#63601 closed enhancement (fixed)
libxslt @1.1.34_6: add a variant for better performance
Reported by: | cooljeanius (Eric Gallager) | Owned by: | ryandesign (Ryan Carsten Schmidt) |
---|---|---|---|
Priority: | Normal | Milestone: | |
Component: | ports | Version: | 2.7.1 |
Keywords: | Cc: | ||
Port: | libxslt |
Description
The xsltproc binary can be very slow sometimes; to deal with that, I would like a variant of libxslt that turns on additional optimization options for increased performance. An example variant can be found in my copy of the libxslt Portfile in my LocalPorts repo: https://github.com/cooljeanius/LocalPorts/blob/6f35e07e73b6aebbf1074fee8b25717b5c4ce3b0/textproc/libxslt/Portfile#L86-L89
Change History (6)
comment:1 Changed 2 years ago by szhorvat (Szabolcs Horvát)
comment:2 Changed 2 years ago by szhorvat (Szabolcs Horvát)
I benchmarked my own use case after adding -O3 -flto=thin
. The timing went from 104 to 96 seconds. That's a 7-8% improvement, not small, but also not huge for my own usecase.
The above suggestion includes many additional optimization options that are unsafe, and I would not be comfortable using. I did not include the patch.
I tried this with the default compiler on macOS 10.14.6.
comment:3 Changed 2 years ago by szhorvat (Szabolcs Horvát)
libxslt was just updated to 1.1.37 in MacPorts. This version performs much better with the default options (85 seconds with the same benchmark). With this version, using -O3 -flto=thin
makes no significant different in performance.
Unless others see a significant effect from optimization options, this issue can be closed.
comment:4 Changed 2 years ago by jmroot (Joshua Root)
Owner: | set to ryandesign |
---|---|
Status: | new → assigned |
comment:5 Changed 14 months ago by ryandesign (Ryan Carsten Schmidt)
Resolution: | → wontfix |
---|---|
Status: | assigned → closed |
Ok thanks for letting us know.
comment:6 Changed 14 months ago by dgilman (David Gilman)
Resolution: | wontfix → fixed |
---|
I would like to second this request, and suggest also turning on link-time optimization when supported. At the moment this is one of the annoying bottlenecks for a project I am working on, and any improvements would be most welcome.