RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 03:33:20 +01:00
Code Issues Projects Releases Wiki Activity

[DAGCombiner] add helper function for store merging of extracts; NFC

Browse Source
This commit is contained in:
Sanjay Patel 2020-07-09 10:33:31 -04:00
parent 44f6080ee5
commit 876f75b66f
1 changed files with 79 additions and 64 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

143
lib/CodeGen/SelectionDAG/DAGCombiner.cpp
View File

@ -719,6 +719,14 @@ namespace {
unsigned NumConsecutiveStores,
EVT MemVT, bool AllowVectors, SDNode *Root);
/// This is a helper function for mergeConsecutiveStores. It is used for
/// store chains that are composed entirely of extracted vector elements.
/// When extracting multiple vector elements, try to store them in one
/// vector store rather than a sequence of scalar stores.
bool tryStoreMergeOfExtracts(SmallVectorImpl<MemOpLink> &StoreNodes,
unsigned NumConsecutiveStores, EVT MemVT,
SDNode *Root);
/// Merge consecutive store operations into a wide store.
/// This optimization uses wide integers or vectors when possible.
/// \return true if stores were merged.
@ -16297,6 +16305,7 @@ bool DAGCombiner::tryStoreMergeOfConstants(
int64_t ElementSizeBytes = MemVT.getStoreSize();
unsigned NumMemElts = MemVT.isVector() ? MemVT.getVectorNumElements() : 1;
bool MadeChange = false;
// Store the constants into memory as one consecutive store.
while (NumConsecutiveStores >= 2) {
LSBaseSDNode *FirstInChain = StoreNodes[0].MemNode;
@ -16412,6 +16421,74 @@ bool DAGCombiner::tryStoreMergeOfConstants(
return MadeChange;
}
bool DAGCombiner::tryStoreMergeOfExtracts(
SmallVectorImpl<MemOpLink> &StoreNodes, unsigned NumConsecutiveStores,
EVT MemVT, SDNode *RootNode) {
LLVMContext &Context = *DAG.getContext();
const DataLayout &DL = DAG.getDataLayout();
unsigned NumMemElts = MemVT.isVector() ? MemVT.getVectorNumElements() : 1;
bool MadeChange = false;
// Loop on Consecutive Stores on success.
while (NumConsecutiveStores >= 2) {
LSBaseSDNode *FirstInChain = StoreNodes[0].MemNode;
unsigned FirstStoreAS = FirstInChain->getAddressSpace();
unsigned FirstStoreAlign = FirstInChain->getAlignment();
unsigned NumStoresToMerge = 1;
for (unsigned i = 0; i < NumConsecutiveStores; ++i) {
// Find a legal type for the vector store.
unsigned Elts = (i + 1) * NumMemElts;
EVT Ty = EVT::getVectorVT(*DAG.getContext(), MemVT.getScalarType(), Elts);
bool IsFast = false;
// Break early when size is too large to be legal.
if (Ty.getSizeInBits() > MaximumLegalStoreInBits)
break;
if (TLI.isTypeLegal(Ty) && TLI.canMergeStoresTo(FirstStoreAS, Ty, DAG) &&
TLI.allowsMemoryAccess(Context, DL, Ty,
*FirstInChain->getMemOperand(), &IsFast) &&
IsFast)
NumStoresToMerge = i + 1;
}
// Check if we found a legal integer type creating a meaningful
// merge.
if (NumStoresToMerge < 2) {
// We know that candidate stores are in order and of correct
// shape. While there is no mergeable sequence from the
// beginning one may start later in the sequence. The only
// reason a merge of size N could have failed where another of
// the same size would not have, is if the alignment has
// improved. Drop as many candidates as we can here.
unsigned NumSkip = 1;
while ((NumSkip < NumConsecutiveStores) &&
