Fix the unwind opcode for new APX registers

[kunalspathak]

APX added 8 new GPR registers and that shifted the base of floating point registers. However the shift was not getting accounted for properly while embedding the unwind code.

Also, while I was there, I realized that we didn't add the mapping of new registers for DWARF CFI unwind.

Fixes:

• Test failure: System.Data.Tests.SqlTypes.SqlMoneyTest.Conversions #112325
• Test failure: System.Data.Tests.SqlTypes.SqlDoubleTest.ArithmeticMethods #112324
• Test failure: JIT/IL_Conformance/Convert/TestConvertFromIntegral/TestConvertFromIntegral.dll #112329

[dotnet-policy-service]

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

[amanasifkhalid]

This might also fix #112286

[saucecontrol]

And #112755 😄

[jakobbotsch]

Any idea why this didn't show up as 0-size diffs in the original PR? We should be diffing unwind codes in SPMI.

Edit: It looks like we don't:

runtime/src/coreclr/tools/superpmi/superpmi/neardiffer.cpp

Lines 1349 to 1381 in 57ed254

	cr2->applyRelocs(&rc, hotCodeBlock_2, hotCodeSize_2, orig_hotCodeBlock_2);
	cr2->applyRelocs(&rc, coldCodeBlock_2, coldCodeSize_2, orig_coldCodeBlock_2);
	cr2->applyRelocs(&rc, roDataBlock_2, roDataSize_2, orig_roDataBlock_2);
	if (!compareCodeSection(mc, cr1, cr2, hotCodeBlock_1, hotCodeSize_1, roDataBlock_1, roDataSize_1,
	orig_hotCodeBlock_1, orig_roDataBlock_1, orig_coldCodeBlock_1, coldCodeSize_1,
	hotCodeBlock_2, hotCodeSize_2, roDataBlock_2, roDataSize_2, orig_hotCodeBlock_2,
	orig_roDataBlock_2, orig_coldCodeBlock_2, coldCodeSize_2))
	return false;
	if (!compareCodeSection(mc, cr1, cr2, coldCodeBlock_1, coldCodeSize_1, roDataBlock_1, roDataSize_1,
	orig_coldCodeBlock_1, orig_roDataBlock_1, orig_hotCodeBlock_1, hotCodeSize_1,
	coldCodeBlock_2, coldCodeSize_2, roDataBlock_2, roDataSize_2, orig_coldCodeBlock_2,
	orig_roDataBlock_2, orig_hotCodeBlock_2, hotCodeSize_2))
	return false;
	if (!compareReadOnlyDataBlock(mc, cr1, cr2, roDataBlock_1, roDataSize_1, orig_roDataBlock_1, roDataBlock_2,
	roDataSize_2, orig_roDataBlock_2))
	return false;
	if (!compareEHInfo(mc, cr1, cr2))
	return false;
	if (!compareGCInfo(mc, cr1, cr2))
	return false;
	if (!compareVars(mc, cr1, cr2))
	return false;
	if (!compareBoundaries(mc, cr1, cr2))
	return false;
	return true;

. I opened #112807 for it.

[kunalspathak]

/azp run runtime-coreclr libraries-jitstress2-jitstressregs, runtime-coreclr r2r-extra

[azure-pipelines]

Azure Pipelines successfully started running 2 pipeline(s).

[kunalspathak]

It seems there are few failures that needs investigation but are unrelated to the change from this PR. I will investigate to make sure and open GH issues, if needed. But this should be ready for review.

r2rextra failures: https://dev.azure.com/dnceng-public/public/_build/results?buildId=959524&view=ms.vss-test-web.build-test-results-tab&runId=25542994&resultId=123196&paneView=debug

[kunalspathak]

@dotnet/jit-contrib

[tannergooding]

Can we add an assert that genIsValidIntReg(reg) for the else case, to help catch any future bugs or expansions in this space?

[BruceForstall]

We could even assert on the actual range of the computed value, since the unwind codes are well-defined and if they change, we want the assert to hit.

[kunalspathak]

let me do it as a follow up PR and merge this as is to unblock weekend runs.

[kunalspathak]

/ba-g osx/arm64 seems to be timeout errors.