[Ffmpeg-devel] [PATCH] fix mpegaudiodec on ARM and benchmark

[Aurelien Jacobs]

On Thu, 24 Aug 2006 00:51:09 +0200
Michael Niedermayer <michaelni at gmx.at> wrote:

> Hi
> 
> On Thu, Aug 24, 2006 at 12:08:22AM +0200, Aurelien Jacobs wrote:
> [...]
> > > >              asm("smull %0, %1, %2, %3     \n\t"\
> > > > -                "mov   %0, %0,     lsr #%4\n\t"\
> > > > -                "add   %1, %0, %1, lsl #%5\n\t"\
> > > > -            : "=r"(lo), "=r"(hi)\
> > > > +                "mov   %0, %0,     lsr %4\n\t"\
> > > > +                "add   %1, %0, %1, lsl %5\n\t"\
> > > > +            : "=&r"(lo), "=&r"(hi)\
> > > >              : "r"(b), "r"(a), "i"(FRAC_BITS), "i"(32-FRAC_BITS));\
> > > >           hi; })
> > > >  #   define MUL64(a,b) ((int64_t)(a) * (int64_t)(b))
> > > > -#   define MULH(a, b) ({ int lo, hi; asm ("smull %0, %1, %2, %3" : "=r"(lo), "=r"(hi) : "r"(b),"r"(a)); hi; })
> > > > +#   define MULH(a, b) ({ int lo, hi; asm ("smull %0, %1, %2, %3" : "=&r"(lo), "=&r"(hi) : "r"(b),"r"(a)); hi; })
> > > 
> > > i think not all 4 of the & are needed, but iam not sure ...
> > 
> > If I remove any one of them, I get a load of messages like this one:
> > {standard input}: Assembler messages:
> > {standard input}:630: rdhi, rdlo and rm must all be different
> > Note that I'm cross-compiling with gcc-4.1 if that's relevant.
> 
> hmm ok, then they probably are needed, feel free to commit the patch ...

done.

> > Setting -march=armv5te (which is exactly what my Xscale is) is quite
> > slower, I don't understand why:
> > BENCHMARKs: VC:   0.000s VO:   0.000s A: 206.553s Sys:   0.438s =  206.991s
> > BENCHMARK%: VC:  0.0000% VO:  0.0000% A: 99.7882% Sys:  0.2118% = 100.0000%
> 
> maybe bugreport to gcc devels?

Previous disscussions on this list about gcc bugreports are not much
encouraging... Maybe I will try...

> > Now I also benchmarked libmad. It's still "slightly" faster !
> > BENCHMARKs: VC:   0.000s VO:   0.000s A:  54.212s Sys:   0.407s =   54.618s
> > BENCHMARK%: VC:  0.0000% VO:  0.0000% A: 99.2554% Sys:  0.7446% = 100.0000%
> > 
> > Then I benchmarked ffmp3 r6050 (patched) with --disable-libavcodec_mpegaudio_hp
> > BENCHMARKs: VC:   0.000s VO:   0.000s A:  78.751s Sys:   0.419s =   79.171s
> > BENCHMARK%: VC:  0.0000% VO:  0.0000% A: 99.4702% Sys:  0.5298% = 100.0000%
> > Pretty impressive ! Not so far from libmad !
> 
> next thing you could try are the different bitstream readers, the default is
> probably not good on arm as it does unaligned reads
> see the #defines at the top of libavcodec/bitstream.h
> note, if anything fails compilation or such please report it ...

Ok, so first the compilation failures. With LIBMPEG2_BITSTREAM_READER there
was a single error, but I'm not sure how to fix it properly:

arm-linux-gnu-gcc -DHAVE_AV_CONFIG_H -I.. -I../libavutil -Wdeclaration-after-statement -O4   -pipe -ffast-math -fomit-frame-pointer -D_REENTRANT -I/usr/include -I/usr/src/DVB/ost/include -I/usr/include/dxr2 -I/usr/local/include/cdda -D_FILE_OFFSET_BITS=64 -D_LARGEFILE_SOURCE -D_ISOC9X_SOURCE    -c -o h263.o h263.c
In file included from mpegvideo.h:30,
                 from h263.c:38:
bitstream.h: In function 'init_get_bits':
bitstream.h:696: warning: assignment discards qualifiers from pointer target type
bitstream.h:701: warning: assignment discards qualifiers from pointer target type
h263.c: In function 'decode_user_data':
h263.c:5802: error: 'GetBitContext' has no member named 'index'

Then with A32_BITSTREAM_READER I had two more errors which are fixed by
the attached patch.

Now the benchmarks:

#define ALT_BITSTREAM_READER
BENCHMARKs: VC:   0.000s VO:   0.000s A:  78.751s Sys:   0.419s =   79.171s
BENCHMARK%: VC:  0.0000% VO:  0.0000% A: 99.4702% Sys:  0.5298% = 100.0000%

#define LIBMPEG2_BITSTREAM_READER
BENCHMARKs: VC:   0.000s VO:   0.000s A:  78.978s Sys:   0.422s =   79.399s
BENCHMARK%: VC:  0.0000% VO:  0.0000% A: 99.4691% Sys:  0.5309% = 100.0000%

#define A32_BITSTREAM_READER
BENCHMARKs: VC:   0.000s VO:   0.000s A:  70.622s Sys:   0.357s =   70.978s
BENCHMARK%: VC:  0.0000% VO:  0.0000% A: 99.4973% Sys:  0.5027% = 100.0000%

We are getting even closer to libmad :-)

What about something like the second attached patch ?

Aurel
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bitstream-reader.diff
Type: text/x-diff
Size: 682 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20060824/66655909/attachment.diff>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: bitstream-reader-arm.diff
Type: text/x-diff
Size: 499 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20060824/66655909/attachment-0001.diff>