[Perf] ARM64 regression in System.Diagnostics.Perf_Activity.ActivityAllocations #68624
Description
Run Information
| Architecture | arm64 |
|---|---|
| OS | ubuntu 18.04 |
| Baseline | b58d8013dbc4f3a65e5e2242fe09a43f455a329f |
| Compare | 84b664a9875067a97fe32b654ff32f23f4001a60 |
| Diff | Diff |
Regressions in System.Diagnostics.Perf_Activity
| Benchmark | Baseline | Test | Test/Base | Test Quality | Edge Detector | Baseline IR | Compare IR | IR Ratio | Baseline ETL | Compare ETL |
|---|---|---|---|---|---|---|---|---|---|---|
| ActivityAllocations - Duration of single invocation | 758.56 ns | 813.42 ns | 1.07 | 0.20 | False | |||||
| ActivityAllocations - Duration of single invocation | 623.66 ns | 715.80 ns | 1.15 | 0.11 | False |
Repro
git clone https://github.com/dotnet/performance.git
python3 .\performance\scripts\benchmarks_ci.py -f net6.0 --filter 'System.Diagnostics.Perf_Activity*'Details
Payloads
Histogram
System.Diagnostics.Perf_Activity.ActivityAllocations(idFormat: W3C)
Description of detection logic
IsRegressionChecked: Marked as regression because the three check build points were 0.05 greater than the baseline.
IsImprovementBase: Marked as not an improvement because the compare was not 5% less than the baseline, or the value was too small.
IsRegressionBase: Marked as regression because the compare was 5% greater than the baseline, and the value was not too small.
IsRegressionChecked: Marked as regression because the three check build points were 0.05 greater than the baseline.
IsRegressionWindowed: Marked as regression because 813.4234915083607 > 802.4856996133994.
IsChangePoint: Marked as a change because one of 4/13/2022 10:02:29 PM, 4/19/2022 4:53:57 AM falls between 4/10/2022 3:30:35 PM and 4/19/2022 4:53:57 AM.
IsRegressionStdDev: Marked as regression because -6.738201722219294 (T) = (0 -836.9530328425158) / Math.Sqrt((284.7705405371702 / (23)) + (1754.8085293566023 / (18))) is less than -2.0226909200346674 = MathNet.Numerics.Distributions.StudentT.InvCDF(0, 1, (23) + (18) - 2, .025) and -0.09216640816227123 = (766.323727398658 - 836.9530328425158) / 766.323727398658 is less than -0.05.
IsImprovementBase: Marked as not an improvement because the compare was not 5% less than the baseline, or the value was too small.
IsChangeEdgeDetector: Marked not as a regression because Edge Detector said so.
```#### System.Diagnostics.Perf_Activity.ActivityAllocations(idFormat: Hierarchical)
```log
Description of detection logic
IsRegressionChecked: Marked as regression because the three check build points were 0.05 greater than the baseline.
IsImprovementBase: Marked as not an improvement because the compare was not 5% less than the baseline, or the value was too small.
IsRegressionBase: Marked as regression because the compare was 5% greater than the baseline, and the value was not too small.
IsRegressionChecked: Marked as regression because the three check build points were 0.05 greater than the baseline.
IsRegressionWindowed: Marked as regression because 715.7983124923339 > 655.6405660025653.
IsChangePoint: Marked as a change because one of 4/13/2022 4:27:35 PM, 4/19/2022 4:53:57 AM falls between 4/10/2022 3:30:35 PM and 4/19/2022 4:53:57 AM.
IsRegressionStdDev: Marked as regression because -11.823299540885158 (T) = (0 -708.1518629941236) / Math.Sqrt((243.4064317595605 / (20)) + (499.0656483304033 / (18))) is less than -2.028094000977961 = MathNet.Numerics.Distributions.StudentT.InvCDF(0, 1, (20) + (18) - 2, .025) and -0.11789003363534732 = (633.4718457872224 - 708.1518629941236) / 633.4718457872224 is less than -0.05.
IsImprovementBase: Marked as not an improvement because the compare was not 5% less than the baseline, or the value was too small.
IsChangeEdgeDetector: Marked not as a regression because Edge Detector said so.
Docs
Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository
Activity
AndyAyersMS commentedon Apr 22, 2022
Possibly from #67920.
AndyAyersMS commentedon Apr 27, 2022
@tarekgh thinks #67920 is not the likely suspect.
Another possibility given that this is arm64 is #66407. SPMI Diffs show there was an impact to
System.Diagnostics.Tracing.EventListener:AddEventSourcethough it was a code size reduction.The regression has persisted since then, so it does not look like noise.
[-][Perf] Changes at 4/14/2022 3:36:10 AM[/-][+][Perf] ARM64 regression in System.Diagnostics.Perf_Activity.ActivityAllocations[/+]25 remaining items
tarekgh commentedon Jun 17, 2022
No, I didn't try arm. But I am wondering now, wouldn't this be more related to JIT then? I cannot think of anything in our changes that can cause such regression in specific architecture.
I would keep the test as it is for now as it looks sensitive to the JIT changes. So, I am seeing the value to keep it. We can consider adding more tests later for real usage scenarios with Activity.
JulieLeeMSFT commentedon Jul 1, 2022
@TIHan PTAL.
TIHan commentedon Jul 5, 2022
@JulieLeeMSFT This regression is occurring in x64 as well, so it's not related to changes in the JIT for ARM64. The regression was also said to be acceptable:
Closing. Anyone is free to re-open this if they deem it critical.