Opened 3 years ago
Last modified 3 years ago
#23327 new defect
Small regression with MPIR-3.0.0
| Reported by: | jdemeyer | Owned by: | |
|---|---|---|---|
| Priority: | major | Milestone: | sage-8.0 |
| Component: | packages: standard | Keywords: | |
| Cc: | vdelecroix, jpflori, fbissey, dimpase | Merged in: | |
| Authors: | Reviewers: | ||
| Report Upstream: | N/A | Work issues: | |
| Branch: | Commit: | ||
| Dependencies: | Stopgaps: |
Description (last modified by )
#22493 has caused a significant regression for padic_base_leaves.py (this was discovered while investigating #23284, it is not necessarily limited to that file).
Before #22493: [142 tests, 178.75 s]
After #22493 and #23209: [142 tests, 181.61 s]
(in both cases, this was a best-out-of-10 timing but the variance is small enough that this is statistically very significant)
Change History (8)
comment:1 Changed 3 years ago by
- Description modified (diff)
comment:2 Changed 3 years ago by
- Cc vdelecroix jpflori fbissey dimpase added
comment:3 Changed 3 years ago by
comment:4 Changed 3 years ago by
- Priority changed from blocker to major
I really cannot see this as the reason not to upgrade MPIR. Consider using ODK funds to get a new CPU where new MPIR shines :-)
comment:5 Changed 3 years ago by
- Priority changed from major to blocker
Maybe we can accept a 10% performance regression, but certainly not without further investigation. The problem may be much worse than it appears to be on the surface.
comment:6 Changed 3 years ago by
in any event, please attach mpir-3.0.0.p0.log of the installation where you see this slowdown.
comment:7 Changed 3 years ago by
Is that the same issue as #23209?
comment:8 Changed 3 years ago by
- Description modified (diff)
- Priority changed from blocker to major
- Summary changed from Regression with MPIR-3.0.0 to Small regression with MPIR-3.0.0
Right, the regression is mostly explained by #23209. But even then, there remains a smaller regression of about 2%.
IMHO 10% performance decrease is not too bad. As well, it's unclear whether it's limited to a particular compiler/OS/hardware combo, or not. One can also say that MPIR has added more assembly codes, and I guess one can as well find a combo that shows a performance boost.