JIT: Cache significant segments computations for layouts in physical promotion #87745
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
…promotion
Before:
JitEnablePhysicalPromotion=1
benchmarks.run_pgo:
Total num SignificantSegments calls: 2543
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
1 .. 1 ===> 312 count ( 37% of total)
2 .. 2 ===> 58 count ( 44% of total)
3 .. 3 ===> 181 count ( 66% of total)
4 .. 5 ===> 191 count ( 90% of total)
6 .. 10 ===> 66 count ( 98% of total)
11 .. 20 ===> 12 count ( 99% of total)
21 .. 35 ===> 4 count (100% of total)
36 .. 50 ===> 0 count (100% of total)
51 .. 75 ===> 0 count (100% of total)
76 .. 100 ===> 0 count (100% of total)
101 .. 150 ===> 0 count (100% of total)
151 .. 250 ===> 0 count (100% of total)
251 .. 500 ===> 0 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
libraries.pmi:
Total num SignificantSegments calls: 28525
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
1 .. 1 ===> 1154 count ( 19% of total)
2 .. 2 ===> 1878 count ( 50% of total)
3 .. 3 ===> 809 count ( 64% of total)
4 .. 5 ===> 782 count ( 77% of total)
6 .. 10 ===> 831 count ( 91% of total)
11 .. 20 ===> 357 count ( 97% of total)
21 .. 35 ===> 101 count ( 98% of total)
36 .. 50 ===> 29 count ( 99% of total)
51 .. 75 ===> 29 count ( 99% of total)
76 .. 100 ===> 8 count (100% of total)
101 .. 150 ===> 0 count (100% of total)
151 .. 250 ===> 0 count (100% of total)
251 .. 500 ===> 0 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
JitEnablePhysicalPromotion=1;JitStressModeNames=STRESS_NO_OLD_PROMOTION
benchmarks.run_pgo:
Total num SignificantSegments calls: 90839
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
1 .. 1 ===> 681 count ( 7% of total)
2 .. 2 ===> 1035 count ( 17% of total)
3 .. 3 ===> 1635 count ( 34% of total)
4 .. 5 ===> 1053 count ( 45% of total)
6 .. 10 ===> 3162 count ( 78% of total)
11 .. 20 ===> 814 count ( 87% of total)
21 .. 35 ===> 931 count ( 96% of total)
36 .. 50 ===> 215 count ( 99% of total)
51 .. 75 ===> 69 count ( 99% of total)
76 .. 100 ===> 4 count ( 99% of total)
101 .. 150 ===> 4 count ( 99% of total)
151 .. 250 ===> 0 count ( 99% of total)
251 .. 500 ===> 11 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
libraries.pmi:
Total num SignificantSegments calls: 277708
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
1 .. 1 ===> 6993 count ( 17% of total)
2 .. 2 ===> 8197 count ( 38% of total)
3 .. 3 ===> 5225 count ( 51% of total)
4 .. 5 ===> 5380 count ( 65% of total)
6 .. 10 ===> 7141 count ( 83% of total)
11 .. 20 ===> 4094 count ( 93% of total)
21 .. 35 ===> 1627 count ( 97% of total)
36 .. 50 ===> 519 count ( 98% of total)
51 .. 75 ===> 292 count ( 99% of total)
76 .. 100 ===> 98 count ( 99% of total)
101 .. 150 ===> 59 count ( 99% of total)
151 .. 250 ===> 11 count ( 99% of total)
251 .. 500 ===> 4 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
After:
benchmarks.run_pgo:
Total num SignificantSegments calls: 915
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
1 .. 1 ===> 741 count ( 89% of total)
2 .. 2 ===> 75 count ( 99% of total)
3 .. 3 ===> 8 count (100% of total)
4 .. 5 ===> 0 count (100% of total)
6 .. 10 ===> 0 count (100% of total)
11 .. 20 ===> 0 count (100% of total)
21 .. 35 ===> 0 count (100% of total)
36 .. 50 ===> 0 count (100% of total)
51 .. 75 ===> 0 count (100% of total)
76 .. 100 ===> 0 count (100% of total)
101 .. 150 ===> 0 count (100% of total)
151 .. 250 ===> 0 count (100% of total)
251 .. 500 ===> 0 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
libraries.pmi:
Total num SignificantSegments calls: 9061
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
