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 44411 - Perf regressions on many benchmarks with AMD Threadripper
Summary: Perf regressions on many benchmarks with AMD Threadripper
Status: RESOLVED FIXED
Alias: None
Product: libraries
Classification: Unclassified
Component: Backend: X86 (show other bugs)
Version: trunk
Hardware: PC Linux
: P enhancement
Assignee: Unassigned LLVM Bugs
URL:
Keywords:
Depends on: 44409 44410 44539
Blocks: release-10.0.0
  Show dependency tree
 
Reported: 2019-12-30 09:30 PST by David Bolvansky
Modified: 2020-02-25 07:55 PST (History)
11 users (show)

See Also:
Fixed By Commit(s):

Attachments
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 David Bolvansky 2019-12-30 09:30:36 PST 
Phoronix ran some benchmarks with Clang 9/10 and GCC 9/10 on AMD. The article shows significant performance regressions in multiple benchmarks.

https://www.phoronix.com/scan.php?page=article&item=gcc-clang-3960x&num=2
Comment 1 Simon Pilgrim 2019-12-30 13:58:36 PST 
This is unlikely to be AMD specific, at the very least it'll probably affect all Haswell-class (AVX2) CPUs - the article says as much:

"unfortunately the release build of LLVM Clang Git was pointing to performance regressions compared to LLVM Clang 9.0 stable. We've also seen Clang 10 running slower than Clang 9 on other systems and will have those additional results out in the coming days."

Is anyone setup with a suitable machine to run a git bisect script to track the regression(s) down?
Comment 2 Hans Wennborg 2020-02-25 07:43:56 PST 
Sounds like the sub-bugs have been fixed or marked as wontfix. I'm guessing we're not looking further into this?
Comment 3 David Bolvansky 2020-02-25 07:55:18 PST 
Right, it should be fine now