[Arm64] Implement more hardware intrinsics from #24794

[echesakov]

1. Implements the following intrinsics as proposed in API Proposal: More SIMD HW Intrinsics #24794:

• AbsoluteDifferenceAdd
• MultiplyExtended, MultiplyExtendedScalar
• PolynomialMultiply
• ReciprocalEstimate, ReciprocalEstimateScalar
• ReciprocalExponentScalar
• ReciprocalSquareRootEstimate, ReciprocalSquareRootEstimateScalar
• ReciprocalSquareRootStep, ReciprocalSquareRootStepScalar
• ReciprocalStep, ReciprocalStepScalar

2. Implements frecpe, frecps, frecpx, frsqrte, frsqrts, urecpe, ursqrte instructions

3. Fixes implementation for fcmeq, fcmge, fcmgt, fcmle, fcmlt (zero immediate) instructions

4. Adds emitter unit tests for the new added instructions and the fixed ones.
   Attached are the outputs for jitDisasm and WinDbg u command

jitDisasm.txt
windbg-u.txt

5. Adds missing flag BaseTypeFromFirstArg for AbsoluteDifference intrinsics

[Dotnet-GitSync-Bot]

Note regarding the new-api-needs-documentation label:

This serves as a reminder for when your PR is modifying a ref *.cs file and adding/modifying public APIs, to please make sure the API implementation in the src *.cs file is documented with triple slash comments, so the PR reviewers can sign off that change.

[echesakov]

@CarolEidt @TamarChristinaArm @tannergooding PTAL

[CarolEidt]

LGTM - thanks!

[echesakov]

LGTM - thanks!

Thank you for review, Carol!

@CarolEidt @tannergooding One additional question:

As you probably noticed one of my last commits was 9e7b2f1 - updating lsraarm64.cpp to reflect that the instructions uaba and saba are RMW.

While that change is sound, ideally, I want this to be specified via the table in hwintrinsiclistarm64.h.
However, since most of the arm64 intrinsics are not RMW this would require marking most of them with HW_Flag_NoRMWSemantics.

I would like to avoid this and propose to invert this flag on arm64, i.e. have no-RMW semantic for intrinsics as default and mark a smaller subset of them with a new flag HW_Flag_HasRMWSemantics.

In other words, I would like to

1. remove definition of HW_Flag_NoRMWSemantics flag on arm64
2. add HW_Flag_HasRMWSemantics
3. make the logic as in 9e7b2f1 to be table-driven.
   I will still need to keep some logic for tgtPrefUse though.

@CarolEidt @tannergooding Will you have any concerns with such change?

[tannergooding]

I'm fine with the underlying flags differing, we have the helper functions that people should be calling to get that info anyways and so we shouldn't really have anything but the table (and those functions) using the flags directly.

[CarolEidt]

I agree with Tanner - it makes sense to have the opposite default for Arm64