summaryrefslogtreecommitdiffstats
path: root/llvm/lib
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/lib')
-rw-r--r--llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp18
-rw-r--r--llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp21
2 files changed, 33 insertions, 6 deletions
diff --git a/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp b/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
index 9dc1cd1feeb..86b6215f09e 100644
--- a/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMTargetTransformInfo.cpp
@@ -523,8 +523,20 @@ unsigned ARMTTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty, OperandValueK
if (Idx != -1)
return LT.first * CostTbl[Idx].Cost;
-
- return TargetTransformInfo::getArithmeticInstrCost(Opcode, Ty, Op1Info,
- Op2Info);
+ unsigned Cost =
+ TargetTransformInfo::getArithmeticInstrCost(Opcode, Ty, Op1Info, Op2Info);
+
+ // This is somewhat of a hack. The problem that we are facing is that SROA
+ // creates a sequence of shift, and, or instructions to construct values.
+ // These sequences are recognized by the ISel and have zero-cost. Not so for
+ // the vectorized code. Because we have support for v2i64 but not i64 those
+ // sequences look particularily beneficial to vectorize.
+ // To work around this we increase the cost of v2i64 operations to make them
+ // seem less beneficial.
+ if (LT.second == MVT::v2i64 &&
+ Op2Info == TargetTransformInfo::OK_UniformConstantValue)
+ Cost += 4;
+
+ return Cost;
}
diff --git a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
index 012521ad014..2797a214212 100644
--- a/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
+++ b/llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
@@ -1013,9 +1013,24 @@ int BoUpSLP::getEntryCost(TreeEntry *E) {
TTI->getCmpSelInstrCost(Opcode, ScalarTy, Builder.getInt1Ty());
VecCost = TTI->getCmpSelInstrCost(Opcode, VecTy, MaskTy);
} else {
- ScalarCost = VecTy->getNumElements() *
- TTI->getArithmeticInstrCost(Opcode, ScalarTy);
- VecCost = TTI->getArithmeticInstrCost(Opcode, VecTy);
+ // Certain instructions can be cheaper to vectorize if they have a
+ // constant second vector operand.
+ TargetTransformInfo::OperandValueKind Op1VK =
+ TargetTransformInfo::OK_AnyValue;
+ TargetTransformInfo::OperandValueKind Op2VK =
+ TargetTransformInfo::OK_UniformConstantValue;
+
+ // Check whether all second operands are constant.
+ for (unsigned i = 0; i < VL.size(); ++i)
+ if (!isa<ConstantInt>(cast<Instruction>(VL[i])->getOperand(1))) {
+ Op2VK = TargetTransformInfo::OK_AnyValue;
+ break;
+ }
+
+ ScalarCost =
+ VecTy->getNumElements() *
+ TTI->getArithmeticInstrCost(Opcode, ScalarTy, Op1VK, Op2VK);
+ VecCost = TTI->getArithmeticInstrCost(Opcode, VecTy, Op1VK, Op2VK);
}
return VecCost - ScalarCost;
}
OpenPOWER on IntegriCloud