SVE: Add Z register aliases

[kunalspathak]

Contributes to #94549

[kunalspathak]

@a74nh @dotnet/jit-contrib

[ghost]

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

Issue Details

---

Contributes to #94549

Author:	kunalspathak
Assignees:	kunalspathak
Labels:	area-CodeGen-coreclr, arch-arm64-sve
Milestone:	-

[a74nh]

LGTM.

[BruceForstall]

Is this going to be more useful than confusing? I guess the Arm manual itself distinguishes V and Z registers, which seems odd, since V registers are also named Q, D, S, H, and B depending on size, and Z registers contain V registers.

It seems like this won't be helpful because you'll be able to use V instead of Z and vice-versa in the code, and there will be no checking.

[kunalspathak]

It seems like this won't be helpful because you'll be able to use V instead of Z and vice-versa in the code, and there will be no checking.

Sure, V and Z are same registers, but I think while reading code, it will be helpful to know if the instruction operates on scalable "Z" register or a fixed-width "V" register specially around instructions that operates on both Z and V registers.

-theEmitter->emitIns_R_R_R(INS_sve_andqv, EA_8BYTE, REG_V0, REG_V0, REG_P0, INS_OPTS_SCALABLE_B);  // IF_SVE_AG_3A /* ANDQV   <Vd>.<T>, <Pg>, <Zn>.<Tb> */
+theEmitter->emitIns_R_R_R(INS_sve_andqv, EA_8BYTE, REG_V0, REG_Z0, REG_P0, INS_OPTS_SCALABLE_B);  // IF_SVE_AG_3A /* ANDQV   <Vd>.<T>, <Pg>, <Zn>.<Tb> */

To that matter, I don't think we use REG_V* explicitely anyway apart from unit test and few other places like unwind, where we can continue to use REG_V*.

[BruceForstall]

I think while reading code, it will be helpful to know if the instruction operates on scalable "Z" register or a fixed-width "V" register specially around instructions that operates on both Z and V registers.

The problem is, if there is no correctness checking, then it's not going to be too useful, IMO.

[kunalspathak]

To that matter, I don't think we use REG_V* explicitely anyway apart from unit test and few other places like unwind, where we can continue to use REG_V*.

Taking account of that and @BruceForstall POV, I will close this PR then.