LLVM Bugzilla is read-only and represents the historical archive of all LLVM issues filled before November 26, 2021. Use github to submit LLVM bugs

Bug 34248 - r311077 - using VPlan in loop vectorizer - generates bad IR for PPC (and probably others)
Summary: r311077 - using VPlan in loop vectorizer - generates bad IR for PPC (and prob...
Status: RESOLVED FIXED
Alias: None
Product: new-bugs
Classification: Unclassified
Component: new bugs (show other bugs)
Version: trunk
Hardware: PC All
: P enhancement
Assignee: ayal.zaks
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2017-08-20 16:14 PDT by Chandler Carruth
Modified: 2017-09-23 10:41 PDT (History)
2 users (show)

See Also:
Fixed By Commit(s): 311849

Attachments
reproducer (8.23 KB, text/plain)
2017-08-20 16:14 PDT, Chandler Carruth 		Details
Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Chandler Carruth 2017-08-20 16:14:56 PDT 
Created attachment 19013 [details]
reproducer

When testing LLVM with the VPlan based vectorization enabled we found lots of issues on non-x86 architectures.

I've attached a reduced IR test case from libpng on PPC64. It assert fails with:

% opt -disable-output -aa-pipeline=default -passes='default<O3>' png.simplified.ll

I'm going to revert the patch while you investigate.
Comment 1 ayal.zaks 2017-08-23 16:04:40 PDT 
Thanks. We've uploaded to D32871 a revised version which fixes the reproducer. Could you please test it pre-commit and let us know if any issues remain, or would you prefer we commit?