Statistics
| Branch: | Revision:

ffmpeg / libavcodec / i386 / dsputil_mmx.c @ 5b21bdab

History | View | Annotate | Download (143 KB)

# Date Author Comment
6810b93a 09/29/2007 10:31 PM Loren Merritt

sse2 version of compute_autocorr().
4x faster than c (somehow, even though doubles only allow 2x simd).
overal flac encoding: 15-50% faster on core2, 4-11% on k8, 3-13% on p4.

Originally committed as revision 10621 to svn://svn.ffmpeg.org/ffmpeg/trunk

7bcc1d5b 08/26/2007 04:10 PM Ramiro Polla

CONFIG_7REGS has been renamed to HAVE_7REGS

Originally committed as revision 10237 to svn://svn.ffmpeg.org/ffmpeg/trunk

90e9e94d 08/26/2007 12:34 PM Michael Niedermayer

workaround gcc bug, untested as my gcc is not complaining

Originally committed as revision 10236 to svn://svn.ffmpeg.org/ffmpeg/trunk

62975029 08/26/2007 01:11 AM Michael Niedermayer

avoid overflow in the 3rd lifting step, this now needs mmx2 at minimum
(patch for plain mmx support is welcome ...)

Originally committed as revision 10226 to svn://svn.ffmpeg.org/ffmpeg/trunk

3e0f7126 08/25/2007 03:20 PM Michael Niedermayer

update mmx code to latest snow changes
note, the code likely can overflow and thus needs some more changes
sse2 updated too but disabled as it is untested

Originally committed as revision 10223 to svn://svn.ffmpeg.org/ffmpeg/trunk

d593e329 08/25/2007 03:00 AM Michael Niedermayer

use 16bit IDWT (a SIMD implementation of it should be >2x faster then with
the old 32bit code)
disable mmx/sse2 optimizations as they need a rewrite now

Originally committed as revision 10218 to svn://svn.ffmpeg.org/ffmpeg/trunk

73f51a4d 07/24/2007 08:54 AM Aurelien Jacobs

help some gcc version to optimize out those functions

Originally committed as revision 9785 to svn://svn.ffmpeg.org/ffmpeg/trunk

674eeb5f 07/10/2007 08:27 PM Aurelien Jacobs

cosmetics: indentation

Originally committed as revision 9582 to svn://svn.ffmpeg.org/ffmpeg/trunk

eb75a698 07/10/2007 08:23 PM Aurelien Jacobs

Avoid linking with h263.c functions when the relevant codecs
are not compiled in.

Originally committed as revision 9581 to svn://svn.ffmpeg.org/ffmpeg/trunk

a00177a9 07/08/2007 11:15 PM Måns Rullgård

make arguments to ssd_int8_vs_int16() const

Originally committed as revision 9548 to svn://svn.ffmpeg.org/ffmpeg/trunk

663deb54 05/20/2007 05:07 AM Zuxy Meng

Remove incorrect comment; MMX2 is preferred over 3DNow! on Athlon

Originally committed as revision 9079 to svn://svn.ffmpeg.org/ffmpeg/trunk

038bfcf9 05/18/2007 08:18 AM Zuxy Meng

3DNow! and SSSE3 optimization to QNS DSP functions; use pmulhrw/pmulhrsw instead of pmulhw

Originally committed as revision 9053 to svn://svn.ffmpeg.org/ffmpeg/trunk

5b0b7054 05/16/2007 11:23 PM Aurelien Jacobs

better separation of vp3dsp functions from dsputil_mmx.c

Originally committed as revision 9039 to svn://svn.ffmpeg.org/ffmpeg/trunk

b550bfaa 05/16/2007 09:51 AM Ronald S. Bultje

Add libavcodec to compiler include flags in order to simplify header
include paths in the source files.
mostly from a patch by Ronald S. Bultje, rbultje ronald.bitfreak net

Originally committed as revision 9034 to svn://svn.ffmpeg.org/ffmpeg/trunk

9b5dc867 05/14/2007 02:28 PM Panagiotis Issaris

Make vp3dsp*.c compilation optional.