(StoreNodes[NumSkip].MemNode->getAlignment() <= FirstStoreAlign))
NumSkip++;
StoreNodes.erase(StoreNodes.begin(), StoreNodes.begin() + NumSkip);
NumConsecutiveStores -= NumSkip;
continue;
}
// Check that we can merge these candidates without causing a cycle.
if (!checkMergeStoreCandidatesForDependencies(StoreNodes, NumStoresToMerge,
RootNode)) {
StoreNodes.erase(StoreNodes.begin(),
StoreNodes.begin() + NumStoresToMerge);
NumConsecutiveStores -= NumStoresToMerge;
continue;
}
MadeChange |= mergeStoresOfConstantsOrVecElts(
StoreNodes, MemVT, NumStoresToMerge, false, true, false);
StoreNodes.erase(StoreNodes.begin(), StoreNodes.begin() + NumStoresToMerge);
NumConsecutiveStores -= NumStoresToMerge;
}
return MadeChange;
}
bool DAGCombiner::mergeConsecutiveStores(StoreSDNode *St) {
if (OptLevel == CodeGenOpt::None || !EnableStoreMerging)
return false;
@ -16485,71 +16562,9 @@ bool DAGCombiner::mergeConsecutiveStores(StoreSDNode *St) {
MemVT, AllowVectors, RootNode);
continue;
}
// When extracting multiple vector elements, try to store them
// in one vector store rather than a sequence of scalar stores.
if (StoreSrc == StoreSource::Extract) {
// Loop on Consecutive Stores on success.
while (NumConsecutiveStores >= 2) {
LSBaseSDNode *FirstInChain = StoreNodes[0].MemNode;
unsigned FirstStoreAS = FirstInChain->getAddressSpace();
unsigned FirstStoreAlign = FirstInChain->getAlignment();
unsigned NumStoresToMerge = 1;
for (unsigned i = 0; i < NumConsecutiveStores; ++i) {
// Find a legal type for the vector store.
unsigned Elts = (i + 1) * NumMemElts;
EVT Ty =
EVT::getVectorVT(*DAG.getContext(), MemVT.getScalarType(), Elts);
bool IsFast = false;
// Break early when size is too large to be legal.
if (Ty.getSizeInBits() > MaximumLegalStoreInBits)
break;
if (TLI.isTypeLegal(Ty) &&
TLI.canMergeStoresTo(FirstStoreAS, Ty, DAG) &&
TLI.allowsMemoryAccess(Context, DL, Ty,
*FirstInChain->getMemOperand(), &IsFast) &&
IsFast)
NumStoresToMerge = i + 1;
}
// Check if we found a legal integer type creating a meaningful
// merge.
if (NumStoresToMerge < 2) {
// We know that candidate stores are in order and of correct
// shape. While there is no mergeable sequence from the
// beginning one may start later in the sequence. The only
// reason a merge of size N could have failed where another of
// the same size would not have, is if the alignment has
// improved. Drop as many candidates as we can here.
unsigned NumSkip = 1;
while (
(NumSkip < NumConsecutiveStores) &&
(StoreNodes[NumSkip].MemNode->getAlignment() <= FirstStoreAlign))
NumSkip++;
StoreNodes.erase(StoreNodes.begin(), StoreNodes.begin() + NumSkip);
NumConsecutiveStores -= NumSkip;
continue;
}
// Check that we can merge these candidates without causing a cycle.
if (!checkMergeStoreCandidatesForDependencies(
StoreNodes, NumStoresToMerge, RootNode)) {
StoreNodes.erase(StoreNodes.begin(),
StoreNodes.begin() + NumStoresToMerge);
NumConsecutiveStores -= NumStoresToMerge;
continue;
}
MadeChange |= mergeStoresOfConstantsOrVecElts(
StoreNodes, MemVT, NumStoresToMerge, false, true, false);
StoreNodes.erase(StoreNodes.begin(),
StoreNodes.begin() + NumStoresToMerge);
NumConsecutiveStores -= NumStoresToMerge;
}
MadeChange |= tryStoreMergeOfExtracts(StoreNodes, NumConsecutiveStores,
MemVT, RootNode);
continue;
}