Update comment and I've pre-commit a mir test https://github.com/llvm/llvm-project/commit/e3e25cfb44bc2a35e3b53d62d37c27b7d13157b6 to demonstrate the situation when `equality is true. The codegen of the test case doesn't change with this patch.

@nemanjai 's solution can make code more compact. Thanks.

Functionality of this patch can be covered by https://reviews.llvm.org/D93191.

Per @RKSimon 's suggestion.

Gonna to add more patterns and refine current code.

Rebased.

Reformat code.

Refined comments.

Gentle ping.

Removed parent revision, since this patch doesn't rely on opt's abs canonicalization.

Rebased on Simon's commit.

Removed the hook and added test cases for other arch.

Pre-committed test case and rebased.

In D91120#2391105, @spatel wrote:

Note that we are planning to canonicalize IR to the abs() intrinsic - D87188.
Assuming that comes through relatively soon (cc @nikic), should we hold off adding what would probably become dead code in the backend?

Ping.

In D89084#2352644, @shchenz wrote:

I guess this opt should also be profitable for opcode SELECT, is there any reason we don't do this?

I think what @efriedma suggests is viable and eases the confusion mentioned by @qiucf .

Add tests to cover zext/trunc.

Tune codegen.

Confirmed with https://alive2.llvm.org/ce/z/8pnUEb

Fix format and add more tests.

LGTM. This solution probes the gap between [stackptr & mask, stackptr] when align is big.

Updated comments again.

Adjust comments, leave function name unchanged.

Per @spatel 's comments.

Updated Summary.