Mirrors/mpv
1
0
Fork 0
mirror of https://github.com/mpv-player/mpv.git synced 2025-12-22 10:57:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

vo_opengl: refactor shader generation (part 1)

Browse Source 
The basic idea is to use dynamically generated shaders instead of a
single monolithic file + a ton of ifdefs. Instead of having to setup
every aspect of it separately (like compiling shaders, setting uniforms,
perfoming the actual rendering steps, the GLSL parts), we generate the
GLSL on the fly, and perform the rendering at the same time. The GLSL
is regenerated every frame, but the actual compiled OpenGL-level shaders
are cached, which makes it fast again. Almost all logic can be in a
single place.

The new code is significantly more flexible, which allows us to improve
the code clarity, performance and add more features easily.

This commit is incomplete. It drops almost all previous code, and
readds only the most important things (some of them actually buggy).
The next commit will complete it - it's separate to preserve authorship
information.
This commit is contained in:
wm4
2015-03-12 21:57:54 +01:00
parent ae6019cbc9
commit e74a4d5bc0
12 changed files with 1305 additions and 2025 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@@ -14,7 +14,6 @@
/input/input_conf.h
/tags
/TAGS
/video/out/gl_video_shaders.h
/video/out/x11_icon.inc
/demux/ebml_defs.c
/demux/ebml_types.h

5
old-makefile
View File

@@ -358,10 +358,6 @@ demux/ebml.c: demux/ebml_defs.c
demux/ebml_defs.c: TOOLS/matroska.pl $(MKVLIB_DEPS)
./$< --generate-definitions > $@
video/out/gl_video.c: video/out/gl_video_shaders.h
video/out/gl_video_shaders.h: TOOLS/file2string.pl video/out/gl_video_shaders.glsl
./$^ >$@
video/out/x11_common.c: video/out/x11_icon.inc
video/out/x11_icon.inc: TOOLS/file2string.pl video/out/x11_icon.bin
./$^ >$@
@@ -475,7 +471,6 @@ clean:
-$(RM) input/input_conf.h
-$(RM) video/out/vdpau_template.c
-$(RM) demux/ebml_types.h demux/ebml_defs.c
-$(RM) video/out/gl_video_shaders.h
-$(RM) video/out/x11_icon.inc
-$(RM) sub/osd_font.h
-$(RM) player/lua/defaults.inc

