ARM64-SVE: Use INC* when incrementing with GetActiveElementCount()/CountXBitElements()

[a74nh]

incp

index += GetActiveElementCount(Sve.CreateTrueMaskUInt32(), loopmask);

Generates:

ptrue p0
cntp    x7, p0, p2.s
add     x6, x7, x6

Instead it should do:

incp    x6, p2.s

decp

Also

index -= GetActiveElementCount(Sve.CreateTrueMaskUInt32(), loopmask);

Should produce:

decp    x6, p2.s

sqincp

Note that the user can also do:

index = Sve.SaturatingIncrementByActiveElementCount(index, loopmask);

This correctly produces:

sqincp   x6,  p2.s

This is currently more optimal than using GetActiveElementCount(). However, the user may not want it to saturate.
(There is nothing to fix here. It is just for reference).

incb/inch/incw/decb/dech/decw

index += Sve.Count32BitElements(SveMaskPattern.ALL);

Produces:

cntw    x7, ALL
add     x6, x7, x6

Instead it should produce:

incw  x6, ALL

The same needs to happen for Count8BitElements(), Count16BitElements(), Count64BitElements(). Also should use the DECx instructions when decrementing.

[dotnet-policy-service]

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