Native geekmaster kindle video player

[geekmaster]

Originally Posted by twobob

Well, with these settings:

usage: ./tones -f 4000 -D hw -c 2 -v -r16000 -mdirect_write
and zcat gmvid.gmv.gz | ./gmplay-1.5a

It is possible to get the two halves to run - happily - on the kindle 3 - without dying.

Now, to get some audio piped through it... any thoughts on how best to "interleave / stream / load / whatever " the data out of ffmpeg? I suppose I better go dig up a man page.

I'm looking for RAW data with ideally RAW wav data strapped to it in some sensible fashion. or at worst something I can decode without trampling on someones IP.

I think I could manage ac3 for example. I think.

Anyways thought I would report the extremely low cpu usage in that mode.
I'll see about getting something other than a stupidly modulated sine out of the demo.

Thanks GM

For compatibility with old gmplay, I was thinking the audio (and any other header info) could overlay part of the image. Either the edges, or for less visual annoyance, just change 8x8 dither blocks to use "non-standard" dither patterns to encode bits. But I would rather avoid hacks and just make a new version of gmplay that supports audio and file headers.

For max decode speed and minimum code, I still want 600x800 pre-dithered video, and let zcat to the decompression.

[twobob]

I was trying to figure out the total data load for each frame.

it's bugger all compared to the video.

I think just sandwiching it on for a quickie test now wouldn't hurt as POC.

Then tidy it up once it's proven to work.

Your thoughts? If I'm honest I'm rather struggling to visualise how to manually get the file I need to get cracking decoding/streaming it.

I guess its going to be in the ffmeg stuff perhaps. I'll try to find something to encode not sure I'll have much in the way of video. donations gratefully accepted. Perhaps I can grab an open license one from somewhere. not my area of expertise I'm afraid.

[geekmaster]

Originally Posted by twobob

I was trying to figure out the total data load for each frame.

it's bugger all compared to the video.

I think just sandwiching it on for a quickie test now wouldn't hurt as POC.

Then tidy it up once it's proven to work.

Your thoughts? If I'm honest I'm rather struggling to visualise how to manually get the file I need to get cracking decoding/streaming it.

I guess its going to be in the ffmeg stuff perhaps. I'll try to find something to encode not sure I'll have much in the way of video. donations gratefully accepted. Perhaps I can grab an open license one from somewhere. not my area of expertise I'm afraid.

On the K5, I successfully played x.264 1080p video in realtime using ffmpeg, piped to raw2gmv, piped to gmplay. I had another instance running in the background that piped audio from ffmpeg to aplay. This works because the K5 returns from eink update calls immediately so you can sleep, inside of being stuck in a busy loop in the system call on the K3. I am glad the newer kindles have less CPU-intensive eink updates.

I think I pubished raw2gmv awhile back. Here:
http://www.mobileread.com/forums/sho....php?p=2074379

You would need to interleave audio in raw2gmv (perhaps stdin plus a name pipe in). Then deinterleave in gmplay.

Or, perhaps use two named pipes (/tmp/gmv and /tmp/gma) for raw2gmv input? I think ffmpeg can output to named pipes.

[twobob]

Originally Posted by geekmaster

On the K5, I successfully played x.264 1080p video in realtime using ffmpeg, piped to raw2gmv, piped to gmplay. I had another instance running in the background that piped audio from ffmpeg to aplay. This works because the K5 returns from eink update calls immediately so you can sleep, inside of being stuck in a busy loop in the system call on the K3. I am glad the newer kindles have less CPU-intensive eink updates.

I think I pubished raw2gmv awhile back. EDIT: here: http://www.mobileread.com/forums/sho....php?p=2145300

Let me be clear, I am not wanting to pipe anything.

I want a single file that I will process with a single piece of code, incorporating your video code and the sound code.

It can be made portable after I get it working on the slowest device, which I would suggest is the device to start on anyways. hardest first and all.

[geekmaster]

Originally Posted by twobob

Let me be clear, I am not wanting to pipe anything.

