Enable StructEnreg by default on all platforms.

[sandreenko]

Diffs are positive and looks similar to x64 diffs merged in #55045 and x86 merged in #55535.

coreclr_tests.pmi.Linux.arm64.checked.mch
Total bytes of delta: -3112
153 total files with Code Size differences (151 improved, 2 regressed), 7 unchanged.

libraries.crossgen2.Linux.arm64.checked.mch
Total bytes of delta: -4460
561 total files with Code Size differences (462 improved, 99 regressed), 49 unchanged.

libraries.pmi.Linux.arm64.checked.mch
Total bytes of delta: -5300
111 total files with Code Size differences (95 improved, 16 regressed), 11 unchanged.

libraries_tests.pmi.Linux.arm64.checked.mch
Total bytes of delta: -324
85 total files with Code Size differences (58 improved, 27 regressed), 17 unchanged.

benchmarks.run.Linux.x64.checked.mch
Total bytes of delta: -390
5 total files with Code Size differences (5 improved, 0 regressed), 0 unchanged.

coreclr_tests.pmi.Linux.x64.checked.mch
Total bytes of delta: -4996
157 total files with Code Size differences (156 improved, 1 regressed), 4 unchanged.

libraries.crossgen2.Linux.x64.checked.mch
Total bytes of delta: -3153
271 total files with Code Size differences (224 improved, 47 regressed), 18 unchanged.

libraries.pmi.Linux.x64.checked.mch
Total bytes of delta: -970
80 total files with Code Size differences (68 improved, 12 regressed), 0 unchanged.

Total bytes of delta: -997
libraries_tests.pmi.Linux.x64.checked.mch
85 total files with Code Size differences (69 improved, 16 regressed), 4 unchanged.

benchmarks.run.windows.arm64.checked.mch
Total bytes of delta: -160
5 total files with Code Size differences (5 improved, 0 regressed), 0 unchanged.

coreclr_tests.pmi.windows.arm64.checked.mch
Total bytes of delta: -3120
153 total files with Code Size differences (151 improved, 2 regressed), 7 unchanged.

libraries.crossgen2.windows.arm64.checked.mch
Total bytes of delta: -4536
572 total files with Code Size differences (472 improved, 100 regressed), 48 unchanged.

libraries.pmi.windows.arm64.checked.mch
Total bytes of delta: -5376 (-5.36% of base)
118 total files with Code Size differences (102 improved, 16 regressed), 11 unchanged.

libraries_tests.pmi.windows.arm64.checked.mch
Total bytes of delta: -308
85 total files with Code Size differences (56 improved, 29 regressed), 18 unchanged.

coreclr_tests.pmi.Linux.arm.checked.mch
Total bytes of delta: -1660
49 total files with Code Size differences (48 improved, 1 regressed), 0 unchanged.

libraries.crossgen2.Linux.arm.checked.mch
Total bytes of delta: -1280
208 total files with Code Size differences (185 improved, 23 regressed), 13 unchanged.

libraries.pmi.Linux.arm.checked.mch
Total bytes of delta: -34
3 total methods with Code Size differences (2 improved, 1 regressed), 0 unchanged.

libraries_tests.pmi.Linux.arm.checked.mch
Total bytes of delta: -118
7 total files with Code Size differences (7 improved, 0 regressed), 4 unchanged.

The regressions are the same:

1. we use different frame size and it could require bigger init code;
2. we use more registers and have to store/restore non-volatile.

The fixes that are included in this PR:

1. genBitCast on arm64/arm32 was not working correctly with regOptional lclVars, did not repro without additional stress pressure, the bug was that if a local was not assigned a register and we had a bitcast on top of it we were marking it as contained and were trying to load directly but for locals the direct load happens in genConsumeRegs that has already happened there and it was skipped because the local was contained;
2. arm32 prespill all struct args, so they are excluded from register allocation;
3. SIMDIntrinsicUpperSave/Restore for SIMD regs when calling conv does not keep them, so if we enregister struct<16> as v8 we have to do bookkeeping for the upper part around each call etc.

the problem with these fixes was that they all were silent bad codegen, I have covered the 3 with new asserts but others are too low in the pipeline to be checked.

Contributes to #43867 and probably closes the scope for 6.0.

[sandreenko]

/azp run runtime-coreclr jitstress, runtime-coreclr outerloop, runtime-coreclr jitstressregs

[azure-pipelines]

Azure Pipelines successfully started running 3 pipeline(s).

[sandreenko]

/azp run runtime-coreclr jitstress, runtime-coreclr outerloop, runtime-coreclr jitstressregs

[azure-pipelines]

Azure Pipelines successfully started running 3 pipeline(s).

[sandreenko]

/azp run runtime-coreclr jitstress2-jitstressregs

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[sandreenko]

/azp run runtime-coreclr jitstress2-jitstressregs, runtime-coreclr jitstress, runtime-coreclr outerloop, runtime-coreclr jitstressregs

[azure-pipelines]

Azure Pipelines successfully started running 4 pipeline(s).

[sandreenko]

PTAL @dotnet/jit-contrib

[kunalspathak]

Does this fix ARM64 issue #35071?

[BruceForstall]

This is the same condition as before; this block appears to be dead code.

[sandreenko]

looks like a merge issue and that is why I am seeing some failures in CI, will fix it soon.

[sandreenko]

fixed, thank you.

[sandreenko]

/azp run runtime-coreclr jitstress2-jitstressregs, runtime-coreclr jitstress, runtime-coreclr outerloop, runtime-coreclr jitstressregs

[azure-pipelines]

Azure Pipelines successfully started running 4 pipeline(s).

[sandreenko]

The UI here does not show the test result (yeah, ci) but:

1. runtime is green;
2. runtime-jitstress is green
3. runtime jitstressregs only one unrelated failure, covered by Exception in test infrastructure - System.InvalidOperationException: Collection was modified after the enumerator was instantiated #11063 (comment)
4. runtime jitstress2-jitstressregs shows one unrelated failure covered by Feed unreliability affecting CI #55449 (comment);

so the tests are passing, I will keep an eye on library stress tests but we have time to fix bugs if anything pops up! Thank Bruce again for the quick review.