OpenMP "feature" causes massive performance issues on multi-cores
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
ufraw (Ubuntu) |
New
|
Undecided
|
Unassigned |
Bug Description
Binary package hint: ufraw
The parallelization of UFRaw is so bad that it is actually (a lot) faster to run it with OMP_NUM_THREADS=1 to disable OpenMP parallelizations on my Core2 Quad. It seems like the parallelization was done over too small chunks to amortize the thread startup cost.
I'd suggest disabling OpenMP support in UFRaw to improve performance. The current version is barely usable, the slowdown is probably (at least) 10x.
ProblemType: Bug
Architecture: amd64
Date: Sun Oct 11 01:10:46 2009
DistroRelease: Ubuntu 9.10
NonfreeKernelMo
Package: ufraw 0.15-2ubuntu1
ProcEnviron:
PATH=(custom, user)
LANG=en_US.UTF-8
SHELL=/bin/bash
ProcVersionSign
SourcePackage: ufraw
Uname: Linux 2.6.31-13-generic x86_64
Using an i7 here and yes, it's deathly slow.
Strange thing is, a few weeks ago this was flying along at a nice speed. Was the OpenMP patch that recent?