Originally committed as revision 9025 to svn://svn.ffmpeg.org/ffmpeg/trunk

1edbfe19 05/12/2007 02:41 AM Loren Merritt

factor sum_abs_dctelem out of dct_sad, and simd it.
sum_abs_dctelem_* alone:
core2: c=186 mmx2=39 sse2=21 ssse3=13 (cycles)
k8: c=163 mmx2=33 sse2=31
p4: c=370 mmx2=60 sse2=60
dct_sad including sum_abs_dctelem_*:
core2: c=405 mmx2=258 sse2=240 ssse3=232...

561f940c 05/12/2007 01:16 AM Loren Merritt

sse2 & ssse3 versions of hadamard. unroll and inline diff_pixels.
core2: before mmx2=193 cycles. after mmx2=174 sse2=122 ssse3=115 (cycles).
k8: before mmx2=205. after mmx2=184 sse2=180.
p4: before mmx2=342. after mmx2=314 sse2=309.

Originally committed as revision 9000 to svn://svn.ffmpeg.org/ffmpeg/trunk

5adf43e4 05/09/2007 01:46 AM Loren Merritt

cosmetics: remove code duplication in hadamard8_diff_mmx

Originally committed as revision 8946 to svn://svn.ffmpeg.org/ffmpeg/trunk

bba5293b 05/08/2007 05:55 PM Loren Merritt

cosmetics: remove duplicate transpose macro

Originally committed as revision 8939 to svn://svn.ffmpeg.org/ffmpeg/trunk

fe037229 04/07/2007 02:10 PM Diego Biurrun

typos

Originally committed as revision 8642 to svn://svn.ffmpeg.org/ffmpeg/trunk

59006372 03/30/2007 07:15 PM Loren Merritt

mmx 16-bit ssd. 2.3x faster svq1 encoding.

Originally committed as revision 8559 to svn://svn.ffmpeg.org/ffmpeg/trunk

d42f8802 02/24/2007 11:58 AM Diego Biurrun

Fix wrong conditional, Snow decoding, not encoding, was SIMD-accelerated.

Originally committed as revision 8116 to svn://svn.ffmpeg.org/ffmpeg/trunk

9dd6c804 01/30/2007 10:31 AM Panagiotis Issaris

Add the const specifier as needed to reduce the number of warnings.

Originally committed as revision 7764 to svn://svn.ffmpeg.org/ffmpeg/trunk

486497e0 11/14/2006 03:18 AM Måns Rullgård

revert bad checkin

Originally committed as revision 7044 to svn://svn.ffmpeg.org/ffmpeg/trunk

be6ed6ff 11/14/2006 03:12 AM Måns Rullgård

move some CFLAGS settings away from config.* writing section

Originally committed as revision 7043 to svn://svn.ffmpeg.org/ffmpeg/trunk

bb54f6ab 11/12/2006 03:34 AM Måns Rullgård

adding more static keywords

Originally committed as revision 6976 to svn://svn.ffmpeg.org/ffmpeg/trunk

e9f1885c 11/03/2006 02:03 AM Michael Niedermayer

optimize H264_DEBLOCK_P0_Q0
2.5% faster filter_mb_fast() on P3

Originally committed as revision 6877 to svn://svn.ffmpeg.org/ffmpeg/trunk

7c428ea6 10/14/2006 05:04 PM Diego Biurrun

Put libmpeg2 IDCT functions under CONFIG_GPL, fixes link failure
with --disable-opts.

Originally committed as revision 6691 to svn://svn.ffmpeg.org/ffmpeg/trunk

c26abfa5 10/11/2006 11:17 PM Diego Biurrun

Rename ABS macro to FFABS.

Originally committed as revision 6666 to svn://svn.ffmpeg.org/ffmpeg/trunk

b78e7197 10/07/2006 03:30 PM Diego Biurrun

Change license headers to say 'FFmpeg' instead of 'this program/this library'
and fix GPL/LGPL version mismatches.

Originally committed as revision 6577 to svn://svn.ffmpeg.org/ffmpeg/trunk

0eb59ddb 10/05/2006 12:23 AM Diego Biurrun

Switch idct_mmx_xvid.c from GPL to LGPL as permitted by the
author, Peter Ross (pross xvid org).

Originally committed as revision 6557 to svn://svn.ffmpeg.org/ffmpeg/trunk

2833fc46 10/01/2006 09:25 PM Loren Merritt

approximate qpel functions: sacrifice some quality for some decoding speed. enabled on B-frames with -lavdopts fast.