202
video/out/gl_osd.c
View File

@@ -55,20 +55,46 @@ static const struct osd_fmt_entry osd_to_gl2_formats[SUBBITMAP_COUNT] = {
struct vertex {
float position[2];
uint8_t color[4];
float texcoord[2];
uint8_t ass_color[4];
};
static const struct gl_vao_entry vertex_vao[] = {
{"vertex_position", 2, GL_FLOAT, false, offsetof(struct vertex, position)},
{"vertex_color", 4, GL_UNSIGNED_BYTE, true, offsetof(struct vertex, color)},
{"vertex_texcoord", 2, GL_FLOAT, false, offsetof(struct vertex, texcoord)},
{"position", 2, GL_FLOAT, false, offsetof(struct vertex, position)},
{"texcoord" , 2, GL_FLOAT, false, offsetof(struct vertex, texcoord)},
{"ass_color", 4, GL_UNSIGNED_BYTE, true, offsetof(struct vertex, ass_color)},
{0}
};
// programs: SUBBITMAP_COUNT elements
struct mpgl_osd *mpgl_osd_init(GL *gl, struct mp_log *log, struct osd_state *osd,
GLuint *programs)
struct mpgl_osd_part {
enum sub_bitmap_format format;
int bitmap_id, bitmap_pos_id;
GLuint texture;
int w, h;
GLuint buffer;
int num_subparts;
struct sub_bitmap *subparts;
struct vertex *vertices;
struct bitmap_packer *packer;
};
struct mpgl_osd {
struct mp_log *log;
struct osd_state *osd;
GL *gl;
bool use_pbo;
bool scaled;
struct mpgl_osd_part *parts[MAX_OSD_PARTS];
const struct osd_fmt_entry *fmt_table;
bool formats[SUBBITMAP_COUNT];
struct gl_vao vao;
// temporary
int stereo_mode;
int display_size[2];
void *scratch;
};
struct mpgl_osd *mpgl_osd_init(GL *gl, struct mp_log *log, struct osd_state *osd)
{
GLint max_texture_size;
gl->GetIntegerv(GL_MAX_TEXTURE_SIZE, &max_texture_size);
@@ -79,7 +105,6 @@ struct mpgl_osd *mpgl_osd_init(GL *gl, struct mp_log *log, struct osd_state *osd
.osd = osd,
.gl = gl,
.fmt_table = osd_to_gl3_formats,
.programs = programs,
.scratch = talloc_zero_size(ctx, 1),
};
@@ -126,6 +151,11 @@ void mpgl_osd_destroy(struct mpgl_osd *ctx)
talloc_free(ctx);
}
void mpgl_osd_set_options(struct mpgl_osd *ctx, bool pbo)
{
ctx->use_pbo = pbo;
}
static bool upload_pbo(struct mpgl_osd *ctx, struct mpgl_osd_part *osd,
struct sub_bitmaps *imgs)
{
@@ -154,8 +184,7 @@ static bool upload_pbo(struct mpgl_osd *ctx, struct mpgl_osd_part *osd,
if (!gl->UnmapBuffer(GL_PIXEL_UNPACK_BUFFER))
success = false;
glUploadTex(gl, GL_TEXTURE_2D, fmt.format, fmt.type, NULL, stride,
bb[0].x, bb[0].y, bb[1].x - bb[0].x, bb[1].y - bb[0].y,
0);
bb[0].x, bb[0].y, bb[1].x - bb[0].x, bb[1].y - bb[0].y, 0);
}
gl->BindBuffer(GL_PIXEL_UNPACK_BUFFER, 0);
@@ -240,11 +269,12 @@ static bool upload_osd(struct mpgl_osd *ctx, struct mpgl_osd_part *osd,
return true;
}
static struct mpgl_osd_part *mpgl_osd_generate(struct mpgl_osd *ctx,
struct sub_bitmaps *imgs)
static void gen_osd_cb(void *pctx, struct sub_bitmaps *imgs)
{
struct mpgl_osd *ctx = pctx;
if (imgs->num_parts == 0 || !ctx->formats[imgs->format])
return NULL;
return;
struct mpgl_osd_part *osd = ctx->parts[imgs->render_index];
@@ -256,83 +286,76 @@ static struct mpgl_osd_part *mpgl_osd_generate(struct mpgl_osd *ctx,
osd->bitmap_id = imgs->bitmap_id;
osd->bitmap_pos_id = imgs->bitmap_pos_id;
osd->num_vertices = 0;
}
osd->num_subparts = osd->packer->count;
return osd->packer->count ? osd : NULL;
MP_TARRAY_GROW(osd, osd->subparts, osd->num_subparts);
memcpy(osd->subparts, imgs->parts,
osd->num_subparts * sizeof(osd->subparts[0]));
}
static void write_quad(struct vertex *va,
static void write_quad(struct vertex *va, float matrix[3][3],
float x0, float y0, float x1, float y1,
float tx0, float ty0, float tx1, float ty1,
float tex_w, float tex_h, const uint8_t color[4])
{
gl_matrix_mul_vec(matrix, &x0, &y0);
gl_matrix_mul_vec(matrix, &x1, &y1);
#define COLOR_INIT {color[0], color[1], color[2], color[3]}
va[0] = (struct vertex){ {x0, y0}, COLOR_INIT, {tx0 / tex_w, ty0 / tex_h} };
va[1] = (struct vertex){ {x0, y1}, COLOR_INIT, {tx0 / tex_w, ty1 / tex_h} };
va[2] = (struct vertex){ {x1, y0}, COLOR_INIT, {tx1 / tex_w, ty0 / tex_h} };
va[3] = (struct vertex){ {x1, y1}, COLOR_INIT, {tx1 / tex_w, ty1 / tex_h} };
va[0] = (struct vertex){ {x0, y0}, {tx0 / tex_w, ty0 / tex_h}, COLOR_INIT };
va[1] = (struct vertex){ {x0, y1}, {tx0 / tex_w, ty1 / tex_h}, COLOR_INIT };
va[2] = (struct vertex){ {x1, y0}, {tx1 / tex_w, ty0 / tex_h}, COLOR_INIT };
va[3] = (struct vertex){ {x1, y1}, {tx1 / tex_w, ty1 / tex_h}, COLOR_INIT };
va[4] = va[2];
va[5] = va[1];
#undef COLOR_INIT
}
static void draw_osd_cb(void *pctx, struct sub_bitmaps *imgs)
static int generate_verts(struct mpgl_osd_part *part, float matrix[3][3])
{
struct mpgl_osd *ctx = pctx;
GL *gl = ctx->gl;
struct mpgl_osd_part *part = mpgl_osd_generate(ctx, imgs);
if (!part)
return;
assert(part->format != SUBBITMAP_EMPTY);
if (!part->num_vertices) {
part->vertices = talloc_realloc(part, part->vertices, struct vertex,
part->packer->count * 6);
int num_vertices = part->num_subparts * 6;
MP_TARRAY_GROW(part, part->vertices, num_vertices);
for (int n = 0; n < part->num_subparts; n++) {
struct sub_bitmap *b = &part->subparts[n];
struct pos pos = part->packer->result[n];
struct vertex *va = part->vertices;
for (int n = 0; n < part->packer->count; n++) {
struct sub_bitmap *b = &imgs->parts[n];
struct pos pos = part->packer->result[n];
// NOTE: the blend color is used with SUBBITMAP_LIBASS only, so it
// doesn't matter that we upload garbage for the other formats
uint32_t c = b->libass.color;
uint8_t color[4] = { c >> 24, (c >> 16) & 0xff,
(c >> 8) & 0xff, 255 - (c & 0xff) };
// NOTE: the blend color is used with SUBBITMAP_LIBASS only, so it
// doesn't matter that we upload garbage for the other formats
uint32_t c = b->libass.color;
uint8_t color[4] = { c >> 24, (c >> 16) & 0xff,
(c >> 8) & 0xff, 255 - (c & 0xff) };
write_quad(&va[part->num_vertices],
b->x, b->y, b->x + b->dw, b->y + b->dh,
pos.x, pos.y, pos.x + b->w, pos.y + b->h,
part->w, part->h, color);
part->num_vertices += 6;
}
write_quad(&va[n * 6], matrix,
b->x, b->y, b->x + b->dw, b->y + b->dh,
pos.x, pos.y, pos.x + b->w, pos.y + b->h,
part->w, part->h, color);
}
return num_vertices;
}
static void draw_part(struct mpgl_osd *ctx, int index, float matrix[3][3])
{
GL *gl = ctx->gl;
struct mpgl_osd_part *part = ctx->parts[index];
int num_vertices = generate_verts(part, matrix);
if (!num_vertices)
return;
gl->Enable(GL_BLEND);
gl->BindTexture(GL_TEXTURE_2D, part->texture);
const int *factors = &blend_factors[part->format][0];
gl->BlendFuncSeparate(factors[0], factors[1], factors[2], factors[3]);
int program = ctx->programs[part->format];
gl->UseProgram(program);
gl_vao_draw_data(&ctx->vao, GL_TRIANGLES, part->vertices, num_vertices);
bool set_offset = ctx->offset[0] != 0.0f || ctx->offset[1] != 0.0f;
if (set_offset) {
gl->Uniform3f(gl->GetUniformLocation(program, "translation"),
ctx->offset[0], ctx->offset[1], 0);
}
gl_vao_draw_data(&ctx->vao, GL_TRIANGLES, part->vertices, part->num_vertices);
if (set_offset)
gl->Uniform3f(gl->GetUniformLocation(program, "translation"), 0, 0, 0);
gl->UseProgram(0);
gl->BindTexture(GL_TEXTURE_2D, 0);
gl->Disable(GL_BLEND);
}
// number of screen divisions per axis (x=0, y=1) for the current 3D mode
@@ -347,26 +370,51 @@ static void get_3d_side_by_side(int stereo_mode, int div[2])
}
}
void mpgl_osd_draw(struct mpgl_osd *ctx, struct mp_osd_res res, double pts,
int stereo_mode)
void mpgl_osd_draw_part(struct mpgl_osd *ctx, int vp_w, int vp_h, int index)
{
GL *gl = ctx->gl;
int div[2];
get_3d_side_by_side(ctx->stereo_mode, div);
gl->Enable(GL_BLEND);
for (int x = 0; x < div[0]; x++) {
for (int y = 0; y < div[1]; y++) {
float matrix[3][3];
gl_matrix_ortho2d(matrix, 0, vp_w, 0, vp_h);
float a_x = ctx->display_size[0] * x;
float a_y = ctx->display_size[1] * y;
matrix[2][0] += a_x * matrix[0][0] + a_y * matrix[1][0];
matrix[2][1] += a_x * matrix[0][1] + a_y * matrix[1][1];
draw_part(ctx, index, matrix);
}
}
}
enum sub_bitmap_format mpgl_osd_get_part_format(struct mpgl_osd *ctx, int index)
{
assert(index >= 0 && index < MAX_OSD_PARTS);
return ctx->parts[index]->format;
}
struct gl_vao *mpgl_osd_get_vao(struct mpgl_osd *ctx)
{
return &ctx->vao;
}
void mpgl_osd_generate(struct mpgl_osd *ctx, struct mp_osd_res res, double pts,
int stereo_mode)
{
for (int n = 0; n < MAX_OSD_PARTS; n++)
ctx->parts[n]->num_subparts = 0;
int div[2];
get_3d_side_by_side(stereo_mode, div);
for (int x = 0; x < div[0]; x++) {
for (int y = 0; y < div[1]; y++) {
struct mp_osd_res s_res = res;
s_res.w /= div[0];
s_res.h /= div[1];
ctx->offset[0] = s_res.w * x;
ctx->offset[1] = s_res.h * y;
osd_draw(ctx->osd, s_res, pts, 0, ctx->formats, draw_osd_cb, ctx);
}
}
struct mp_osd_res s_res = res;
ctx->display_size[0] = s_res.w = s_res.w / div[0];
ctx->display_size[1] = s_res.h = s_res.h / div[1];
gl->Disable(GL_BLEND);
osd_draw(ctx->osd, s_res, pts, 0, ctx->formats, gen_osd_cb, ctx);
ctx->stereo_mode = stereo_mode;
}

39
video/out/gl_osd.h
View File

@@ -7,38 +7,15 @@
#include "gl_utils.h"
#include "sub/osd.h"
struct mpgl_osd_part {
enum sub_bitmap_format format;
int bitmap_id, bitmap_pos_id;
GLuint texture;
int w, h;
GLuint buffer;
int num_vertices;
void *vertices;
struct bitmap_packer *packer;
};
struct mpgl_osd {
struct mp_log *log;
struct osd_state *osd;
GL *gl;
bool use_pbo;
bool scaled;
struct mpgl_osd_part *parts[MAX_OSD_PARTS];
const struct osd_fmt_entry *fmt_table;
bool formats[SUBBITMAP_COUNT];
struct gl_vao vao;
GLuint *programs; // SUBBITMAP_COUNT elements
// temporary
float offset[2];
void *scratch;
};
struct mpgl_osd *mpgl_osd_init(GL *gl, struct mp_log *log, struct osd_state *osd,
GLuint *programs);
struct mpgl_osd *mpgl_osd_init(GL *gl, struct mp_log *log, struct osd_state *osd);
void mpgl_osd_destroy(struct mpgl_osd *ctx);
void mpgl_osd_draw(struct mpgl_osd *ctx, struct mp_osd_res res, double pts,
int stereo_mode);
void mpgl_osd_set_options(struct mpgl_osd *ctx, bool pbo);
void mpgl_osd_generate(struct mpgl_osd *ctx, struct mp_osd_res res, double pts,
int stereo_mode);
enum sub_bitmap_format mpgl_osd_get_part_format(struct mpgl_osd *ctx, int index);
struct gl_vao *mpgl_osd_get_vao(struct mpgl_osd *ctx);
void mpgl_osd_draw_part(struct mpgl_osd *ctx, int vp_w, int vp_h, int index);
#endif

