January 2012 - ffmpeg-devel - lists.ffmpeg.org

manually hacked configure options increase h264 decoding speed significantly
by madshi 09 Feb '12

09 Feb '12

Hey guys, recently I've noticed that ffdshow decodes faster than my own ffmpeg dlls. After some digging I found out that ffdshow uses a couple of custom configure settings that noticably improve h264 decoding performance. So now I'm wondering why these options are not activated by default. Is there a good reason for that? I'm testing on win32, using MingW/GCC for compiling. Here's the list of changes I'm currently doing manually to config.h: #define HAVE_CMOV 1 #define HAVE_EBP_AVAILABLE 1 #define HAVE_FAST_CLZ 0 #define HAVE_FAST_CMOV 1 #define HAVE_ISATTY 0 #define HAVE_MEMALIGN 1 These are all set differently. The biggest performance gain comes from activating HAVE_EBP_AVAILABLE, but the others seem to help a little bit, too. Furthermore ffdshow also uses a modified mem.c with "__mingw_aligned_malloc", "__mingw_aligned_realloc" and "__mingw_aligned_free". Probably the modified mem.c is necessary for being able to set HAVE_MEMALIGN? Not sure. I can't claim to understand what these options do exactly and why they help h264 decoding performance, but I thought I should post it here and ask for feedback. If there's no disadvantage to these modified options, maybe ffmpeg can use these options by default? Best regards, Mathias.

6 19

[PATCH] doc/filters.texi: merge setdar and setsar documentation
by Stefano Sabatini 06 Feb '12

06 Feb '12

Most of the documentation is common to both filters, and I'm going to add more shared docs. Factorize. --- doc/filters.texi | 60 +++++++++++++++++++++-------------------------------- 1 files changed, 24 insertions(+), 36 deletions(-) diff --git a/doc/filters.texi b/doc/filters.texi index 7d008bc..338632a 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -2323,10 +2323,10 @@ select='gte(t\,10)*lte(t\,20)*eq(pict_type\,I)' select='isnan(prev_selected_t)+gte(t-prev_selected_t\,10)' @end example -@anchor{setdar} -@section setdar +@section setdar, setsar -Set the Display Aspect Ratio for the filter output video. +The @code{setdar} filter sets the Display Aspect Ratio for the filter +output video. This is done by changing the specified Sample (aka Pixel) Aspect Ratio, according to the following equation: @@ -2337,12 +2337,23 @@ the video frame. Also the display aspect ratio set by this filter may be changed by later filters in the filterchain, e.g. in case of scaling or if another "setdar" or a "setsar" filter is applied. -The filter accepts a parameter string which represents the wanted -display aspect ratio. -The parameter can be a floating point number string, or an expression -of the form @var{num}:@var{den}, where @var{num} and @var{den} are the -numerator and denominator of the aspect ratio. -If the parameter is not specified, it is assumed the value "0:1". +The @code{setsar} filter sets the Sample (aka Pixel) Aspect Ratio for +the filter output video. + +Note that as a consequence of the application of this filter, the +output display aspect ratio will change according to the equation +above. + +Keep in mind that the sample aspect ratio set by this filter may be +changed by later filters in the filterchain, e.g. if another "setsar" +or a "setdar" filter is applied. + +The @code{setdar} and @code{setsar} filters accept a parameter string +which represents the wanted display aspect ratio. The parameter can +be a floating point number string, or an expression of the form +@var{num}:@var{den}, where @var{num} and @var{den} are the numerator +and denominator of the aspect ratio. If the parameter is not +specified, it is assumed the value "0:1". For example to change the display aspect ratio to 16:9, specify: @example @@ -2351,7 +2362,10 @@ setdar=16:9 setdar=1.77777 @end example -See also the @ref{setsar} filter documentation. +For example to change the sample aspect ratio to 10:11, specify: +@example +setsar=10:11 +@end example @section setpts @@ -2404,32 +2418,6 @@ setpts=N/(25*TB) setpts='1/(25*TB) * (N + 0.05 * sin(N*2*PI/25))' @end example -@anchor{setsar} -@section setsar - -Set the Sample (aka Pixel) Aspect Ratio for the filter output video. - -Note that as a consequence of the application of this filter, the -output display aspect ratio will change according to the following -equation: -@math{DAR = HORIZONTAL_RESOLUTION / VERTICAL_RESOLUTION * SAR} - -Keep in mind that the sample aspect ratio set by this filter may be -changed by later filters in the filterchain, e.g. if another "setsar" -or a "setdar" filter is applied. - -The filter accepts a parameter string which represents the wanted -sample aspect ratio. -The parameter can be a floating point number string, or an expression -of the form @var{num}:@var{den}, where @var{num} and @var{den} are the -numerator and denominator of the aspect ratio. -If the parameter is not specified, it is assumed the value "0:1". - -For example to change the sample aspect ratio to 10:11, specify: -@example -setsar=10:11 -@end example - @section settb Set the timebase to use for the output frames timestamps. -- 1.7.5.4

2 2

[PATCH] lavu: introduce av_parse_ratio() and use it in ffmpeg and lavfi/aspect
by Stefano Sabatini 06 Feb '12

06 Feb '12

Simplify and provide ratio parsing consistency. --- ffmpeg.c | 34 ++++++++-------------------------- libavfilter/vf_aspect.c | 25 ++++++------------------- libavutil/parseutils.c | 30 ++++++++++++++++++++++++++---- libavutil/parseutils.h | 15 +++++++++++++++ 4 files changed, 55 insertions(+), 49 deletions(-) diff --git a/ffmpeg.c b/ffmpeg.c index 6be1fcb..c9c997d 100644 --- a/ffmpeg.c +++ b/ffmpeg.c @@ -3137,30 +3137,6 @@ static int opt_pad(const char *opt, const char *arg) return -1; } -static double parse_frame_aspect_ratio(const char *arg) -{ - int x = 0, y = 0; - double ar = 0; - const char *p; - char *end; - - p = strchr(arg, ':'); - if (p) { - x = strtol(arg, &end, 10); - if (end == p) - y = strtol(end + 1, &end, 10); - if (x > 0 && y > 0) - ar = (double)x / (double)y; - } else - ar = strtod(arg, NULL); - - if (!ar) { - av_log(NULL, AV_LOG_FATAL, "Incorrect aspect ratio specification.\n"); - exit_program(1); - } - return ar; -} - static int opt_video_channel(const char *opt, const char *arg) { av_log(NULL, AV_LOG_WARNING, "This option is deprecated, use -channel.\n"); @@ -3990,8 +3966,14 @@ static OutputStream *new_video_stream(OptionsContext *o, AVFormatContext *oc) } MATCH_PER_STREAM_OPT(frame_aspect_ratios, str, frame_aspect_ratio, oc, st); - if (frame_aspect_ratio) - ost->frame_aspect_ratio = parse_frame_aspect_ratio(frame_aspect_ratio); + if (frame_aspect_ratio) { + AVRational q; + if (av_parse_ratio(&q, frame_aspect_ratio, 255) < 0 || q.num <= 0 || q.den <= 0) { + av_log(NULL, AV_LOG_FATAL, "Invalid aspect ratio: %s\n", frame_aspect_ratio); + exit_program(1); + } + ost->frame_aspect_ratio = av_q2d(q); + } video_enc->bits_per_raw_sample = frame_bits_per_raw_sample; MATCH_PER_STREAM_OPT(frame_pix_fmts, str, frame_pix_fmt, oc, st); diff --git a/libavfilter/vf_aspect.c b/libavfilter/vf_aspect.c index e3f9299..6029a6c 100644 --- a/libavfilter/vf_aspect.c +++ b/libavfilter/vf_aspect.c @@ -24,6 +24,7 @@ */ #include "libavutil/mathematics.h" +#include "libavutil/parseutils.h" #include "avfilter.h" typedef struct { @@ -33,32 +34,18 @@ typedef struct { static av_cold int init(AVFilterContext *ctx, const char *args, void *opaque) { AspectContext *aspect = ctx->priv; - double ratio; - int64_t gcd; - char c = 0; + int ret; if (args) { - if (sscanf(args, "%d:%d%c", &aspect->aspect.num, &aspect->aspect.den, &c) != 2) - if (sscanf(args, "%lf%c", &ratio, &c) == 1) - aspect->aspect = av_d2q(ratio, 100); - - if (c || aspect->aspect.num <= 0 || aspect->aspect.den <= 0) { + if ((ret = av_parse_ratio(&aspect->aspect, args, 100)) < 0 || + aspect->aspect.num < 0 || aspect->aspect.den <= 0) { av_log(ctx, AV_LOG_ERROR, "Invalid string '%s' for aspect ratio.