[wasm] NET8 previews performance tracking

[radekdoulik]

Issues to look at (most important at top):

AOT related issues
dotnet/perf-autofiling-issues#14019 223 regressions, probably AOT and generics
dotnet/perf-autofiling-issues#13755 23 regressions, might show opportunities for SIMD improvements
dotnet/perf-autofiling-issues#19230 3 regressions, utf16 widening code change. It now uses Vector128.Widen instead of 2 calls to WidenLower and WidenUpper. This should result in theory in the same code as Widen is calling these too and returning pair. There might be something like additional store in the emitted code. To chase further, try to compare before and after emitted code of the WidenAsciiToUtf16 method.

NET8 Preview1-Preview6 comparison - choose previews branches, NET8 p1 and p6 in our case, and one of wasm RunConfig's
https://msit.powerbi.com/groups/me/apps/54e0e83f-07bc-45bf-87b7-a7677ff3af2a/reports/b7f22b49-2eb8-4608-933a-13305cc8fab6/ReportSection587f36b937cea0cbe025?experience=power-bi

interpreter

only few regressions, overall the performance is improved

AOT

there are 200+ regressions from dotnet/perf-autofiling-issues#14019

still mostly improvements in AOT as well

Blazor size regression

Tagging subscribers to this area: @directhex
See info in area-owners.md if you want to be subscribed.

Issue Details

---

Issues to look at (most important at top):

AOT related issues
dotnet/perf-autofiling-issues#14019 223 regressions, probably AOT and generics
dotnet/perf-autofiling-issues#13755 23 regressions, might show opportunities for SIMD improvements
dotnet/perf-autofiling-issues#19230 3 regressions, utf16 widening code change

NET8 Preview1-Preview6 comparison - choose previews branches, NET8 p1 and p6 in our case, and one of wasm RunConfig's
https://msit.powerbi.com/groups/me/apps/54e0e83f-07bc-45bf-87b7-a7677ff3af2a/reports/b7f22b49-2eb8-4608-933a-13305cc8fab6/ReportSection587f36b937cea0cbe025?experience=power-bi

interpreter

only few regressions, overall the performance is improved

AOT

there are 200+ regressions from dotnet/perf-autofiling-issues#14019

still mostly improvements in AOT as well

Blazor size regression

Author:	radekdoulik
Assignees:	radekdoulik
Labels:	area-Infrastructure-mono
Milestone:	8.0.0

[xtqqczze]

dotnet/perf-autofiling-issues#19230 3 regressions, utf16 widening code change. It now uses Vector128.Widen instead of 2 calls to WidenLower and WidenUpper. This should result in theory in the same code as Widen is calling these too and returning pair. There might be something like additional store in the emitted code. To chase further, try to compare before and after emitted code of the WidenAsciiToUtf16 method.

#87737