Thanks @Yi-Hong.Lyu @nemanjai. The test was in https://reviews.llvm.org/D72574 and moved to this patch.

Addressed review comments.

Addressed review comments above.

Addressed comments above.

Thanks @dsanders. Comment addressed.

Commandeer the patch as Stefan is on vacation.

Addressed Nemanja and Daniel's comment to give Inst type uint64_t

Commandeer this patch as Stefan is on vacation.

Addressed the comments by

• Add "DisableNonVolatileCR" into "PPCFunctionInfo" as a new property, add the functions to set and check it.
• Set the new member in the constructor based on the value of the option.
• Clean up redundant "#2" in the test case.

Though the logic in this new function looks consistent with original code with optimized condition check, it is still not that readable to me. Agree with @nemanjai 's suggestion, adding conceptual functions for"SingleRLWINM", "SingleRLWIMI", "SingleRLDICL" and "PairOfRLDICL" will make it more readable. Also added comments for variable re-def issues.

@Yi-Hong.Lyu For PowerPC, CR2, CR3, CR4 are non-volatile CR fields according to ELFv2ABI, which will not hold for other ABIs.

Thanks @xbolva00 for your comment.

Since the patch is approved, I will hold it for another 3 days if any comment and commit it on Monday.

ping

Addressed review comments from @nemanjai

Overall looks good to me. Some minor changes needed.

• Merged to latest code base and compiled with -O3 and PGO
• Collected stats for SPECInt and SPECFP benchmarks (SPEC2017) with baseline and patch.

• With the feature enabled, found performance gain for SPEC benchmarks, e.g. 2.1% for perlbench_r and 1.6% for povray_r.

Collected stats below for SPECInt and SPECFP benchmarks (SPEC2017) with this feature enabled, compiled with -O3 and PGO.

Thanks @nemanjai 's review. Addressed all the comments and updated the test cases accordingly.

Addressed @nemanjai 's review comment.

Thanks @lei for the review. Updated the patch to r

1. Resolve the indentation issues
2. Modify the variable name to be consistent with current naming convention.

@jsji Yes, please ignore diff 216684. Thanks you for the remind.

Thanks @efriedma for the review and remind.

ping

ping

After re-basing the patch with latest master ToT, there is a LIT failure for the KnowCRBitSpill.ll. As per discussed and reviewed in the meeting, the CRSET test case behavior change is as expected. The test case check logic is updated accordingly.

Thanks all for the review!

Ping..