RISC-V EVL tail folding failure on SPEC CPU 2017 525.x264_r #122681
Assignees
Labels
Comments
|
@llvm/issue-subscribers-backend-risc-v Author: Luke Lau (lukel97)
Split out from the discussion here: https://github.com//pull/122458#issuecomment-2585713670
On RISC-V with It's been failing since at least 6ad0dcf, detected from this LNT run: https://lnt.lukelau.me/db_default/v4/nts/89 |
lukel97
added a commit
that referenced
this issue
Jan 14, 2025
…22885) This fixes a miscompilation extracted from 525.x264_r, where we were failing to update the runtime VF of a VPReverseVectorPointerRecipe. We were removing a use of VF whilst iterating over the users() iterator, which messed up the iterator in-flight and caused us to miss some recipes. This fixes it by copying the users into a SmallVector first. Fixes #122681 Fixes #122682
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Split out from the discussion here: #122458 (comment)
On RISC-V with
-march=rva22u64_v -O3 -flto -mllvm -force-tail-folding-style=data-with-evl -mllvm -prefer-predicate-over-epilogue=predicate-else-scalar-epilogue, the SPEC CPU 2017 525.x264_r benchmark fails in the train dataset, likely due to a miscompile.It's been failing since at least 6ad0dcf, detected from this LNT run: https://lnt.lukelau.me/db_default/v4/nts/89
The text was updated successfully, but these errors were encountered: