summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorBin Jin <bjin1990@gmail.com>2015-10-26 22:43:48 +0000
committerwm4 <wm4@nowhere>2015-11-05 17:38:20 +0100
commit4c43c30421b1d713b7a17b437e381fe1efd01902 (patch)
tree073551ba56ccc34be9c093eabab62cb09b1ca496
parent7438f208c37deb1a30df54278a6d81227038f33e (diff)
downloadmpv-4c43c30421b1d713b7a17b437e381fe1efd01902.tar.bz2
mpv-4c43c30421b1d713b7a17b437e381fe1efd01902.tar.xz
vo_opengl: add Super-xBR filter for upscaling
Add the Super-xBR filter for image doubling, and the prescaling framework to support it. The shader code was ported from MPDN extensions project, with modification to process luma only. This commit is largely inspired by code from #2266, with `gl_transform_trans()` authored by @haasn taken directly.
-rw-r--r--DOCS/man/vo.rst38
-rw-r--r--video/out/opengl/superxbr.c234
-rw-r--r--video/out/opengl/superxbr.h36
-rw-r--r--video/out/opengl/utils.c14
-rw-r--r--video/out/opengl/utils.h2
-rw-r--r--video/out/opengl/video.c216
-rw-r--r--video/out/opengl/video.h4
-rw-r--r--wscript_build.py1
8 files changed, 538 insertions, 7 deletions
diff --git a/DOCS/man/vo.rst b/DOCS/man/vo.rst
index b2a6cd7045..36ca35d51f 100644
--- a/DOCS/man/vo.rst
+++ b/DOCS/man/vo.rst
@@ -549,6 +549,44 @@ Available video output drivers are:
feature doesn't work correctly with different scale factors in
different directions.
+ ``prescale=<filter>``
+ This option provides non-convolution-based filters for upscaling. These
+ filters resize the video to multiple of the original size (all currently
+ supported prescalers can only perform image doubling in a single pass).
+ Generally another convolution based filter (the main scaler) will be
+ applied after prescaler to match the target display size.
+
+ ``none``
+ Disable all prescalers. This is the default.
+
+ ``superxbr``
+ A relatively fast prescaler originally developed for pixel art.
+
+ Some parameters can be tuned with ``superxbr-sharpness`` and
+ ``superxbr-edge-strength`` options.
+
+ Note that all the filters above are designed (or implemented) to process
+ luma plane only and probably won't work as intended for video in RGB
+ format.
+
+ ``prescale-passes=<1..5>``
+ The number of passes to apply the prescaler (defaults to be 1). Setting
+ it to 2 will perform a 4x upscaling.
+
+ ``prescale-downscaling-threshold=<0..32>``
+ This option prevents "overkill" use of prescalers, which can be caused
+ by misconfiguration, or user trying to play a video with much larger
+ size. With this option, user can specify the maximal allowed downscaling
+ ratio in both dimension. To satisfy it, the number of passes for
+ prescaler will be reduced, and if necessary prescaler could also be
+ disabled.
+
+ The default value is 2.0, and should be able to prevent most seemingly
+ unreasonable use of prescalers. Most user would probably want to set it
+ to a smaller value between 1.0 and 1.5 for better performance.
+
+ A value less than 1.0 will disable the check.
+
``pre-shaders=<files>``, ``post-shaders=<files>``, ``scale-shader=<file>``
Custom GLSL fragment shaders.
diff --git a/video/out/opengl/superxbr.c b/video/out/opengl/superxbr.c
new file mode 100644
index 0000000000..a88c058a38
--- /dev/null
+++ b/video/out/opengl/superxbr.c
@@ -0,0 +1,234 @@
+/*
+ * This file is part of mpv.
+ *
+ * mpv 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; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * mpv 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 mpv. If not, see <http://www.gnu.org/licenses/>.