1 .. 1 ===> 4172 count ( 69% of total)
2 .. 2 ===> 1105 count ( 88% of total)
3 .. 3 ===> 416 count ( 95% of total)
4 .. 5 ===> 214 count ( 98% of total)
6 .. 10 ===> 67 count ( 99% of total)
11 .. 20 ===> 3 count ( 99% of total)
21 .. 35 ===> 1 count (100% of total)
36 .. 50 ===> 0 count (100% of total)
51 .. 75 ===> 0 count (100% of total)
76 .. 100 ===> 0 count (100% of total)
101 .. 150 ===> 0 count (100% of total)
151 .. 250 ===> 0 count (100% of total)
251 .. 500 ===> 0 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
JitEnablePhysicalPromotion=1;JitStressModeNames=STRESS_NO_OLD_PROMOTION
benchmarks.run_pgo:
Total num SignificantSegments calls: 15082
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
1 .. 1 ===> 6709 count ( 69% of total)
2 .. 2 ===> 1735 count ( 87% of total)
3 .. 3 ===> 462 count ( 92% of total)
4 .. 5 ===> 471 count ( 97% of total)
6 .. 10 ===> 237 count (100% of total)
11 .. 20 ===> 0 count (100% of total)
21 .. 35 ===> 0 count (100% of total)
36 .. 50 ===> 0 count (100% of total)
51 .. 75 ===> 0 count (100% of total)
76 .. 100 ===> 0 count (100% of total)
101 .. 150 ===> 0 count (100% of total)
151 .. 250 ===> 0 count (100% of total)
251 .. 500 ===> 0 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
libraries.pmi:
Total uncached SignificantSegments calls: 68938
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
1 .. 1 ===> 24323 count ( 61% of total)
2 .. 2 ===> 8687 count ( 83% of total)
3 .. 3 ===> 3357 count ( 91% of total)
4 .. 5 ===> 2378 count ( 97% of total)
6 .. 10 ===> 789 count ( 99% of total)
11 .. 20 ===> 105 count ( 99% of total)
21 .. 35 ===> 1 count (100% of total)
36 .. 50 ===> 0 count (100% of total)
51 .. 75 ===> 0 count (100% of total)
76 .. 100 ===> 0 count (100% of total)
101 .. 150 ===> 0 count (100% of total)
151 .. 250 ===> 0 count (100% of total)
251 .. 500 ===> 0 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
dotnet-issue-labeler
bot
added
the
area-CodeGen-coreclr
ghost
assigned 
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch Issue DetailsAlso removes the fixed bit vector debug checking code as I felt it had served its purpose and I didn't want to further complicate the caching. This should save a lot of JIT-EE calls from physical promotion. JitEnablePhysicalPromotion=1
benchmarks.run_pgo:
-Total num SignificantSegments calls: 2543
+Total num SignificantSegments calls: 915
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
- 1 .. 1 ===> 312 count ( 37% of total)
- 2 .. 2 ===> 58 count ( 44% of total)
- 3 .. 3 ===> 181 count ( 66% of total)
- 4 .. 5 ===> 191 count ( 90% of total)
- 6 .. 10 ===> 66 count ( 98% of total)
- 11 .. 20 ===> 12 count ( 99% of total)
- 21 .. 35 ===> 4 count (100% of total)
+ 1 .. 1 ===> 741 count ( 89% of total)
+ 2 .. 2 ===> 75 count ( 99% of total)
+ 3 .. 3 ===> 8 count (100% of total)
+ 4 .. 5 ===> 0 count (100% of total)
+ 6 .. 10 ===> 0 count (100% of total)
+ 11 .. 20 ===> 0 count (100% of total)
+ 21 .. 35 ===> 0 count (100% of total)
36 .. 50 ===> 0 count (100% of total)
51 .. 75 ===> 0 count (100% of total)
76 .. 100 ===> 0 count (100% of total)
@@ -19,19 +19,19 @@ Total num SignificantSegments calls: 2543
501 .. 1000 ===> 0 count (100% of total)
libraries.pmi:
-Total num SignificantSegments calls: 28525
+Total num SignificantSegments calls: 9061
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
- 1 .. 1 ===> 1154 count ( 19% of total)
- 2 .. 2 ===> 1878 count ( 50% of total)
- 3 .. 3 ===> 809 count ( 64% of total)
- 4 .. 5 ===> 782 count ( 77% of total)
- 6 .. 10 ===> 831 count ( 91% of total)
- 11 .. 20 ===> 357 count ( 97% of total)
- 21 .. 35 ===> 101 count ( 98% of total)