Originally committed as revision 6412 to svn://svn.ffmpeg.org/ffmpeg/trunk

62bb489b 09/27/2006 07:54 PM Måns Rullgård

add some #ifdef CONFIG_ENCODERS/DECODERS

Originally committed as revision 6356 to svn://svn.ffmpeg.org/ffmpeg/trunk

2a2311be 09/14/2006 10:13 PM Aurelien Jacobs

disable vp3 mmx idct for theora files to avoid artifacts
(see theora-a4_v6-k250-s0_2.ogg)

Originally committed as revision 6253 to svn://svn.ffmpeg.org/ffmpeg/trunk

7f889a76 09/14/2006 12:38 AM Diego Biurrun

Remove the LGPL exception clause as discussed on ffmpeg-devel
and move the dependent code under CONFIG_GPL.

Originally committed as revision 6248 to svn://svn.ffmpeg.org/ffmpeg/trunk

1dac8fea 09/12/2006 08:58 PM Aurelien Jacobs

Enables back the mmx/sse optimized version of the vp3 idct.
It generates different md5sum than the reference C implementation,
but no visual difference, so enabled only when bitexact is not set.

Originally committed as revision 6241 to svn://svn.ffmpeg.org/ffmpeg/trunk

0e176c3e 09/02/2006 11:10 PM Måns Rullgård

remove redundant declarations

Originally committed as revision 6153 to svn://svn.ffmpeg.org/ffmpeg/trunk

3e20143e 08/28/2006 09:33 AM Loren Merritt

mmx implementation of deblocking strength decision.
2-3% faster h264.

Originally committed as revision 6113 to svn://svn.ffmpeg.org/ffmpeg/trunk

4454dc1b 08/12/2006 04:37 PM John Dalgliesh

Support for MacIntel, last part: balign directives
Determines whether .align's arg is power-of-two or not, then defines ASMALIGN appropriately in config.h. Changes all .baligns to ASMALIGNs.
Patch by John Dalgliesh % johnd AH defyne P org %
Original thread:...

06972056 08/11/2006 06:19 PM Loren Merritt

vorbis simd tweaks

Originally committed as revision 5983 to svn://svn.ffmpeg.org/ffmpeg/trunk

1f1aa1d9 08/10/2006 08:24 PM Michael Niedermayer

convert vector_fmul_reverse_sse2 and vector_fmul_add_add_sse2 to sse
please complain if they are slower on sse2 cpus ...

Originally committed as revision 5976 to svn://svn.ffmpeg.org/ffmpeg/trunk

eb4825b5 08/10/2006 07:06 PM Loren Merritt

sse and 3dnow implementations of float->int conversion and mdct windowing.
15% faster vorbis.

Originally committed as revision 5975 to svn://svn.ffmpeg.org/ffmpeg/trunk

ee5df927 08/09/2006 06:37 AM Loren Merritt

emms -> femms

Originally committed as revision 5965 to svn://svn.ffmpeg.org/ffmpeg/trunk

21bb884f 08/08/2006 12:03 PM Michael Niedermayer

change vorbis_inverse_coupling_sse2() so it works on sse1 cpus

Originally committed as revision 5957 to svn://svn.ffmpeg.org/ffmpeg/trunk

cd035a60 08/03/2006 07:09 AM Loren Merritt

10l, vorbis_inverse_coupling_sse() was really 3dnow

Originally committed as revision 5903 to svn://svn.ffmpeg.org/ffmpeg/trunk

2dac4acf 08/03/2006 03:18 AM Loren Merritt

sse & sse2 implementations of vorbis channel coupling.
9% faster vorbis (on a K8).

Originally committed as revision 5898 to svn://svn.ffmpeg.org/ffmpeg/trunk

595e7bd9 07/29/2006 08:45 AM Stefan Gehrer

some MMX optimizations for the CAVS decoder

Originally committed as revision 5846 to svn://svn.ffmpeg.org/ffmpeg/trunk

5ced7b80 07/19/2006 09:49 AM Michael Niedermayer

disable the vp3 mmx and sse2 idcts, their output doesnt match the c idct (tested with -f crc) and the theora spec does not allow different idcts not to mention the difference is quite vissible ...

