1.0~rc1-16 , Re: help with mplayer bug 431139
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 hi, and thanks everybody for the help so far; I uploaded mplayer 1.0~rc1-16 ; it was built on the alpha buildd with the '-mcpu=ev5' option. If somebody has a CPU without MVI, may please s/he test if s/he can play a mpeg file such as http://samples.mplayerhq.hu/MPEG2/dvd.mpeg ? a. A Mennucc ha scritto: hi a bug http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=431139 was posted against mplayer sometime ago: it seems that the binary contains a itoft instruction , and this is illegal on some older alpha CPUs what should I do ? add a specific-mcpu=... line ? what if I do? would code then be too slow? thanks in advance a. -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG2qCi9B/tjjP8QKQRAjE7AJ9jSumcmWjG4KS2/KkwQJsMQZjwtACgo+vi DyS9qrlZsfiZkpccnfTQhnU= =g4f3 -END PGP SIGNATURE- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: help with mplayer bug 431139
Hello, On Sun, Aug 26, 2007 at 04:15:27PM -0700, Steve Langasek wrote: On Sun, Aug 26, 2007 at 09:12:27PM +0200, A Mennucc wrote: By looking in http://buildd.debian.org/fetch.cgi?pkg=mplayer;ver=1.0%7Erc1-15;arch=alpha;stamp=1185972985 I deduced that indeed configure detects the CPU of the buildd , as follows Checking for GCC CPU optimization abilities ... ev67 Checking for MVI instruction support in GCC ... yes This means that mplayer is compiled with 'gcc -mcpu=ev67 ' ; and also that CAN_COMPILE_ALPHA_MVI is set, and this enables some specific code in libmpeg2/ (I don't know what MVI is , alas - but it looks as something that not all CPUs have) I don't know what MVI is either, but according to http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/2003-September/020895.html it's not supported on ev56 so should be disabled, or else enabled only as an optional subarch optimization within the package. According to my vague memories it is Motion Video Instructions, which I find here as well: http://www.alasir.com/articles/alpha_history/alpha_21164_21164pc.html Looking at http://en.wikipedia.org/wiki/DEC_Alpha it supports starts at 21164PC. Greetings Helge -- Dr. Helge Kreutzmann, Dipl.-Phys. [EMAIL PROTECTED] gpg signed mail preferred 64bit GNU powered http://www.itp.uni-hannover.de/~kreutzm Help keep free software libre: http://www.ffii.de/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: help with mplayer bug 431139
This is all correct, but again, the mplayer code explicitly checks the cputype at runtime and dyanmically uses the most efficient code. So, it should be possible to build for all systems (ev5) and yet maintain the performance of MVI for systems that support it. ...tom Helge Kreutzmann wrote: According to my vague memories it is Motion Video Instructions, which I find here as well: http://www.alasir.com/articles/alpha_history/alpha_21164_21164pc.html Looking at http://en.wikipedia.org/wiki/DEC_Alpha it supports starts at 21164PC. Greetings Helge -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: help with mplayer bug 431139
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 hi Steve and all, I am currently tryin to figure out what I should set in '-mcpu' : is it DEB_HOST_GNU_CPU ? And, just for the record... Steve Langasek ha scritto: On Sun, Aug 26, 2007 at 09:12:27PM +0200, A Mennucc wrote: So for starters, build with: DEB_HOST_GNU_TYPE := $(shell dpkg-architecture -qDEB_HOST_GNU_TYPE) DEB_BUILD_GNU_TYPE := $(shell dpkg-architecture -qDEB_BUILD_GNU_TYPE) this part above is already in debian/rules , ifeq ($(DEB_BUILD_GNU_TYPE), $(DEB_HOST_GNU_TYPE)) conf_args += --build $(DEB_BUILD_GNU_TYPE) else conf_args += --build $(DEB_BUILD_GNU_TYPE) --host $(DEB_HOST_GNU_TYPE) endif ./configure $(conf_args) [...] but this part is not, since ./configure in mplayer is not autoconf-generated, and there is no --build or --host option I don't know what MVI is either, but according to http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/2003-September/020895.html it's not supported on ev56 so should be disabled, or else enabled only as an optional subarch optimization within the package. thanks for the info a. -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG0qiI9B/tjjP8QKQRAmwaAKCFS+f85PMzXOR2BLCOKeaifuHirwCfddje hss4LKd23Jxx4WGS7NqN3pI= =JXLD -END PGP SIGNATURE- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
