From e74a4d5bc0b101fbfb371942c00d3a77267dc4a6 Mon Sep 17 00:00:00 2001 From: wm4 Date: Thu, 12 Mar 2015 21:57:54 +0100 Subject: vo_opengl: refactor shader generation (part 1) 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. --- video/out/gl_osd.c | 200 ++-- video/out/gl_osd.h | 39 +- video/out/gl_utils.c | 496 +++++++++- video/out/gl_utils.h | 39 +- video/out/gl_video.c | 1925 +++++++++++++-------------------------- video/out/gl_video.h | 4 +- video/out/gl_video_shaders.glsl | 542 ----------- video/out/vo_opengl.c | 5 +- video/out/vo_opengl_cb.c | 20 +- 9 files changed, 1280 insertions(+), 1990 deletions(-) delete mode 100644 video/out/gl_video_shaders.glsl (limited to 'video/out') diff --git a/video/out/gl_osd.c b/video/out/gl_osd.c index 117d7feedb..0ab85f59c4 100644 --- a/video/out/gl_osd.c +++ b/video/out/gl_osd.c @@ -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; + int num_vertices = part->num_subparts * 6; + MP_TARRAY_GROW(part, part->vertices, num_vertices); - struct mpgl_osd_part *part = mpgl_osd_generate(ctx, imgs); - if (!part) - return; + 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; - assert(part->format != SUBBITMAP_EMPTY); + // 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) }; - if (!part->num_vertices) { - part->vertices = talloc_realloc(part, part->vertices, struct vertex, - part->packer->count * 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); + } - struct vertex *va = part->vertices; + return num_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) }; - - 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; - } - } +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); - 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); + gl_vao_draw_data(&ctx->vao, GL_TRIANGLES, part->vertices, 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; - - gl->Enable(GL_BLEND); - int div[2]; - get_3d_side_by_side(stereo_mode, div); + get_3d_side_by_side(ctx->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); + 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); } } +} - gl->Disable(GL_BLEND); +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); + + 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]; + + osd_draw(ctx->osd, s_res, pts, 0, ctx->formats, gen_osd_cb, ctx); + ctx->stereo_mode = stereo_mode; } diff --git a/video/out/gl_osd.h b/video/out/gl_osd.h index 5c34911e22..0acd200ab4 100644 --- a/video/out/gl_osd.h +++ b/video/out/gl_osd.h @@ -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 diff --git a/video/out/gl_utils.c b/video/out/gl_utils.c index 80ec840582..ca2fef10bf 100644 --- a/video/out/gl_utils.c +++ b/video/out/gl_utils.c @@ -25,6 +25,7 @@ #include #include #include +#include #include #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) { - bool res = true; - 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; + + 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); +} diff --git a/video/out/gl_utils.h b/video/out/gl_utils.h index 1934396afe..a1bb2ecafb 100644 --- a/video/out/gl_utils.h +++ b/video/out/gl_utils.h @@ -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 diff --git a/video/out/gl_video.c b/video/out/gl_video.c index 5ddb6c5cad..a52bd82020 100644 --- a/video/out/gl_video.c +++ b/video/out/gl_video.c @@ -41,11 +41,6 @@ #include "bitmap_packer.h" #include "dither.h" -static const char vo_opengl_shaders[] = -// Generated from gl_video_shaders.glsl -#include "video/out/gl_video_shaders.h" -; - // Pixel width of 1D lookup textures. #define LOOKUP_TEXTURE_SIZE 256 @@ -70,14 +65,21 @@ static const char *const fixed_scale_filters[] = { int filter_sizes[] = {2, 4, 6, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, 60, 64, 0}; +struct vertex_pt { + float x, y; +}; + struct vertex { - float position[2]; - float texcoord[2]; + struct vertex_pt position; + struct vertex_pt texcoord[4]; }; static const struct gl_vao_entry vertex_vao[] = { - {"vertex_position", 2, GL_FLOAT, false, offsetof(struct vertex, position)}, - {"vertex_texcoord", 2, GL_FLOAT, false, offsetof(struct vertex, texcoord)}, + {"position", 2, GL_FLOAT, false, offsetof(struct vertex, position)}, + {"texcoord0", 2, GL_FLOAT, false, offsetof(struct vertex, texcoord[0])}, + {"texcoord1", 2, GL_FLOAT, false, offsetof(struct vertex, texcoord[1])}, + {"texcoord2", 2, GL_FLOAT, false, offsetof(struct vertex, texcoord[2])}, + {"texcoord3", 2, GL_FLOAT, false, offsetof(struct vertex, texcoord[3])}, {0} }; @@ -85,6 +87,7 @@ struct texplane { int w, h; int tex_w, tex_h; GLint gl_internal_format; + GLenum gl_target; GLenum gl_format; GLenum gl_type; GLuint gl_texture; @@ -102,11 +105,15 @@ struct video_image { struct scaler { int index; const char *name; + double scale_factor; float params[2]; float antiring; + + bool initialized; struct filter_kernel *kernel; GLuint gl_lut; - const char *lut_name; + GLenum gl_target; + struct fbotex sep_fbo; bool insufficient; // kernel points here @@ -121,6 +128,13 @@ struct fbosurface { #define FBOSURFACES_MAX 2 +struct src_tex { + GLuint gl_tex; + GLenum gl_target; + int tex_w, tex_h; + struct mp_rect src; +}; + struct gl_video { GL *gl; @@ -131,13 +145,12 @@ struct gl_video { int depth_g; int texture_16bit_depth; // actual bits available in 16 bit textures + struct gl_shader_cache *sc; + GLenum gl_target; // texture target (GL_TEXTURE_2D, ...) for video and FBOs struct gl_vao vao; - GLuint osd_programs[SUBBITMAP_COUNT]; - GLuint indirect_program, scale_sep_program, final_program, inter_program; - struct osd_state *osd_state; struct mpgl_osd *osd; double osd_pts; @@ -146,8 +159,6 @@ struct gl_video { bool use_lut_3d; GLuint dither_texture; - float dither_quantization; - float dither_center; int dither_size; struct mp_image_params real_image_params; // configured format @@ -159,7 +170,6 @@ struct gl_video { bool is_yuv, is_rgb, is_packed_yuv; bool has_alpha; char color_swizzle[5]; - float chroma_fix[2]; float input_gamma, conv_gamma; float user_gamma; @@ -169,8 +179,9 @@ struct gl_video { struct video_image image; struct fbotex indirect_fbo; // RGB target - struct fbotex scale_sep_fbo; // first pass when doing 2 pass scaling + struct fbotex chroma_merge_fbo; struct fbosurface surfaces[FBOSURFACES_MAX]; + size_t surface_idx; // state for luma (0) and chroma (1) scalers @@ -179,9 +190,6 @@ struct gl_video { // true if scaler is currently upscaling bool upscaling; - // reinit_rendering must be called - bool need_reinit_rendering; - bool is_interpolated; struct mp_csp_equalizer video_eq; @@ -192,8 +200,11 @@ struct gl_video { struct mp_rect src_rect; // displayed part of the source video struct mp_rect dst_rect; // video rectangle on output window struct mp_osd_res osd_rect; // OSD size/margins - int vp_x, vp_y, vp_w, vp_h; // GL viewport - bool vp_vflipped; + int vp_w, vp_h; + + // temporary during rendering + struct src_tex pass_tex[4]; + bool use_indirect; int frames_rendered; @@ -203,8 +214,6 @@ struct gl_video { struct gl_hwdec *hwdec; bool hwdec_active; - - void *scratch; }; struct fmt_entry { @@ -323,6 +332,7 @@ const struct gl_video_opts