[Perf] Windows/arm64: Regressions in IfStatements

[performanceautofiler]

Run Information

Architecture	arm64
OS	Windows 10.0.19041
Baseline	7d5efbb9e10b6d8beb91c90cbdefd7360869cece
Compare	0e24ea7c2a0436a8f2bf83e8f5981ec035518b99
Diff	Diff

Regressions in IfStatements.IfStatements

Benchmark	Baseline	Test	Test/Base	Test Quality	Edge Detector	Baseline IR	Compare IR	IR Ratio	Baseline ETL	Compare ETL
AndAnd - Duration of single invocation	50.85 μs	60.02 μs	1.18	0.00	False
AndAndAnd - Duration of single invocation	48.76 μs	51.73 μs	1.06	0.00	False

Test Report

Repro

git clone https://github.com/dotnet/performance.git
py .\performance\scripts\benchmarks_ci.py -f net6.0 --filter 'IfStatements.IfStatements*'

Related Issues

Regressions

Improvements

Payloads

Baseline
Compare

Histogram

Edge Detector Info

Collection Data

IfStatements.IfStatements.AndAnd

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.
IsRegressionWindowed: Marked as regression because 60.016950830140495 > 53.430863129058444.
IsChangePoint: Marked as a change because one of 10/11/2022 1:45:10 PM, 11/1/2022 8:41:52 AM, 11/3/2022 4:18:41 AM falls between 10/25/2022 1:30:53 PM and 11/3/2022 4:18:41 AM.
IsRegressionStdDev: Marked as regression because -701.6727127857075 (T) = (0 -60020.57172055355) / Math.Sqrt((3457.9591975832996 / (39)) + (568.9918365919992 / (7))) is less than -2.0153675744421933 = MathNet.Numerics.Distributions.StudentT.InvCDF(0, 1, (39) + (7) - 2, .025) and -0.17980682074412616 = (50873.21980618611 - 60020.57172055355) / 50873.21980618611 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.
IsChangeEdgeDetector: Marked not as a regression because Edge Detector said so.

```#### IfStatements.IfStatements.AndAndAnd

```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.
IsRegressionWindowed: Marked as regression because 51.733831401766004 > 51.209566442757016.
IsChangePoint: Marked as a change because one of 11/1/2022 8:41:52 AM, 11/3/2022 4:18:41 AM falls between 10/25/2022 1:30:53 PM and 11/3/2022 4:18:41 AM.
IsRegressionStdDev: Marked as regression because -578.8484547464244 (T) = (0 -51739.81218834986) / Math.Sqrt((497.56047658656183 / (39)) + (91.36335693339602 / (7))) is less than -2.0153675744421933 = MathNet.Numerics.Distributions.StudentT.InvCDF(0, 1, (39) + (7) - 2, .025) and -0.06026238051797967 = (48799.06440052408 - 51739.81218834986) / 48799.06440052408 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.
IsChangeEdgeDetector: Marked not as a regression because Edge Detector said so.

Docs

Profiling workflow for dotnet/runtime repository
Benchmarking workflow for dotnet/runtime repository

### Run Information

Architecture	arm64
OS	Windows 10.0.19041
Baseline	7d5efbb9e10b6d8beb91c90cbdefd7360869cece
Compare	0e24ea7c2a0436a8f2bf83e8f5981ec035518b99
Diff	Diff

[dotnet-issue-labeler]

I couldn't figure out the best area label to add to this issue. If you have write-permissions please help me learn by adding exactly one area label.

[kunalspathak]

#73472 (comment)
@a74nh

[EgorBo]

Same regression on Ampere arm64 Ubuntu - dotnet/perf-autofiling-issues#9668

[a74nh]

I'm seeing the same slowdown.

Without if conversion. =~50us

        12000003          and     w3, w0, #1
        12000024          and     w4, w1, #1
        2A040063          orr     w3, w3, w4
        12000044          and     w4, w2, #1
        2A040063          orr     w3, w3, w4
        35000043          cbnz    w3, G_M12418_IG04
        528000A0          mov     w0, #5

With if conversion. =~60us

        12000003          and     w3, w0, #1
        12000024          and     w4, w1, #1
        2A040063          orr     w3, w3, w4
        35000083          cbnz    w3, G_M12418_IG04
        528000A3          mov     w3, #5
        7200005F          tst     w2, #1
        1A800060          csel    w0, w3, w0, eq

.... But with #77728 =~43us

        12000003          and     w3, w0, #1
        12000024          and     w4, w1, #1
        2A040063          orr     w3, w3, w4
        12000044          and     w4, w2, #1
        2A040063          orr     w3, w3, w4
        528000A4          mov     w4, #5
        7100007F          cmp     w3, #0
        1A800080          csel    w0, w4, w0, eq

I think this is because with the new patch, the if conversion happens after optimise bools. I could pull that part out of 77728 and submit it separately, or we could just wait for 77728 to get merged?

[a74nh]

I think this is because with the new patch, the if conversion happens after optimise bools

... Yes, it is.

The IR is a chain of JTRUEs.

• Without If conversion: Optimize bools turns the sequence into a single JTRUE statement.

• With If conversion: If conversion converts the final JTRUE, and ignores the others. Then optimize bools takes the remaining JTRUEs and combines them

• With the If conversion moved later: Optimize bools turns the sequence into a single JTRUE statement. Then If conversion converts this single block.

[a74nh]

Hoping this should be fixed now that #77728 is merged

[a74nh]

Not sure how to link the latest graphs here. But they drop quite nicely with the latest HEAD:

And And
50.85 μs -> 60.02 μs -> 41.39 μs

And And And
48.76 μs -> 51.73 μs -> 46.22 μs

@kunalspathak - should ok to close this now.

[jakobbotsch]

Closing since this was fixed by #77728.