+ *
+ * You can alternatively redistribute this file and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ */
+
+#include "superxbr.h"
+
+#include <assert.h>
+
+#define GLSL(x) gl_sc_add(sc, #x "\n");
+#define GLSLF(...) gl_sc_addf(sc, __VA_ARGS__)
+#define GLSLH(x) gl_sc_hadd(sc, #x "\n");
+#define GLSLHF(...) gl_sc_haddf(sc, __VA_ARGS__)
+
+struct superxbr_opts {
+ float sharpness;
+ float edge_strength;
+};
+
+const struct superxbr_opts superxbr_opts_def = {
+ .sharpness = 1.0f,
+ .edge_strength = 1.0f,
+};
+
+#define OPT_BASE_STRUCT struct superxbr_opts
+const struct m_sub_options superxbr_conf = {
+ .opts = (const m_option_t[]) {
+ OPT_FLOATRANGE("sharpness", sharpness, 0, 0.0, 2.0),
+ OPT_FLOATRANGE("edge-strength", edge_strength, 0, 0.0, 1.0),
+ {0}
+ },
+ .size = sizeof(struct superxbr_opts),
+ .defaults = &superxbr_opts_def,
+};
+
+/*
+
+ ******* Super XBR Shader *******
+
+ Copyright (c) 2015 Hyllian - sergiogdb@gmail.com
+
+ Permission is hereby granted, free of charge, to any person obtaining a copy
+ of this software and associated documentation files (the "Software"), to deal
+ in the Software without restriction, including without limitation the rights
+ to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ copies of the Software, and to permit persons to whom the Software is
+ furnished to do so, subject to the following conditions:
+
+ The above copyright notice and this permission notice shall be included in
+ all copies or substantial portions of the Software.
+
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ THE SOFTWARE.
+
+*/
+
+void pass_superxbr(struct gl_shader_cache *sc, int planes, int tex_num,
+ int step, const struct superxbr_opts *conf,
+ struct gl_transform *transform)
+{
+ assert(0 <= step && step < 2);
+ GLSLF("// superxbr (tex %d, step %d)\n", tex_num, step + 1);
+
+ if (!conf)
+ conf = &superxbr_opts_def;
+
+ if (step == 0) {
+ *transform = (struct gl_transform){{{2.0,0.0}, {0.0,2.0}}, {-0.5,-0.5}};
+
+ GLSLH(#define wp1 2.0)
+ GLSLH(#define wp2 1.0)
+ GLSLH(#define wp3 -1.0)
+ GLSLH(#define wp4 4.0)
+ GLSLH(#define wp5 -1.0)
+ GLSLH(#define wp6 1.0)
+
+ GLSLHF("#define weight1 (%f*1.29633/10.0)\n", conf->sharpness);
+ GLSLHF("#define weight2 (%f*1.75068/10.0/2.0)\n", conf->sharpness);
+
+ GLSLH(#define Get(x, y) (texture(tex, pos + (vec2(x, y) - vec2(0.25, 0.25)) / tex_size)[plane]))
+ } else {
+ *transform = (struct gl_transform){{{1.0,0.0}, {0.0,1.0}}, {0.0,0.0}};
+
+ GLSLH(#define wp1 2.0)
+ GLSLH(#define wp2 0.0)
+ GLSLH(#define wp3 0.0)
+ GLSLH(#define wp4 0.0)
+ GLSLH(#define wp5 0.0)
+ GLSLH(#define wp6 0.0)
+
+ GLSLHF("#define weight1 (%f*1.75068/10.0)\n", conf->sharpness);
+ GLSLHF("#define weight2 (%f*1.29633/10.0/2.0)\n", conf->sharpness);
+
+ GLSLH(#define Get(x, y) (texture(tex, pos + (vec2((x) + (y) - 1, (y) - (x))) / tex_size)[plane]))
+ }
+ GLSLH(float df(float A, float B)
+ {
+ return abs(A-B);
+ })
+
+ GLSLH(float d_wd(float b0, float b1, float c0, float c1, float c2,
+ float d0, float d1, float d2, float d3, float e1,
+ float e2, float e3, float f2, float f3)
+ {
+ return (wp1*(df(c1,c2) + df(c1,c0) + df(e2,e1) + df(e2,e3)) +
+ wp2*(df(d2,d3) + df(d0,d1)) +
+ wp3*(df(d1,d3) + df(d0,d2)) +
+ wp4*df(d1,d2) +
+ wp5*(df(c0,c2) + df(e1,e3)) +
+ wp6*(df(b0,b1) + df(f2,f3)));
+ })
+
+ GLSLH(float hv_wd(float i1, float i2, float i3, float i4,
+ float e1, float e2, float e3, float e4)
+ {
+ return (wp4*(df(i1,i2)+df(i3,i4)) +
+ wp1*(df(i1,e1)+df(i2,e2)+df(i3,e3)+df(i4,e4)) +
+ wp3*(df(i1,e2)+df(i3,e4)+df(e1,i2)+df(e3,i4)));
+ })
+
+ GLSLHF("float superxbr(sampler2D tex, vec2 pos, vec2 tex_size, int plane) {\n");
+
+ if (step == 0) {
+ GLSLH(vec2 dir = fract(pos * tex_size) - 0.5;)
+
+ // Optimization: Discard (skip drawing) unused pixels, except those
+ // at the edge.
