[perf] Group inline-related properties in `ComputedStyle` [chromium/src : main]
Koji Ishii (Gerrit)
Koji Ishii added 1 comment![]( "Open in Gerrit")
Related details
- Anders Hartvoll Ruud
Code-Coverage
Code-Owners
Code-Review
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Steinar H Gunderson (Gerrit)
Steinar H Gunderson added 1 comment![]( "Open in Gerrit")
If we're not better on Speedometer, and code size goes up… why are we doing this? :-) Are we showing better on some other benchmark (e.g. the style perftest)? Are we reducing memory usage? Some other unstated benefit?
Related details
- Anders Hartvoll Ruud
- Koji Ishii
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
chromeperf@appspot.gserviceaccount.com (Gerrit)
Message from chrom...@appspot.gserviceaccount.com![]( "Open in Gerrit")
📍 Job mac-m1_mini_2020-perf/speedometer3 complete.
See results at: https://pinpoint-dot-chromeperf.appspot.com/job/1478ad9e710000
Steinar H Gunderson (Gerrit)
Steinar H Gunderson added 1 comment![]( "Open in Gerrit")
If we're not better on Speedometer, and code size goes up… why are we doing this? :-) Are we showing better on some other benchmark (e.g. the style perftest)? Are we reducing memory usage? Some other unstated benefit?
FWIW, I pooled all of your 750 Speedometer runs, and it still didn't show up as statistically significant on any of the subtests (nor the total score), so I think we can pretty clearly say there's no Speedometer win here.
Memory usage on the style perftest appears to be ever so slightly worse (in practice probably the same). Performance on the style perftest appears to be in the noise.
Koji Ishii (Gerrit)
Koji Ishii added 1 comment![]( "Open in Gerrit")
Steinar H GundersonIf we're not better on Speedometer, and code size goes up… why are we doing this? :-) Are we showing better on some other benchmark (e.g. the style perftest)? Are we reducing memory usage? Some other unstated benefit?
FWIW, I pooled all of your 750 Speedometer runs, and it still didn't show up as statistically significant on any of the subtests (nor the total score), so I think we can pretty clearly say there's no Speedometer win here.
Memory usage on the style perftest appears to be ever so slightly worse (in practice probably the same). Performance on the style perftest appears to be in the noise.
Thanks for taking a look at this. The goal is to improve the memory locality in the line breaker and inline layout.
I'm also hoping to improve the memory in style, as inline-related properties are more likely to be set together, but I'm not familiar with style perf tests. Thanks for running it and finding that there's a slight regression. Can you tell me how to run it by myself, to experiment other combinations?
I was wondering whether this should focus on line breaker only, or include inline layout. If this is showing a regressed result, I'd like to try focusing on line breaker.
Related details
- Anders Hartvoll Ruud
- Steinar H Gunderson
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Steinar H Gunderson (Gerrit)
Steinar H Gunderson added 1 comment![]( "Open in Gerrit")
Steinar H GundersonIf we're not better on Speedometer, and code size goes up… why are we doing this? :-) Are we showing better on some other benchmark (e.g. the style perftest)? Are we reducing memory usage? Some other unstated benefit?
Koji IshiiFWIW, I pooled all of your 750 Speedometer runs, and it still didn't show up as statistically significant on any of the subtests (nor the total score), so I think we can pretty clearly say there's no Speedometer win here.
Memory usage on the style perftest appears to be ever so slightly worse (in practice probably the same). Performance on the style perftest appears to be in the noise.
Steinar H GundersonThanks for taking a look at this. The goal is to improve the memory locality in the line breaker and inline layout.
I'm also hoping to improve the memory in style, as inline-related properties are more likely to be set together, but I'm not familiar with style perf tests. Thanks for running it and finding that there's a slight regression. Can you tell me how to run it by myself, to experiment other combinations?
I was wondering whether this should focus on line breaker only, or include inline layout. If this is showing a regressed result, I'd like to try focusing on line breaker.
FWIW, I generally get disappointingly little win from memory locality optimizations in S3, even more so on macOS. The data sets generally fit too well in the M1's large and fast L2 cache.
You can run the style perftest by ./out/Release/blink_perf_tests and looking at the GCAlloc lines. If you want to get more stable results (Oilpan leaks a tiny bit of behavior between tests, unfortunately), you can run one test per process, e.g.:
```
for X in ECommerce Encyclopedia Extension News Search Social1 Social2 Sports Video; do printf "%-20s " $X; ./out/Release/blink_perf_tests --gtest_filter=StyleCalcPerfTest.$X 2>&1 | grep -i gcalloc; done
```
You can also use `--measure-computed-style-memory`, which simply counts itself instead of asking Oilpan. This won't tell you about GC patterns, though.
Related details
- Anders Hartvoll Ruud
- Koji Ishii
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Koji Ishii (Gerrit)
Koji Ishii added 1 comment![]( "Open in Gerrit")
Thanks for many ideas. I'll turn this to WIP for further investigations.
Related details
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
chromeperf@appspot.gserviceaccount.com (Gerrit)
Message from chrom...@appspot.gserviceaccount.com![]( "Open in Gerrit")
📍 Job mac-m1_mini_2020-perf/speedometer3 complete.
See results at: https://pinpoint-dot-chromeperf.appspot.com/job/1563c39e710000
Related details
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |