I don't have too much time before I have to go, but I mostly worked on new SSE code for the subtitle texture preparation functions. I've also fixed a "text falloff" problem with bitmap subtitles (still needs better aspect ratio correction).
I'll try to involve myself with the development of a better subtitle renderer if I can come to terms with the rest of the people involved (I'm asking for a lot, and I won't cooperate if the subtitle renderer isn't fundamentally changed to finally produce quality output). It's at the xy-vsfilter tracker, and I'll try to read that thread later on (lot of text, no time for it now). http://code.google.com/p/xy-vsfilter/issues/detail?id=40
@ForceX: A difficult situation, and pretty hard to analyze as well. It could be the combination with the FlipEx present mode (which is pretty new). Only I don't remember exactly when I implemented the FlipEx mode (I believe I did mention it somewhere in this thread). I'll take a look at creating a debug build later, maybe that will help to test.
@Hera: With all present modes or some in particular? I believe you were having issues with subtitles before, this has been solved?
I've worked out the (efficient) programming for a set of Lanczos filters and Bézier curves. Regarding the Lanczos filters, I know the base properties of a truncated windowed sinc function, and I was not very surprised when I simulated values for input:
For a simulated value of t = .312 , the weights for the function are:
-3-> 0.01353790110500565952473035467006
-2-> -0.01482150141104846335254841219284
-1-> -0.08627089893640504238179223189432
0-> 0.81387680430780980122826820168191
1-> 0.31372932636473093833162302177254
2-> -0.02780500481463081933622775622197
3-> -0.02055291682427676817514069470158
4-> 0.00582486728065545855547543976717
The weight sum for a Lanczos2 {-1, 0, 1, 2} is: 1.0135302269215048778418712353382 .
The weight sum for a Lanczos3 {-2, -1, 0, 1, 2, 3} is: 0.97815580868617964631418212844374 .
The weight sum for a Lanczos4 {-3, -2, -1, 0, 1, 2, 3, 4} is: 0.99751857707184076439438792288097 .
The current set of filters and the Bézier curves always sum up to 1 (those are simply built that way). My question is: does anyone know of a way to compensate the weight factors of truncated sinc functions? Also, up how big should the filter window be? I've seen versions taking 16 pixels in each sampling direction (Lanczos8, 256 pixels weighed per output pixel), that's just a lot for a filter that weights the last set of pixels by less than ±0.001 (it's even less for the less dramatic base windowed functions than Lanczos).
Lanczos4:// (C) 2011 Jan-Willem Krans (janwillem32 <at> hotmail.com)
// This file is part of Video pixel shader pack.
// This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 2.
// This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
// You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
// prototype Lanczos4 height resizer
// This shader should be run as a screen space pixel shader if you are up-scaling.
// This shader should not be run as a screen space pixel shader if you are down-scaling.
// This shader requires compiling with ps_2_a, but higher is better, see http://en.wikipedia.org/wiki/Pixel_shader to look up what PS version your video card supports.
// This shader is meant to work with linear RGB input and output. Regular R'G'B' with a video gamma encoding will have to be converted with the linear gamma shaders to work properly.
// Use this shader to scale the height of an image by Lanczos4 interpolation.
// fractions, either decimal or not, are allowed
// set the magnification factor
#define Magnify (4/3.)
sampler s0;
float2 c0;
float2 c1;
#define sp(a, b) float4 a = tex2D(s0, float2(tex.x, coord+b*c1.y));
#define PI acos(-1)// this will generate the number PI with full precision
float4 main(float2 tex : TEXCOORD0) : COLOR
{
float coord = (tex.y/Magnify+.5-.5/Magnify)*c0.y;// assign the output position, normalized to texture height in pixels
float t = frac(coord);// calculate the difference between the output pixel and the original surrounding two pixels
coord = (coord-t+.5)*c1.y;// adjust sampling matrix to put the output pixel in the interval [Q1, Q2)
sp(Q3, 0)// nearest original pixel to the top
if (t) {
sp(Q0, -3) sp(Q1, -2) sp(Q2, -1) sp(Q4, 1) sp(Q5, 2) sp(Q6, 3) sp(Q7, 4)// original pixels
float4 wset0 = float4(3, 2, 1, 0)+t;
float4 wset1 = float4(1, 2, 3, 4)-t;
float4 w0 = sin(wset0*PI)*sin(wset0*PI/2.)/(wset0*wset0*PI*PI/2.);
float4 w1 = sin(wset1*PI)*sin(wset1*PI/2.)/(wset1*wset1*PI*PI/2.);
return w0.x*Q0+w0.y*Q1+w0.z*Q2+w0.w*Q3+w1.x*Q4+w1.y*Q5+w1.z*Q6+w1.w*Q7;}// interpolation output
return Q3;// case t == 0 is required to return sample Q3, because of a possible division by 0
}
// (C) 2011 Jan-Willem Krans (janwillem32 <at> hotmail.com)
// This file is part of Video pixel shader pack.
// This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 2.
// This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
// You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
// prototype Lanczos4 width resizer
// This shader should be run as a screen space pixel shader if you are up-scaling.
// This shader should not be run as a screen space pixel shader if you are down-scaling.
// This shader requires compiling with ps_2_a, but higher is better, see http://en.wikipedia.org/wiki/Pixel_shader to look up what PS version your video card supports.
// This shader is meant to work with linear RGB input and output. Regular R'G'B' with a video gamma encoding will have to be converted with the linear gamma shaders to work properly.
// Use this shader to scale the width of an image by Lanczos4 interpolation.
// fractions, either decimal or not, are allowed
// set the magnification factor
#define Magnify (4/3.)
sampler s0;
float c0;
float c1;
#define sp(a, b) float4 a = tex2D(s0, float2(coord+b*c1, tex.y));
#define PI acos(-1)// this will generate the number PI with full precision
float4 main(float2 tex : TEXCOORD0) : COLOR
{
float coord = (tex.x/Magnify+.5-.5/Magnify)*c0;// assign the output position, normalized to texture height in pixels
float t = frac(coord);// calculate the difference between the output pixel and the original surrounding two pixels
coord = (coord-t+.5)*c1;// adjust sampling matrix to put the output pixel in the interval [Q1, Q2)
sp(Q3, 0)// nearest original pixel to the left
if (t) {
sp(Q0, -3) sp(Q1, -2) sp(Q2, -1) sp(Q4, 1) sp(Q5, 2) sp(Q6, 3) sp(Q7, 4)// original pixels
float4 wset0 = float4(3, 2, 1, 0)+t;
float4 wset1 = float4(1, 2, 3, 4)-t;
float4 w0 = sin(wset0*PI)*sin(wset0*PI/2.)/(wset0*wset0*PI*PI/2.);
float4 w1 = sin(wset1*PI)*sin(wset1*PI/2.)/(wset1*wset1*PI*PI/2.);
return w0.x*Q0+w0.y*Q1+w0.z*Q2+w0.w*Q3+w1.x*Q4+w1.y*Q5+w1.z*Q6+w1.w*Q7;}// interpolation output
return Q3;// case t == 0 is required to return sample Q3, because of a possible division by 0
}