Originally committed as revision 5788 to svn://svn.ffmpeg.org/ffmpeg/trunk

98d417cb 07/05/2006 07:31 PM Måns Rullgård

#define SBUTTERFLY outside CONFIG_ENCODERS

Originally committed as revision 5628 to svn://svn.ffmpeg.org/ffmpeg/trunk

9c39071d 07/03/2006 10:52 AM Luca Abeni

Move REG_* macros from libavcodec/i386/mmx.h to libavutil/x86_cpu.h

Originally committed as revision 5595 to svn://svn.ffmpeg.org/ffmpeg/trunk

75ca1a5f 04/05/2006 04:13 AM Loren Merritt

gmc_mmx tweaks

Originally committed as revision 5269 to svn://svn.ffmpeg.org/ffmpeg/trunk

703c8195 04/04/2006 09:23 AM Loren Merritt

mmx implementation of 3-point GMC. (5x faster than C)

Originally committed as revision 5265 to svn://svn.ffmpeg.org/ffmpeg/trunk

513fbd8e 03/23/2006 08:16 PM Loren Merritt

prefetch pixels for future motion compensation. 2-5% faster h264.

Originally committed as revision 5203 to svn://svn.ffmpeg.org/ffmpeg/trunk

fdd30579 03/22/2006 10:05 PM Loren Merritt

added mmx implementation of h264_chroma_mc2

Originally committed as revision 5200 to svn://svn.ffmpeg.org/ffmpeg/trunk

e8600e5e 03/22/2006 12:08 PM Robert Edele

add MMX and SSE versions of ff_snow_inner_add_yblock
Patch by Robert Edele < yartrebo AH earthlink POIS net >
Original Thread:
Date: Mar 22, 2006 3:24 AM
Subject: [Ffmpeg-devel] [PATCH] snow mmx + sse2 part 5

Originally committed as revision 5197 to svn://svn.ffmpeg.org/ffmpeg/trunk

2c9a0285 03/21/2006 09:51 PM Robert Edele

snow mmx+sse2 optimizations, part 4

Patch by Robert Edele, yartrebo <<at>> earthlink <<dot>> net

Originally committed as revision 5191 to svn://svn.ffmpeg.org/ffmpeg/trunk

4567b4bd 03/20/2006 10:27 PM Robert Edele

Add the mmx and sse2 implementations of ff_snow_vertical_compose().
Patch by Robert Edele < yartrebo AH earthlink POIS net >
Original thread:
Date: Mar 20, 2006 5:54 PM
Subject: [Ffmpeg-devel] [PATCH] snow mmx + sse2 part 3

Originally committed as revision 5185 to svn://svn.ffmpeg.org/ffmpeg/trunk

548a1c8a 03/07/2006 10:45 PM Loren Merritt

h264_idct8_add_mmx

Originally committed as revision 5123 to svn://svn.ffmpeg.org/ffmpeg/trunk

6da971f1 03/07/2006 10:33 PM Loren Merritt

h264_idct_add only needs mmx1

Originally committed as revision 5122 to svn://svn.ffmpeg.org/ffmpeg/trunk

ef9d1d15 02/10/2006 06:55 AM Loren Merritt

h264: special case dc-only idct. ~1% faster overall

Originally committed as revision 4971 to svn://svn.ffmpeg.org/ffmpeg/trunk

68b51e58 02/05/2006 01:35 PM Steve L'Homme

MSVC-compatible _align8/_align16 declaration
patch by Steve Lhomme, steve .dot. lhomme .at. free .dot. fr

Originally committed as revision 4942 to svn://svn.ffmpeg.org/ffmpeg/trunk

5509bffa 01/12/2006 10:43 PM Diego Biurrun

Update licensing information: The FSF changed postal address.

Originally committed as revision 4842 to svn://svn.ffmpeg.org/ffmpeg/trunk

bb270c08 12/22/2005 01:10 AM Diego Biurrun

COSMETICS: tabs --> spaces, some prettyprinting

Originally committed as revision 4764 to svn://svn.ffmpeg.org/ffmpeg/trunk

115329f1 12/17/2005 06:14 PM Diego Biurrun

COSMETICS: Remove all trailing whitespace.

Originally committed as revision 4749 to svn://svn.ffmpeg.org/ffmpeg/trunk

ea15df80 11/12/2005 05:23 AM Loren Merritt

use sse16_sse2() in nsse

Originally committed as revision 4688 to svn://svn.ffmpeg.org/ffmpeg/trunk

a6624e21 10/27/2005 06:45 AM Loren Merritt

faster h264_chroma_mc8_mmx, added h264_chroma_mc4_mmx.
2-4% overall speedup.

