This is an archive of the discontinued LLVM Phabricator instance.

Differential D101061

[SLP]Skip undefs trying to find perfect/shuffled tree entries matching.
ClosedPublic

Authored by ABataev on Apr 22 2021, 6:46 AM.

Details

Reviewers
RKSimon
spatel
sdesmalen
Commits
rG18c61fc498c7: [SLP]Skip undefs trying to find perfect/shuffled tree entries matching.
Summary

We can skip check for undefs trying to find perfect/shuffled tree
entries matching, they can be ignored completely improving the final
cost/vectorization results.

Diff Detail

Unit TestsFailed

TimeTest
0 msx64 debian > libomptarget.mapping::declare_mapper_nested_default_mappers_array.cpp

Event Timeline

ABataev created this revision.Apr 22 2021, 6:46 AM
Herald added a subscriber: hiraditya. · View Herald TranscriptApr 22 2021, 6:46 AM
ABataev requested review of this revision.Apr 22 2021, 6:46 AM
Herald added a project: Restricted Project. · View Herald TranscriptApr 22 2021, 6:46 AM
Harbormaster completed remote builds in B100256: Diff 339608.Apr 22 2021, 8:31 AM
RKSimon accepted this revision.Apr 22 2021, 8:38 AM

LGTM

This revision is now accepted and ready to land.Apr 22 2021, 8:38 AM
This revision was landed with ongoing or failed builds.Apr 22 2021, 8:59 AM
Closed by commit rG18c61fc498c7: [SLP]Skip undefs trying to find perfect/shuffled tree entries matching. (authored by ABataev). · Explain Why
This revision was automatically updated to reflect the committed changes.
ABataev added a commit: rG18c61fc498c7: [SLP]Skip undefs trying to find perfect/shuffled tree entries matching..

Revision Contents

PathSize
llvm/
lib/
Transforms/
Vectorize/
2 lines
test/
Transforms/
SLPVectorizer/
X86/
2 lines

Diff 339608

llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp

  • This file is larger than 256 KB, so syntax highlighting is disabled by default.
Show First 20 Lines • Show All 4,260 Lines • ▼ Show 20 LinesBoUpSLP::isGatherShuffledEntry(const TreeEntry *TE, SmallVectorImpl<int> &Mask,
assert(VLIt != VectorizableTree.end() && assert(VLIt != VectorizableTree.end() &&
"Gathered values should be in the tree."); "Gathered values should be in the tree.");
Mask.assign(TE->Scalars.size(), UndefMaskElem); Mask.assign(TE->Scalars.size(), UndefMaskElem);
Entries.clear(); Entries.clear();
DenseMap<const TreeEntry *, int> Used; DenseMap<const TreeEntry *, int> Used;
int NumShuffles = 0; int NumShuffles = 0;
for (int I = 0, E = TE->Scalars.size(); I < E; ++I) { for (int I = 0, E = TE->Scalars.size(); I < E; ++I) {
Value *V = TE->Scalars[I]; Value *V = TE->Scalars[I];
if (isa<UndefValue>(V))
continue;
const TreeEntry *VTE = getTreeEntry(V); const TreeEntry *VTE = getTreeEntry(V);
if (!VTE) { if (!VTE) {
// Check if it is used in one of the gathered entries. // Check if it is used in one of the gathered entries.
const auto *It = const auto *It =
find_if(make_range(VectorizableTree.begin(), VLIt), find_if(make_range(VectorizableTree.begin(), VLIt),
[V](const std::unique_ptr<TreeEntry> &EntryPtr) { [V](const std::unique_ptr<TreeEntry> &EntryPtr) {
return EntryPtr->State == TreeEntry::NeedToGather && return EntryPtr->State == TreeEntry::NeedToGather &&
is_contained(EntryPtr->Scalars, V); is_contained(EntryPtr->Scalars, V);
▲ Show 20 Lines • Show All 3,771 Lines • Show Last 20 Lines

llvm/test/Transforms/SLPVectorizer/X86/remark_extract_broadcast.ll

Show All 39 Lines
; CHECK-NEXT: store <8 x i16> [[TMP10]], <8 x i16>* [[TMP11]], align 2 ; CHECK-NEXT: store <8 x i16> [[TMP10]], <8 x i16>* [[TMP11]], align 2
; CHECK-NEXT: ret void ; CHECK-NEXT: ret void
; ;
; YAML: Pass: slp-vectorizer ; YAML: Pass: slp-vectorizer
; YAML-NEXT: Name: StoresVectorized ; YAML-NEXT: Name: StoresVectorized
; YAML-NEXT: Function: fextr ; YAML-NEXT: Function: fextr
; YAML-NEXT: Args: ; YAML-NEXT: Args:
; YAML-NEXT: - String: 'Stores SLP vectorized with cost ' ; YAML-NEXT: - String: 'Stores SLP vectorized with cost '
; YAML-NEXT: - Cost: '1' ; YAML-NEXT: - Cost: '-1'
; YAML-NEXT: - String: ' and with tree size ' ; YAML-NEXT: - String: ' and with tree size '
; YAML-NEXT: - TreeSize: '4' ; YAML-NEXT: - TreeSize: '4'
entry: entry:
%LD = load <8 x i16>, <8 x i16>* undef %LD = load <8 x i16>, <8 x i16>* undef
%V0 = extractelement <8 x i16> %LD, i32 0 %V0 = extractelement <8 x i16> %LD, i32 0
br label %t br label %t
Show All 34 Lines