496
video/out/gl_utils.c
View File

@@ -25,6 +25,7 @@
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
#include <stdarg.h>
#include <assert.h>
#include "common/common.h"
@@ -290,14 +291,6 @@ void gl_vao_unbind(struct gl_vao *vao)
}
}
void gl_vao_bind_attribs(struct gl_vao *vao, GLuint program)
{
GL *gl = vao->gl;
for (int n = 0; vao->entries[n].name; n++)
gl->BindAttribLocation(program, n, vao->entries[n].name);
}
// Draw the vertex data (as described by the gl_vao_entry entries) in ptr
// to the screen. num is the number of vertexes. prim is usually GL_TRIANGLES.
// If ptr is NULL, then skip the upload, and use the data uploaded with the
@@ -320,24 +313,47 @@ void gl_vao_draw_data(struct gl_vao *vao, GLenum prim, void *ptr, size_t num)
}
// Create a texture and a FBO using the texture as color attachments.
// gl_target: GL_TEXTURE_2D
// gl_filter: GL_LINEAR
// iformat: texture internal format
// Returns success.
bool fbotex_init(struct fbotex *fbo, GL *gl, struct mp_log *log, int w, int h,
GLenum gl_target, GLenum gl_filter, GLenum iformat)
GLenum iformat)
{
assert(!fbo->fbo);
assert(!fbo->texture);
return fbotex_change(fbo, gl, log, w, h, iformat, 0);
}
// Like fbotex_init(), except it can be called on an already initialized FBO;
// and if the parameters are the same as the previous call, do not touch it.
// flags can be 0, or a combination of FBOTEX_FUZZY_W and FBOTEX_FUZZY_H.
// Enabling FUZZY for W or H means the w or h does not need to be exact.
bool fbotex_change(struct fbotex *fbo, GL *gl, struct mp_log *log, int w, int h,
GLenum iformat, int flags)
{
bool res = true;
assert(!fbo->fbo);
assert(!fbo->texture);
int cw = w, ch = h;
if ((flags & FBOTEX_FUZZY_W) && cw < fbo->tex_w)
cw = fbo->tex_w;
if ((flags & FBOTEX_FUZZY_H) && ch < fbo->tex_h)
ch = fbo->tex_h;
if (fbo->tex_w == cw && fbo->tex_h == ch && fbo->iformat == iformat)
return true;
if (flags & FBOTEX_FUZZY_W)
w = MP_ALIGN_UP(w, 256);
if (flags & FBOTEX_FUZZY_H)
h = MP_ALIGN_UP(h, 256);
GLenum filter = fbo->tex_filter;
*fbo = (struct fbotex) {
.gl = gl,
.vp_w = w,
.vp_h = h,
.tex_w = w,
.tex_h = h,
.iformat = iformat,
};
mp_verbose(log, "Create FBO: %dx%d\n", fbo->tex_w, fbo->tex_h);
@@ -347,19 +363,20 @@ bool fbotex_init(struct fbotex *fbo, GL *gl, struct mp_log *log, int w, int h,
gl->GenFramebuffers(1, &fbo->fbo);
gl->GenTextures(1, &fbo->texture);
gl->BindTexture(gl_target, fbo->texture);
gl->TexImage2D(gl_target, 0, iformat, fbo->tex_w, fbo->tex_h, 0,
gl->BindTexture(GL_TEXTURE_2D, fbo->texture);
gl->TexImage2D(GL_TEXTURE_2D, 0, iformat, fbo->tex_w, fbo->tex_h, 0,
GL_RGBA, GL_UNSIGNED_BYTE, NULL);
gl->TexParameteri(gl_target, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
gl->TexParameteri(gl_target, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
gl->TexParameteri(gl_target, GL_TEXTURE_MIN_FILTER, gl_filter);
gl->TexParameteri(gl_target, GL_TEXTURE_MAG_FILTER, gl_filter);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_CLAMP_TO_EDGE);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_CLAMP_TO_EDGE);
gl->BindTexture(GL_TEXTURE_2D, 0);
fbotex_set_filter(fbo, filter ? filter : GL_LINEAR);
glCheckError(gl, log, "after creating framebuffer texture");
gl->BindFramebuffer(GL_FRAMEBUFFER, fbo->fbo);
gl->FramebufferTexture2D(GL_FRAMEBUFFER, GL_COLOR_ATTACHMENT0,
gl_target, fbo->texture, 0);
GL_TEXTURE_2D, fbo->texture, 0);
GLenum err = gl->CheckFramebufferStatus(GL_FRAMEBUFFER);
if (err != GL_FRAMEBUFFER_COMPLETE) {
@@ -375,6 +392,19 @@ bool fbotex_init(struct fbotex *fbo, GL *gl, struct mp_log *log, int w, int h,
return res;
}
void fbotex_set_filter(struct fbotex *fbo, GLenum tex_filter)
{
GL *gl = fbo->gl;
if (fbo->tex_filter != tex_filter && fbo->texture) {
gl->BindTexture(GL_TEXTURE_2D, fbo->texture);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, tex_filter);
gl->TexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, tex_filter);
gl->BindTexture(GL_TEXTURE_2D, 0);
}
fbo->tex_filter = tex_filter;
}
void fbotex_uninit(struct fbotex *fbo)
{
GL *gl = fbo->gl;
@@ -386,8 +416,16 @@ void fbotex_uninit(struct fbotex *fbo)
}
}
// Standard parallel 2D projection, except y1 < y0 means that the coordinate
// system is flipped, not the projection.
void gl_matrix_ortho2d(float m[3][3], float x0, float x1, float y0, float y1)
{
if (y1 < y0) {
float t = y0;
y0 = t - y1;
y1 = t;
}
memset(m, 0, 9 * sizeof(float));
m[0][0] = 2.0f / (x1 - x0);
m[1][1] = 2.0f / (y1 - y0);
@@ -422,3 +460,417 @@ void gl_set_debug_logger(GL *gl, struct mp_log *log)
}
}
}
#define SC_ENTRIES 10
#define SC_UNIFORM_ENTRIES 20
enum uniform_type {
UT_invalid,
UT_i,
UT_f,
UT_m,
};
struct sc_uniform {
char *name;
enum uniform_type type;
const char *glsl_type;
int size;
GLint loc;
union {
GLfloat f[9];
GLint i[4];
} v;
};
struct sc_entry {
GLuint gl_shader;
// the following fields define the shader's contents
char *key; // vertex+frag shader (mangled)
struct gl_vao *vao;
};
struct gl_shader_cache {
GL *gl;
struct mp_log *log;
// this is modified during use (gl_sc_add() etc.)
char *text;
struct gl_vao *vao;
struct sc_entry entries[SC_ENTRIES];
int num_entries;
struct sc_uniform uniforms[SC_UNIFORM_ENTRIES];
int num_uniforms;
};
struct gl_shader_cache *gl_sc_create(GL *gl, struct mp_log *log)
{
struct gl_shader_cache *sc = talloc_ptrtype(NULL, sc);
*sc = (struct gl_shader_cache){
.gl = gl,
.log = log,
.text = talloc_strdup(sc, ""),
};
return sc;
}
void gl_sc_reset(struct gl_shader_cache *sc)
{
sc->text[0] = '\0';
for (int n = 0; n < sc->num_uniforms; n++)
talloc_free(sc->uniforms[n].name);
sc->num_uniforms = 0;
}
static void sc_flush_cache(struct gl_shader_cache *sc)
{
for (int n = 0; n < sc->num_entries; n++) {
struct sc_entry *e = &sc->entries[n];
sc->gl->DeleteProgram(e->gl_shader);
talloc_free(e->key);
}
sc->num_entries = 0;
}
void gl_sc_destroy(struct gl_shader_cache *sc)
{
gl_sc_reset(sc);
sc_flush_cache(sc);
talloc_free(sc);
}
void gl_sc_add(struct gl_shader_cache *sc, const char *text)
{
sc->text = talloc_strdup_append(sc->text, text);
}
void gl_sc_addf(struct gl_shader_cache *sc, const char *textf, ...)
{
va_list ap;
va_start(ap, textf);
ta_xvasprintf_append(&sc->text, textf, ap);
va_end(ap);
}
static struct sc_uniform *find_uniform(struct gl_shader_cache *sc,
const char *name)
{
for (int n = 0; n < sc->num_uniforms; n++) {
if (strcmp(sc->uniforms[n].name, name) == 0)
return &sc->uniforms[n];
}
// not found -> add it
assert(sc->num_uniforms < SC_UNIFORM_ENTRIES); // just don't have too many
struct sc_uniform *new = &sc->uniforms[sc->num_uniforms++];
*new = (struct sc_uniform) { .loc = -1, .name = talloc_strdup(NULL, name) };
return new;
}
void gl_sc_uniform_sampler(struct gl_shader_cache *sc, char *name, GLenum target,
int unit)
{
struct sc_uniform *u = find_uniform(sc, name);
u->type = UT_i;
u->size = 1;
switch (target) {
case GL_TEXTURE_1D: u->glsl_type = "sampler1D"; break;
case GL_TEXTURE_2D: u->glsl_type = "sampler2D"; break;
case GL_TEXTURE_RECTANGLE: u->glsl_type = "sampler2DRect"; break;
case GL_TEXTURE_3D: u->glsl_type = "sampler3D"; break;
default: abort();
}
u->v.i[0] = unit;
}
void gl_sc_uniform_f(struct gl_shader_cache *sc, char *name, GLfloat f)
{
struct sc_uniform *u = find_uniform(sc, name);
u->type = UT_f;
u->size = 1;
u->glsl_type = "float";
u->v.f[0] = f;
}
void gl_sc_uniform_vec2(struct gl_shader_cache *sc, char *name, GLfloat f[2])
{
struct sc_uniform *u = find_uniform(sc, name);
u->type = UT_f;
u->size = 2;
u->glsl_type = "vec2";
u->v.f[0] = f[0];
u->v.f[1] = f[1];
}
void gl_sc_uniform_vec3(struct gl_shader_cache *sc, char *name, GLfloat f[3])
{
struct sc_uniform *u = find_uniform(sc, name);
u->type = UT_f;
u->size = 3;
u->glsl_type = "vec3";
u->v.f[0] = f[0];
u->v.f[1] = f[1];
u->v.f[2] = f[2];
}
static void transpose2x2(float r[2 * 2])
{
MPSWAP(float, r[0+2*1], r[1+2*0]);
}
void gl_sc_uniform_mat2(struct gl_shader_cache *sc, char *name,
bool transpose, GLfloat *v)