Originally committed as revision 4666 to svn://svn.ffmpeg.org/ffmpeg/trunk

b926572a 10/09/2005 11:38 PM Loren Merritt

h264 mmx weighted prediction. up to 3% overall speedup.

Originally committed as revision 4630 to svn://svn.ffmpeg.org/ffmpeg/trunk

5693c083 09/30/2005 02:31 AM Loren Merritt

sse2 16x16 sum squared diff (306=>268 cycles on a K8)
faster 8x8 mmx ssd (77=>70 cycles)

Originally committed as revision 4623 to svn://svn.ffmpeg.org/ffmpeg/trunk

12e96681 09/21/2005 09:17 PM Michael Niedermayer

replace a few mov + psrlq with pshufw, there are more cases which could benefit from this but they would require us to duplicate some functions ...
the trick is from various places (my own code in libpostproc, a patch on the x264 list, ...)

Originally committed as revision 4608 to svn://svn.ffmpeg.org/ffmpeg/trunk

cd7af76d 09/10/2005 07:30 PM Reimar Döffinger

Fix compile without CONFIG_GPL, misplaced #endif caused a missing }.

Originally committed as revision 4575 to svn://svn.ffmpeg.org/ffmpeg/trunk

84740d59 09/10/2005 05:01 PM Michael Niedermayer

xvids mmx&mmx2 idcts
needed to decode xvid without some minor artefacts
under #ifdef CONFIG_GPL of course

Originally committed as revision 4572 to svn://svn.ffmpeg.org/ffmpeg/trunk

79396ac6 09/06/2005 09:25 PM Måns Rullgård

Kill some compiler warnings. Compiled code verified identical after changes.

Originally committed as revision 4567 to svn://svn.ffmpeg.org/ffmpeg/trunk

d2bb7db1 06/02/2005 08:45 PM Loren Merritt

sort H.264 mmx dsp functions into their own file

Originally committed as revision 4338 to svn://svn.ffmpeg.org/ffmpeg/trunk

c26ae41d 06/01/2005 09:19 PM Michael Niedermayer

adding a few const

Originally committed as revision 4337 to svn://svn.ffmpeg.org/ffmpeg/trunk

1d62fc85 05/31/2005 10:48 PM Loren Merritt

MMX for H.264 iDCT (adapted from x264)

Originally committed as revision 4329 to svn://svn.ffmpeg.org/ffmpeg/trunk

3072f0cb 05/25/2005 08:05 AM Zoltán Hidvégi

MMX code for (put|avg)_h264_chroma_mc8

Originally committed as revision 4305 to svn://svn.ffmpeg.org/ffmpeg/trunk

5cf08f23 05/18/2005 09:17 AM Loren Merritt

H.264 deblocking optimizations (mmx for chroma_bS4 case, convert existing cases to 8-bit math)

Originally committed as revision 4271 to svn://svn.ffmpeg.org/ffmpeg/trunk

5773a746 05/17/2005 06:28 PM Michael Niedermayer

porting the mmx&sse2 (sse2 untested) vp3 idcts to the lavc idct API

Originally committed as revision 4260 to svn://svn.ffmpeg.org/ffmpeg/trunk

b178f758 05/17/2005 09:11 AM Michael Niedermayer

disabling vp3 mmx&mmx2 idcts, they must be ported over to the lavc idct API, ill port the vp3 c idct

Originally committed as revision 4255 to svn://svn.ffmpeg.org/ffmpeg/trunk

c998bdd9 05/08/2005 06:48 PM Michael Niedermayer

fix PIC

Originally committed as revision 4204 to svn://svn.ffmpeg.org/ffmpeg/trunk

42251a2a 04/25/2005 01:01 AM Loren Merritt

MMX for H.264 deblocking filter

Originally committed as revision 4158 to svn://svn.ffmpeg.org/ffmpeg/trunk

4d9ae03b 02/07/2005 05:09 PM Martin Drab

optimization and gcc 4.0 bug workaround patch by (Martin Drab >drab kepler.fjfi.cvut cz<)

