @test19a_vec(
; CHECK-NEXT: [[B_MASK:%.*]] = and <2 x i39> %A, <i39 -4, i39 -4>
; CHECK-NEXT: [[C:%.*]] = icmp eq <2 x i39> [[B_MASK]], <i39 -4, i39 -4>
; CHECK-NEXT: [[C:%.*]] = icmp ugt <2 x i39> %A, <i39 -5, i39 -5>
; CHECK-NEXT: ret <2 x i1> [[C]]
;
%B = ashr <2 x i39> %A, <i39 2, i39 2>

8
test/Transforms/InstCombine/icmp.ll
View File

@ -1815,11 +1815,9 @@ define i1 @icmp_and_X_-16_eq-16(i32 %X) {
ret i1 %cmp
}
; FIXME: Vectors should fold the same way.
define <2 x i1> @icmp_and_X_-16_eq-16_vec(<2 x i32> %X) {
; CHECK-LABEL: @icmp_and_X_-16_eq-16_vec(
; CHECK-NEXT: [[AND:%.*]] = and <2 x i32> %X, <i32 -16, i32 -16>
; CHECK-NEXT: [[CMP:%.*]] = icmp eq <2 x i32> [[AND]], <i32 -16, i32 -16>
; CHECK-NEXT: [[CMP:%.*]] = icmp ugt <2 x i32> %X, <i32 -17, i32 -17>
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%and = and <2 x i32> %X, <i32 -16, i32 -16>
@ -1837,11 +1835,9 @@ define i1 @icmp_and_X_-16_ne-16(i32 %X) {
ret i1 %cmp
}
; FIXME: Vectors should fold the same way.
define <2 x i1> @icmp_and_X_-16_ne-16_vec(<2 x i32> %X) {
; CHECK-LABEL: @icmp_and_X_-16_ne-16_vec(
; CHECK-NEXT: [[AND:%.*]] = and <2 x i32> %X, <i32 -16, i32 -16>
; CHECK-NEXT: [[CMP:%.*]] = icmp ne <2 x i32> [[AND]], <i32 -16, i32 -16>
; CHECK-NEXT: [[CMP:%.*]] = icmp ult <2 x i32> %X, <i32 -16, i32 -16>
; CHECK-NEXT: ret <2 x i1> [[CMP]]
;
%and = and <2 x i32> %X, <i32 -16, i32 -16>

4
test/Transforms/InstCombine/shift.ll
View File

@ -373,11 +373,9 @@ define i1 @test19a(i32 %A) {
ret i1 %C
}
; FIXME: Vectors should fold the same way.
define <2 x i1> @test19a_vec(<2 x i32> %A) {
; CHECK-LABEL: @test19a_vec(
; CHECK-NEXT: [[B_MASK:%.*]] = and <2 x i32> %A, <i32 -4, i32 -4>
; CHECK-NEXT: [[C:%.*]] = icmp eq <2 x i32> [[B_MASK]], <i32 -4, i32 -4>
; CHECK-NEXT: [[C:%.*]] = icmp ugt <2 x i32> %A, <i32 -5, i32 -5>
; CHECK-NEXT: ret <2 x i1> [[C]]
;
%B = ashr <2 x i32> %A, <i32 2, i32 2>