{
struct sc_uniform *u = find_uniform(sc, name);
u->type = UT_m;
u->size = 2;
u->glsl_type = "mat2";
for (int n = 0; n < 4; n++)
u->v.f[n] = v[n];
if (transpose)
transpose2x2(&u->v.f[0]);
}
static void transpose3x3(float r[3 * 3])
{
MPSWAP(float, r[0+3*1], r[1+3*0]);
MPSWAP(float, r[0+3*2], r[2+3*0]);
MPSWAP(float, r[1+3*2], r[2+3*1]);
}
void gl_sc_uniform_mat3(struct gl_shader_cache *sc, char *name,
bool transpose, GLfloat *v)
{
struct sc_uniform *u = find_uniform(sc, name);
u->type = UT_m;
u->size = 3;
u->glsl_type = "mat3";
for (int n = 0; n < 9; n++)
u->v.f[n] = v[n];
if (transpose)
transpose3x3(&u->v.f[0]);
}
// This will call glBindAttribLocation() on the shader before it's linked
// (OpenGL requires this to happen before linking). Basically, it associates
// the input variable names with the fields in the vao.
// The vertex shader is setup such that the elements are available as fragment
// shader variables using the names in the vao entries, which "position" being
// set to gl_Position.
void gl_sc_set_vao(struct gl_shader_cache *sc, struct gl_vao *vao)
{
sc->vao = vao;
}
static const char *vao_glsl_type(const struct gl_vao_entry *e)
{
// pretty dumb... too dumb, but works for us
switch (e->num_elems) {
case 1: return "float";
case 2: return "vec2";
case 3: return "vec3";
case 4: return "vec4";
default: abort();
}
}
// Assumes program is current (gl->UseProgram(program)).
static void update_uniform(GL *gl, GLuint program, struct sc_uniform *u)
{
GLint loc = gl->GetUniformLocation(program, u->name);
if (loc < 0)
return;
switch (u->type) {
case UT_i:
assert(u->size == 1);
gl->Uniform1i(loc, u->v.i[0]);
break;
case UT_f:
switch (u->size) {
case 1: gl->Uniform1f(loc, u->v.f[0]); break;
case 2: gl->Uniform2f(loc, u->v.f[0], u->v.f[1]); break;
case 3: gl->Uniform3f(loc, u->v.f[0], u->v.f[1], u->v.f[2]); break;
case 4: gl->Uniform4f(loc, u->v.f[0], u->v.f[1], u->v.f[2], u->v.f[3]); break;
default: abort();
}
break;
case UT_m:
switch (u->size) {
case 2: gl->UniformMatrix2fv(loc, 1, GL_FALSE, &u->v.f[0]); break;
case 3: gl->UniformMatrix3fv(loc, 1, GL_FALSE, &u->v.f[0]); break;
default: abort();
}
break;
default:
abort();
}
}
static void compile_attach_shader(struct gl_shader_cache *sc, GLuint program,
GLenum type, const char *source)
{
GL *gl = sc->gl;
GLuint shader = gl->CreateShader(type);
gl->ShaderSource(shader, 1, &source, NULL);
gl->CompileShader(shader);
GLint status;
gl->GetShaderiv(shader, GL_COMPILE_STATUS, &status);
GLint log_length;
gl->GetShaderiv(shader, GL_INFO_LOG_LENGTH, &log_length);
int pri = status ? (log_length > 1 ? MSGL_V : MSGL_DEBUG) : MSGL_ERR;
const char *typestr = type == GL_VERTEX_SHADER ? "vertex" : "fragment";
if (mp_msg_test(sc->log, pri)) {
MP_MSG(sc, pri, "%s shader source:\n", typestr);
mp_log_source(sc->log, pri, source);
}
if (log_length > 1) {
GLchar *logstr = talloc_zero_size(NULL, log_length + 1);
gl->GetShaderInfoLog(shader, log_length, NULL, logstr);
MP_MSG(sc, pri, "%s shader compile log (status=%d):\n%s\n",
typestr, status, logstr);
talloc_free(logstr);
}
gl->AttachShader(program, shader);
gl->DeleteShader(shader);
}
static void link_shader(struct gl_shader_cache *sc, GLuint program)
{
GL *gl = sc->gl;
gl->LinkProgram(program);
GLint status;
gl->GetProgramiv(program, GL_LINK_STATUS, &status);
GLint log_length;
gl->GetProgramiv(program, GL_INFO_LOG_LENGTH, &log_length);
int pri = status ? (log_length > 1 ? MSGL_V : MSGL_DEBUG) : MSGL_ERR;
if (mp_msg_test(sc->log, pri)) {
GLchar *logstr = talloc_zero_size(NULL, log_length + 1);
gl->GetProgramInfoLog(program, log_length, NULL, logstr);
MP_MSG(sc, pri, "shader link log (status=%d): %s\n", status, logstr);
talloc_free(logstr);
}
}
static GLuint create_program(struct gl_shader_cache *sc, const char *vertex,
const char *frag)
{
GL *gl = sc->gl;
MP_VERBOSE(sc, "recompiling a shader program:\n");
mp_log_source(sc->log, MSGL_V, sc->text);
GLuint prog = gl->CreateProgram();
compile_attach_shader(sc, prog, GL_VERTEX_SHADER, vertex);
compile_attach_shader(sc, prog, GL_FRAGMENT_SHADER, frag);
for (int n = 0; sc->vao->entries[n].name; n++) {
char vname[80];
snprintf(vname, sizeof(vname), "vertex_%s", sc->vao->entries[n].name);
gl->BindAttribLocation(prog, n, vname);
}
link_shader(sc, prog);
return prog;
}
#define ADD(x, ...) (x) = talloc_asprintf_append(x, __VA_ARGS__)
// 1. Generate vertex and fragment shaders from the fragment shader text added
// with gl_sc_add(). The generated shader program is cached (based on the
// text), so actual compilation happens only the first time.
// 2. Update the uniforms set with gl_sc_uniform_*.
// 3. Make the new shader program current (glUseProgram()).
// 4. Reset the sc state and prepare for a new shader program. (All uniforms
// and fragment operations needed for the next program have to be re-added.)
void gl_sc_gen_shader_and_reset(struct gl_shader_cache *sc)
{
GL *gl = sc->gl;
void *tmp = talloc_new(NULL);
assert(sc->vao);
// set up shader text (header + uniforms + body)
char *header = talloc_asprintf(tmp, "#version %d%s\n", gl->glsl_version,
gl->es >= 300 ? " es" : "");
if (gl->es)
ADD(header, "precision mediump float;\n");
char *vert_in = gl->glsl_version >= 130 ? "in" : "attribute";
char *vert_out = gl->glsl_version >= 130 ? "out" : "varying";
char *frag_in = gl->glsl_version >= 130 ? "in" : "varying";
// vertex shader: we don't use the vertex shader, so just setup a dummy,
// which passes through the vertex array attributes.
char *vert_head = talloc_strdup(tmp, header);
char *vert_body = talloc_strdup(tmp, "void main() {\n");
char *frag_vaos = talloc_strdup(tmp, "");
for (int n = 0; sc->vao->entries[n].name; n++) {
const struct gl_vao_entry *e = &sc->vao->entries[n];
const char *glsl_type = vao_glsl_type(e);
if (strcmp(e->name, "position") == 0) {
// setting raster pos. requires setting gl_Position magic variable
assert(e->num_elems == 2 && e->type == GL_FLOAT);
ADD(vert_head, "%s vec2 position;\n", vert_in);
ADD(vert_body, "gl_Position = vec4(position, 1.0, 1.0);\n");
} else {
ADD(vert_head, "%s %s vertex_%s;\n", vert_in, glsl_type, e->name);
ADD(vert_head, "%s %s %s;\n", vert_out, glsl_type, e->name);
ADD(vert_body, "%s = vertex_%s;\n", e->name, e->name);
ADD(frag_vaos, "%s %s %s;\n", frag_in, glsl_type, e->name);
}
}
ADD(vert_body, "}\n");
char *vert = talloc_asprintf(tmp, "%s%s", vert_head, vert_body);
// fragment shader; still requires adding used uniforms and VAO elements
char *frag = talloc_strdup(tmp, header);
ADD(frag, "#define RG %s\n", gl->mpgl_caps & MPGL_CAP_TEX_RG ? "rg" : "ra");
if (gl->glsl_version >= 130) {
ADD(frag, "#define texture1D texture\n");
ADD(frag, "#define texture3D texture\n");
ADD(frag, "out vec4 out_color;\n");
}
ADD(frag, "%s", frag_vaos);
for (int n = 0; n < sc->num_uniforms; n++) {
struct sc_uniform *u = &sc->uniforms[n];
ADD(frag, "uniform %s %s;\n", u->glsl_type, u->name);
}
ADD(frag, "void main() {\n");
ADD(frag, "%s", sc->text);
// we require _all_ frag shaders to write to a "vec4 color"
if (gl->glsl_version >= 130) {
ADD(frag, "out_color = color;\n");
} else {
ADD(frag, "gl_FragColor = color;\n");
}
ADD(frag, "}\n");
char *key = talloc_asprintf(tmp, "%s%s", vert, frag);
struct sc_entry *entry = NULL;
for (int n = 0; n < sc->num_entries; n++) {
if (strcmp(key, sc->entries[n].key) == 0) {
entry = &sc->entries[n];
break;
}
}
if (!entry) {
if (sc->num_entries == SC_ENTRIES)
sc_flush_cache(sc);
entry = &sc->entries[sc->num_entries++];
*entry = (struct sc_entry){.key = talloc_strdup(NULL, key)};
}
// build vertex shader from vao
if (!entry->gl_shader)
entry->gl_shader = create_program(sc, vert, frag);
gl->UseProgram(entry->gl_shader);
// For now we set the uniforms every time. This is probably bad, and we
// should switch to caching them.
for (int n = 0; n < sc->num_uniforms; n++)
update_uniform(gl, entry->gl_shader, &sc->uniforms[n]);
talloc_free(tmp);
gl_sc_reset(sc);
}