- 36 .. 50 ===> 29 count ( 99% of total)
- 51 .. 75 ===> 29 count ( 99% of total)
- 76 .. 100 ===> 8 count (100% of total)
+ 1 .. 1 ===> 4172 count ( 69% of total)
+ 2 .. 2 ===> 1105 count ( 88% of total)
+ 3 .. 3 ===> 416 count ( 95% of total)
+ 4 .. 5 ===> 214 count ( 98% of total)
+ 6 .. 10 ===> 67 count ( 99% of total)
+ 11 .. 20 ===> 3 count ( 99% of total)
+ 21 .. 35 ===> 1 count (100% of total)
+ 36 .. 50 ===> 0 count (100% of total)
+ 51 .. 75 ===> 0 count (100% of total)
+ 76 .. 100 ===> 0 count (100% of total)
101 .. 150 ===> 0 count (100% of total)
151 .. 250 ===> 0 count (100% of total)
251 .. 500 ===> 0 count (100% of total)
@@ -39,39 +39,39 @@ Total num SignificantSegments calls: 28525
JitEnablePhysicalPromotion=1;JitStressModeNames=STRESS_NO_OLD_PROMOTION
benchmarks.run_pgo:
-Total num SignificantSegments calls: 90839
+Total num SignificantSegments calls: 15082
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
- 1 .. 1 ===> 681 count ( 7% of total)
- 2 .. 2 ===> 1035 count ( 17% of total)
- 3 .. 3 ===> 1635 count ( 34% of total)
- 4 .. 5 ===> 1053 count ( 45% of total)
- 6 .. 10 ===> 3162 count ( 78% of total)
- 11 .. 20 ===> 814 count ( 87% of total)
- 21 .. 35 ===> 931 count ( 96% of total)
- 36 .. 50 ===> 215 count ( 99% of total)
- 51 .. 75 ===> 69 count ( 99% of total)
- 76 .. 100 ===> 4 count ( 99% of total)
- 101 .. 150 ===> 4 count ( 99% of total)
- 151 .. 250 ===> 0 count ( 99% of total)
- 251 .. 500 ===> 11 count (100% of total)
+ 1 .. 1 ===> 6709 count ( 69% of total)
+ 2 .. 2 ===> 1735 count ( 87% of total)
+ 3 .. 3 ===> 462 count ( 92% of total)
+ 4 .. 5 ===> 471 count ( 97% of total)
+ 6 .. 10 ===> 237 count (100% of total)
+ 11 .. 20 ===> 0 count (100% of total)
+ 21 .. 35 ===> 0 count (100% of total)
+ 36 .. 50 ===> 0 count (100% of total)
+ 51 .. 75 ===> 0 count (100% of total)
+ 76 .. 100 ===> 0 count (100% of total)
+ 101 .. 150 ===> 0 count (100% of total)
+ 151 .. 250 ===> 0 count (100% of total)
+ 251 .. 500 ===> 0 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
libraries.pmi:
-Total num SignificantSegments calls: 277708
+Total num SignificantSegments calls: 68938
(Per context) SignificantSegments calls
<= 0 ===> 0 count ( 0% of total)
- 1 .. 1 ===> 6993 count ( 17% of total)
- 2 .. 2 ===> 8197 count ( 38% of total)
- 3 .. 3 ===> 5225 count ( 51% of total)
- 4 .. 5 ===> 5380 count ( 65% of total)
- 6 .. 10 ===> 7141 count ( 83% of total)
- 11 .. 20 ===> 4094 count ( 93% of total)
- 21 .. 35 ===> 1627 count ( 97% of total)
- 36 .. 50 ===> 519 count ( 98% of total)
- 51 .. 75 ===> 292 count ( 99% of total)
- 76 .. 100 ===> 98 count ( 99% of total)
- 101 .. 150 ===> 59 count ( 99% of total)
- 151 .. 250 ===> 11 count ( 99% of total)
- 251 .. 500 ===> 4 count (100% of total)
+ 1 .. 1 ===> 24323 count ( 61% of total)
+ 2 .. 2 ===> 8687 count ( 83% of total)
+ 3 .. 3 ===> 3357 count ( 91% of total)
+ 4 .. 5 ===> 2378 count ( 97% of total)
+ 6 .. 10 ===> 789 count ( 99% of total)
+ 11 .. 20 ===> 105 count ( 99% of total)
+ 21 .. 35 ===> 1 count (100% of total)
+ 36 .. 50 ===> 0 count (100% of total)
+ 51 .. 75 ===> 0 count (100% of total)
+ 76 .. 100 ===> 0 count (100% of total)
+ 101 .. 150 ===> 0 count (100% of total)
+ 151 .. 250 ===> 0 count (100% of total)
+ 251 .. 500 ===> 0 count (100% of total)
501 .. 1000 ===> 0 count (100% of total)
|
|
cc @dotnet/jit-contrib PTAL @EgorBo |
EgorBo
approved these changes
Jun 19, 2023
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.
Also removes the fixed bit vector debug checking code as I felt it had served its purpose and I didn't want to further complicate the caching.
This should save a lot of JIT-EE calls from physical promotion.