Consolidate importer spilling code #72291
Merged
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
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
ghost
added
the
community-contribution
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsFixes #72133.
|
SingleAccretion
changed the title
SingleAccretion
force-pushed
the
More-Spilling-Fixes-Upstream
branch
2 times, most recently
from
9771593
to
230661b
Compare
|
@dotnet/jit-contrib |
AndyAyersMS
reviewed
Jul 18, 2022
The comment states we don't need it, which is incorrect. Diffs are improvements because we block forward substitution of calls into "ASG(BLK(ADDR(LCL_VAR<field>, ...)))", which allows morph to leave the "can be replaced with its field" local alone.
Spill "glob refs" on stores to "aliased" locals.
SingleAccretion
force-pushed
the
More-Spilling-Fixes-Upstream
branch
from
ee37049
to
a22fb7c
Compare
AndyAyersMS
approved these changes
Jul 23, 2022
jkotas
added a commit
to jkotas/runtime
that referenced
this pull request
Jul 24, 2022
This reverts commit c624626.
jkotas
added a commit
that referenced
this pull request
Jul 24, 2022
SingleAccretion
added a commit
to SingleAccretion/runtime
that referenced
this pull request
Jul 24, 2022
* Add tests * Fix losing GLOB_REF on the LHS The comment states we don't need it, which is incorrect. Diffs are improvements because we block forward substitution of calls into "ASG(BLK(ADDR(LCL_VAR<field>, ...)))", which allows morph to leave the "can be replaced with its field" local alone. * Prospective fix Spill "glob refs" on stores to "aliased" locals. * Delete now-not-necessary code * Fix up asserts * Clean out '(unsigned)CHECK_SPILL_ALL/NONE' casts * Don't manually spill for 'st[s]fld' * Revert 'Clean out '(unsigned)CHECK_SPILL_ALL/NONE' casts'
AndyAyersMS
pushed a commit
that referenced
this pull request
Jul 26, 2022
* Consolidate importer spilling code (#72291) * Add tests * Fix losing GLOB_REF on the LHS The comment states we don't need it, which is incorrect. Diffs are improvements because we block forward substitution of calls into "ASG(BLK(ADDR(LCL_VAR<field>, ...)))", which allows morph to leave the "can be replaced with its field" local alone. * Prospective fix Spill "glob refs" on stores to "aliased" locals. * Delete now-not-necessary code * Fix up asserts * Clean out '(unsigned)CHECK_SPILL_ALL/NONE' casts * Don't manually spill for 'st[s]fld' * Revert 'Clean out '(unsigned)CHECK_SPILL_ALL/NONE' casts' * Fix assignments done via return buffers The mistake in logic was that the only trees which could modify unaliased locals are assignments, which is not true, calls can do that as well. One day we will move the return buffer handling out of importer, but until then, special handling is required. An alternative fix would have been to bring back the explicit "impSpillLclRefs" to "stloc/starg" code, but that would contradict the overall goal of consolidating the spilling logic.
SingleAccretion
added a commit
to SingleAccretion/runtime
that referenced
this pull request
Aug 22, 2022
* Consolidate importer spilling code (dotnet#72291) * Add tests * Fix losing GLOB_REF on the LHS The comment states we don't need it, which is incorrect. Diffs are improvements because we block forward substitution of calls into "ASG(BLK(ADDR(LCL_VAR<field>, ...)))", which allows morph to leave the "can be replaced with its field" local alone. * Prospective fix Spill "glob refs" on stores to "aliased" locals. * Delete now-not-necessary code * Fix up asserts * Clean out '(unsigned)CHECK_SPILL_ALL/NONE' casts * Don't manually spill for 'st[s]fld' * Revert 'Clean out '(unsigned)CHECK_SPILL_ALL/NONE' casts' * Fix assignments done via return buffers The mistake in logic was that the only trees which could modify unaliased locals are assignments, which is not true, calls can do that as well. One day we will move the return buffer handling out of importer, but until then, special handling is required. An alternative fix would have been to bring back the explicit "impSpillLclRefs" to "stloc/starg" code, but that would contradict the overall goal of consolidating the spilling logic.
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.
Fixes #72133.
Diffs are mostly regressions, with some improvements due to the new logic being less conservative in some cases (not spilling
ADDR(LCL_VAR)s as much).The most common cause of regression is from
initobj. Essentially, for IL like below:We will now get this IR:
Instead of:
This is, of course, because the new code (correctly) presumes
<some load>could be aliased by the store.A solution to this could be to enhance the
impILConsumesAddrlogic to account forinitobjs. This, however, is both risky and non-trivial, as there are dependencies (e. g. the single-def logic) onlvHasLdAddrOp. Essentially, it would requires treatingldloca; initobjas equivalent tostlocvery early on.