please test, Re: help with mplayer bug 431139
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 hi I prepared a patch that may solve the problem ; but unfortunately I cannot test it, since I do not have access to alpha machines. The source is http://tonelli.sns.it/pub/mplayer/sid/mplayer_1.0~rc1-16.dsc http://tonelli.sns.it/pub/mplayer/sid/mplayer_1.0~rc1-16.diff.gz http://tonelli.sns.it/pub/mplayer/mplayer_1.0~rc1.orig.tar.gz (it is signed). If someone can test it, I would appreciate a. A Mennucc ha scritto: hi a bug http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=431139 was posted against mplayer sometime ago: it seems that the binary contains a itoft instruction , and this is illegal on some older alpha CPUs what should I do ? add a specific-mcpu=... line ? what if I do? would code then be too slow? thanks in advance a. -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG0rtu9B/tjjP8QKQRAiJRAJ9NNr12LouJLipk4oqN+XIH2vuxvQCgn1lk DqHIuRMhBRoJS/D7pVfvHs0= =/5mI -END PGP SIGNATURE- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: help with mplayer bug 431139
Again, I looked at the code, there should be no reason do disable MVI - the code already checks the CPU type and runs the appropriate code for those accelerated instructions. You should be able to keep ALPHA_CAN_BUILD_MVI, but just pull down the cpu type to ev5, or alternatively, keep the cpu type where it is, but add: -mno-cix -mno-fix -mno-bwx to disable the generation of incompatible instructions in the non-protected code. ...tom but this part is not, since ./configure in mplayer is not autoconf-generated, and there is no --build or --host option I don't know what MVI is either, but according to http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/2003-September/020895.html it's not supported on ev56 so should be disabled, or else enabled only as an optional subarch optimization within the package. thanks for the info a. -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: help with mplayer bug 431139
On Mon, Aug 27, 2007 at 12:33:44PM +0200, A Mennucc wrote: I am currently tryin to figure out what I should set in '-mcpu' : is it DEB_HOST_GNU_CPU ? Er, preferably nothing at all. There's no good reason to be overriding gcc's default optimization here; your package should be targetting the same instruction set as the rest of the Debian architecture, whatever that happens to be at the time. At the moment, I believe -mcpu=ev5 happens to be the correct option. ifeq ($(DEB_BUILD_GNU_TYPE), $(DEB_HOST_GNU_TYPE)) conf_args += --build $(DEB_BUILD_GNU_TYPE) else conf_args += --build $(DEB_BUILD_GNU_TYPE) --host $(DEB_HOST_GNU_TYPE) endif ./configure $(conf_args) [...] but this part is not, since ./configure in mplayer is not autoconf-generated, and there is no --build or --host option Oh, spew. Can someone hit mplayer upstream with a very large cluebat, please? On Mon, Aug 27, 2007 at 01:54:22PM +0200, A Mennucc wrote: I prepared a patch that may solve the problem ; but unfortunately I cannot test it, since I do not have access to alpha machines. The source is http://tonelli.sns.it/pub/mplayer/sid/mplayer_1.0~rc1-16.dsc http://tonelli.sns.it/pub/mplayer/sid/mplayer_1.0~rc1-16.diff.gz http://tonelli.sns.it/pub/mplayer/mplayer_1.0~rc1.orig.tar.gz (it is signed). If someone can test it, I would appreciate Trying to build this package fails for me on alpha at: Checking for inttypes.h (required) ... no Checking for bitypes.h (inttypes.h predecessor) ... Error: Cannot find header either inttypes.h or bitypes.h (see DOCS/HTML/en/faq.html). The failure in configure.log is: cc -O4 -march=ev5 -mcpu=ev5 -pipe -ffast-math -fomit-frame-pointer-o /tmp/mplayer-conf-12199-5493.o /tmp/mplayer-conf-21897-5493.c cc1: error: unrecognized command line option -march=ev5 ldd /tmp/mplayer-conf-12199-5493.o ldd: /tmp/mplayer-conf-12199-5493.o: No such file or directory Cheers, -- Steve Langasek Give me a lever long enough and a Free OS Debian Developer to set it on, and I can move the world. [EMAIL PROTECTED] http://www.debian.org/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: help with mplayer bug 431139
