Opened 15 years ago

Closed 15 years ago

Last modified 15 years ago

#17218 closed enhancement (wontfix)

octave optimized?

Reported by: hans@… Owned by: stechert@…
Priority: Low Milestone:
Component: ports Version: 1.6.0
Keywords: Cc: alakazam@…, MarcusCalhoun-Lopez (Marcus Calhoun-Lopez)
Port: octave

Description

Octave on my laptop is seeing a much-reduced speedup with sparse matrices when compared to linux on similar hardware.

x = rand(1000,1000); y = x; y(y<0.99) = 0; z = sparse(y); tic; x*x; toc tic; x*y; toc tic; x*z; toc

I see times on the order of 0.68 for x*x and x*y, and 0.62 for x*z, but on linux my friend (with a macbook pro - I have a macbook) sees on the order of 0.9 for x*x, x*y and 0.15 for x*z. I'm happy general matrix multiplication seems faster with the macports version, but I'd love to see a more substantial speedup for sparse matrices. I don't know if there might be some compile-time options for optimization to this or the sparse matrix library, or if there might be something to explain it away in the architecture, hardware, or OS, but I put it out there FWIW.

Change History (10)

comment:1 Changed 15 years ago by MarcusCalhoun-Lopez (Marcus Calhoun-Lopez)

When comparing, which version of octave are you using?
The reason I ask is the recent upgrade of SuiteSparse (r41628) changed the optimization from -O2 to -O3.

comment:2 Changed 15 years ago by MarcusCalhoun-Lopez (Marcus Calhoun-Lopez)

Cc: mcalhoun@… added

Cc Me!

comment:3 Changed 15 years ago by hans@…

3.0.3

With 3.0.2, I had horrendous sparse matrix performance (on those same benchmarks, x*z took on the order of 2.5 seconds)

comment:4 Changed 15 years ago by MarcusCalhoun-Lopez (Marcus Calhoun-Lopez)

Good to know that there was improvement.
The compile options are the ones recommended by the SuiteSparse package.

It seems to me that a careful analysis of octave performance would be a significant undertaking.

comment:5 Changed 15 years ago by jmroot (Joshua Root)

Cc: alakazam@… added
Milestone: Port BugsPort Enhancements
Owner: changed from macports-tickets@… to stechert@…

comment:6 Changed 15 years ago by hans@…

Yes, it probably would. If appropriate optimizations are enabled during octave and suitesparse compilation then that's as much as can be expected of a package maintainer.

comment:7 Changed 15 years ago by MarcusCalhoun-Lopez (Marcus Calhoun-Lopez)

It turns out that in a couple of SuiteSparse Makefiles, the optimization was set to -O (not -O3).
This is fixed in r42051.
Perhaps this will fix your problem.

comment:8 Changed 15 years ago by hans@…

Doesn't seem to affect it much, but thanks.

comment:9 Changed 15 years ago by MarcusCalhoun-Lopez (Marcus Calhoun-Lopez)

Resolution: wontfix
Status: newclosed

Unfortunately, I don't think there is much more that will be done on this issue.
I will therefore close the ticket.

comment:10 Changed 15 years ago by (none)

Milestone: Port Enhancements

Milestone Port Enhancements deleted

Note: See TracTickets for help on using tickets.