gl_video_opts_def = { .sigmoid_center = 0.75, .sigmoid_slope = 6.5, .scalers = { "bilinear", "bilinear" }, + .dscaler = "bilinear", .scaler_params = {{NAN, NAN}, {NAN, NAN}}, .scaler_radius = {3, 3}, .alpha_mode = 2, @@ -431,10 +441,12 @@ const struct m_sub_options gl_video_conf = { }; static void uninit_rendering(struct gl_video *p); -static void delete_shaders(struct gl_video *p); +static void uninit_scaler(struct gl_video *p, int scaler_unit); static void check_gl_features(struct gl_video *p); static bool init_format(int fmt, struct gl_video *init); -static double get_scale_factor(struct gl_video *p); + +#define GLSL(x) gl_sc_add(p->sc, #x "\n"); +#define GLSLF(...) gl_sc_addf(p->sc, __VA_ARGS__) static const struct fmt_entry *find_tex_format(GL *gl, int bytes_per_comp, int n_channels) @@ -467,977 +479,34 @@ void gl_video_set_debug(struct gl_video *p, bool enable) gl_set_debug_logger(gl, enable ? p->log : NULL); } -// Draw a textured quad. -// x0, y0, x1, y1 = destination coordinates of the quad in pixels -// tx0, ty0, tx1, ty1 = source texture coordinates in pixels -// tex_w, tex_h = size of the texture in pixels -// flags = bits 0-1: rotate, bits 2: flip vertically -static void draw_quad(struct gl_video *p, - float x0, float y0, float x1, float y1, - float tx0, float ty0, float tx1, float ty1, - float tex_w, float tex_h, int flags) -{ - if (p->gl_target != GL_TEXTURE_2D) - tex_w = tex_h = 1.0f; - - if (flags & 4) { - float tmp = ty0; - ty0 = ty1; - ty1 = tmp; - } - - struct vertex va[4] = { - { {x0, y0}, {tx0 / tex_w, ty0 / tex_h} }, - { {x0, y1}, {tx0 / tex_w, ty1 / tex_h} }, - { {x1, y0}, {tx1 / tex_w, ty0 / tex_h} }, - { {x1, y1}, {tx1 / tex_w, ty1 / tex_h} }, - }; - - int rot = flags & 3; - while (rot--) { - static const int perm[4] = {1, 3, 0, 2}; - struct vertex vb[4]; - memcpy(vb, va, sizeof(vb)); - for (int n = 0; n < 4; n++) - memcpy(va[n].texcoord, vb[perm[n]].texcoord, sizeof(float[2])); - } - - gl_vao_draw_data(&p->vao, GL_TRIANGLE_STRIP, va, 4); - - debug_check_gl(p, "after rendering"); -} - -static void transpose3x3(float r[3][3]) -{ - MPSWAP(float, r[0][1], r[1][0]); - MPSWAP(float, r[0][2], r[2][0]); - MPSWAP(float, r[1][2], r[2][1]); -} - -static void update_uniforms(struct gl_video *p, GLuint program) -{ - GL *gl = p->gl; - GLint loc; - - if (program == 0) - return; - - gl->UseProgram(program); - - struct mp_csp_params cparams = MP_CSP_PARAMS_DEFAULTS; - cparams.gray = p->is_yuv && !p->is_packed_yuv && p->plane_count == 1; - cparams.input_bits = p->image_desc.component_bits; - cparams.texture_bits = (cparams.input_bits + 7) & ~7; - mp_csp_set_image_params(&cparams, &p->image_params); - mp_csp_copy_equalizer_values(&cparams, &p->video_eq); - if (p->image_desc.flags & MP_IMGFLAG_XYZ) { - cparams.colorspace = MP_CSP_XYZ; - cparams.input_bits = 8; - cparams.texture_bits = 8; - } - - loc = gl->GetUniformLocation(program, "transform"); - if (loc >= 0 && p->vp_w > 0 && p->vp_h > 0) { - float matrix[3][3]; - int vvp[2] = {p->vp_h, 0}; - if (p->vp_vflipped) - MPSWAP(int, vvp[0], vvp[1]); - gl_matrix_ortho2d(matrix, 0, p->vp_w, vvp[0], vvp[1]); - gl->UniformMatrix3fv(loc, 1, GL_FALSE, &matrix[0][0]); - } - - loc = gl->GetUniformLocation(program, "colormatrix"); - if (loc >= 0) { - struct mp_cmat m = {{{0}}}; - if (p->image_desc.flags & MP_IMGFLAG_XYZ) { - // Hard-coded as relative colorimetric for now, since this transforms - // from the source file's D55 material to whatever color space our - // projector/display lives in, which should be D55 for