This is an archive of the discontinued LLVM Phabricator instance.

[NFC][OpenMP] Prepare amdgpu plugin for asynchronous implementation of target region launch
AbandonedPublic

Authored by carlo.bertolli on Dec 7 2021, 8:50 AM.

Details

Summary

At present, amdgpu plugin merges both asynchronous and synchronous kernel launch implementations into a single synchronous version.
This patch prepares the plugin for asynchronous implementation by:

  • Separating the control flow path of asynchronous and synchronous kernel launch functions
  • Privatize actual kernel launch code (valid in both cases) into an anonymous namespace base function

Diff Detail

Event Timeline

carlo.bertolli created this revision.Dec 7 2021, 8:50 AM
carlo.bertolli requested review of this revision.Dec 7 2021, 8:50 AM

This rendered poorly in phab diff. Could we move the functions around separately to the code change?

carlo.bertolli abandoned this revision.Dec 7 2021, 10:41 AM

Abandoning this one for readability issues and moving to two separate patches.
First one:
https://reviews.llvm.org/D115267