\n", args); - return AVERROR(EINVAL); + return ret; } - gcd = av_gcd(FFABS(aspect->aspect.num), FFABS(aspect->aspect.den)); - if (gcd) { - aspect->aspect.num /= gcd; - aspect->aspect.den /= gcd; - } + av_log(ctx, AV_LOG_INFO, "a:%d/%d\n", aspect->aspect.num, aspect->aspect.den); } - - if (aspect->aspect.den == 0) - aspect->aspect = (AVRational) {0, 1}; - - av_log(ctx, AV_LOG_INFO, "a:%d/%d\n", aspect->aspect.num, aspect->aspect.den); return 0; } diff --git a/libavutil/parseutils.c b/libavutil/parseutils.c index 2649e3b..32c50bd 100644 --- a/libavutil/parseutils.c +++ b/libavutil/parseutils.c @@ -31,6 +31,31 @@ #include "random_seed.h" #include "parseutils.h" +int av_parse_ratio(AVRational *q, const char *str, int max) +{ + char c; + int ret; + int64_t gcd; + + if (sscanf(str, "%d:%d%c", &q->num, &q->den, &c) != 2) { + double d; + ret = av_expr_parse_and_eval(&d, str, NULL, NULL, + NULL, NULL, NULL, NULL, + NULL, 0, NULL); + if (ret < 0) + return AVERROR(EINVAL); + *q = av_d2q(d, max); + } + + gcd = av_gcd(FFABS(q->num), FFABS(q->den)); + if (gcd) { + q->num /= gcd; + q->den /= gcd; + } + + return 0; +} + typedef struct { const char *abbr; int width, height; @@ -124,7 +149,6 @@ int av_parse_video_rate(AVRational *rate, const char *arg) { int i, ret; int n = FF_ARRAY_ELEMS(video_rate_abbrs); - double res; /* First, we check our abbreviation table */ for (i = 0; i < n; ++i) @@ -134,10 +158,8 @@ int av_parse_video_rate(AVRational *rate, const char *arg) } /* Then, we try to parse it as fraction */ - if ((ret = av_expr_parse_and_eval(&res, arg, NULL, NULL, NULL, NULL, NULL, NULL, - NULL, 0, NULL)) < 0) + if ((ret = av_parse_ratio(rate, arg, 1001000)) < 0) return ret; - *rate = av_d2q(res, 1001000); if (rate->num <= 0 || rate->den <= 0) return AVERROR(EINVAL); return 0; diff --git a/libavutil/parseutils.h b/libavutil/parseutils.h index 2a74a06..3b18b59 100644 --- a/libavutil/parseutils.h +++ b/libavutil/parseutils.h @@ -29,6 +29,21 @@ */ /** + * Parse str and store the parsed ratio in q. + * + * Note tha a ratio with infinite (1/0) or negative value is + * considered value, so you should check on the returned value if you + * want to exclude those values. + * + * @param[in,out] q pointer to the AVRational which will contain the ratio + * @param[in] str the string to parse: it has to be a string in the format + * num:den, a float number or an expression + * @param[in] max the maximum allowed numerator and denominator + * @return >= 0 on success, a negative error code otherwise + */ +int av_parse_ratio(AVRational *q, const char *str, int max); + +/** * Parse str and put in width_ptr and height_ptr the detected values. * * @param[in,out] width_ptr pointer to the variable which will contain the detected -- 1.7.5.4

