Perf Utf8Encoding for ARM64 comparison

[carlossanlop]

The purpose of this PR is to verify the performance of the recent AdvSimd.Arm64 improvements added to Utf16Utility.Validation.cs.

The UTF8Encoding.GetByteCount method reaches the Utf16Utility.Validation.cs lines 82, 150 and 190, when called with the contents of the recently added text files containing long unicode strings.

cc @kunalspathak @pgovind @adamsitnik

[kunalspathak]

LGTM. Thanks!

[carlossanlop]

Results:

SSE2 OFF vs SSE2 ON

Off: $Env:COMPlus_EnableSSE2=0
On: $Env:COMPlus_EnableSSE2=1

❯ D:\runtime\.dotnet\dotnet.exe run --base C:\users\carlos\Desktop\sse_off\ --diff C:\users\carlos\Desktop\sse_on\ --threshold 0.001%
summary:
better: 20, geomean: 1.854
total diff: 20

No Slower results for the provided threshold = 0.001% and noise filter = 0.3ns.

| Faster                                                                | base/diff | Base Median (ns) | Diff Median (ns) | Modality|
| --------------------------------------------------------------------- | ---------:| ----------------:| ----------------:| --------:|
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: Greek)              |     10.49 |        154142.40 |         14700.10 |         |
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: Cyrillic)           |      8.97 |        104052.54 |         11601.99 |         |
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: Chinese)            |      5.33 |         72346.37 |         13583.33 |         |
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: EnglishAllAscii)    |      4.04 |         29311.31 |          7260.16 |         |
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: EnglishMostlyAscii) |      3.35 |         98971.88 |         29549.56 |         |

AdvSimd OFF vs AdvSimd ON

Off: export COMPlus_EnableArm64AdvSimd=0
On: export COMPlus_EnableArm64AdvSimd=1

root@calopearm:/home/calope/performance/src/tools/ResultsComparer# dn run -c release --base /home/calope/arm_off/ --diff /home/calope/arm_on/ --threshold 0.001%summary:
better: 13, geomean: 1.772
total diff: 13

No Slower results for the provided threshold = 0.001% and noise filter = 0.3ns.

| Faster                                                                | base/diff | Base Median (ns) | Diff Median (ns) | Modality|
| --------------------------------------------------------------------- | ---------:| ----------------:| ----------------:| -------- |
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: Greek)              |      3.50 |        295751.19 |         84604.67 |         |
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: Cyrillic)           |      2.76 |        225379.32 |         81806.99 | several?|
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: Chinese)            |      2.52 |        209528.09 |         83099.90 |         |
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: EnglishAllAscii)    |      1.56 |         64962.94 |         41614.67 |         |
| System.Text.Perf_Utf8Encoding.GetByteCount(Input: EnglishMostlyAscii) |      1.16 |        228289.04 |        196492.73 |         |

[pgovind]

LGTM! Nice numbers!

[adamsitnik]

LGTM, thank you @carlossanlop !

[pgovind]

I'm just wondering if this will break the mono CI too.

[pgovind]

Maybe not. I just saw that @adamsitnik fixed the mono issue with a NoMono category. Nice!

[adamsitnik]

I'm just wondering if this will break the mono CI too.

This one not because it's not using the forbidden constructor :D