Optimise AdvSimd GetIndexOfFirstNonAsciiByte

[ylpoonlg]

The main loop body was calling GetIndexOfFirstNonAsciiByteInLane_AdvSimd to calculate the index of the non-ASCII character within the vector on every iteration, but the index is only needed when we have found a non-ASCII character. Instead, we can use VectorContainsNonAsciiChar for the checks in the loop, which is much cheaper.

Performance results

A simple benchmark is created to directly measure GetIndexOfFirstNonAsciiByte, with different buffer sizes and positions of the first non-ASCII character.

Run on Neoverse-V2 (lower is better)

Method	Size	NonAscii	Version	Mean	Error	Ratio
GetNonAscii	10	Middle	Before	1.970 us	0.5263 us	1.000
GetNonAscii	10	Middle	After	1.182 us	0.0783 us	0.600
GetNonAscii	10	None	Before	1.903 us	0.4852 us	1.000
GetNonAscii	10	None	After	1.216 us	0.0420 us	0.639
GetNonAscii	10	Start	Before	1.894 us	0.4932 us	1.000
GetNonAscii	10	Start	After	1.190 us	0.0794 us	0.628
GetNonAscii	100	Middle	Before	5.881 us	0.1196 us	1.000
GetNonAscii	100	Middle	After	3.226 us	0.0100 us	0.549
GetNonAscii	100	None	Before	7.085 us	0.2668 us	1.000
GetNonAscii	100	None	After	3.062 us	0.0550 us	0.432
GetNonAscii	100	Start	Before	1.863 us	0.2965 us	1.000
GetNonAscii	100	Start	After	1.535 us	0.0214 us	0.824
GetNonAscii	10000	Middle	Before	250.345 us	2.6488 us	1.000
GetNonAscii	10000	Middle	After	96.068 us	1.8411 us	0.384
GetNonAscii	10000	None	Before	463.014 us	6.9630 us	1.000
GetNonAscii	10000	None	After	199.708 us	0.8972 us	0.431
GetNonAscii	10000	Start	Before	1.990 us	0.2821 us	1.000
GetNonAscii	10000	Start	After	1.561 us	0.0709 us	0.784

cc @dotnet/arm64-contrib @SwapnilGaikwad

[SwapnilGaikwad]

cc: @EgorBo @a74nh @JulieLeeMSFT