I want a single file that I will process with a single piece of code, incorporating your video code and the sound code.

It can be made portable after I get it working on the slowest device, which I would suggest is the device to start on anyways. hardest first and all.

Yes! I love monolithic noob-friendly .c files. You can tell from the other stuff I published.

BTW, I have a parallel dithering algorithm now that does 32-bits (4-pixels) per operation. The ARM processors prefer 32-bit ops. I have not finished coding it, but on paper it looks good.

[twobob]

Originally Posted by geekmaster

Yes! I love monolithic noob-friendly .c files. You can tell from the other stuff I published.

BTW, I have a parallel dithering algorithm now that does 32-bits (4-pixels) per operation. The ARM processors prefer 32-bit ops. I have not finished coding it, but on paper it looks good.

Yes I was digging through the crunching algorithms you posted. I linked one of the dsp ones myself. Excellent resource, if a little deep for the non maths chops chap.

Well. any pointers to getting the file out of ffmpeg gratefully accepted, I'm reading the man page...

monolithic, yes. Single c file.

I'll play with obfuscated "call" compression later, ALSA is wordy and there is not much avoiding it but I've seen a few nice shorthand examples that I may be able to weld - ahem - glue - into the final result.

[geekmaster]

Crap. I have tons of copy/paste problems in my VM. That link is self-referential. Now I have to search again. A few posts before the raw2gmv download it shows ffmpeg commands to get video raw out of it pipe to raw2gmv...

Okay, here is my video transcoder sub-thread:
http://www.mobileread.com/forums/sho....php?p=2074379

EDIT: Also fixed in the previous post, but not in twobob's quoted copy of the old post.

[twobob]

Originally Posted by geekmaster

Crap. I have tons of copy/paste problems in my VM. That link is self-referential. Now I have to search again. A few posts before the raw2gmv download it shows ffmpeg commands to get video raw out of it pipe to raw2gmv...

wow you really are reading my mind today.

Except I was thinking cant raw to GMV be adjusted to also encode the audio info into the frames?

And to reiterate, for future reference you can assume I have read all your posts on this subject, probably 4 times now.

So I'm thinking

ffmpeg -i /BLAH/sintel-2048-surround.mp4 \
-pix_fmt gray -f rawvideo -vf pad=800:600:0:75 -s 800x450 -|./raw2gmv|nc -l 5555

Should be ffmpeg -i /BLAH/OurTest.mp4 \
-pix_fmt gray -someAudioFlags? -f rawvideo? -vf pad=800:600:0:75 -s 800x450 -|./raw2gmvWithAudio > SomeResultingHelpfulFile.gmv

Am I on the right lines?

[geekmaster]

To do simultaneous dither and render, look at the animation demos. I split them apart here so I could post a tiny video player that did not contain the transcoder, in the tradition of linux tools that you just pipe together into a long one-liner.

My algorithms keep evolving in small increments, each time I post a new demo.

The "no dither table" algorithm was my original creation, using logic tools including Karnaugh maps and DeMorgan's theorem (usually used for digital logic hardware simplification). It was an evolutionary step beyond my "formula 42" dithering that supported all kindle models in that formula. The key idea was to make it cache-friendly by elimination of conditional branching logic that can flush the cache, and later to avoid small cache flushes caused by RAM-resident table lookups. The latest branch-free formula also contains contrast and brightness adjustments. In the future it will do all that, and do 4-pixels per operation at no extra cost (and perhaps some savings making it more than 4 times faster).

[twobob]

Piping introduces an unknown overhead though doesn't it?

I thought the pipe stream needed to be handled in some way internally that put about a .4 second delay in each section using a pipe. (Which, while fine for many purposes, isn't the kind of thing I like to think about when doing audio. All the more reason to use audio striped into the datastream.)

I am probably entirely wrong. I often am.

Anyway, I'm digging. Thanks.

[geekmaster]

Piping is bad because multiprocessing has a relatively large overhead. That is why I am a fan of Simon Tatham's (brain damaging) coroutines, that are evolved from Duff's Device (which I also use a lot). Of course, assembler language stack swaps are better than coroutines, but not portable.