+ GLSLH(vec2 dist = tex_size * min(pos, vec2(1.0) - pos);)
+ GLSLH(if (dir.x * dir.y < 0 && dist.x > 1 && dist.y > 1)
+ return 0.0;)
+
+ GLSLH(if (dir.x < 0 || dir.y < 0 || dist.x < 1 || dist.y < 1)
+ return texture(tex, pos - dir / tex_size)[plane];)
+ } else {
+ GLSLH(vec2 dir = fract(pos * tex_size / 2) - 0.5;)
+ GLSLH(if (dir.x * dir.y > 0)
+ return texture(tex, pos)[plane];)
+ }
+
+ GLSLH(float P0 = Get(-1,-1);
+ float P1 = Get( 2,-1);
+ float P2 = Get(-1, 2);
+ float P3 = Get( 2, 2);
+
+ float B = Get( 0,-1);
+ float C = Get( 1,-1);
+ float D = Get(-1, 0);
+ float E = Get( 0, 0);
+ float F = Get( 1, 0);
+ float G = Get(-1, 1);
+ float H = Get( 0, 1);
+ float I = Get( 1, 1);
+
+ float F4 = Get(2, 0);
+ float I4 = Get(2, 1);
+ float H5 = Get(0, 2);
+ float I5 = Get(1, 2);)
+
+/*
+ P1
+ |P0|B |C |P1| C F4 |a0|b1|c2|d3|
+ |D |E |F |F4| B F I4 |b0|c1|d2|e3| |e1|i1|i2|e2|
+ |G |H |I |I4| P0 E A I P3 |c0|d1|e2|f3| |e3|i3|i4|e4|
+ |P2|H5|I5|P3| D H I5 |d0|e1|f2|g3|
+ G H5
+ P2
+*/
+
+ /* Calc edgeness in diagonal directions. */
+ GLSLH(float d_edge = (d_wd( D, B, G, E, C, P2, H, F, P1, H5, I, F4, I5, I4 ) -
+ d_wd( C, F4, B, F, I4, P0, E, I, P3, D, H, I5, G, H5 ));)
+
+ /* Calc edgeness in horizontal/vertical directions. */
+ GLSLH(float hv_edge = (hv_wd(F, I, E, H, C, I5, B, H5) -
+ hv_wd(E, F, H, I, D, F4, G, I4));)
+
+ /* Filter weights. Two taps only. */
+ GLSLH(vec4 w1 = vec4(-weight1, weight1+0.5, weight1+0.5, -weight1);
+ vec4 w2 = vec4(-weight2, weight2+0.25, weight2+0.25, -weight2);)
+
+ /* Filtering and normalization in four direction generating four colors. */
+ GLSLH(float c1 = dot(vec4(P2, H, F, P1), w1);
+ float c2 = dot(vec4(P0, E, I, P3), w1);
+ float c3 = dot(vec4( D+G, E+H, F+I, F4+I4), w2);
+ float c4 = dot(vec4( C+B, F+E, I+H, I5+H5), w2);)
+
+ GLSLHF("float limits = %f + 0.000001;\n", conf->edge_strength);
+ GLSLH(float edge_strength = smoothstep(0.0, limits, abs(d_edge));)
+
+ /* Smoothly blends the two strongest directions(one in diagonal and the
+ * other in vert/horiz direction). */
+ GLSLHF("float color = mix(mix(c1, c2, step(0.0, d_edge)),"
+ "mix(c3, c4, step(0.0, hv_edge)), 1 - %f);\n",
+ conf->edge_strength);
+ /* Anti-ringing code. */
+ GLSLH(float min_sample = min(min(E, F), min(H, I));
+ float max_sample = max(max(E, F), max(H, I));
+ float aux = color;
+ color = clamp(color, min_sample, max_sample);)
+ GLSLHF("color = mix(aux, color, 1-2.0*abs(%f-0.5));\n", conf->edge_strength);
+
+ GLSLH(return color;)
+
+ GLSLHF("}"); // superxbr()
+
+ GLSL(vec4 color = vec4(1.0);)
+
+ for (int i = 0; i < planes; i++) {
+ GLSLF("color[%d] = superxbr(texture%d, texcoord%d, texture_size%d, %d);\n",
+ i, tex_num, tex_num, tex_num, i);
+ }
+}
diff --git a/video/out/opengl/superxbr.h b/video/out/opengl/superxbr.h
new file mode 100644
index 0000000000..46f15fd269
--- /dev/null
+++ b/video/out/opengl/superxbr.h
@@ -0,0 +1,36 @@
+/*
+ * This file is part of mpv.