a proper - // home cinema setup either way. - mp_get_xyz2rgb_coeffs(&cparams, p->csp_src, - MP_INTENT_RELATIVE_COLORIMETRIC, &m); - } else { - mp_get_yuv2rgb_coeffs(&cparams, &m); - } - transpose3x3(m.m); // GLES2 can not transpose in glUniformMatrix3fv - gl->UniformMatrix3fv(loc, 1, GL_FALSE, &m.m[0][0]); - loc = gl->GetUniformLocation(program, "colormatrix_c"); - gl->Uniform3f(loc, m.c[0], m.c[1], m.c[2]); - } - - gl->Uniform1f(gl->GetUniformLocation(program, "input_gamma"), - p->input_gamma); - - gl->Uniform1f(gl->GetUniformLocation(program, "conv_gamma"), - p->conv_gamma); - - // Coefficients for the sigmoidal transform are taken from the - // formula here: http://www.imagemagick.org/Usage/color_mods/#sigmoidal - float sig_center = p->opts.sigmoid_center; - float sig_slope = p->opts.sigmoid_slope; - - // This function needs to go through (0,0) and (1,1) so we compute the - // values at 1 and 0, and then scale/shift them, respectively. - float sig_offset = 1.0/(1+expf(sig_slope * sig_center)); - float sig_scale = 1.0/(1+expf(sig_slope * (sig_center-1))) - sig_offset; - - gl->Uniform1f(gl->GetUniformLocation(program, "sig_center"), sig_center); - gl->Uniform1f(gl->GetUniformLocation(program, "sig_slope"), sig_slope); - gl->Uniform1f(gl->GetUniformLocation(program, "sig_scale"), sig_scale); - gl->Uniform1f(gl->GetUniformLocation(program, "sig_offset"), sig_offset); - - gl->Uniform1f(gl->GetUniformLocation(program, "inv_gamma"), - 1.0f / p->user_gamma); - - for (int n = 0; n < p->plane_count; n++) { - char textures_n[32]; - char textures_size_n[32]; - snprintf(textures_n, sizeof(textures_n), "texture%d", n); - snprintf(textures_size_n, sizeof(textures_size_n), "textures_size[%d]", n); - - gl->Uniform1i(gl->GetUniformLocation(program, textures_n), n); - if (p->gl_target == GL_TEXTURE_2D) { - gl->Uniform2f(gl->GetUniformLocation(program, textures_size_n), - p->image.planes[n].tex_w, p->image.planes[n].tex_h); - } else { - // Makes the pixel size calculation code think they are 1x1 - gl->Uniform2f(gl->GetUniformLocation(program, textures_size_n), 1, 1); - } - } - - loc = gl->GetUniformLocation(program, "chroma_div"); - if (loc >= 0) { - int xs = p->image_desc.chroma_xs; - int ys = p->image_desc.chroma_ys; - gl->Uniform2f(loc, 1.0 / (1 << xs), 1.0 / (1 << ys)); - } - - gl->Uniform2f(gl->GetUniformLocation(program, "chroma_fix"), - p->chroma_fix[0], p->chroma_fix[1]); - - loc = gl->GetUniformLocation(program, "chroma_center_offset"); - if (loc >= 0) { - int chr = p->opts.chroma_location; - if (!chr) - chr = p->image_params.chroma_location; - int cx, cy; - mp_get_chroma_location(chr, &cx, &cy); - // By default texture coordinates are such that chroma is centered with - // any chroma subsampling. If a specific direction is given, make it - // so that the luma and chroma sample line up exactly. - // For 4:4:4, setting chroma location should have no effect at all. - // luma sample size (in chroma coord. space) - float ls_w = 1.0 / (1 << p->image_desc.chroma_xs); - float ls_h = 1.0 / (1 << p->image_desc.chroma_ys); - // move chroma center to luma center (in chroma coord. space) - float o_x = ls_w < 1 ? ls_w * -cx / 2 : 0; - float o_y = ls_h < 1 ? ls_h * -cy / 2 : 0; - int c = p->gl_target == GL_TEXTURE_2D ? 