Add patch to fix volatile access bug #595
Conversation
There was a problem hiding this comment.
I know you're in a hurry, but could you please at least put something in the commit message that allows people to find out what the patch is about without having to go and read the message in the subsidiary patch? "#111679" by itself means someone has to manually make up a URL to go and look that up in the LLVM project (and has to work out from context that it is the LLVM project, because the PR number is too big to be picolibc).
A full URL to the PR would be useful. And maybe a one-sentence summary of what the fix is about.
| From: =?UTF-8?q?Dominik=20W=C3=B3jt?= <[email protected]> | ||
| Date: Mon, 16 Oct 2023 11:35:48 +0200 | ||
| Subject: [libc++] tests with picolibc: xfail two remaining tests | ||
| Subject: [PATCH 1/9] [libc++] tests with picolibc: xfail two remaining tests |
There was a problem hiding this comment.
You should remaster the patch series using format-patch -k, to avoid getting these [PATCH 1/9] prefixes. Goes with the git am -k we advise for reapplying them.
(If you didn't do either, then git am without -k would also discard square-bracketed prefixes you actually wanted to keep, which there are lots of in LLVM.)
We have a customer affected by this bug, which was fixed after LLVM 19, and we'd like to do a release of LLVM-ET 19 with the fix. LLVM PR: llvm/llvm-project#111679 Original commit message: [DAGISel] Keep flags when converting FP load/store to integer (#111679) This DAG combine replaces a floating-point load/store pair which has no other uses with an integer one, but did not copy the memory operand flags to the new instructions, resulting in it dropping the volatile flag. This optimisation is still valid if one or both of the instructions is volatile, so we can copy over the whole MachineMemOperand to generate volatile integer loads and stores where needed.
ostannard
force-pushed
the
volatile-bug-llvm19
branch
from
3022296 to
db19a7a
Compare
ostannard
changed the title
We have a customer affected by this bug, which was fixed after LLVM 19, and we'd like to do a release of LLVM-ET 19 with the fix.
LLVM PR: llvm/llvm-project#111679
Original commit message:
[DAGISel] Keep flags when converting FP load/store to integer (#111679)
This DAG combine replaces a floating-point load/store pair which has no other uses with an integer one, but did not copy the memory operand flags to the new instructions, resulting in it dropping the volatile flag. This optimisation is still valid if one or both of the instructions is volatile, so we can copy over the whole MachineMemOperand to generate volatile integer loads and stores where needed.