39
video/out/gl_utils.h
View File

@@ -66,23 +66,54 @@ void gl_vao_init(struct gl_vao *vao, GL *gl, int stride,
void gl_vao_uninit(struct gl_vao *vao);
void gl_vao_bind(struct gl_vao *vao);
void gl_vao_unbind(struct gl_vao *vao);
void gl_vao_bind_attribs(struct gl_vao *vao, GLuint program);
void gl_vao_draw_data(struct gl_vao *vao, GLenum prim, void *ptr, size_t num);
struct fbotex {
GL *gl;
GLuint fbo;
GLuint texture;
int tex_w, tex_h; // size of .texture
int vp_x, vp_y, vp_w, vp_h; // viewport of fbo / used part of the texture
GLenum iformat;
GLenum tex_filter;
int tex_w, tex_h; // size of .texture
};
bool fbotex_init(struct fbotex *fbo, GL *gl, struct mp_log *log, int w, int h,
GLenum gl_target, GLenum gl_filter, GLenum iformat);
GLenum iformat);
void fbotex_uninit(struct fbotex *fbo);
bool fbotex_change(struct fbotex *fbo, GL *gl, struct mp_log *log, int w, int h,
GLenum iformat, int flags);
#define FBOTEX_FUZZY_W 1
#define FBOTEX_FUZZY_H 2
void fbotex_set_filter(struct fbotex *fbo, GLenum gl_filter);
void gl_matrix_ortho2d(float m[3][3], float x0, float x1, float y0, float y1);
static inline void gl_matrix_mul_vec(float m[3][3], float *x, float *y)
{
float vx = *x, vy = *y;
*x = vx * m[0][0] + vy * m[1][0] + m[2][0];
*y = vx * m[0][1] + vy * m[1][1] + m[2][1];
}
void gl_set_debug_logger(GL *gl, struct mp_log *log);
struct gl_shader_cache;
struct gl_shader_cache *gl_sc_create(GL *gl, struct mp_log *log);
void gl_sc_destroy(struct gl_shader_cache *sc);
void gl_sc_add(struct gl_shader_cache *sc, const char *text);
void gl_sc_addf(struct gl_shader_cache *sc, const char *textf, ...);
void gl_sc_uniform_sampler(struct gl_shader_cache *sc, char *name, GLenum target,
int unit);
void gl_sc_uniform_f(struct gl_shader_cache *sc, char *name, GLfloat f);
void gl_sc_uniform_vec2(struct gl_shader_cache *sc, char *name, GLfloat f[2]);
void gl_sc_uniform_vec3(struct gl_shader_cache *sc, char *name, GLfloat f[3]);
void gl_sc_uniform_mat2(struct gl_shader_cache *sc, char *name,
bool transpose, GLfloat *v);
void gl_sc_uniform_mat3(struct gl_shader_cache *sc, char *name,
bool transpose, GLfloat *v);
void gl_sc_set_vao(struct gl_shader_cache *sc, struct gl_vao *vao);
void gl_sc_gen_shader_and_reset(struct gl_shader_cache *sc);
void gl_sc_reset(struct gl_shader_cache *sc);
#endif

1973
video/out/gl_video.c
View File

File diff suppressed because it is too large Load Diff

4
video/out/gl_video.h
View File

@@ -73,9 +73,9 @@ void gl_video_set_output_depth(struct gl_video *p, int r, int g, int b);
void gl_video_set_lut3d(struct gl_video *p, struct lut3d *lut3d);
void gl_video_upload_image(struct gl_video *p, struct mp_image *img);
void gl_video_render_frame(struct gl_video *p, int fbo, struct frame_timing *t);
void gl_video_resize(struct gl_video *p, struct mp_rect *window,
void gl_video_resize(struct gl_video *p, int vp_w, int vp_h,
struct mp_rect *src, struct mp_rect *dst,
struct mp_osd_res *osd, bool vflip);
struct mp_osd_res *osd);
void gl_video_get_colorspace(struct gl_video *p, struct mp_image_params *params);
struct mp_csp_equalizer;
struct mp_csp_equalizer *gl_video_eq_ptr(struct gl_video *p);

