This is an archive of the discontinued LLVM Phabricator instance.

[OpenMP] Increase attributor iterations on the GPU
ClosedPublic

Authored by jhuber6 on Jun 25 2021, 7:44 AM.

Details

Summary

Increase the number of attributor iterations on a GPU target. I forgot to
change this in D104416.

Diff Detail

Event Timeline

jhuber6 created this revision.Jun 25 2021, 7:44 AM
jhuber6 requested review of this revision.Jun 25 2021, 7:44 AM
Herald added a reviewer: baziotis. · View Herald Transcript
Herald added a project: Restricted Project. · View Herald Transcript
ormris removed a subscriber: ormris.Jun 25 2021, 1:19 PM
jdoerfert accepted this revision.Jun 25 2021, 10:26 PM

LG. Please clang format. Since this was already discussed and approved, feel free to submit such things.

This revision is now accepted and ready to land.Jun 25 2021, 10:26 PM
This revision was landed with ongoing or failed builds.Jun 28 2021, 5:51 AM
This revision was automatically updated to reflect the committed changes.
DavidSpickett added inline comments.
llvm/lib/Transforms/IPO/OpenMPOpt.cpp
2671

Building with GCC I get a warning here:

llvm-project/llvm/lib/Transforms/IPO/OpenMPOpt.cpp:2671:37: warning: address of function 'isOpenMPDevice' will always evaluate to 'true' [-Wpointer-bool-conversion]
  unsigned MaxFixponitIterations = (isOpenMPDevice) ? 128 : 32;
                                    ^~~~~~~~~~~~~~  ~
/home/david.spickett/llvm-project/llvm/lib/Transforms/IPO/OpenMPOpt.cpp:2671:37: note: prefix with the address-of operator to silence this warning
  unsigned MaxFixponitIterations = (isOpenMPDevice) ? 128 : 32;
                                    ^
                                    &

Also going by the patch before this it should be MaxFixpointIterations. ponit -> point

jhuber6 added inline comments.Jun 28 2021, 6:48 AM
llvm/lib/Transforms/IPO/OpenMPOpt.cpp
2671

Should be fixed upstream, thanks for pointing out the typo.