RepoMirrors/mpv
1
0
Fork 0
mirror of https://github.com/mpv-player/mpv synced 2025-02-12 18:07:12 +00:00
Code Issues Releases Wiki Activity

vo_opengl: unroll ewa_lanczos to avoid looping and unnecessary samples

Browse Source 
This speeds up performance by a factor of something like 10%,
since it omits unnecessary checks.

This will also make adding anti-ringing easier.
This commit is contained in:
Niklas Haas 2015-01-18 17:41:49 +01:00
parent f5e48f0235
commit 6c250505fe
2 changed files with 29 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
video/out/gl_video.c
View File

@ -954,9 +954,29 @@ static void shader_setup_scaler(char **shader, struct scaler *scaler, int pass)
APPENDF(shader, "#define DEF_SCALER%d \\\n ", unit); APPENDF(shader, "#define DEF_SCALER%d \\\n ", unit);
char lut_fn[40]; char lut_fn[40];
if (scaler->kernel->polar) { if (scaler->kernel->polar) {
int radius = (int)scaler->kernel->radius;
// SAMPLE_CONVOLUTION_POLAR_R(NAME, R, LUT) // SAMPLE_CONVOLUTION_POLAR_R(NAME, R, LUT)
APPENDF(shader, "SAMPLE_CONVOLUTION_POLAR_R(%s, %d, %s)\n", APPENDF(shader, "SAMPLE_CONVOLUTION_POLAR_R(%s, %d, %s, WEIGHTS%d)\n",
name, (int)scaler->kernel->radius, lut_tex); name, radius, lut_tex, unit);
// Pre-compute unrolled weights matrix
APPENDF(shader, "#define WEIGHTS%d(LUT) \\\n ", unit);
for (int y = 1-radius; y <= radius; y++) {
for (int x = 1-radius; x <= radius; x++) {
// Since we can't know the subpixel position in advance,
// assume a worst case scenario.
int yy = y > 0 ? y-1 : y;
int xx = x > 0 ? x-1 : x;
double d = sqrt(xx*xx + yy*yy);
// Samples outside the radius are unnecessary
if (d < radius) {
APPENDF(shader, "SAMPLE_POLAR(LUT, %f, %d, %d) \\\n ",
(double)radius, x, y);
}
}
}
APPENDF(shader, "\n");
} else { } else {
if (size == 2 || size == 6) { if (size == 2 || size == 6) {
snprintf(lut_fn, sizeof(lut_fn), "weights%d", size); snprintf(lut_fn, sizeof(lut_fn), "weights%d", size);

15
video/out/gl_video_shaders.glsl
View File

@ -298,21 +298,20 @@ float[6] weights6(sampler2D lookup, float f) {
return res; \ return res; \
} }
#define SAMPLE_POLAR(LUT, R, X, Y) \
w = texture1D(LUT, length(vec2(X, Y) - fcoord)/R).r; \
wsum += w; \
res += w * texture(tex, base + pt * vec2(X, Y)); \
#define SAMPLE_CONVOLUTION_POLAR_R(NAME, R, LUT) \ #define SAMPLE_CONVOLUTION_POLAR_R(NAME, R, LUT, WEIGHTS_FN) \
vec4 NAME(VIDEO_SAMPLER tex, vec2 texsize, vec2 texcoord) { \ vec4 NAME(VIDEO_SAMPLER tex, vec2 texsize, vec2 texcoord) { \
vec2 pt = vec2(1.0) / texsize; \ vec2 pt = vec2(1.0) / texsize; \
vec2 fcoord = fract(texcoord * texsize - vec2(0.5)); \ vec2 fcoord = fract(texcoord * texsize - vec2(0.5)); \
vec2 base = texcoord - fcoord * pt; \ vec2 base = texcoord - fcoord * pt; \
vec4 res = vec4(0); \ vec4 res = vec4(0); \
float wsum = 0; \ float wsum = 0; \
for (int y = 1-R; y <= R; y++) { \ float w; \
for (int x = 1-R; x <= R; x++) { \ WEIGHTS_FN(LUT); \
float w = texture1D(LUT, length(vec2(x,y) - fcoord)/R).r; \
wsum += w; \
res += w * texture(tex, base + pt * vec2(x, y)); \
} \
} \
return res / wsum; \ return res / wsum; \
} }