This is an archive of the discontinued LLVM Phabricator instance.

Differential D112125

A hot fix for a reduction parallel pattern of OpenMP backend
ClosedPublic

Authored by MikeDvorskiy on Oct 20 2021, 12:58 AM.

Details

Reviewers
ldionne
jdoerfert
nadiasvertex
Group Reviewers
Restricted Project
Commits
rG8f5cb64c7147: [pstl] A hot fix for a reduction parallel pattern of OpenMP backend
Summary

A hot fix for OpenMP backend; + "exit" condition for recursion of "__parallel_reduce_body"

Diff Detail

Event Timeline

MikeDvorskiy created this revision.Oct 20 2021, 12:58 AM
Herald added subscribers: guansong, yaxunl. · View Herald TranscriptOct 20 2021, 12:58 AM
MikeDvorskiy requested review of this revision.Oct 20 2021, 12:58 AM
Herald added a reviewer: jdoerfert. · View Herald TranscriptOct 20 2021, 12:58 AM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added a reviewer: Restricted Project. · View Herald Transcript
Herald added subscribers: libcxx-commits, sstefan1. · View Herald Transcript
Harbormaster completed remote builds in B129674: Diff 380867.Oct 20 2021, 1:31 AM
MikeDvorskiy updated this revision to Diff 380881.Oct 20 2021, 2:27 AM

clang format

Harbormaster completed remote builds in B129681: Diff 380881.Oct 20 2021, 2:55 AM
MikeDvorskiy updated this revision to Diff 380901.EditedOct 20 2021, 4:53 AM

A hot performance fix for parallel reduce pattern of OpenMP backend. The recursion should be limited by a grain size - __default_chunk_size.

Harbormaster completed remote builds in B129695: Diff 380901.Oct 20 2021, 5:16 AM
nadiasvertex accepted this revision.Oct 20 2021, 11:34 AM
nadiasvertex added a subscriber: nadiasvertex.

Thank you! I fixed that bug once, but in the various changes I missed it the last time.

ldionne accepted this revision.Oct 27 2021, 7:16 AM
This revision is now accepted and ready to land.Oct 27 2021, 7:16 AM
Closed by commit rG8f5cb64c7147: [pstl] A hot fix for a reduction parallel pattern of OpenMP backend (authored by MikeDvorskiy). · Explain WhyOct 27 2021, 8:53 AM
This revision was automatically updated to reflect the committed changes.
MikeDvorskiy added a commit: rG8f5cb64c7147: [pstl] A hot fix for a reduction parallel pattern of OpenMP backend.

Revision Contents

PathSize
pstl/
include/
pstl/
internal/
omp/
3 lines

Diff 380867

pstl/include/pstl/internal/omp/parallel_reduce.h

Context not available.
__parallel_reduce_body(_RandomAccessIterator __first, _RandomAccessIterator __last, _Value __identity, __parallel_reduce_body(_RandomAccessIterator __first, _RandomAccessIterator __last, _Value __identity,
_RealBody __real_body, _Reduction __reduce) _RealBody __real_body, _Reduction __reduce)
{ {
if(__last - __first < 2)
Lint: Pre-merge checks
Inline Actions

clang-format: please reformat the code

-    if(__last - __first < 2)
+    if (__last - __first < 2)
Lint: Pre-merge checks: clang-format: please reformat the code ``` - if(__last - __first < 2) + if (__last…
return __real_body(__first, __last, __identity);
auto __middle = __first + ((__last - __first) / 2); auto __middle = __first + ((__last - __first) / 2);
_Value __v1(__identity), __v2(__identity); _Value __v1(__identity), __v2(__identity);
__parallel_invoke_body( __parallel_invoke_body(
Context not available.