On Mon Feb 02 2015 at 8:34:07 AM Andreas Cadhalpun < andreas.cadhal...@googlemail.com> wrote:
> Hi James, > > On 02.02.2015 11:45, James Cowgill wrote: > > It failed with: > >> /tmp/ccAK6rnm.s: Assembler messages: > >> /tmp/ccAK6rnm.s:15: Warning: the `dspr2' extension requires MIPS64 > revision 2 or greater > >> /tmp/ccAK6rnm.s:17316: Error: opcode not supported on this processor: > mips64 (mips64) `ext $5,$2,4,4' > [...] > >> common.mak:49: recipe for target 'libavcodec/aacdec.o' failed > >> make[2]: *** [libavcodec/aacdec.o] Error 1 > > > > Which I find curious because ext is a mips(32|64)r2 instruction and I > > thought I had just disabled those :) > > Probably some #ifdef guards are missing somewhere. > > > Like Timothy said, yes they use some pointer arithmetic so they will > > need porting to MIPS64. > > > > Also since the Debian mips port is hard-float, you should be able to > > enable the FPU instruction even on mips and mipsel (it fails at the > > moment). > > I had asked Nedeljko Babic (from Imgtech) about the test failures on > mips(el). He wasn't sure if the Debian mips buildds (Loongson 3A) have a > FPU and thus advised to disable all the optimizations there. > > > I can have a look at all this and submit some patches upstream when > > they're ready (don't know how long it will take to do this though). > > That'd be much appreciated, thanks. You can send patches to > ffmpeg-de...@ffmpeg.org. > > Until then, I'm just going to apply your patch for debian/rules to > disable these optimizations on mips64el. > > Best regards, > Andreas >
