Improve loop inversion #52347
Merged
Improve loop inversion #52347
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
When doing loop inversion, we duplicate the condition block to the top of the loop to create a fall-through zero trip test. Improve this by redirecting all incoming branches to the condition block that appear to be coming from outside the potential loop body to branch to the new duplicated condition block. This improves the ability of the loop recognizer to find loops, whereas before we would reject the loops as "multi-entry". There are good diffs where more loops are detected, leading to better optimization and more loop alignment. There are also asm diffs regressions.
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
|
Perf results on BenchmarksGame and BenchI shows a big improvement in BenchE (which was the goal), and no obvious above-the-noise diff in the others.
|
|
There are some asserts that need to be addressed: |
|
/azp run runtime-coreclr jitstress |
|
Azure Pipelines successfully started running 1 pipeline(s). |
x64 diffsasm.aspnet.run.windows.x64.checked.4 Detail diffsasm.benchmarks.run.windows.x64.checked.3 Detail diffsasm.libraries.crossgen.windows.x64.checked.3 Detail diffsasm.libraries.crossgen2.windows.x64.checked.2 Detail diffsasm.libraries.pmi.windows.x64.checked.2 Detail diffsasm.tests.pmi.windows.x64.checked Detail diffsasm.tests_libraries.pmi.windows.x64.checked Detail diffsx86 diffsasm.benchmarks.run.windows.x86.checked.1 Detail diffsasm.libraries.crossgen.windows.x86.checked.1 Detail diffsasm.libraries.crossgen2.windows.x86.checked.1 Detail diffsasm.libraries.pmi.windows.x86.checked.1 Detail diffsasm.tests.pmi.windows.x86.checked Detail diffsasm.tests_libraries.pmi.windows.x86.checked.1 Detail diffs |
|
@AndyAyersMS @dotnet/jit-contrib PTAL |
AndyAyersMS
approved these changes
May 6, 2021
| // Redirect the predecessor to the new block. | ||
| JITDUMP("Redirecting non-loop " FMT_BB " -> " FMT_BB " to " FMT_BB " -> " FMT_BB "\n", predBlock->bbNum, | ||
| bTest->bbNum, predBlock->bbNum, bNewCond->bbNum); | ||
| optRedirectBlock(predBlock, &blockMap, /*updatePreds*/ true); |
There was a problem hiding this comment.
Since we expect optRedirectBlock to actually do something, I wonder if it should return a bool indicating that it made changes..?
Or, perhaps verify after all this redirecting that the block and the cloned block now have the expected numbers of preds?
There was a problem hiding this comment.
Hmmm... I think the only way it could fail to redirect is if the predecessor is BBJ_NONE, which shouldn't occur. I suppose your suggestion is just to assert this. (Other types like BBJ_THROW, BBJ_RETURN, should never be predecessors.)
It turns out if it ends up not redirecting the pred, it's not fatal anyway, it's just sub-optimal.
|
test failure is #51543 |
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.
When doing loop inversion, we duplicate the condition block to the
top of the loop to create a fall-through zero trip test. Improve
this by redirecting all incoming branches to the condition block
that appear to be coming from outside the potential loop body
to branch to the new duplicated condition block. This improves the
ability of the loop recognizer to find loops, whereas before we
would reject the loops as "multi-entry".
There are good diffs where more loops are detected, leading to
better optimization and more loop alignment.