+ *
+ * mpv 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; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * mpv 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 mpv. If not, see <http://www.gnu.org/licenses/>.
+ *
+ * You can alternatively redistribute this file and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ */
+
+#ifndef MP_GL_SUPERXBR_H
+#define MP_GL_SUPERXBR_H
+
+#include "common.h"
+#include "utils.h"
+
+extern const struct superxbr_opts superxbr_opts_def;
+extern const struct m_sub_options superxbr_conf;
+
+void pass_superxbr(struct gl_shader_cache *sc, int planes, int tex_num,
+ int step, const struct superxbr_opts *conf,
+ struct gl_transform *transform);
+
+#endif
diff --git a/video/out/opengl/utils.c b/video/out/opengl/utils.c
index ac36bdffc4..0026090df9 100644
--- a/video/out/opengl/utils.c
+++ b/video/out/opengl/utils.c
@@ -435,6 +435,18 @@ void gl_transform_ortho(struct gl_transform *t, float x0, float x1,
t->t[1] = -(y1 + y0) / (y1 - y0);
}
+// Apply the effects of one transformation to another, transforming it in the
+// process. In other words: post-composes t onto x
+void gl_transform_trans(struct gl_transform t, struct gl_transform *x)
+{
+ float x00 = x->m[0][0], x01 = x->m[0][1], x10 = x->m[1][0], x11 = x->m[1][1];
+ x->m[0][0] = t.m[0][0] * x00 + t.m[0][1] * x10;
+ x->m[1][0] = t.m[0][0] * x01 + t.m[0][1] * x11;
+ x->m[0][1] = t.m[1][0] * x00 + t.m[1][1] * x10;
+ x->m[1][1] = t.m[1][0] * x01 + t.m[1][1] * x11;
+ gl_transform_vec(t, &x->t[0], &x->t[1]);
+}
+
static void GLAPIENTRY gl_debug_cb(GLenum source, GLenum type, GLuint id,
GLenum severity, GLsizei length,
const GLchar *message, const void *userParam)
@@ -462,7 +474,7 @@ void gl_set_debug_logger(GL *gl, struct mp_log *log)
}
}
-#define SC_ENTRIES 16
+#define SC_ENTRIES 32
#define SC_UNIFORM_ENTRIES 20
enum uniform_type {
diff --git a/video/out/opengl/utils.h b/video/out/opengl/utils.h
index ffcaa4da0f..8682366a24 100644
--- a/video/out/opengl/utils.h
+++ b/video/out/opengl/utils.h
@@ -117,6 +117,8 @@ static inline void gl_transform_rect(struct gl_transform t, struct mp_rect_f *r)
gl_transform_vec(t, &r->x1, &r->y1);
}
+void gl_transform_trans(struct gl_transform t, struct gl_transform *x);
+
void gl_set_debug_logger(GL *gl, struct mp_log *log);
struct gl_shader_cache;
diff --git a/video/out/opengl/video.c b/video/out/opengl/video.c
index 55fafad96e..b69330d1a9 100644
--- a/video/out/opengl/video.c
+++ b/video/out/opengl/video.c
@@ -38,6 +38,7 @@
#include "hwdec.h"
#include "osd.h"
#include "stream/stream.h"
+#include "superxbr.h"
#include "video_shaders.h"
#include "video/out/filter_kernels.h"
#include "video/out/aspect.h"
@@ -48,6 +49,12 @@
// Pixel width of 1D lookup textures.
