>> On 07/17/2011 01:14 PM, Reinhard Tartler wrote: >>> cOn Sun, Jul 17, 2011 at 12:36:23 (CEST), David Kuehling wrote: [..] >>> >>> Grepping over a copy of the ffmpeg sources, I find ldl in >>> inline-assembler fragments in various files, such as intreadwrite.h. >>> These are #ifdefed checking for ARCH_MIPS64, which is set by the >>> configure script. >>> >>> Unfortunately the configure script checks 'uname -m' for the >>> architecture, which might be mips64 when running debian on a 64-bit >>> kernel (this happens, e.g. when running with the loongson kernels >>> shipped by debian). >>> >>> Forcing ./configure --arch=mips/mipsel might help. >> >> That makes sense to me. I'll add --arch=$(DEB_BUILD_CPU) to the >> configure line for the next upload. >> > Why aren't the builds done under 'linux32'? Doing that would cause > uname to return the proper values for an o32 build. The complete userspace *is* o32, just the kernel is not. I think that is a pretty valid way to run a system, compiling the kernel for mips64 gives better performance on those machines that can run mips64 code. A mips64 kernel can run 32 and 64-bit architecture binaries, and has to pick one description when asked via 'uname -m'. The 'setarch' tool can be used to configure which architecture that is. I.e. athough my machine usually returns 'mips64' on 'uname -m', after running, 'setarch mips' it returns just 'mips'. Maybe that'd be a a cleaner way to fix the problem for all package builds? cheers, David -- GnuPG public key: http://dvdkhlng.users.sourceforge.net/dk.gpg Fingerprint: B17A DC95 D293 657B 4205 D016 7DEF 5323 C174 7D40
Attachment:
pgpN9iHGX4rM6.pgp
Description: PGP signature