wazevo(amd64): reserve dx when calling memmove to support go 1.24 #2378
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Signed-off-by: Anuraag Agrawal <[email protected]>
anuraaga
force-pushed
the
go-124-memmove
branch
from
f571a5c to
2f093ff
Compare
mathetake
reviewed
Feb 17, 2025
| @@ -1918,6 +1918,9 @@ func (m *machine) lowerCall(si *ssa.Instruction) { | |||
| for i := regalloc.RealReg(0); i < 16; i++ { | |||
| m.insert(m.allocateInstr().asDefineUninitializedReg(regInfo.RealRegToVReg[xmm0+i])) | |||
| } | |||
| // Since Go 1.24 it may also use DX, which is not reserved for the function call's 3 args. | |||
| // https://github.com/golang/go/blob/go1.24.0/src/runtime/memmove_amd64.s#L123 | |||
| m.insert(m.allocateInstr().asDefineUninitializedReg(regInfo.RealRegToVReg[rdx])) | |||
There was a problem hiding this comment.
instead of uninitialized, but shouldn't we need to put the proper value on it? otherwise it seems to me it would behave unexpectedly
There was a problem hiding this comment.
maybe looking at the memmove signature change see what's the expected value on rdx
There was a problem hiding this comment.
(this makes me feel like this might not be the amd64 only issue)
mathetake
approved these changes
Feb 17, 2025
anuraaga
added a commit
to anuraaga/wazero
that referenced
this pull request
Feb 18, 2025
…tratelabs#2378) Signed-off-by: Anuraag Agrawal <[email protected]>
reltuk
added a commit
to dolthub/go-icu-regex
that referenced
this pull request
Aug 20, 2025
Pick up bug fix for tetratelabs/wazero#2378
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes #2375
Go 1.24 changed memmove implementation on amd64, which resulted in an extra register allocation. This makes sure to reserve it for memmove invocation. While this could be a regression on go 1.23, it seems not significant enough to have version-specific code for this especially since it will go out-of-date soon.
Originally, @mathetake had suggested vendoring in the working go 1.23's memmove implementation, which both fixes and ensures forward compatibility, but I couldn't let go of the 30% perf gain from the change. After this, someone may want to try vendoring 1.24's implementation (it's not trivial due to needing to read many cpu flags), for forward compatibility, but before it I would check whether it's worth using the direct call to save on host function call overhead if implemented with normal
copy.golang/go@601ea46