542
video/out/gl_video_shaders.glsl
View File

@@ -1,542 +0,0 @@
/*
* 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.
*/
// Note that this file is not directly passed as shader, but run through some
// text processing functions, and in fact contains multiple vertex and fragment
// shaders.
// inserted at the beginning of all shaders
#!section prelude
#ifdef GL_ES
precision mediump float;
#endif
// GLSL 1.20 compatibility layer
// texture() should be assumed to always map to texture2D()
#if __VERSION__ >= 130
# define texture1D texture
# define texture3D texture
# define DECLARE_FRAGPARMS \
out vec4 out_color;
#else
# define texture texture2D
# define DECLARE_FRAGPARMS
# define out_color gl_FragColor
# define in varying
#endif
#if HAVE_RG
#define RG rg
#else
#define RG ra
#endif
// Earlier GLSL doesn't support mix() with bvec
#if __VERSION__ >= 130
vec3 srgb_expand(vec3 v)
{
return mix(v / vec3(12.92), pow((v + vec3(0.055))/vec3(1.055), vec3(2.4)),
lessThanEqual(vec3(0.04045), v));
}
vec3 srgb_compand(vec3 v)
{
return mix(v * vec3(12.92), vec3(1.055) * pow(v, vec3(1.0/2.4)) - vec3(0.055),
lessThanEqual(vec3(0.0031308), v));
}
vec3 bt2020_expand(vec3 v)
{
return mix(v / vec3(4.5), pow((v + vec3(0.0993))/vec3(1.0993), vec3(1.0/0.45)),
lessThanEqual(vec3(0.08145), v));
}
vec3 bt2020_compand(vec3 v)
{
return mix(v * vec3(4.5), vec3(1.0993) * pow(v, vec3(0.45)) - vec3(0.0993),
lessThanEqual(vec3(0.0181), v));
}
#endif
#!section vertex_all
#if __VERSION__ < 130
# undef in
# define in attribute
# define out varying
#endif
uniform mat3 transform;
uniform vec3 translation;
#if HAVE_3DTEX
uniform sampler3D lut_3d;
#endif
uniform mat3 cms_matrix; // transformation from file's gamut to bt.2020
in vec2 vertex_position;
in vec4 vertex_color;
out vec4 color;
in vec2 vertex_texcoord;
out vec2 texcoord;
void main() {
vec3 position = vec3(vertex_position, 1) + translation;
#ifndef FIXED_SCALE
position = transform * position;
#endif
gl_Position = vec4(position, 1);
color = vertex_color;
// Although we are not scaling in linear light, both 3DLUT and SRGB still
// operate on linear light inputs so we have to convert to it before
// either step can be applied.
#ifdef USE_OSD_LINEAR_CONV_BT1886
color.rgb = pow(color.rgb, vec3(1.961));
#endif
#ifdef USE_OSD_LINEAR_CONV_SRGB
color.rgb = srgb_expand(color.rgb);
#endif
#ifdef USE_OSD_CMS_MATRIX
// Convert to the right target gamut first (to BT.709 for sRGB,
// and to BT.2020 for 3DLUT). Normal clamping here as perceptually
// accurate colorimetry is probably not worth the performance trade-off
// here.
color.rgb = clamp(cms_matrix * color.rgb, 0.0, 1.0);
#endif
#ifdef USE_OSD_3DLUT
color.rgb = pow(color.rgb, vec3(1.0/2.4)); // linear -> 2.4 3DLUT space
color = vec4(texture3D(lut_3d, color.rgb).rgb, color.a);
#endif
#ifdef USE_OSD_SRGB
color.rgb = srgb_compand(color.rgb);
#endif
texcoord = vertex_texcoord;
}
#!section frag_osd_libass
uniform sampler2D texture0;
in vec2 texcoord;
in vec4 color;
DECLARE_FRAGPARMS
void main() {
out_color = vec4(color.rgb, color.a * texture(texture0, texcoord).r);
}
#!section frag_osd_rgba
uniform sampler2D texture0;
in vec2 texcoord;
DECLARE_FRAGPARMS
void main() {
out_color = texture(texture0, texcoord).bgra;
}
#!section frag_video
uniform VIDEO_SAMPLER texture0;
uniform VIDEO_SAMPLER texture1;
uniform VIDEO_SAMPLER texture2;
uniform VIDEO_SAMPLER texture3;
uniform vec2 textures_size[4];
uniform vec2 chroma_center_offset;
uniform vec2 chroma_div;
uniform vec2 chroma_fix;
uniform sampler2D lut_2d_c;
uniform sampler2D lut_2d_l;
#if HAVE_1DTEX
uniform sampler1D lut_1d_c;
uniform sampler1D lut_1d_l;
#endif
#if HAVE_3DTEX
uniform sampler3D lut_3d;
#endif
uniform sampler2D dither;
uniform mat3 colormatrix;
uniform vec3 colormatrix_c;
uniform mat3 cms_matrix;
uniform mat2 dither_trafo;
uniform float inv_gamma;
uniform float input_gamma;
uniform float conv_gamma;
uniform float sig_center;
uniform float sig_slope;
uniform float sig_scale;
uniform float sig_offset;
uniform float dither_quantization;
uniform float dither_center;
uniform float filter_param1_l;
uniform float filter_param1_c;
uniform float antiring_factor;
uniform vec2 dither_size;
uniform float inter_coeff;
in vec2 texcoord;
DECLARE_FRAGPARMS
#define CONV_NV12 1
#define CONV_PLANAR 2
vec4 sample_bilinear(VIDEO_SAMPLER tex, vec2 texsize, vec2 texcoord, float param1) {
return texture(tex, texcoord);
}
#define SAMPLE_TRIVIAL(tex, texsize, texcoord) texture(tex, texcoord)
// Explanation how bicubic scaling with only 4 texel fetches is done:
// http://www.mate.tue.nl/mate/pdfs/10318.pdf
// 'Efficient GPU-Based Texture Interpolation using Uniform B-Splines'
// Explanation why this algorithm normally always blurs, even with unit scaling:
// http://bigwww.epfl.ch/preprints/ruijters1001p.pdf
// 'GPU Prefilter for Accurate Cubic B-spline Interpolation'
vec4 calcweights(float s) {
vec4 t = vec4(-0.5, 0.1666, 0.3333, -0.3333) * s + vec4(1, 0, -0.5, 0.5);
t = t * s + vec4(0, 0, -0.5, 0.5);
t = t * s + vec4(-0.6666, 0, 0.8333, 0.1666);
vec2 a = vec2(1, 1) / vec2(t.z, t.w);
t.xy = t.xy * a + vec2(1, 1);
t.x = t.x + s;
t.y = t.y - s;
return t;
}
vec4 sample_bicubic_fast(VIDEO_SAMPLER tex, vec2 texsize, vec2 texcoord, float param1) {
vec2 pt = 1.0 / texsize;
vec2 fcoord = fract(texcoord * texsize + vec2(0.5, 0.5));
vec4 parmx = calcweights(fcoord.x);
vec4 parmy = calcweights(fcoord.y);
vec4 cdelta;
cdelta.xz = parmx.RG * vec2(-pt.x, pt.x);
cdelta.yw = parmy.RG * vec2(-pt.y, pt.y);
// first y-interpolation
vec4 ar = texture(tex, texcoord + cdelta.xy);
vec4 ag = texture(tex, texcoord + cdelta.xw);
vec4 ab = mix(ag, ar, parmy.b);
// second y-interpolation
vec4 br = texture(tex, texcoord + cdelta.zy);
vec4 bg = texture(tex, texcoord + cdelta.zw);
vec4 aa = mix(bg, br, parmy.b);
// x-interpolation
return mix(aa, ab, parmx.b);
}
#if HAVE_ARRAYS
float[2] weights2(sampler2D lookup, float f) {
vec2 c = texture(lookup, vec2(0.5, f)).RG;
return float[2](c.r, c.g);
}
float[6] weights6(sampler2D lookup, float f) {
vec4 c1 = texture(lookup, vec2(0.25, f));
vec4 c2 = texture(lookup, vec2(0.75, f));
return float[6](c1.r, c1.g, c1.b, c2.r, c2.g, c2.b);
}
// For N=n*4 with n>1.
#define WEIGHTS_N(NAME, N) \
float[N] NAME(sampler2D lookup, float f) { \
float r[N]; \
for (int n = 0; n < N / 4; n++) { \
vec4 c = texture(lookup, \
vec2(1.0 / (N / 2) + n / float(N / 4), f)); \
r[n * 4 + 0] = c.r; \
r[n * 4 + 1] = c.g; \
r[n * 4 + 2] = c.b; \
r[n * 4 + 3] = c.a; \
} \
return r; \
}
// The DIR parameter is (0, 1) or (1, 0), and we expect the shader compiler to
// remove all the redundant multiplications and additions, and also to unroll
// the loop and remove the conditional completely
#define SAMPLE_CONVOLUTION_SEP_N(NAME, DIR, N, LUT, WEIGHTS_FUNC, ANTIRING) \
vec4 NAME(VIDEO_SAMPLER tex, vec2 texsize, vec2 texcoord) { \
vec2 pt = (vec2(1.0) / texsize) * DIR; \
float fcoord = dot(fract(texcoord * texsize - vec2(0.5)), DIR); \
vec2 base = texcoord - fcoord * pt - pt * vec2(N / 2 - 1); \
float weights[N] = WEIGHTS_FUNC(LUT, fcoord); \
vec4 res = vec4(0); \
vec4 hi = vec4(0); \
vec4 lo = vec4(1); \
for (int n = 0; n < N; n++) { \
vec4 c = texture(tex, base + pt * vec2(n)); \
res += vec4(weights[n]) * c; \
if (n == N/2-1 || n == N/2) { \
lo = min(lo, c); \
hi = max(hi, c); \
} \
} \
return mix(res, clamp(res, lo, hi), ANTIRING); \
}
#define SAMPLE_CONVOLUTION_N(NAME, N, LUT, WEIGHTS_FUNC) \
vec4 NAME(VIDEO_SAMPLER tex, vec2 texsize, vec2 texcoord) { \
vec2 pt = vec2(1.0) / texsize; \
vec2 fcoord = fract(texcoord * texsize - vec2(0.5)); \
vec2 base = texcoord - fcoord * pt - pt * vec2(N / 2 - 1); \
vec4 res = vec4(0); \
float w_x[N] = WEIGHTS_FUNC(LUT, fcoord.x); \
float w_y[N] = WEIGHTS_FUNC(LUT, fcoord.y); \
for (int y = 0; y < N; y++) { \
vec4 line = vec4(0); \
for (int x = 0; x < N; x++) \
line += vec4(w_x[x]) * texture(tex, base + pt * vec2(x, y));\
res += vec4(w_y[y]) * line; \
} \
return res; \
}
#define SAMPLE_POLAR_HELPER(LUT, R, X, Y) \
w = texture1D(LUT, length(vec2(X, Y) - fcoord)/R).r; \