#define LOOKUP_TEXTURE_SIZE 256
+// Maximal number of passes that prescaler can be applied.
+#define MAX_PRESCALE_PASSES 5
+
+// Maximal number of steps each pass of prescaling contains
+#define MAX_PRESCALE_STEPS 2
+
// scale/cscale arguments that map directly to shader filter routines.
// Note that the convolution filters are not included in this list.
static const char *const fixed_scale_filters[] = {
@@ -166,6 +173,7 @@ struct gl_video {
struct fbotex blend_subs_fbo;
struct fbotex unsharp_fbo;
struct fbotex output_fbo;
+ struct fbotex deband_fbo;
struct fbosurface surfaces[FBOSURFACES_MAX];
// these are duplicated so we can keep rendering back and forth between
@@ -173,6 +181,8 @@ struct gl_video {
struct fbotex pre_fbo[2];
struct fbotex post_fbo[2];
+ struct fbotex prescale_fbo[MAX_PRESCALE_PASSES][MAX_PRESCALE_STEPS];
+
int surface_idx;
int surface_now;
int frames_drawn;
@@ -192,6 +202,7 @@ struct gl_video {
// temporary during rendering
struct src_tex pass_tex[TEXUNIT_VIDEO_NUM];
int texture_w, texture_h;
+ struct gl_transform texture_offset; // texture transform without rotation
bool use_linear;
bool use_normalized_range;
float user_gamma;
@@ -330,6 +341,8 @@ const struct gl_video_opts gl_video_opts_def = {
.alpha_mode = 2,
.background = {0, 0, 0, 255},
.gamma = 1.0f,
+ .prescale_passes = 1,
+ .prescale_downscaling_threshold = 2.0f,
};
const struct gl_video_opts gl_video_opts_hq_def = {
@@ -353,6 +366,8 @@ const struct gl_video_opts gl_video_opts_hq_def = {
.blend_subs = 0,
.pbo = 1,
.deband = 1,
+ .prescale_passes = 1,
+ .prescale_downscaling_threshold = 2.0f,
};
static int validate_scaler_opt(struct mp_log *log, const m_option_t *opt,
@@ -429,6 +444,12 @@ const struct m_sub_options gl_video_conf = {
OPT_FLAG("deband", deband, 0),
OPT_SUBSTRUCT("deband", deband_opts, deband_conf, 0),
OPT_FLOAT("sharpen", unsharp, 0),
+ OPT_CHOICE("prescale", prescale, 0, ({"none", 0}, {"superxbr", 1})),
+ OPT_INTRANGE("prescale-passes",
+ prescale_passes, 0, 1, MAX_PRESCALE_PASSES),
+ OPT_FLOATRANGE("prescale-downscaling-threshold",
+ prescale_downscaling_threshold, 0, 0.0, 32.0),
+ OPT_SUBSTRUCT("superxbr", superxbr_opts, superxbr_conf, 0),
OPT_REMOVED("approx-gamma", "this is always enabled now"),
OPT_REMOVED("cscale-down", "chroma is never downscaled"),
@@ -463,6 +484,7 @@ static void check_gl_features(struct gl_video *p);
static bool init_format(int fmt, struct gl_video *init);
static void gl_video_upload_image(struct gl_video *p, struct mp_image *mpi);
static void assign_options(struct gl_video_opts *dst, struct gl_video_opts *src);
+static void get_scale_factors(struct gl_video *p, double xy[2]);
#define GLSL(x) gl_sc_add(p->sc, #x "\n");
#define GLSLF(...) gl_sc_addf(p->sc, __VA_ARGS__)
@@ -580,12 +602,18 @@ static void uninit_rendering(struct gl_video *p)
fbotex_uninit(&p->indirect_fbo);
fbotex_uninit(&p->blend_subs_fbo);
fbotex_uninit(&p->unsharp_fbo);
+ fbotex_uninit(&p->deband_fbo);
for (int n = 0; n < 2; n++) {
fbotex_uninit(&p->pre_fbo[n]);
fbotex_uninit(&p->post_fbo[n]);
}
+ for (int pass = 0; pass < MAX_PRESCALE_PASSES; pass++) {
+ for (int step = 0; step < MAX_PRESCALE_STEPS; step++)
+ fbotex_uninit(&p->prescale_fbo[pass][step]);
+ }
+
for (int n = 0; n < FBOSURFACES_MAX; n++)
fbotex_uninit(&p->surfaces[n].fbotex);
@@ -1126,6 +1154,125 @@ static void pass_sample(struct gl_video *p, int src_tex, struct scaler *scaler,
GLSL(color.a = 1.0;)
}
+// Get the number of passes for prescaler, with given display size.
