Don't use lvDoNotEnregister in optCopyProp_LclVarScore.

[sandreenko]

The value of lvDoNotEnregister is not finalized when we call optCopyProp_LclVarScore during global morph, so we should not rely on it.

It causes many asm diffs when we change what we mark as lvDoNotEnregister so I want to avoid this noise by merging this first.

Note: there is another call to optCopyProp_LclVarScore during optCopyProp where such check could make sense but it is guarded with

runtime/src/coreclr/jit/copyprop.cpp

Lines 199 to 202 in bfb8d6b

	if (lvaTable[lclNum].lvDoNotEnregister != lvaTable[newLclNum].lvDoNotEnregister)
	{
	continue;
	}

so also not needed.

The diffs are small and are mostly noise.

SuperPMI diffs

benchmarks.run.windows.x64.checked.mch
Total bytes of delta: 32
1 total files with Code Size differences (0 improved, 1 regressed)

libraries.pmi.windows.x64.checked.mch
Total bytes of delta: 3
27 total files with Code Size differences (16 improved, 11 regressed)

libraries_tests.pmi.windows.x64.checked.mch
Total bytes of delta: -26
2 total files with Code Size differences (2 improved, 0 regressed)

coreclr_tests.pmi.Linux.arm64.checked.mch
Total bytes of delta: -8
1 total files with Code Size differences (1 improved, 0 regressed)

libraries.pmi.Linux.arm64.checked.mch
Total bytes of delta: -20
27 total files with Code Size differences (16 improved, 11 regressed)

libraries_tests.pmi.Linux.arm64.checked.mch
Total bytes of delta: -56
9 total files with Code Size differences (7 improved, 2 regressed)

coreclr_tests.pmi.Linux.x64.checked.mch
Total bytes of delta: -3
2 total files with Code Size differences (1 improved, 1 regressed)

libraries.pmi.Linux.x64.checked.mch
Total bytes of delta: -21
29 total files with Code Size differences (18 improved, 11 regressed)

libraries_tests.pmi.Linux.x64.checked.mch
Total bytes of delta: -107
11 total files with Code Size differences (9 improved, 2 regressed)

coreclr_tests.pmi.windows.arm64.checked.mch
Total bytes of delta: -8
1 total files with Code Size differences (1 improved, 0 regressed)

libraries.pmi.windows.arm64.checked.mch
Total bytes of delta: -20
27 total files with Code Size differences (16 improved, 11 regressed)

libraries_tests.pmi.windows.arm64.checked.mch
Total bytes of delta: -44
9 total files with Code Size differences (7 improved, 2 regressed)

benchmarks.run.windows.x86.checked.mch
Total bytes of delta: -734
17 total files with Code Size differences (16 improved, 1 regressed)

coreclr_tests.pmi.windows.x86.checked.mch
Total bytes of delta: -111
35 total files with Code Size differences (11 improved, 24 regressed)

libraries.pmi.windows.x86.checked.mch
Total bytes of delta: 86
5 total files with Code Size differences (0 improved, 5 regressed)

libraries_tests.pmi.windows.x86.checked.mch
Total bytes of delta: -1748
80 total files with Code Size differences (78 improved, 2 regressed)

libraries.pmi.Linux.arm.checked.mch
Total bytes of delta: -10
32 total files with Code Size differences (17 improved, 15 regressed)

libraries_tests.pmi.Linux.arm.checked.mch
Total bytes of delta: -56
7 total files with Code Size differences (6 improved, 1 regressed)

[sandreenko]

PTAL @BruceForstall @dotnet/jit-contrib

[BruceForstall]

LGTM.

Should lvDoNotEnregister be a PhasedVar<bool> (at least in DEBUG)?

[sandreenko]

Should lvDoNotEnregister be a PhasedVar (at least in DEBUG)?

I don't think it is possible now, we are sure that it is set only after lowering, the check was added in #52802

but we read it early and often to guess about perf transformations or avoid checks that are not necessary. I don't think we could finalize its value earlier without pulling target dependencies from lowering to morph so I don't see a good way to make it a phased variable, but I will think about it for 7.0.

[kunalspathak]

https://pvscmdupload.blob.core.windows.net/reports/allTestHistory/refs/heads/main_arm64_Windows%2010.0.19041/System.Collections.CtorFromCollectionNonGeneric(Int32).ArrayList(Size%3a%20512).html

There seems to be some regression on Arm64 and this PR is one of the commit in the diff. Can you double check?