FYI there are some benchmarks in https://reviews.llvm.org/D64082 that are not submitted but could be used as basis for making decisions for this patch.

Clarify top comment.

Add comment for buffer size.

Name magical constant.

Be evel less permissive as to what we allow the compiler to see.

In D64082#1566955, @spatel wrote:

Thanks for working on this! I'm not familiar with how the benchmarking framework works, so someone else should definitely have a look.

Does the framework automatically account for and filter out noisy results? I'm guessing that tiny memcmp() will have a lot of run-to-run variation.

Address review comments.

Thanks Roman.

test-suite benchmarks: https://reviews.llvm.org/D64082

For information, I had to roll this back at it breaks sanitizers. Sanitizer passes insert some nobuiltin attributes on all memcmp calls to prevent expansion and still be able to intercept them (maybeMarkSanitizerLibraryCallNoBuiltin), and these passes are added last in opt:

clang-format

I encourage adding small memcmp tests to test-suite as a follow-up, so we know that things won't break going forward.

Rebase on r364384.

Thanks.

Remove asm testing.

Split off options refactoring to r364281, rebase.

Ping. Any other concerns ?

Amazing work Roman ! A few nits on the form, I'll let others more knowledgeable about AMD architectures comment on the model itself.

Thanks for the review.

Fix formatting

• Mark ldmxscr as writing to memory. Unfortunately the DAG relies on intrinsics writing to memory to determine side-effet-ness.
• Mark stmxscr as having side effects, as discusse din the review.
• Update tests.

Ping

In D62896#1531118, @craig.topper wrote:

A side effecting instruction doesn't prevent other non-side effecting instructions from moving across it.

In D62896#1530895, @craig.topper wrote:

Don’t we need stmxcsr to have side effects to keep it from reordering relative to ldmxcsr?

There are a bunch of tests to update, but I'd like to get opinions on this before going further. I'm not sure why the properties were not already there. Did I miss something ?

Thanks.

Make ExpandMemCmp preserve the DomTree.

Thanks.

Thanks.

Rebase

add missing inline

In D60318#1509938, @RKSimon wrote:

@courbet Please can you ensure you have EXPENSIVE_CHECKS enabled in all your builds before going any further - you've been breaking the buildbots for well over a week now and you still haven't fixed the underlying issue.

Thanks.

Address comments.

remove debug

Thanks.

Do we have a test for this? I don't think it is a problem but if we do we could update it. Otherwise, LGTM.