Fix wrong constant folding for bswap16 #67726
Closed
Conversation
This file contains 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
The semantics of bswap16 is currently to swap the lower 2 bytes and leave the upper bytes alone. We need to keep the same behavior when constant folding or we can quickly end up discarding necessary casts. Fix dotnet#67723
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
|
Hmm, in fact on ARM64 we emit |
|
Tagging subscribers to this area: @JulieLeeMSFT Issue DetailsThe semantics of bswap16 is currently to swap the lower 2 bytes and Fix #67723 cc @dotnet/jit-contrib @aromaa
|
|
I've pushed a commit that disables constant folding for |
|
Small number of diffs. There are a few occurrences, but many of the diffs are duplicates of the same functions. |
|
A few other solutions come to mind if we want to continue to be able to constant fold:
|
jakobbotsch
added a commit
to jakobbotsch/runtime
that referenced
this pull request
Apr 12, 2022
Currently the JIT's constant folding (gtFoldExprConst and VNs EvalOpSpecialized) assumes that BSWAP16 zero extends into the upper 16 bits. This was not the case, and in fact the behavior of BSWAP16 depended on platform. Normally this would not be a problem since we always insert normalizing casts when creating BSWAP16 nodes, however VN was smart enough to remove this cast in some cases (see the test). Change the semantics of BSWAP16 nodes to zero extend into the upper 16 bits to match constant folding, and add a small peephole to avoid inserting this normalization in the common case where it is not necessary. Fixes dotnet#67723 Subsumes dotnet#67726
|
Subsumed by #67903 which implements bullet 3 above instead. |
jakobbotsch
added a commit
that referenced
this pull request
Apr 20, 2022
Currently the JIT's constant folding (gtFoldExprConst and VNs EvalOpSpecialized) assumes that BSWAP16 zero extends into the upper 16 bits. This was not the case, and in fact the behavior of BSWAP16 depended on platform. Normally this would not be a problem since we always insert normalizing casts when creating BSWAP16 nodes, however VN was smart enough to remove this cast in some cases (see the test). Change the semantics of BSWAP16 nodes to zero extend into the upper 16 bits to match constant folding, and add a small peephole to avoid inserting this normalization in the common case where it is not necessary. Fixes #67723 Subsumes #67726
directhex
pushed a commit
to directhex/runtime
that referenced
this pull request
Apr 21, 2022
Currently the JIT's constant folding (gtFoldExprConst and VNs EvalOpSpecialized) assumes that BSWAP16 zero extends into the upper 16 bits. This was not the case, and in fact the behavior of BSWAP16 depended on platform. Normally this would not be a problem since we always insert normalizing casts when creating BSWAP16 nodes, however VN was smart enough to remove this cast in some cases (see the test). Change the semantics of BSWAP16 nodes to zero extend into the upper 16 bits to match constant folding, and add a small peephole to avoid inserting this normalization in the common case where it is not necessary. Fixes dotnet#67723 Subsumes dotnet#67726
ghost
locked as resolved and limited conversation to collaborators
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
bswap16currently usesrev16on ARM architectures andror <16 bit reg>, 8on xarch. The behavior of these are not the same so consider the upper 16 bits to be undefined and disable constant folding of these.Fix #67723
cc @dotnet/jit-contrib @aromaa