The official guide to swscale for confused developers.

Originally committed as revision 15316 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
Michael Niedermayer 2008-09-14 02:38:47 +00:00
parent 70735a3f9e
commit 38d174b375
1 changed files with 98 additions and 0 deletions

98
doc/swscale.txt Normal file
View File

@ -0,0 +1,98 @@
The official guide to swscale for confused developers.
========================================================
Current (simplified) Architecture:
---------------------------------
Input
v
_______OR_________
/ \
/ \
special converter [Input to YUV converter]
| |
| (8bit YUV 4:4:4 / 4:2:2 / 4:2:0 / 4:0:0 )
| |
| v
| Horizontal scaler
| |
| (15bit YUV 4:4:4 / 4:2:2 / 4:2:0 / 4:1:1 / 4:0:0 )
| |
| v
| Vertical scaler and output converter
| |
v v
output
Swscale has 2 scaler pathes, each side must be capable to handle
slices, that is consecutive non overlapping rectangles of dimension
(0,slice_top) - (picture_width, slice_bottom)
special converter
This generally are unscaled converters of common
formats, like YUV 4:2:0/4:2:2 -> RGB15/16/24/32. Though it could also
in principle contain scalers optimized for specific common cases.
Main path
The main path is used when no special converter can be used, the code
is designed as a destination line pull architecture. That is for each
output line the vertical scaler pulls lines from a ring buffer that
when the line is unavailable pulls it from the horizontal scaler and
input converter of the current slice.
When no more output can be generated as lines from a next slice would
be needed then all remaining lines in the current slice are converted
and horizontally scaled and put in the ring buffer.
[this is done for luma and chroma, each with possibly different numbers
of lines per picture]
Input to YUV Converter
When the input to the main path is not planar 8bit per component yuv or
8bit gray then it is converted to planar 8bit YUV, 2 sets of converters
exist for this currently one performing horizontal downscaling by 2
before the convertion and the other leaving the full chroma resolution
but being slightly slower. The scaler will try to preserve full chroma
here when the output uses it, its possible to force full chroma with
SWS_FULL_CHR_H_INP though even for cases where the scaler thinks its
useless.
Horizontal scaler
There are several horizontal scalers, a special case worth mentioning is
the fast bilinear scaler that is made of runtime generated mmx2 code
using specially tuned pshufw instructions.
The remaining scalers are specially tuned for various filter lengths
they scale 8bit unsigned planar data to 16bit signed planar data.
Future >8bit per component inputs will need to add a new scaler here
that preserves the input precission.
Vertical scaler and output converter
There is a large number of combined vertical scalers+output converters
Some are:
* unscaled output converters
* unscaled output converters that average 2 chroma lines
* bilinear converters (C, MMX and accurate MMX)
* arbitrary filter length converters (C, MMX and accurate MMX)
And
* Plain C 8bit 4:2:2 YUV -> RGB converters using LUTs
* Plain C 17bit 4:4:4 YUV -> RGB converters using multiplies
* MMX 11bit 4:2:2 YUV -> RGB converters
* Plain C 16bit Y -> 16bit gray
...
RGB with less than 8bit per component uses dither to improve the
subjective quality and low frequency accuracy.
Filter coefficients:
--------------------
There are several different scalers (bilinear, bicubic, lanczos, area, sinc, ...)
Their coefficients are calculated in initFilter().
Horinzontal filter coeffs have a 1.0 point at 1<<14, vertical ones at 1<<12.
The 1.0 points have been choosen to maximize precission while leaving a
little headroom for convolutional filters like sharpening filters and
minimizing SIMD instructions needed to apply them.
It would be trivial to use a different 1.0 point if some specific scaler
would benefit from it.
Also as already hinted at initFilter() accepts an optional convolutional
filter as input that can be used for contrast, saturation, blur, sharpening
shift, chroma vs. luma shift, ...