[llvm-commits] [llvm] r170328 - /llvm/trunk/lib/Transforms/Scalar/SROA.cpp
Chandler Carruth
chandlerc at gmail.com
Mon Dec 17 05:41:22 PST 2012
Author: chandlerc
Date: Mon Dec 17 07:41:21 2012
New Revision: 170328
URL: http://llvm.org/viewvc/llvm-project?rev=170328&view=rev
Log:
Hoist the insertVector helper to be a static helper.
This will allow its use inside of memcpy rewriting as well. This routine
is more complex than extractVector, and some of its uses are not 100%
where I want them to be so there is still some work to do here.
While this can technically change the output in some cases, it shouldn't
be a change that matters -- IE, it can leave some dead code lying around
that prior versions did not, etc.
Yet another step in the refactorings leading up to the solution to the
last component of PR14478.
Modified:
llvm/trunk/lib/Transforms/Scalar/SROA.cpp
Modified: llvm/trunk/lib/Transforms/Scalar/SROA.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Transforms/Scalar/SROA.cpp?rev=170328&r1=170327&r2=170328&view=diff
==============================================================================
--- llvm/trunk/lib/Transforms/Scalar/SROA.cpp (original)
+++ llvm/trunk/lib/Transforms/Scalar/SROA.cpp Mon Dec 17 07:41:21 2012
@@ -2251,6 +2251,56 @@
return V;
}
+static Value *insertVector(IRBuilder<> &IRB, Value *Old, Value *V,
+ unsigned BeginIndex, const Twine &Name) {
+ VectorType *VecTy = cast<VectorType>(Old->getType());
+ assert(VecTy && "Can only insert a vector into a vector");
+
+ VectorType *Ty = dyn_cast<VectorType>(V->getType());
+ if (!Ty) {
+ // Single element to insert.
+ V = IRB.CreateInsertElement(Old, V, IRB.getInt32(BeginIndex),
+ Name + ".insert");
+ DEBUG(dbgs() << " insert: " << *V << "\n");
+ return V;
+ }
+
+ assert(Ty->getNumElements() <= VecTy->getNumElements() &&
+ "Too many elements!");
+ if (Ty->getNumElements() == VecTy->getNumElements()) {
+ assert(V->getType() == VecTy && "Vector type mismatch");
+ return V;
+ }
+ unsigned EndIndex = BeginIndex + Ty->getNumElements();
+
+ // When inserting a smaller vector into the larger to store, we first
+ // use a shuffle vector to widen it with undef elements, and then
+ // a second shuffle vector to select between the loaded vector and the
+ // incoming vector.
+ SmallVector<Constant*, 8> Mask;
+ Mask.reserve(VecTy->getNumElements());
+ for (unsigned i = 0; i != VecTy->getNumElements(); ++i)
+ if (i >= BeginIndex && i < EndIndex)
+ Mask.push_back(IRB.getInt32(i - BeginIndex));
+ else
+ Mask.push_back(UndefValue::get(IRB.getInt32Ty()));
+ V = IRB.CreateShuffleVector(V, UndefValue::get(V->getType()),
+ ConstantVector::get(Mask),
+ Name + ".expand");
+ DEBUG(dbgs() << " shuffle1: " << *V << "\n");
+
+ Mask.clear();
+ for (unsigned i = 0; i != VecTy->getNumElements(); ++i)
+ if (i >= BeginIndex && i < EndIndex)
+ Mask.push_back(IRB.getInt32(i));
+ else
+ Mask.push_back(IRB.getInt32(i + VecTy->getNumElements()));
+ V = IRB.CreateShuffleVector(V, Old, ConstantVector::get(Mask),
+ Name + "insert");
+ DEBUG(dbgs() << " shuffle2: " << *V << "\n");
+ return V;
+}
+
namespace {
/// \brief Visitor to rewrite instructions using a partition of an alloca to
/// use a new alloca.