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 hi I add another info By looking in http://buildd.debian.org/fetch.cgi?pkg=mplayer;ver=1.0%7Erc1-15;arch=alpha;stamp=1185972985 I deduced that indeed configure detects the CPU of the buildd , as follows Checking for GCC CPU optimization abilities ... ev67 Checking for MVI instruction support in GCC ... yes This means that mplayer is compiled with 'gcc -mcpu=ev67 ' ; and also that CAN_COMPILE_ALPHA_MVI is set, and this enables some specific code in libmpeg2/ (I don't know what MVI is , alas - but it looks as something that not all CPUs have) So the simplest workaround seems to set -mcpu to some lower CPU and to disable MVI... a. -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG0dCb9B/tjjP8QKQRAnVNAJ9/rHMi2j49z1Uow3ZiCCT9prHTEACfQRso GpZPah4Qx7lpzUKnj34GUe0= =8jJN -END PGP SIGNATURE- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: help with mplayer bug 431139
Alternatively the code could have checks for the appropriate features by using the amask instruction and chose the correct code sequences - probably a win in the MVI cases, unclear about the itoft cases. It would have to be looked at to see what code path tricks could be played. Is this code in mplayer or libmpeg2? (or is libmpeg2 included in mplayer!?) ...tom A Mennucc wrote: This means that mplayer is compiled with 'gcc -mcpu=ev67 ' ; and also that CAN_COMPILE_ALPHA_MVI is set, and this enables some specific code in libmpeg2/ (I don't know what MVI is , alas - but it looks as something that not all CPUs have) So the simplest workaround seems to set -mcpu to some lower CPU and to disable MVI... a. -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFG0dCb9B/tjjP8QKQRAnVNAJ9/rHMi2j49z1Uow3ZiCCT9prHTEACfQRso GpZPah4Qx7lpzUKnj34GUe0= =8jJN -END PGP SIGNATURE- -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: help with mplayer bug 431139
On Sun, Aug 26, 2007 at 09:12:27PM +0200, A Mennucc wrote: I add another info By looking in http://buildd.debian.org/fetch.cgi?pkg=mplayer;ver=1.0%7Erc1-15;arch=alpha;stamp=1185972985 I deduced that indeed configure detects the CPU of the buildd , as follows Checking for GCC CPU optimization abilities ... ev67 Checking for MVI instruction support in GCC ... yes So for starters, build with: DEB_HOST_GNU_TYPE := $(shell dpkg-architecture -qDEB_HOST_GNU_TYPE) DEB_BUILD_GNU_TYPE := $(shell dpkg-architecture -qDEB_BUILD_GNU_TYPE) ifeq ($(DEB_BUILD_GNU_TYPE), $(DEB_HOST_GNU_TYPE)) conf_args += --build $(DEB_BUILD_GNU_TYPE) else conf_args += --build $(DEB_BUILD_GNU_TYPE) --host $(DEB_HOST_GNU_TYPE) endif ./configure $(conf_args) [...] alpha is not the only architecture where your package will have problems if you allow configure to incorrectly select an -mcpu option at build time based on the build host's capabilities. This means that mplayer is compiled with 'gcc -mcpu=ev67 ' ; and also that CAN_COMPILE_ALPHA_MVI is set, and this enables some specific code in libmpeg2/ (I don't know what MVI is , alas - but it looks as something that not all CPUs have) I don't know what MVI is either, but according to http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/2003-September/020895.html it's not supported on ev56 so should be disabled, or else enabled only as an optional subarch optimization within the package. Thanks, -- Steve Langasek Give me a lever long enough and a Free OS Debian Developer to set it on, and I can move the world. [EMAIL PROTECTED] http://www.debian.org/ -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]
Re: help with mplayer bug 431139
Turns out it was simple to find the code in question. The code already checks the amask bits for MVI availability with CAN_COMPILE_ALPHA_MVI . So, there should be a way to keep CAN_COMPILE_ALPHA_MVI set but keep the rest of the program from using ev67 specific instructions - the mvi acceleration already has the appropriate guards in place such that the mvi accelerated code will not execute on cpus that do not support it. gcc needs to be set such that the mvi primitives are available, but general code generation is ev5. ...tom The libmpeg2 in A Mennucc wrote: This means that mplayer is compiled with 'gcc -mcpu=ev67 ' ; and also that CAN_COMPILE_ALPHA_MVI is set, and this enables some specific code in libmpeg2/ (I don't know what MVI is , alas - but it looks as something that not all CPUs have) So the simplest workaround seems to set -mcpu to some lower CPU and to disable MVI... -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of unsubscribe. Trouble? Contact [EMAIL PROTECTED]