summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
diff options
context:
space:
mode:
authorCraig Topper <craig.topper@intel.com>2017-06-22 16:23:30 +0000
committerCraig Topper <craig.topper@intel.com>2017-06-22 16:23:30 +0000
commitdffbbcb3fd42124842a8ae2cae18d40874d81f0b (patch)
treec93a55597568ce81a9a497d1af6afd5f44068471 /llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
parenta690e3cea23b17715fdbcecb24ae634ebc103fa5 (diff)
downloadbcm5719-llvm-dffbbcb3fd42124842a8ae2cae18d40874d81f0b.tar.gz
bcm5719-llvm-dffbbcb3fd42124842a8ae2cae18d40874d81f0b.zip
[InstCombine] Teach foldSelectICmpAndOr to recognize (select (icmp slt (trunc (X)), 0), Y, (or Y, C2))
Summary: InstCombine likes to turn (icmp eq (and X, C1), 0) into (icmp slt (trunc (X)), 0) sometimes. This breaks foldSelectICmpAndOr's ability to recognize (select (icmp eq (and X, C1), 0), Y, (or Y, C2))->(or (shl (and X, C1), C3), y). This patch tries to recover this. I had to flip around some of the early out checks so that I could create a new And instruction during the compare processing without it possibly never getting used. Reviewers: spatel, majnemer, davide Reviewed By: spatel Subscribers: llvm-commits Differential Revision: https://reviews.llvm.org/D34184 llvm-svn: 306029
Diffstat (limited to 'llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp')
-rw-r--r--llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp49
1 files changed, 38 insertions, 11 deletions
diff --git a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
index 993c095dc6f..33951e66497 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -317,19 +317,44 @@ static Value *foldSelectICmpAndOr(const SelectInst &SI, Value *TrueVal,
Value *FalseVal,
InstCombiner::BuilderTy *Builder) {
const ICmpInst *IC = dyn_cast<ICmpInst>(SI.getCondition());
- if (!IC || !IC->isEquality() || !SI.getType()->isIntegerTy())
+ if (!IC || !SI.getType()->isIntegerTy())
return nullptr;
Value *CmpLHS = IC->getOperand(0);
Value *CmpRHS = IC->getOperand(1);
- if (!match(CmpRHS, m_Zero()))
- return nullptr;
+ Value *V;
+ unsigned C1Log;
+ bool IsEqualZero;
+ bool NeedAnd = false;
+ if (IC->isEquality()) {
+ if (!match(CmpRHS, m_Zero()))
+ return nullptr;
+
+ const APInt *C1;
+ if (!match(CmpLHS, m_And(m_Value(), m_Power2(C1))))
+ return nullptr;
+
+ V = CmpLHS;
+ C1Log = C1->logBase2();
+ IsEqualZero = IC->getPredicate() == ICmpInst::ICMP_EQ;
+ } else if (IC->getPredicate() == ICmpInst::ICMP_SLT ||
+ IC->getPredicate() == ICmpInst::ICMP_SGT) {
+ // We also need to recognize (icmp slt (trunc (X)), 0) and
+ // (icmp sgt (trunc (X)), -1).
+ IsEqualZero = IC->getPredicate() == ICmpInst::ICMP_SGT;
+ if ((IsEqualZero && !match(CmpRHS, m_AllOnes())) ||
+ (!IsEqualZero && !match(CmpRHS, m_Zero())))
+ return nullptr;
- Value *X;
- const APInt *C1;
- if (!match(CmpLHS, m_And(m_Value(X), m_Power2(C1))))
+ if (!match(CmpLHS, m_OneUse(m_Trunc(m_Value(V)))))
+ return nullptr;
+
+ C1Log = CmpLHS->getType()->getScalarSizeInBits() - 1;
+ NeedAnd = true;
+ } else {
return nullptr;
+ }
const APInt *C2;
bool OrOnTrueVal = false;
@@ -340,15 +365,11 @@ static Value *foldSelectICmpAndOr(const SelectInst &SI, Value *TrueVal,
if (!OrOnFalseVal && !OrOnTrueVal)
return nullptr;
- Value *V = CmpLHS;
Value *Y = OrOnFalseVal ? TrueVal : FalseVal;
- unsigned C1Log = C1->logBase2();
unsigned C2Log = C2->logBase2();
- ICmpInst::Predicate Pred = IC->getPredicate();
- bool NeedXor = (Pred == ICmpInst::ICMP_NE && OrOnFalseVal) ||
- (Pred == ICmpInst::ICMP_EQ && OrOnTrueVal);
+ bool NeedXor = (!IsEqualZero && OrOnFalseVal) || (IsEqualZero && OrOnTrueVal);
bool NeedShift = C1Log != C2Log;
bool NeedZExtTrunc = Y->getType()->getIntegerBitWidth() !=
V->getType()->getIntegerBitWidth();
@@ -359,6 +380,12 @@ static Value *foldSelectICmpAndOr(const SelectInst &SI, Value *TrueVal,
(IC->hasOneUse() + Or->hasOneUse()))
return nullptr;
+ if (NeedAnd) {
+ // Insert the AND instruction on the input to the truncate.
+ APInt C1 = APInt::getOneBitSet(V->getType()->getScalarSizeInBits(), C1Log);
+ V = Builder->CreateAnd(V, ConstantInt::get(V->getType(), C1));
+ }
+
if (C2Log > C1Log) {
V = Builder->CreateZExtOrTrunc(V, Y->getType());
V = Builder->CreateShl(V, C2Log - C1Log);
OpenPOWER on IntegriCloud