4 16

[PATCH]v408 decoder and encoder
by Carl Eugen Hoyos 04 Feb '12

04 Feb '12

Hi! Fixes the remaining sample from ticket #470. Please comment, Carl Eugen

3 12

[PATCH 1/3] web: add the security page link to the header
by Michael Niedermayer 04 Feb '12

04 Feb '12

--- src/template_head2 | 3 ++- 1 files changed, 2 insertions(+), 1 deletions(-) diff --git a/src/template_head2 b/src/template_head2 index 63f27ea..e30b5f1 100644 --- a/src/template_head2 +++ b/src/template_head2 @@ -22,7 +22,8 @@ <a href="contact.html">Contact</a> | <a href="consulting.html">Consulting</a> | <a href="projects.html">Projects</a> | -<a href="legal.html">Legal</a> +<a href="legal.html">Legal</a> | +<a href="security.html">Security</a> </div> <div id="body"> -- 1.7.5.4

5 12

warning fixes when compiling with MinGW
by Vincent Torri 03 Feb '12

03 Feb '12

Hey, Here is a patch that fixes some warnings when compiling with mingw: 1) lseek is already defined (os_support.h). 2) some error macros too (network.h) please review Vincent Torri PS: please CC me as i have not subscribed to the ML

2 1

dvd/vob "decoder" part 1: the system itself
by erik＠arawix.com 03 Feb '12

03 Feb '12

Hello; Included are the files, and patches needed for the dvd decoder. I threw away my original system on trying to insert packets into the main stream (as the "attachment/data" type stream). The reason for that is that the dvd data stream, that contains commands for the client can change/insert packets "in the past". Example; a user seeks back. The DVD ifo system will insert packets containing "change the subtitle color" so it works ok for the displayed cell/chapter. LibAV doesn't expect new packets inserted "backwards" in the stream, and the buffer system gets all messed up because packet sizes are off. Changing to a new system means my "hack" on decoding the packet header isn't needed any more, and except for stream flagging the mpegps demuxer should be unchanged. My attachment decoder is not needed anymore (for those who looked at it, only dvd_protocol.c/h are still needed). The new system uses a callback, that the user registers. This is, after much experimenting, the best way I could come up with. The callback though is registered per protocol, and thus changes are needed in "gasp" URLContext. Reason for keeping it per context? Well, a user can be reading from several dvd devices at the same time if he/she uses the avlibs. Or some end devices (like the ps3) send several request to open a file at the same time. If one callback is registered for the whole library, this messes up the system. Whats send in the callback? Basically AVDVDParseContext packets. They contain a type (the enum effdvd_Type ) that tells the end-user app what should be done, and a pts timestamp in dvd time (90000/tick). The packets should be handled at that time. There are some packets which are obvious (like: new subtitle palette) and some that are not. The most important one is effdvd_Reset_IFO. effdvd_Reset_IFO asks the end-user app: clear your streams, flag all streams you are using as invalid (more later on that), and be prepared to re-init all your codec decoders. Why does one need to do that? Because a vob file 1 might contain audio data in ac3 with 2 channels, while a vob 2 might have it in 6 channels. Or the resolution of the video stream might change (from dvd pal to dvd ntsc widescreen for example). First; DVD's may contain up to 64 streams. One of the patches changes the MAX_STREAMS constant to match that. Second; Stream re-init. I implemented this by adding a flag to AVStream (at the end, discard_flags). Set this to one in end-user app, and this will tell the demuxer that this stream has been discarded by user (user closed all codecs, etc). It works by telling the demuxer that once it dedects a certain packet type, and normally does a av_new_stream; don't re-add the stream to the end of the new list (pushing more and more streams on the list). Example; suppose the discard_flag is set to 1 for the video stream (and we only have one). When a new packet is received by the demuxer, its loop will not find a valid video stream anymore. It thus will create a new one, but will fill the slot for the "old" video stream with the new video stream. Don't know if this text makes sense, bit a quick look at the code should be helpful. Why not do this in the protocol reader? Basically 1) because the protocol doesn't and shouldn't know about the streams and 2) because the end-user needs to close and free his/her decoders. Once that's done, the demuxer doesn't even know. Next; chapter seeking. I added a new flag for that: #define AVSEEK_FLAG_DIRECT 8 ///< seek directly, skipping any stamps and translation #define AVSEEK_FLAG_CHAPTER 16 ///< seek to chapter, if supported Using that will basically use av_seek_frame() to override a search like it does for AVSEEK_FLAG_BYTE. The arguments timestamp will thus position chapters, instead of timestamps. Summary; whats the total change; 1) The whole setup adds a protocol handler called ffdvd. User inits his open file with av_open_input_file ("dvdread:e:/") or av_open_input_file("dvdread:/cdrom1/") 2) User registers a callback by using url_fset_protocol_callback(), a new function in avio.c. I tried to make it generic. 3) User polls streams as before for audio/video 4) User can get callback events from the protocol, with timestamps. 5) Events get examined for the timestamps on the protocol packet, and handled upon. 6) When needed, these events are executed, and for some functions a notify is send back (example, when user selects a button dvd_protocol_select_button() is called). 7) That's it. How well does this all work? I build a example app separate from ffmpeg, on which rather big changes are going to be needed. Note that these changes have almost nothing to do with the dvd protocol handler. They are mostly implementing the "we should look at this/this is wrong/etc" comments (example; discarding new streams). I have ffplay ok, but will need to work with someone on ffmpeg. Source is. Offcourse available. I am attaching the main files here, and the most important patches. Since I am not sure on how the continue on this, please tell me how this needs to be handled (ege do patches first, what order,. Etc. ) If the system is accepted, at least one patch needs a major version bump. I should probably split up some patches too. Note 2; I need to talk to the people for libdvdread, since I fixed a few problems in their code (and added one function to read attributes that its not available now dvdnav_get_video_attributes()). To have this working, we would offcourse also need to depend on libdvdread (if the user enabled it by ./configure, changes for makesfiles and that needed). Note 3: we will probably need some commandline extensions to specify seeking to a chapter, etc. for ffmpeg I have worked on and of on this for the better part of 2 months figuring out the best and least intrusive way. If this all gets rejected, I am fine with that. If it doesn't, also. Just let me know. Erik