+static int get_prescale_passes(struct gl_video *p)
+{
+ if (!p->opts.prescale)
+ return 0;
+ // The downscaling threshold check is turned off.
+ if (p->opts.prescale_downscaling_threshold < 1.0f)
+ return p->opts.prescale_passes;
+
+ double scale_factors[2];
+ get_scale_factors(p, scale_factors);
+
+ int passes = 0;
+ for (; passes < p->opts.prescale_passes; passes ++) {
+ // The scale factor happens to be the same for superxbr and nnedi3.
+ scale_factors[0] /= 2;
+ scale_factors[1] /= 2;
+
+ if (1.0f / scale_factors[0] > p->opts.prescale_downscaling_threshold)
+ break;
+ if (1.0f / scale_factors[1] > p->opts.prescale_downscaling_threshold)
+ break;
+ }
+
+ return passes;
+}
+
+// apply pre-scalers
+static void pass_prescale(struct gl_video *p, int src_tex_num, int dst_tex_num,
+ int planes, int w, int h, int passes,
+ float tex_mul, struct gl_transform *offset)
+{
+ *offset = (struct gl_transform){{{1.0,0.0}, {0.0,1.0}}, {0.0,0.0}};
+
+ int tex_num = src_tex_num;
+
+ // Happens to be the same for superxbr and nnedi3.
+ const int steps_per_pass = 2;
+
+ for (int pass = 0; pass < passes; pass++) {
+ for (int step = 0; step < steps_per_pass; step++) {
+ struct gl_transform transform = {{{0}}};
+
+ switch(p->opts.prescale) {
+ case 1:
+ pass_superxbr(p->sc, planes, tex_num, step,
+ p->opts.superxbr_opts, &transform);
+ break;
+ default:
+ abort();
+ }
+
+ if (tex_mul != 1.0) {
+ GLSLF("color *= %f;\n", tex_mul);
+ tex_mul = 1.0;
+ }
+
+ gl_transform_trans(transform, offset);
+
+ w *= (int)transform.m[0][0];
+ h *= (int)transform.m[1][1];
+
+ finish_pass_fbo(p, &p->prescale_fbo[pass][step],
+ w, h, dst_tex_num, 0);
+ tex_num = dst_tex_num;
+ }
+ }
+}
+
+// Prescale the planes from the main textures.
+static bool pass_prescale_luma(struct gl_video *p, float tex_mul,
+ struct gl_transform *chromafix,
+ struct gl_transform *transform,
+ struct src_tex *prescaled_tex,
+ int *prescaled_planes)
+{
+ // number of passes to apply prescaler, can be zero.
+ int prescale_passes = get_prescale_passes(p);
+
+ if (prescale_passes == 0)
+ return false;
+
+ p->use_normalized_range = true;
+
+ // estimate a safe upperbound of planes being prescaled on texture0.
+ *prescaled_planes = p->is_yuv ? 1 :
+ (!p->color_swizzle[0] || p->color_swizzle[3] == 'a') ? 3 : 4;
+
+ struct src_tex tex_backup[4];
+ for (int i = 0; i < 4; i++)
+ tex_backup[i] = p->pass_tex[i];
+
+ if (p->opts.deband) {
+ // apply debanding before upscaling.
+ pass_sample_deband(p->sc, p->opts.deband_opts, 0, p->gl_target,
+ tex_mul, p->texture_w, p->texture_h, &p->lfg);
+ finish_pass_fbo(p, &p->deband_fbo, p->texture_w,
+ p->texture_h, 0, 0);
+ tex_backup[0] = p->pass_tex[0];
+ }
+
+ // process texture0 and store the result in texture4.
+ pass_prescale(p, 0, 4, *prescaled_planes, p->texture_w, p->texture_h,
+ prescale_passes, p->opts.deband ? 1.0 : tex_mul, transform);
+
+ // correct the chromafix under new transform.
+ chromafix->t[0] -= transform->t[0] / transform->m[0][0];
+ chromafix->t[1] -= transform->t[1] / transform->m[1][1];
+
+ // restore the first four texture.
