Re: [CinCVS] [PATCH] make configure detect and work on amd64
On Tuesday 06 March 2007 21:04, Johannes Sixt wrote: On Tuesday 06 March 2007 00:44, [EMAIL PROTECTED] wrote: what about attached patch for the mmx part? Looks good. I'd appreciate to have this cross-checked by others with x86_64 hardware. Please test both --disable-mmx and --enable-mmx. The patch is now in SVN. -- Hannes ___ Cinelerra mailing list Cinelerra@skolelinux.no https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra
Re: [CinCVS] [PATCH] make configure detect and work on amd64
On Tuesday 06 March 2007 00:44, [EMAIL PROTECTED] wrote: what about attached patch for the mmx part? Looks good. I'd appreciate to have this cross-checked by others with x86_64 hardware. Please test both --disable-mmx and --enable-mmx. -- Hannes ___ Cinelerra mailing list Cinelerra@skolelinux.no https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra
Re: [CinCVS] [PATCH] make configure detect and work on amd64
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Johannes Sixt schrieb: - Adding all those optimization options, -funroll-loops -minline-all-stringops etc. to the default compile line is a big NO-NO. It may be convenient for you, but it is not good practice in general. Is there any harm done if I remove them? Typically and in most cases, it is just sufficient to use -O3 The compiler has sensible defaults for most cases. In the Patch I see the line CPU_CFLAGS= -march=athlon64 -funroll-loops -minline-all-stringops -ffast-math -mmmx -msse -msse2 - -msse3 -m3dnow -DHAVE_MMX -DUSE_MMX -DHAVE_3dnow $CPU_CFLAGS This is a lot of redundant and superfluos information. As far as I am informed, the only thing you *really* should switch on for AMD64 is the switch -msse3 This includes all previous SSE levels and superseedes MMX and 3DNow! So, at least the line should be for X86_64: CPU_CFLAGS= -march=athlon64 -msse3 .. What are the -DHAVE_MMX -DUSE_MMX -DHAVE_3dnow good for? Does anyone know? Hermann V. -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.5 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFF4DvfZbZrB6HelLIRAkEeAKCWrmHpTlP2CGbRuntAOx/DvqJs4gCgpG8O vNQPfM12yzwr/UweJI4FziQ= =X6rh -END PGP SIGNATURE- ___ Cinelerra mailing list Cinelerra@skolelinux.no https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra
Re: [CinCVS] [PATCH] make configure detect and work on amd64
In the Patch I see the line CPU_CFLAGS= -march=athlon64 -funroll-loops -minline-all-stringops -ffast-math -mmmx -msse -msse2 - -msse3 -m3dnow -DHAVE_MMX -DUSE_MMX -DHAVE_3dnow $CPU_CFLAGS This is a lot of redundant and superfluos information. As far as I am informed, the only thing you *really* should switch on for AMD64 is the switch -msse3 Note that not every amd64 support sse3, mine for example doesn't. Note also that x86_64 *DOES NOT* mean amd64, that can also mean core 2 duo, which afaik doesnt support 3dnow. Also, are you sure that -ffast-math is safe here ? I dont like autodecting cpu at configure time because that, most of the time, leads to headaches. Why not honouring CFLAGS variable and that's all ? After all, that's the person who compiles that might know best what his system is. -DHAVE_MMX -DUSE_MMX -DHAVE_3dnow are some defines used in some source files to enable or not asm optimizations. I had tried that once on my amd64 but those were either useless (didnt enable anything at all, that was probably some old switchs that are unused now) or didn't compile (probably because it's nasm asm that is for x86_32). You can grep for them to see where they are used and what for ;) Regards, Alexis. pgpUeOVpKB1Um.pgp Description: PGP signature
Re: [CinCVS] [PATCH] make configure detect and work on amd64
On Feb 17, 2007, at 21:07:58, Johannes Sixt wrote: Thanks, but: - There was talk recently that compilation fails on amd_64 unless --enable-mmx=no. Your patch sets it to 'yes' if --enable-mmx was not set. I had to fix this particular problem. Has something changed? See https://init.linpro.no/pipermail/skolelinux.no/cinelerra/2007- January/009702.html http://repo.or.cz/w/cinelerra_cv.git? a=commitdiff;h=f7ca8d92e1c0cb0ef90cbc17506d388dd5cb0622 http://repo.or.cz/w/cinelerra_cv.git? a=commitdiff;h=052cb77e7d96c19ae260f696baa9d40c6e2313cc i was using libavcodec from ubuntu, not the one coming with cinelerra - Is libx264_pic a commonly used name or specific to ubuntu? its also present in debian, not sure about others. - I'd like to move the x86_64 case above the powerpc case. Ok? sure - Adding all those optimization options, -funroll-loops -minline- all-stringops etc. to the default compile line is a big NO-NO. It may be convenient for you, but it is not good practice in general. Is there any harm done if I remove them? i guess you can remove them too - You define an AM_CONDITIONAL named CPU_x86_64 that is not used anywhere. Ok to remove it? its not used right now so it might be ok to remove it. - The assignment libx264_pic=yes must read libx264=yes. I'll fix that. ok j ___ Cinelerra mailing list Cinelerra@skolelinux.no https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra
[CinCVS] [PATCH] make configure detect and work on amd64
Hi, trying to compile cinelerra on amd64 / ubuntu edgy i had some problems i wanted to use ffmpeg from the repositories, same for x264. with attached patch, * libmpeg3/video/mmxidct.S is only used on x86_32 * adds amd64 CFLAGS to use all amd64 features if configure is run with --enable-mmx * use libx264_pic if available instead of libx264 and if called with --with-pic j cinelerra.amd64.compile.patch Description: Binary data
Re: [CinCVS] [PATCH] make configure detect and work on amd64
On Saturday 17 February 2007 17:03, [EMAIL PROTECTED] wrote: Hi, trying to compile cinelerra on amd64 / ubuntu edgy i had some problems i wanted to use ffmpeg from the repositories, same for x264. with attached patch, * libmpeg3/video/mmxidct.S is only used on x86_32 * adds amd64 CFLAGS to use all amd64 features if configure is run with --enable-mmx * use libx264_pic if available instead of libx264 and if called with --with-pic Thanks, but: - There was talk recently that compilation fails on amd_64 unless --enable-mmx=no. Your patch sets it to 'yes' if --enable-mmx was not set. I had to fix this particular problem. Has something changed? See https://init.linpro.no/pipermail/skolelinux.no/cinelerra/2007-January/009702.html http://repo.or.cz/w/cinelerra_cv.git?a=commitdiff;h=f7ca8d92e1c0cb0ef90cbc17506d388dd5cb0622 http://repo.or.cz/w/cinelerra_cv.git?a=commitdiff;h=052cb77e7d96c19ae260f696baa9d40c6e2313cc - Is libx264_pic a commonly used name or specific to ubuntu? - I'd like to move the x86_64 case above the powerpc case. Ok? - Adding all those optimization options, -funroll-loops -minline-all-stringops etc. to the default compile line is a big NO-NO. It may be convenient for you, but it is not good practice in general. Is there any harm done if I remove them? - You define an AM_CONDITIONAL named CPU_x86_64 that is not used anywhere. Ok to remove it? - The assignment libx264_pic=yes must read libx264=yes. I'll fix that. -- Hannes ___ Cinelerra mailing list Cinelerra@skolelinux.no https://init.linpro.no/mailman/skolelinux.no/listinfo/cinelerra