c = texture(tex, base + pt * vec2(X, Y)); \
wsum += w; \
res += vec4(w) * c;
#define SAMPLE_POLAR_PRIMARY(LUT, R, X, Y) \
SAMPLE_POLAR_HELPER(LUT, R, X, Y) \
lo = min(lo, c); \
hi = max(hi, c);
#define SAMPLE_POLAR_POTENTIAL(LUT, R, X, Y) \
if (length(vec2(X, Y) - fcoord)/R < 1.0) { \
SAMPLE_POLAR_HELPER(LUT, R, X, Y) \
}
#define SAMPLE_CONVOLUTION_POLAR_R(NAME, R, LUT, WEIGHTS_FN, ANTIRING) \
vec4 NAME(VIDEO_SAMPLER tex, vec2 texsize, vec2 texcoord) { \
vec2 pt = vec2(1.0) / texsize; \
vec2 fcoord = fract(texcoord * texsize - vec2(0.5)); \
vec2 base = texcoord - fcoord * pt; \
vec4 res = vec4(0.0); \
vec4 lo = vec4(1.0); \
vec4 hi = vec4(0.0); \
float wsum = 0.0; \
float w; \
vec4 c; \
WEIGHTS_FN(LUT); \
res = res / vec4(wsum); \
return mix(res, clamp(res, lo, hi), ANTIRING); \
}
#endif /* HAVE_ARRAYS */
#ifdef DEF_SCALER0
DEF_SCALER0
#endif
#ifdef DEF_SCALER1
DEF_SCALER1
#endif
// Unsharp masking
vec4 sample_sharpen3(VIDEO_SAMPLER tex, vec2 texsize, vec2 texcoord, float param1) {
vec2 pt = 1.0 / texsize;
vec2 st = pt * 0.5;
vec4 p = texture(tex, texcoord);
vec4 sum = texture(tex, texcoord + st * vec2(+1, +1))
+ texture(tex, texcoord + st * vec2(+1, -1))
+ texture(tex, texcoord + st * vec2(-1, +1))
+ texture(tex, texcoord + st * vec2(-1, -1));
return p + (p - 0.25 * sum) * param1;
}
vec4 sample_sharpen5(VIDEO_SAMPLER tex, vec2 texsize, vec2 texcoord, float param1) {
vec2 pt = 1.0 / texsize;
vec2 st1 = pt * 1.2;
vec4 p = texture(tex, texcoord);
vec4 sum1 = texture(tex, texcoord + st1 * vec2(+1, +1))
+ texture(tex, texcoord + st1 * vec2(+1, -1))
+ texture(tex, texcoord + st1 * vec2(-1, +1))
+ texture(tex, texcoord + st1 * vec2(-1, -1));
vec2 st2 = pt * 1.5;
vec4 sum2 = texture(tex, texcoord + st2 * vec2(+1, 0))
+ texture(tex, texcoord + st2 * vec2( 0, +1))
+ texture(tex, texcoord + st2 * vec2(-1, 0))
+ texture(tex, texcoord + st2 * vec2( 0, -1));
vec4 t = p * 0.859375 + sum2 * -0.1171875 + sum1 * -0.09765625;
return p + t * param1;
}
void main() {
vec2 chr_texcoord = texcoord;
#ifdef USE_CHROMA_FIX
chr_texcoord = chr_texcoord * chroma_fix;
#endif
#ifdef USE_RECTANGLE
chr_texcoord = chr_texcoord * chroma_div;
#else
// Texture coordinates are [0,1], and chroma plane coordinates are
// magically rescaled.
#endif
chr_texcoord = chr_texcoord + chroma_center_offset;
#ifndef USE_CONV
#define USE_CONV 0
#endif
#ifndef USE_LINEAR_INTERPOLATION
#define USE_LINEAR_INTERPOLATION 0
#endif
#if USE_LINEAR_INTERPOLATION == 1
vec4 acolor = mix(
texture(texture0, texcoord),
texture(texture1, texcoord),
inter_coeff);
#elif USE_CONV == CONV_PLANAR
vec4 acolor = vec4(SAMPLE(texture0, textures_size[0], texcoord).r,
SAMPLE_C(texture1, textures_size[1], chr_texcoord).r,
SAMPLE_C(texture2, textures_size[2], chr_texcoord).r,
1.0);
#elif USE_CONV == CONV_NV12
vec4 acolor = vec4(SAMPLE(texture0, textures_size[0], texcoord).r,
SAMPLE_C(texture1, textures_size[1], chr_texcoord).RG,
1.0);
#else
vec4 acolor = SAMPLE(texture0, textures_size[0], texcoord);
#endif
#ifdef USE_COLOR_SWIZZLE
acolor = acolor. USE_COLOR_SWIZZLE ;
#endif
#ifdef USE_ALPHA_PLANE
acolor.a = SAMPLE(texture3, textures_size[3], texcoord).r;
#endif
vec3 color = acolor.rgb;
float alpha = acolor.a;
#ifdef USE_INPUT_GAMMA
// Pre-colormatrix input gamma correction (eg. for MP_IMGFLAG_XYZ)
color = pow(color, vec3(input_gamma));
#endif
#ifdef USE_COLORMATRIX
// Conversion from Y'CbCr or other spaces to RGB
color = mat3(colormatrix) * color + colormatrix_c;
#endif
#ifdef USE_CONV_GAMMA
// Post-colormatrix converted gamma correction (eg. for MP_IMGFLAG_XYZ)
color = pow(color, vec3(conv_gamma));
#endif
#ifdef USE_CONST_LUMA
// Conversion from C'rcY'cC'bc to R'Y'cB' via the BT.2020 CL system:
// C'bc = (B'-Y'c) / 1.9404 | C'bc <= 0
// = (B'-Y'c) / 1.5816 | C'bc > 0
//
// C'rc = (R'-Y'c) / 1.7184 | C'rc <= 0
// = (R'-Y'c) / 0.9936 | C'rc > 0
//
// as per the BT.2020 specification, table 4. This is a non-linear
// transformation because (constant) luminance receives non-equal
// contributions from the three different channels.
color.br = color.br * mix(vec2(1.5816, 0.9936), vec2(1.9404, 1.7184),
lessThanEqual(color.br, vec2(0))) + color.gg;
// Expand channels to camera-linear light. This shader currently just
// assumes everything uses the BT.2020 12-bit gamma function, since the
// difference between 10 and 12-bit is negligible for anything other than
// 12-bit content.
color = bt2020_expand(color);
// Calculate the green channel from the expanded RYcB
// The BT.2020 specification says Yc = 0.2627*R + 0.6780*G + 0.0593*B
color.g = (color.g - 0.2627*color.r - 0.0593*color.b)/0.6780;
// Re-compand to receive the R'G'B' result, same as other systems
color = bt2020_compand(color);
#endif
#ifdef USE_COLORMATRIX
// CONST_LUMA involves numbers outside the [0,1] range so we make sure
// to clip here, after the (possible) USE_CONST_LUMA calculations are done,
// instead of immediately after the colormatrix conversion.
color = clamp(color, 0.0, 1.0);
#endif
// If we are scaling in linear light (SRGB or 3DLUT option enabled), we
// expand our source colors before scaling. We distinguish between
// BT.1886 (typical video files) and sRGB (typical image files).
#ifdef USE_LINEAR_LIGHT_BT1886
// This calculation is derived from the BT.1886 recommendation which
// is itself derived from the curves of typical CRT monitors. It claims
// that a correct video playback environment should have a pure power
// curve transfer function (in contrast to the complex BT.709 function)
// with a gamma value of 2.40, but this includes the typical gamma boost
// of ~1.2 for dark viewing environments. The figure used here instead
// (1.961) is therefore a pure power curve but without the boost, which
// is a very close approximation of the true BT.709 function.
color = pow(color, vec3(1.961));
#endif
#ifdef USE_LINEAR_LIGHT_SRGB
// This is not needed for most sRGB content since we can use GL_SRGB to
// directly sample RGB texture in linear light, but for things which are
// also sRGB but in a different format (such as JPEG's YUV), we need
// to convert to linear light manually.
color = srgb_expand(color);
#endif
#ifdef USE_SIGMOID
color = sig_center - log(1.0/(color * sig_scale + sig_offset) - 1.0)/sig_slope;
#endif
// Image upscaling happens roughly here
#ifdef USE_SIGMOID_INV
// Inverse of USE_SIGMOID
color = (1.0/(1.0 + exp(sig_slope * (sig_center - color))) - sig_offset) / sig_scale;
#endif
#ifdef USE_CMS_MATRIX
// Convert to the right target gamut first (to BT.709 for sRGB,
// and to BT.2020 for 3DLUT).
color = cms_matrix * color;
#endif
// Clamp to the target gamut. This clamp is needed because the gamma
// functions are not well-defined outside this range, which is related to
// the fact that they're not representable on the target device.
// TODO: Desaturate colorimetrically; this happens automatically for
// 3dlut targets but not for sRGB mode. Not sure if this is a requirement.
color = clamp(color, 0.0, 1.0);
#ifdef USE_INV_GAMMA
// User-defined gamma correction factor (via the gamma sub-option)
color = pow(color, vec3(inv_gamma));
#endif
#ifdef USE_3DLUT
// For the 3DLUT we are arbitrarily using 2.4 as input gamma to reduce
// the amount of rounding errors, so we pull up to that space first and
// then pass it through the 3D texture.
color = pow(color, vec3(1.0/2.4));
color = texture3D(lut_3d, color).rgb;
#endif
#ifdef USE_SRGB
// Adapt and compand from the linear BT2020 source to the sRGB output
color = srgb_compand(color);
#endif
#ifdef USE_INV_BT1886
color = pow(color, vec3(1.0/1.961));
#endif
#ifdef USE_DITHER
vec2 dither_pos = gl_FragCoord.xy / dither_size;
#ifdef USE_TEMPORAL_DITHER
dither_pos = dither_trafo * dither_pos;
#endif
float dither_value = texture(dither, dither_pos).r;
color = floor(color * dither_quantization + dither_value + dither_center) /
dither_quantization;
#endif
#ifdef USE_ALPHA_BLEND
color = color * alpha;
#endif
#ifdef USE_ALPHA
out_color = vec4(color, alpha);
#else
out_color = vec4(color, 1.0);
#endif
}

