This is an archive of the discontinued LLVM Phabricator instance.

[PowerPC] Improve error message on MASSV pass
ClosedPublic

Authored by masoud.ataei on Jul 15 2021, 12:51 PM.

Details

Summary

Problem:
If user add -mno-altivec along with -mllvm -vector-library=MASSV regardless of compiling P8 or P9 (or other targets), they will get confusing message.

Solution:
Updated the error message to mention vectorization is needed be enable when using MASSV pass.

Diff Detail

Event Timeline

masoud.ataei created this revision.Jul 15 2021, 12:51 PM
masoud.ataei requested review of this revision.Jul 15 2021, 12:51 PM
Whitney accepted this revision.Jul 15 2021, 1:04 PM
This revision is now accepted and ready to land.Jul 15 2021, 1:04 PM
bmahjour added inline comments.Jul 15 2021, 1:24 PM
llvm/lib/Target/PowerPC/PPCLowerMASSVEntries.cpp
86

what about Power 10? should it say "Power8 and above" or "Power8 or later"?

masoud.ataei added inline comments.Jul 15 2021, 1:55 PM
llvm/lib/Target/PowerPC/PPCLowerMASSVEntries.cpp
86

Currently there is no Power10 specific entry on Linux MASS library (there are on AIX). So, if you use -mcpu=pwr10 you are actually getting P9 specific functions (like __xl_expd2_P9).

Do you still think the message should say "Power8 and above"?

bmahjour added inline comments.Jul 15 2021, 2:11 PM
llvm/lib/Target/PowerPC/PPCLowerMASSVEntries.cpp
86

Sure, but MASSV is still supported on P10. The fact that we resolve to p9 entries in some cases is an implementation detail that is not relevant to the purpose of this message. I'd prefer "Power8 and later".

xgupta closed this revision.Jul 22 2021, 4:08 AM
xgupta added a subscriber: xgupta.

Seems patch is committed in https://reviews.llvm.org/rGee2068b30ecf297c004c06eedd7e1063c67a279c, revision should be close now?

Seems patch is committed in https://reviews.llvm.org/rGee2068b30ecf297c004c06eedd7e1063c67a279c, revision should be close now?

Yes, thanks for closing it.

Seems patch is committed in https://reviews.llvm.org/rGee2068b30ecf297c004c06eedd7e1063c67a279c, revision should be close now?

Yes, thanks for closing it.

The tools that automatically close reviews are very particular about the format of the commit message. In this case, there is an extra space at the beginning of the line that's meant to start with Differential Revision:, that's probably why auto-close didn't work.