@@ -2519,52 +2569,6 @@
return !LI.isVolatile() && !IsPtrAdjusted;
}
- Value *insertVector(IRBuilder<> &IRB, Value *V,
- unsigned BeginIndex, unsigned EndIndex) {
- assert(VecTy && "Can only insert a vector into a vector alloca");
- unsigned NumElements = EndIndex - BeginIndex;
- assert(NumElements <= VecTy->getNumElements() && "Too many elements!");
-
- if (NumElements == VecTy->getNumElements())
- return convertValue(TD, IRB, V, VecTy);
-
- LoadInst *LI = IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
- getName(".load"));
- if (NumElements == 1) {
- V = IRB.CreateInsertElement(LI, V, IRB.getInt32(BeginIndex),
- getName(".insert"));
- DEBUG(dbgs() << " insert: " << *V << "\n");
- return V;
- }
-
- // When inserting a smaller vector into the larger to store, we first
- // use a shuffle vector to widen it with undef elements, and then
- // a second shuffle vector to select between the loaded vector and the
- // incoming vector.
- SmallVector<Constant*, 8> Mask;
- Mask.reserve(VecTy->getNumElements());
- for (unsigned i = 0; i != VecTy->getNumElements(); ++i)
- if (i >= BeginIndex && i < EndIndex)
- Mask.push_back(IRB.getInt32(i - BeginIndex));
- else
- Mask.push_back(UndefValue::get(IRB.getInt32Ty()));
- V = IRB.CreateShuffleVector(V, UndefValue::get(V->getType()),
- ConstantVector::get(Mask),
- getName(".expand"));
- DEBUG(dbgs() << " shuffle1: " << *V << "\n");
-
- Mask.clear();
- for (unsigned i = 0; i != VecTy->getNumElements(); ++i)
- if (i >= BeginIndex && i < EndIndex)
- Mask.push_back(IRB.getInt32(i));
- else
- Mask.push_back(IRB.getInt32(i + VecTy->getNumElements()));
- V = IRB.CreateShuffleVector(V, LI, ConstantVector::get(Mask),
- getName("insert"));
- DEBUG(dbgs() << " shuffle2: " << *V << "\n");
- return V;
- }
-
bool rewriteVectorizedStoreInst(IRBuilder<> &IRB, Value *V,
StoreInst &SI, Value *OldOp) {
unsigned BeginIndex = getIndex(BeginOffset);
@@ -2579,7 +2583,9 @@
V = convertValue(TD, IRB, V, PartitionTy);
// Mix in the existing elements.
- V = insertVector(IRB, V, BeginIndex, EndIndex);
+ Value *Old = IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
+ getName(".load"));
+ V = insertVector(IRB, Old, V, BeginIndex, getName(".vec"));
StoreInst *Store = IRB.CreateAlignedStore(V, &NewAI, NewAI.getAlignment());
Pass.DeadInsts.insert(&SI);
@@ -2771,10 +2777,17 @@
Value *Splat = getIntegerSplat(IRB, II.getValue(),
TD.getTypeSizeInBits(ElementTy)/8);
- if (NumElements > 1)
+ if (NumElements > 1) {
Splat = getVectorSplat(IRB, Splat, NumElements);
- V = insertVector(IRB, Splat, BeginIndex, EndIndex);
+ Type *SplatVecTy = VectorType::get(ElementTy, NumElements);
+ if (Splat->getType() != SplatVecTy)
+ Splat = convertValue(TD, IRB, Splat, SplatVecTy);
+ }
+
+ Value *Old = IRB.CreateAlignedLoad(&NewAI, NewAI.getAlignment(),
+ getName(".oldload"));
+ V = insertVector(IRB, Old, Splat, BeginIndex, getName(".vec"));
} else if (IntTy) {
// If this is a memset on an alloca where we can widen stores, insert the
// set integer.
More information about the llvm-commits
mailing list