Opened 16 years ago
Last modified 16 years ago
#4 new enhancement
use SSE2 to speed up multiplication
| Reported by: | zooko | Owned by: | somebody |
|---|---|---|---|
| Priority: | major | Milestone: | |
| Component: | component1 | Version: | |
| Keywords: | Cc: | ||
| Launchpad Bug: |
Description
Jack Lloyd contributed a patch that speeds up encoding using SSE2 instructions (on some platforms). Unfortunately there is this mysterious problem where it actually causes it to slow down when it is executed from Python! So this patch probably shouldn't be applied as-is, but someone should investigate and figure out why it causes a slowdown when invoked from Python but not from C.
http://allmydata.org/pipermail/tahoe-dev/2009-February/001066.html
Note: See
TracTickets for help on using
tickets.
Oh, I see that Jack's mail also says that zfec takes 3 times as long to decode something if it is being invoked from Python as if it is being invoked from C! That's another mystery. It isn't actually a high priority, because the CPU usage due to zfec is much smaller than the CPU usage due to other things such as the Python interpreter, in all measurements that I've seen, but still, it would be great for someone to investigate.