[Ffmpeg-devel] [PATCH] faad decoding error return

Mike Melanson mike
Wed Mar 8 02:42:14 CET 2006


Jindrich Makovicka wrote:
> well... what about dropping half a megabyte for a start? ;)
> 
> iq_table[i] = pow(i, 4.0/3.0);

	That's a good start. That would obviate the need for iq_table.h (~250 
KB). Other obvious tables:

* sine_win.h (~150 KB) which is a bunch of sin() values
* cfft_tab.h (~117 KB) has a bunch of complex constants for FFT (cos() 
and sin() values)
* ic_predict.h (~15 KB) has a table of inverse exponent (right term?) 
values (1/2, 1/4, 1/8, etc.), as well as another table I am not sure of 
(named mnt_table, 128 values descending from .953  -> .479)
* kbd_win.h (~85 KB) has tables named kbd wich run from 0.0 -> 1.0; what 
does kbd mean?
* sbr_noise.h (~37 KB) comes from the official spec but must be based 
off of some known random function with a known seed
* sbr_qmf_c.h (~22 KB) contains a qmf array-- does anyone know what that 
means?
* ssr_win.h (~14 KB) more sine and kbd (?) tables

Wondering why the author(s) chose to transport these tables, I can only 
assume they were targeting platforms that had no floating point units 
and did not even have the support libraries to emulate the functions in 
software. Or perhaps on platforms with flawed FP functionality. I think 
FFmpeg at least assumes that sin() and cos() will be available. And we 
traditionally have been apathetic toward platforms deemed "broken". So 
it shouldn't be a problem if we want to generate the tables at runtime.

	Then there are the Huffman tables. There is no getting around having to 
transport those. However, the FAAD source uses some roundabout process 
for decoding them (or perhaps just codifies a double cascaded Huffman 
table; I am still working through the details). I wonder if we can unify 
the tables and let FFmpeg's internal Huffman facilities worry about the 
heavy lifting?

	These are just some things I have been investigating. I wanted to let 
people know that the AAC task might actually be manageable. :)

-- 
	-Mike Melanson





More information about the ffmpeg-devel mailing list