Sorry I forgot to follow this up sooner. Looking at it again I think you are right so will drop this patch.

LGTM

LGTM, just some comments/notes on what changes for others looking at this.

I'm going to submit this, since it has been approved for a while and covers a decent subset of the features without invasive changes. We can keep discussing any future alterations in D121768.

Your method can not get rid of non-trivial feature like useAA, which has an virtual interface.

Then I think it's important to have the flexibility to customize the interface, at least for now.

This patch makes a distinction between trivial/nontrivial members & interface b/c some targets has some tricky interfaces, e.g

bool hasCMov() const { return HasCMov || X86SSELevel >= SSE1 || is64Bit(); }
bool useAA() const override { return UseAA; }
bool hasLAHFSAHF() const { return HasLAHFSAHF64 || !is64Bit(); }

It seems that you patch handle them incorrectly by now.

Address review comments

This looks remarkably similar to an ARM/AArch64 patch I've had up for a couple of weeks: D120906

LGTM

Ping. Anyone have thoughts on this approach?

LGTM

Thanks for picking this up, LGTM

LGTM, please give @nickdesaulniers some time to respond though. I do agree that iterating over the features repeatedly is less than ideal, but also that this patch is probably not the place to try to fix it.

LGTM

This renames the following instructions that were captured by the scheduler regex: CPYi8, CPYi16, CPYi32, CPYi64.

Should they be setting or using NZCV though?

Use Sched<[]> rather than hasNoSchedulingInfo. This requried updating the regexes used by various scheduling models to apply InstRW to CPY* instructions so that they do not capture the new MOPS instructions.

SET->CPY

Combine V8 and V9 macros

Add 8.8 and 9.3; patches adding these were merged first in the end.

Add mayLoad/mayStore/hasSideEffects

Rebase

Minor fix after rebase

Rebase

Thanks for the review @nickdesaulniers. Adding a few more reviewers just to check that I'm not missing something.

I've updated the description with documentation.

Remove redundancies in arm tests

Remove crypto extensions and add SVE2 for AArch64, as per D109517

Thanks for the comments

Use pipe in second test

Tweak description and test case.

LGTM

Makes sense to me

According to D95749, the rationale for allowing ELFOSABI_NONE with SHF_GNU_RETAIN is to keep consistent behaviour with STT_GNU_IFUNC and STB_GNU_UNIQUE, which are permitted under ELFOSABI_NONE. @MaskRay Should we perhaps revisit that decision?