23.x [RISCV] Scheduling/Fusion Patches - #561
Conversation
Sam Elliott (lenary)
changed the title
d4a8184 to
a26f37a
Compare
|
This is not mergeable until the LTO+Linker Scripts is mergeable, but for the moment I'm using this to get a draft toolchain for testing by Harsh. |
This is not strictly true. I'm going to rebase the PR to get rid of the LTO+Linker Scripts once the fusions are all working correctly. |
This adds to patches corresponding to upstream patches: - llvm/llvm-project#203139 (and its follow-up fix, llvm/llvm-project#209934) - llvm/llvm-project#209542 (and its follow-up fix, llvm/llvm-project#213815) Both were rejected by the upstream release managers. Signed-off-by: Sam Elliott <aelliott@qti.qualcomm.com>
661c1fc to
026ecee
Compare
|
These have been tested on internal workloads and are working. |
Jonathon Penix (jonathonpenix)
left a comment
Jonathon Penix (jonathonpenix)
left a comment
There was a problem hiding this comment.
Can't comment on the patches themselves, but LGTM as far as adding the patchfiles, etc.
|
Doing some internal build runs of this. |
|
Internal runs came back clean. |
|
We're going to do a few more specialised internal runs to check we don't get more crashes. |
Pankaj Gode (pgodeq)
left a comment
Pankaj Gode (pgodeq)
left a comment
There was a problem hiding this comment.
Sam had shared the build artifact to run test runs. They look clean.
I had also used updated wrappers/clang-config files. Included the tune flags for latency and fusion, in the clang-config files, as well.
62a670c
into
qualcomm:release/qualcomm-software/23.x
3 participants
This adds to patches corresponding to upstream patches:
Both were rejected by the upstream release managers.