+ for (int i = 0; i < 4; i++)
+ p->pass_tex[i] = tex_backup[i];
+
+ // backup texture4 for later use.
+ *prescaled_tex = p->pass_tex[4];
+
+ return true;
+}
+
// sample from video textures, set "color" variable to yuv value
static void pass_read_video(struct gl_video *p)
{
@@ -1136,6 +1283,16 @@ static void pass_read_video(struct gl_video *p)
tx_bits = (in_bits + 7) & ~7;
float tex_mul = ((1 << tx_bits) - 1.0) / ((1 << in_bits) - 1.0);
+ struct src_tex prescaled_tex;
+ struct gl_transform offset = {{{0}}};
+ int prescaled_planes;
+
+ bool prescaled = pass_prescale_luma(p, tex_mul, &chromafix, &offset,
+ &prescaled_tex, &prescaled_planes);
+
+ const int scale_factor_x = prescaled ? (int)offset.m[0][0] : 1;
+ const int scale_factor_y = prescaled ? (int)offset.m[1][1] : 1;
+
bool color_defined = false;
if (p->plane_count > 1) {
// Chroma processing (merging -> debanding -> scaling)
@@ -1174,10 +1331,11 @@ static void pass_read_video(struct gl_video *p)
}
// Sample either directly or by upscaling
- if (p->image_desc.flags & MP_IMGFLAG_SUBSAMPLED) {
+ if ((p->image_desc.flags & MP_IMGFLAG_SUBSAMPLED) || prescaled) {
GLSLF("// chroma scaling\n");
pass_sample(p, 1, &p->scaler[2], cscale, 1.0,
- p->texture_w, p->texture_h, chromafix);
+ p->texture_w * scale_factor_x,
+ p->texture_h * scale_factor_y, chromafix);
GLSL(vec2 chroma = color.xy;)
color_defined = true; // pass_sample defines vec4 color
} else {
@@ -1200,12 +1358,20 @@ static void pass_read_video(struct gl_video *p)
// stuff
GLSL(vec4 main;)
GLSLF("{\n");
- if (p->opts.deband) {
+ if (!prescaled && p->opts.deband) {
pass_sample_deband(p->sc, p->opts.deband_opts, 0, p->gl_target, tex_mul,
p->texture_w, p->texture_h, &p->lfg);
p->use_normalized_range = true;
} else {
- GLSL(vec4 color = texture(texture0, texcoord0);)
+ if (!prescaled) {
+ GLSL(vec4 color = texture(texture0, texcoord0);)
+ } else {
+ // just use bilinear for non-essential planes.
+ GLSLF("vec4 color = texture(texture0, "
+ "texcoord0 + vec2(%f,%f) / texture_size0);\n",
+ -offset.t[0] / scale_factor_x,
+ -offset.t[1] / scale_factor_y);
+ }
if (p->use_normalized_range)
GLSLF("color *= %f;\n", tex_mul);
}
@@ -1214,13 +1380,33 @@ static void pass_read_video(struct gl_video *p)
// Set up the right combination of planes
GLSL(color = main;)
+ if (prescaled) {
+ // Restore texture4 and merge it into the main texture.
+ p->pass_tex[4] = prescaled_tex;
+
+ const char* planes_to_copy = "abgr" + 4 - prescaled_planes;
+ GLSLF("color.%s = texture(texture4, texcoord4).%s;\n",
+ planes_to_copy, planes_to_copy);
+
+ p->texture_w *= scale_factor_x;
+ p->texture_h *= scale_factor_y;
+ gl_transform_trans(offset, &p->texture_offset);
+ }
if (p->plane_count > 1)
GLSL(color.yz = chroma;)
if (p->has_alpha && p->plane_count >= 4) {
- GLSL(color.a = texture(texture3, texcoord3).r;)
+ if (!prescaled) {
+ GLSL(color.a = texture(texture3, texcoord3).r;)
+ } else {
+ GLSLF("color.a = texture(texture3, "
+ "texcoord3 + vec2(%f,%f) / texture_size3).r;",
+ -offset.t[0] / scale_factor_x,
+ -offset.t[1] / scale_factor_y);
+ }
if (p->use_normalized_range)
GLSLF("color.a *= %f;\n", tex_mul);
}
+
}
// yuv conversion, and any other conversions before main up/down-scaling
@@ -1327,6 +1513,8 @@ static void compute_src_transform(struct gl_video *p, struct gl_transform *tr,
oy = p->src_rect.y0;
struct gl_transform transform = {{{sx,0.0}, {0.0,sy}}, {ox,oy}};
+ gl_transform_trans(p->texture_offset, &transform);
+
int xc = 0, yc = 1;
*vp_w = p->dst_rect.x1 - p->dst_rect.x0,
*vp_h = p->dst_rect.y1 - p->dst_rect.y0;
@@ -1348,14 +1536,23 @@ static void pass_scale_main(struct gl_video *p)
// Figure out the main scaler.
