JIT: Generalize FlowGraphNaturalLoop::AnalyzeIteration#97327
Merged
jakobbotsch merged 1 commit intodotnet:mainfrom Jan 23, 2024
Merged
JIT: Generalize FlowGraphNaturalLoop::AnalyzeIteration#97327jakobbotsch merged 1 commit intodotnet:mainfrom
FlowGraphNaturalLoop::AnalyzeIteration#97327jakobbotsch merged 1 commit intodotnet:mainfrom
Conversation
Consider all exit edges as potential tests on a IV variable, and try to recognize induction from those. Previously we looked only at the lexically bottom most block. Overall this unlocks only a pretty minor number of new cases which then leads to a bit more cloning. Size regressions are expected because of that.
ghost
added
the
area-CodeGen-coreclr
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsConsider all exit edges as potential tests on a IV variable, and try to recognize induction from those. Previously we looked only at the lexically bottom most block. Overall this unlocks only a pretty minor number of new cases which then leads to a bit more cloning. Size regressions are expected because of that. Based on #97325. Should unblock #97191.
|
Contributor
Assembly diffsAssembly diffs for linux/arm64 ran on windows/x64Diffs are based on 2,501,261 contexts (1,003,806 MinOpts, 1,497,455 FullOpts). MISSED contexts: base: 3,546 (0.14%), diff: 3,956 (0.16%) Overall (+654,368 bytes)
FullOpts (+654,368 bytes)
Assembly diffs for linux/x64 ran on windows/x64Diffs are based on 2,595,036 contexts (1,052,329 MinOpts, 1,542,707 FullOpts). MISSED contexts: base: 3,596 (0.14%), diff: 3,599 (0.14%) Overall (+847,760 bytes)
FullOpts (+847,760 bytes)
Assembly diffs for osx/arm64 ran on windows/x64Diffs are based on 2,262,764 contexts (930,876 MinOpts, 1,331,888 FullOpts). MISSED contexts: base: 2,925 (0.13%), diff: 3,201 (0.14%) Overall (+527,124 bytes)
FullOpts (+527,124 bytes)
Assembly diffs for windows/arm64 ran on windows/x64Diffs are based on 2,318,293 contexts (931,543 MinOpts, 1,386,750 FullOpts). MISSED contexts: base: 2,587 (0.11%), diff: 2,601 (0.11%) Overall (+731,948 bytes)
FullOpts (+731,948 bytes)
Assembly diffs for windows/x64 ran on windows/x64Diffs are based on 2,492,949 contexts (983,689 MinOpts, 1,509,260 FullOpts). MISSED contexts: base: 3,859 (0.15%), diff: 3,862 (0.15%) Overall (+879,433 bytes)
FullOpts (+879,433 bytes)
Details here Assembly diffs for linux/arm ran on windows/x86Diffs are based on 2,237,735 contexts (827,812 MinOpts, 1,409,923 FullOpts). MISSED contexts: base: 74,052 (3.20%), diff: 74,543 (3.22%) Overall (+922,034 bytes)
FullOpts (+922,034 bytes)
Assembly diffs for windows/x86 ran on windows/x86Diffs are based on 2,299,179 contexts (841,817 MinOpts, 1,457,362 FullOpts). MISSED contexts: base: 2,090 (0.09%), diff: 2,191 (0.10%) Overall (+281,351 bytes)
FullOpts (+281,351 bytes)
Details here Throughput diffsThroughput diffs for linux/arm64 ran on windows/x64Overall (+0.01% to +0.37%)
FullOpts (+0.01% to +0.50%)
Throughput diffs for linux/x64 ran on windows/x64Overall (+0.02% to +0.42%)
FullOpts (+0.02% to +0.55%)
Throughput diffs for osx/arm64 ran on windows/x64Overall (+0.01% to +0.40%)
FullOpts (+0.01% to +0.60%)
Throughput diffs for windows/arm64 ran on windows/x64Overall (+0.01% to +0.55%)
FullOpts (+0.01% to +0.83%)
Throughput diffs for windows/x64 ran on windows/x64Overall (+0.02% to +0.53%)
FullOpts (+0.02% to +0.74%)
Details here Throughput diffs for linux/arm ran on windows/x86Overall (+0.01% to +0.59%)
FullOpts (+0.01% to +0.77%)
Throughput diffs for windows/x86 ran on windows/x86Overall (+0.01% to +0.27%)
FullOpts (+0.01% to +0.34%)
Details here Throughput diffs for linux/arm64 ran on linux/x64Overall (+0.02% to +0.38%)
FullOpts (+0.02% to +0.51%)
Throughput diffs for linux/x64 ran on linux/x64Overall (+0.02% to +0.43%)
FullOpts (+0.02% to +0.56%)
Details here |
Member
Author
|
/azp run runtime-coreclr jitstress, runtime-coreclr libraries-jitstress |
|
Azure Pipelines successfully started running 2 pipeline(s). |
Member
Author
|
cc @dotnet/jit-contrib PTAL @BruceForstall Should be ready assuming CI looks good. Diffs are quite substantial due to the new cloning, but it looks like the same methods are repeated a bunch of times if you look at the jit-analyze output, so probably the results are a bit overblown. As usual loop cloning profitability heuristics is probably the right way to deal with this. |
BruceForstall
approved these changes
Jan 22, 2024
Member
Author
tmds
pushed a commit
to tmds/runtime
that referenced
this pull request
Jan 23, 2024
Consider all exit edges as potential tests on a IV variable, and try to recognize induction from those. Previously we looked only at the lexically bottom most block. Overall this unlocks only a pretty minor number of new cases which then leads to a bit more cloning. Size regressions are expected because of that.
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
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
3 participants
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.
Consider all exit edges as potential tests on a IV variable, and try to recognize induction from those. Previously we looked only at the lexically bottom most block.
Overall this unlocks only a pretty minor number of new cases which then leads to a bit more cloning. Size regressions are expected because of that.
Based on #97325. Should unblock #97191.