6 10

Contract work
by Rene von Olten 02 Feb '12

02 Feb '12

Hello Developers, I am a freelance web designer based in France and for one of my clients (a non-commercial educational/game web site for a church organization) I'm looking for the following solution: They have videos hosted on Apache/linux server, 5 to 60 mins duration These videos need to be visitor personalized, each video contains a static scene showing a bulletin board or school board and the name of the visitor needs to be displayed on this school board Videos are encoded on local PC in h264 using Telestream Episode, they are flexible on this and can change to any other video encoder. Important is that they code the videos on local computers and then upload to their remote Apache server. When creating the customized video, to avoid long render times the existing videos should not be re-encoded, only the part with the bulletin board should be re-encoded and the final video segments need to be merged into one video file which can be downloaded by the user or viewed in browser flash or html5 video player. At first for development/testing purposes I will have 3 video segments available at my own server: 1) First part of video 2) Part of video which is static and contains the bulletin board. 3) Last part of video The following steps need to be applied to these segments: Steps: 1) (for example) generate a still image that consists of the user name, for example "David". 2) Merge this still image with the 2nd part of the video so that "David" appears on the bulletin board. This can be a simple merge, no need to worry about alpha channels, masks, etc. (maybe later), encode to new merged output. If you can skip the creating of the still image and merge "David" into the video file directly, great. 3) Merge the 3 parts of the video into one large file without re-encoding part 1 and 3 At a later stage the splitting of the video into three sections before merging the visitor name should be done on the server as well, so there will be one stock video available, your solution splits this into three parts, re-renders part 2 to insert the name, and then merges the 3 parts into 1 video file again. Please let me know if you think you can create the above solution for us and an indication of your rate for this. Regards, René von Olten

2 1

[PATCH] doc: document amerge filter as an alternative for the -map_channel limitation.
by Clément Bœsch 02 Feb '12

02 Feb '12

From: Clément Bœsch <clement.boesch(a)smartjog.com> --- doc/ffmpeg.texi | 19 +++++++++++++++++++ 1 files changed, 19 insertions(+), 0 deletions(-) diff --git a/doc/ffmpeg.texi b/doc/ffmpeg.texi index 83a45a8..f4df331 100644 --- a/doc/ffmpeg.texi +++ b/doc/ffmpeg.texi @@ -833,6 +833,25 @@ possible, for example, to turn two separate mono streams into a single stereo stream. However spliting a stereo stream into two single channel mono streams is possible. +If you need this feature, a possible workaround is to use the @emph{amerge} +filter. For example, if you need to merge a media (here @file{input.mkv}) with +6 mono audio streams into one single 5.1 channel audio stream: + +@example +ffmpeg -f lavfi -i " +amovie=input.mkv:si=0 [a0]; +amovie=input.mkv:si=1 [a1]; +amovie=input.mkv:si=2 [a2]; +amovie=input.mkv:si=3 [a3]; +amovie=input.mkv:si=4 [a4]; +amovie=input.mkv:si=5 [a5]; +[a0][a1] amerge [x0]; +[x0][a2] amerge [x1]; +[x1][a3] amerge [x2]; +[x2][a4] amerge [x3]; +[x3][a5] amerge" -c:a pcm_s16le output.mkv +@end example + @item -map_metadata[:@var{metadata_type}][:@var{index}] @var{infile}[:@var{metadata_type}][:@var{index}] (@emph{output,per-metadata}) Set metadata information of the next output file from @var{infile}. Note that those are file indices (zero-based), not filenames. -- 1.7.7.3

3 18

lavfi: audio filters cleanup and fixes
by Clément Bœsch 02 Feb '12

02 Feb '12

Hi, Thanks to Stefano I was able to understand why -t parameter got ignored: the API needs you to copy props from in to out samples buffer ref, and reset changing props after each filter_samples. In the case of the aresample filter, there is also a need to scale the PTS and redefine the number of samples according to the result of the resampling. The filter is still broken with some MP3, but it works with "clean" input. I'd like to push this patchset ASAP if no one object since it fixes a few things and simplifies a lot the filters.

3 12