5
video/out/vo_opengl.c
View File

@@ -93,12 +93,11 @@ static void resize(struct gl_priv *p)
MP_VERBOSE(vo, "Resize: %dx%d\n", vo->dwidth, vo->dheight);
struct mp_rect wnd = {0, 0, vo->dwidth, vo->dheight};
struct mp_rect src, dst;
struct mp_osd_res osd;
vo_get_src_dst_rects(vo, &src, &dst, &osd);
gl_video_resize(p->renderer, &wnd, &src, &dst, &osd, false);
gl_video_resize(p->renderer, vo->dwidth, -vo->dheight, &src, &dst, &osd);
vo->want_redraw = true;
}
@@ -198,7 +197,7 @@ static int query_format(struct vo *vo, int format)
static void video_resize_redraw_callback(struct vo *vo, int w, int h)
{
struct gl_priv *p = vo->priv;
gl_video_resize_redraw(p->renderer, w, h);
gl_video_resize_redraw(p->renderer, w, -h);
}

20
video/out/vo_opengl_cb.c
View File

@@ -66,7 +66,7 @@ struct mpv_opengl_cb_context {
int queued_frames;
struct mp_image_params img_params;
bool reconfigured;
struct mp_rect wnd;
int vp_w, vp_h;
bool flip;
bool force_update;
bool imgfmt_supported[IMGFMT_END - IMGFMT_START];
@@ -282,28 +282,22 @@ int mpv_opengl_cb_render(struct mpv_opengl_cb_context *ctx, int fbo, int vp[4])
ctx->force_update |= ctx->reconfigured;
int h = vp[3];
bool flip = h < 0 && h > INT_MIN;
if (flip)
h = -h;
struct mp_rect wnd = {vp[0], vp[1], vp[0] + vp[2], vp[1] + h};
if (wnd.x0 != ctx->wnd.x0 || wnd.y0 != ctx->wnd.y0 ||
wnd.x1 != ctx->wnd.x1 || wnd.y1 != ctx->wnd.y1 ||
ctx->flip != flip)
int vp_w = vp[2], vp_h = vp[3];
if (ctx->vp_w != vp_w || ctx->vp_h != vp_h)
ctx->force_update = true;
if (ctx->force_update && vo) {
ctx->force_update = false;
ctx->wnd = wnd;
ctx->flip = flip;
ctx->vp_w = vp_w;
ctx->vp_h = vp_h;
struct mp_rect src, dst;
struct mp_osd_res osd;
mp_get_src_dst_rects(ctx->log, &ctx->vo_opts, vo->driver->caps,
&ctx->img_params, wnd.x1 - wnd.x0, wnd.y1 - wnd.y0,
&ctx->img_params, vp_w, abs(vp_h),
1.0, &src, &dst, &osd);
gl_video_resize(ctx->renderer, &wnd, &src, &dst, &osd, !ctx->flip);
gl_video_resize(ctx->renderer, vp_w, vp_h, &src, &dst, &osd);
}
if (ctx->reconfigured)

4
wscript_build.py
View File

@@ -53,10 +53,6 @@ def build(ctx):
source = "etc/input.conf",
target = "input/input_conf.h")
ctx.file2string(
source = "video/out/gl_video_shaders.glsl",
target = "video/out/gl_video_shaders.h")
ctx.file2string(
source = "sub/osd_font.otf",
target = "sub/osd_font.h")