Originally committed as revision 3945 to svn://svn.ffmpeg.org/ffmpeg/trunk

053dea12 10/11/2004 02:19 AM Aurelien Jacobs

adapting existing mmx/mmx2/sse/3dnow optimizations so they work on x86_64 patch by (Aurelien Jacobs <aurel at gnuage dot org>)

Originally committed as revision 3578 to svn://svn.ffmpeg.org/ffmpeg/trunk

178fcca8 09/25/2004 11:18 PM Michael Niedermayer

1/2 resolution decoding

Originally committed as revision 3507 to svn://svn.ffmpeg.org/ffmpeg/trunk

e69538fa 09/08/2004 06:49 PM Michael Niedermayer

h264_qpel8_hv_lowpass_mmx2/3dnow

Originally committed as revision 3446 to svn://svn.ffmpeg.org/ffmpeg/trunk

e772bb8a 09/08/2004 06:25 PM Michael Niedermayer

h264_qpel4_hv_lowpass_mmx2/3dnow
alignment fixes

Originally committed as revision 3445 to svn://svn.ffmpeg.org/ffmpeg/trunk

56d8bd56 09/08/2004 12:52 AM Michael Niedermayer

optimization

Originally committed as revision 3440 to svn://svn.ffmpeg.org/ffmpeg/trunk

a6e39f45 09/07/2004 10:56 PM Michael Niedermayer

optimization

Originally committed as revision 3439 to svn://svn.ffmpeg.org/ffmpeg/trunk

ed8ffdf4 09/07/2004 05:22 PM Michael Niedermayer

optimization

Originally committed as revision 3438 to svn://svn.ffmpeg.org/ffmpeg/trunk

437525c4 09/07/2004 01:48 AM Michael Niedermayer

h264 luma motion compensation in mmx2/3dnow

Originally committed as revision 3437 to svn://svn.ffmpeg.org/ffmpeg/trunk

d6af6b03 09/06/2004 03:17 AM Michael Niedermayer

10000l fix and use more mmx2/3dnow code for mpeg4 qpel which has been written and commited long time ago but appearently never used, qpel motion compensation is 5% faster
now

Originally committed as revision 3435 to svn://svn.ffmpeg.org/ffmpeg/trunk

1ec4df0f 06/08/2004 02:13 AM Michael Niedermayer

sse8 and nsse in mmx

Originally committed as revision 3206 to svn://svn.ffmpeg.org/ffmpeg/trunk

e96682e6 05/18/2004 05:09 PM Michael Niedermayer

some of the warning fixes by (Michael Roitzsch <mroi at users dot sourceforge dot net>)

Originally committed as revision 3140 to svn://svn.ffmpeg.org/ffmpeg/trunk

7daabccb 04/27/2004 04:06 AM Mike Melanson

move the 0x80 vector outside of the function, thus saving the compiler
the trouble of having to initialize each byte on the stack individually

Originally committed as revision 3083 to svn://svn.ffmpeg.org/ffmpeg/trunk

f9ed9d85 04/27/2004 03:58 AM Mike Melanson

separate out put_signed_pixels_clamped() into its own function and
implement an optimized MMX version of the function

Originally committed as revision 3082 to svn://svn.ffmpeg.org/ffmpeg/trunk

116824d0 04/26/2004 12:20 AM Mike Melanson

reorganize and simplify the VP3 IDCT stuff

Originally committed as revision 3071 to svn://svn.ffmpeg.org/ffmpeg/trunk

38acbc3c 04/25/2004 03:33 AM Mike Melanson

hook up support for SSE2-optimized VP3 IDCT

Originally committed as revision 3064 to svn://svn.ffmpeg.org/ffmpeg/trunk

01456e8e 03/08/2004 02:18 AM Mike Melanson

use optimized VP3 functions where appropriate

Originally committed as revision 2857 to svn://svn.ffmpeg.org/ffmpeg/trunk

5c0513bd 02/29/2004 10:10 PM Dmitry Baryshkov

attribute used patch by (mitya at school dot ioffe dot ru (Dmitry Baryshkov))

Originally committed as revision 2830 to svn://svn.ffmpeg.org/ffmpeg/trunk

364a1797 02/06/2004 02:12 AM Michael Niedermayer

quantizer noise shaping optimization

Originally committed as revision 2750 to svn://svn.ffmpeg.org/ffmpeg/trunk