Change CFLAGS to -march=native?
Most recent distributions should have a 4.x compiler and support the -march=native CFLAG?
e.g. ffmpeg compiles with -march=x86_64 -mtune=generic O3
Others with -march=x86_64 -mtune=generic O2
The upstream versions of all the included PTS tests are used. The lobbying should be done to the upstream projects to change their defaults.
I am a little lost here.
What also absolute annoying is, that kernel.org removed the file worldofpadman.run and http://thilo.kickchat.com/download/ moved it into a folder called 'old'
There is no way to run the test for padman and nearly every test I want to run contains it Just hangs there forever...
//edit: or lets say the node ftp://ftp.bm.kernel.org is not reachable...
//edit: it shows md5 checksum failure (guess it just downloads the 404 error), so I am going to download and save worldofpadman.run into ./phoronix-test-suite/installed-tests/pts/padman-1.1.0
Last edited by disi; 10-25-2011 at 01:23 PM.
Upstreams are usually striving for safe defaults. A binary compiled with -march=native only runs on one specific subarchitecture, -Wall -Werror might have no problems with the compiler version of the developers, but might break with others... If you want your funroll-loops you have to add them yourself.
Originally Posted by Michael
I am not talking about ricer stuff. But since the binary is build on the system, which is going to be tested, why not support the processor features?
Originally Posted by AnonymousCoward
e.g. Amd64 vs. Bulldozer or other newer CPU
My second question was more in the direction, who to contact e.g. about the padman problem?
Sure, if you compile a program for one specific system, add -march=native. But if you're an upstream, once you notice that people who don't know what they're doing want to compile your stuff, you keep your defaults sane and your support workload low.
Originally Posted by disi