double xy[2];
get_scale_factors(p, xy);
+
+ // actual scale factor should be divided by the scale factor of prescaling.
+ xy[0] /= p->texture_offset.m[0][0];
+ xy[1] /= p->texture_offset.m[1][1];
+
bool downscaling = xy[0] < 1.0 || xy[1] < 1.0;
bool upscaling = !downscaling && (xy[0] > 1.0 || xy[1] > 1.0);
double scale_factor = 1.0;
struct scaler *scaler = &p->scaler[0];
struct scaler_config scaler_conf = p->opts.scaler[0];
- if (p->opts.scaler_resizes_only && !downscaling && !upscaling)
+ if (p->opts.scaler_resizes_only && !downscaling && !upscaling) {
scaler_conf.kernel.name = "bilinear";
+ // bilinear is going to be used, just remove all sub-pixel offsets.
+ p->texture_offset.t[0] = (int)p->texture_offset.t[0];
+ p->texture_offset.t[1] = (int)p->texture_offset.t[1];
+ }
if (downscaling && p->opts.scaler[1].kernel.name) {
scaler_conf = p->opts.scaler[1];
scaler = &p->scaler[1];
@@ -1644,6 +1841,7 @@ static void pass_render_frame(struct gl_video *p)
// initialize the texture parameters
p->texture_w = p->image_params.w;
p->texture_h = p->image_params.h;
+ p->texture_offset = (struct gl_transform){{{1.0,0.0}, {0.0,1.0}}, {0.0,0.0}};
if (p->opts.dumb_mode)
return;
@@ -2509,12 +2707,18 @@ static void assign_options(struct gl_video_opts *dst, struct gl_video_opts *src)
talloc_free(dst->pre_shaders);
talloc_free(dst->post_shaders);
talloc_free(dst->deband_opts);
+ talloc_free(dst->superxbr_opts);
*dst = *src;
if (src->deband_opts)
dst->deband_opts = m_sub_options_copy(NULL, &deband_conf, src->deband_opts);
+ if (src->superxbr_opts) {
+ dst->superxbr_opts = m_sub_options_copy(NULL, &superxbr_conf,
+ src->superxbr_opts);
+ }
+
for (int n = 0; n < 4; n++) {
dst->scaler[n].kernel.name =
(char *)handle_scaler_opt(dst->scaler[n].kernel.name, n == 3);
diff --git a/video/out/opengl/video.h b/video/out/opengl/video.h
index 2d9e9abfca..0ed7b7cb41 100644
--- a/video/out/opengl/video.h
+++ b/video/out/opengl/video.h
@@ -98,6 +98,10 @@ struct gl_video_opts {
int deband;
struct deband_opts *deband_opts;
float unsharp;
+ int prescale;
+ int prescale_passes;
+ float prescale_downscaling_threshold;
+ struct superxbr_opts *superxbr_opts;
};
extern const struct m_sub_options gl_video_conf;
diff --git a/wscript_build.py b/wscript_build.py
index c410c3724d..b6836a7beb 100644
--- a/wscript_build.py
+++ b/wscript_build.py
@@ -325,6 +325,7 @@ def build(ctx):
( "video/out/opengl/hwdec_vdpau.c", "vdpau-gl-x11" ),
( "video/out/opengl/lcms.c", "gl" ),
( "video/out/opengl/osd.c", "gl" ),
+ ( "video/out/opengl/superxbr.c", "gl" ),
( "video/out/opengl/utils.c", "gl" ),
( "video/out/opengl/video.c", "gl" ),
( "video/out/opengl/video_shaders.c", "gl" ),