The piping is handy for initial testing (small evolutionary changes to existing code). Once it works, it can be refactored into a monolithic loop (perhaps using a state-machine if it gets too complicated). For monolithic loops that contain subfunctions that need to be performed at different rates, I use a DDA (digital differential algorithm) to do rate conversion to decide when to fire various periodic events.

For really slow primitive processors with tiny RAM but big ROM, I do all my timing with a huge jump table, and interleave all my periodic events evenly through it (dithered to prevent too many events firing on the same clock interrupt). I also do deferred non-maskable interrupt processing after saving the critical measurement. You know -- how to squeeze jaw-dropping performance out of tiny resources.

[twobob]

Originally Posted by geekmaster

To do simultaneous dither and render, look at the animation demos. I split them apart here so I could post a tiny video player that did not contain the transcoder, in the tradition of linux tools that you just pipe together into a long one-liner.

My algorithms keep evolving in small increments, each time I post a new demo.

The "no dither table" algorithm was my original creation, using logic tools including Carnaugh maps and DeMorgan's theorem (usually used for digital logic hardware simplification). It was an evolutionary step beyond my "formula 42" dithering that supported all kindle models in that formula. The key idea was to make it cache-friendly by elimination of conditional branching logic that can flush the cache, and later to avoid small cache flushes caused by RAM-resident table lookups.

Yes, an excellent, and possibly, unique implementation of the concepts.

I do have a very fantastic audio book on its way to throw some fancy maths at and I have more than my fair share of experience in hacking working system together once I have the files. I'm just struggling to visualise exactly the right settings in ffmpeg.

Ah well I'm playing now, Ill figure it out
I too aim to evolve the demos to be more robust and informative and the audio thread to contain at least enough data to get most audio requiring apps off the ground.

It's really not that much info out there at the moment in terms of great demos.
I have trawled google into the floor trying to find further implementations or polling for alsa. Frankly I'm happy to rub shoulders with giants on this one. It's a bit of an arcane art and the calls are all low level stuff but I find that comforting since it will probably result in a few neat ways to do the basic stuff we want and then the rest can be ignored. rather than some unwieldy do-it-all setup system that can not be circumnavigated.

ffmpeg here I come.

[twobob]

Okay I have my output.raw from ffmpeg.
Next up the wav output I guess.

EDIT: FOR THE FUTURE READER I am reading here:
http://ffmpeg.org/ffmpeg.html

Okay so this looks pretty promising:

‘-attach filename (output)’

Add an attachment to the output file.

This is supported by a few formats like Matroska for e.g. fonts used in rendering subtitles. Attachments are implemented as a specific type of stream, so this option will add a new stream to the file.

It is then possible to use per-stream options on this stream in the usual way. Attachment streams created with this option will be created after all the other streams (i.e. those created with -map or automatic mappings).

ffmpeg -i INPUT -attach Mystuff.wav

Am I right in thinking I could weld - err - glue the audio like this? Just a thought, or maybe i'll just cross encode them by hand in the absence of other routes.

So as to not go too OT I will shift the remainder of this dialog to:
http://www.mobileread.com/forums/sho...=183528&page=7

Thanks

[geekmaster]

Just for reference, I posted this in the other thread:

#!/bin/sh
./ffmpeg -d -i $1 -an -pix_fmt gray -sws_flags neighbor -f rawvideo -s 800x600 -|./raw2gmv|./gmplay &
./ffmpeg -d -i $1 -vn -f s16le -acodec pcm_s16le -ar 44100 -|aplay -f cd

Note that it has an "&" on one line, so it runs them both in parallel. This is a common method to output both sound and video from ffmpeg.

[geekmaster]

Originally Posted by twobob

...
So as to not go too OT I will shift the remainder of this dialog to:
http://www.mobileread.com/forums/sho...=183528&page=7

Thanks

Some, discussing A/V issues in a video player thread seems a better place than an "Oops, my bad!" thread. New thread title perhaps? You can ask a site mode to rename it...