This is an archive of the discontinued LLVM Phabricator instance.

X86Vector: remove rsqrt Integration test
AbandonedPublic

Authored by cota on Oct 8 2021, 12:00 PM.

Details

Summary

This integration test (i.e. running the generated bitcode) is causing
more trouble than it's worth. We just want to ensure that the
right intrinsic (@llvm.x86.avx.rsqrt.ps.256) is called; what happens
afterwards (i.e. what LLVM generates and on what hardware it runs)
does not concern us and we therefore should not test it.

Given that an existing test in mlir/test/Target/LLVMIR/x86vector.mlir
already checks that we generate the right intrinsic, delete this
integration test.

Diff Detail

Event Timeline

cota created this revision.Oct 8 2021, 12:00 PM
cota requested review of this revision.Oct 8 2021, 12:00 PM
aartbik accepted this revision.Oct 8 2021, 12:40 PM

Bye old friend!

This revision is now accepted and ready to land.Oct 8 2021, 12:40 PM
cota added a comment.Oct 8 2021, 1:24 PM

After discussing with Mehdi and Aart, I'm withdrawing this change in favor of https://reviews.llvm.org/D111461

cota abandoned this revision.Oct 8 2021, 1:24 PM