1 : 0; - gl->Uniform2f(loc, o_x / FFMAX(p->image.planes[1].w * c, 1), - o_y / FFMAX(p->image.planes[1].h * c, 1)); - } - - gl->Uniform2f(gl->GetUniformLocation(program, "dither_size"), - p->dither_size, p->dither_size); - - gl->Uniform1i(gl->GetUniformLocation(program, "lut_3d"), TEXUNIT_3DLUT); - - loc = gl->GetUniformLocation(program, "cms_matrix"); - if (loc >= 0) { - float cms_matrix[3][3] = {{0}}; - // Hard-coded to relative colorimetric - for a BT.2020 3DLUT we expect - // the input to be actual BT.2020 and not something red- or blueshifted, - // and for sRGB monitors we most likely want relative scaling either way. - mp_get_cms_matrix(p->csp_src, p->csp_dest, MP_INTENT_RELATIVE_COLORIMETRIC, cms_matrix); - gl->UniformMatrix3fv(loc, 1, GL_TRUE, &cms_matrix[0][0]); - } - - for (int n = 0; n < 2; n++) { - const char *lut = p->scalers[n].lut_name; - if (lut) - gl->Uniform1i(gl->GetUniformLocation(program, lut), - TEXUNIT_SCALERS + n); - } - - gl->Uniform1i(gl->GetUniformLocation(program, "dither"), TEXUNIT_DITHER); - gl->Uniform1f(gl->GetUniformLocation(program, "dither_quantization"), - p->dither_quantization); - gl->Uniform1f(gl->GetUniformLocation(program, "dither_center"), - p->dither_center); - - float sparam1_l = p->opts.scaler_params[0][0]; - float sparam1_c = p->opts.scaler_params[1][0]; - gl->Uniform1f(gl->GetUniformLocation(program, "filter_param1_l"), - isnan(sparam1_l) ? 0.5f : sparam1_l); - gl->Uniform1f(gl->GetUniformLocation(program, "filter_param1_c"), - isnan(sparam1_c) ? 0.5f : sparam1_c); - - gl->Uniform3f(gl->GetUniformLocation(program, "translation"), 0, 0, 0); - - gl->UseProgram(0); - - debug_check_gl(p, "update_uniforms()"); -} - -static void update_all_uniforms(struct gl_video *p) -{ - for (int n = 0; n < SUBBITMAP_COUNT; n++) - update_uniforms(p, p->osd->programs[n]); - update_uniforms(p, p->indirect_program); - update_uniforms(p, p->scale_sep_program); - update_uniforms(p, p->final_program); - update_uniforms(p, p->inter_program); -} - -#define SECTION_HEADER "#!section " - -static char *get_section(void *talloc_ctx, struct bstr source, - const char *section) -{ - char *res = talloc_strdup(talloc_ctx, ""); - bool copy = false; - while (source.len) { - struct bstr line = bstr_strip_linebreaks(bstr_getline(source, &source)); - if (bstr_eatstart(&line, bstr0(SECTION_HEADER))) { - copy = bstrcmp0(line, section) == 0; - } else if (copy) { - res = talloc_asprintf_append_buffer(res, "%.*s\n", BSTR_P(line)); - } - } - return res; -} - -static char *t_concat(void *talloc_ctx, const char *s1, const char *s2) -{ - return talloc_asprintf(talloc_ctx, "%s%s", s1, s2); -} - -static GLuint create_shader(struct gl_video *p, GLenum type, const char *header, - const char *source) -{ - GL *gl = p->gl; - - void *tmp = talloc_new(NULL); - const char *full_source = t_concat(tmp, header, source); - - GLuint shader = gl->CreateShader(type); - gl->ShaderSource(shader, 1, &full_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(p->log, pri)) { - MP_MSG(p, pri, "%s shader source:\n", typestr); - mp_log_source(p->log, pri, full_source); - } - if (log_length > 1) { - GLchar *logstr = talloc_zero_size(tmp, log_length + 1); - gl->GetShaderInfoLog(shader, log_length, NULL, logstr); - MP_MSG(p, pri, "%s shader compile log (status=%d):\n%s\n", - typestr, status, logstr); - } - - talloc_free(tmp); - - return shader; -} - -static void prog_create_shader(struct gl_video *p, GLuint program, GLenum type, - const char *header, const char *source) -{ - GL *gl = p->gl; - GLuint shader = create_shader(p, type, header, source); - gl->AttachShader(program, shader); - gl->DeleteShader(shader); -} - -static void link_shader(struct gl_video *p, GLuint program) -{ - GL *gl = p->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(p->log, pri)) { - GLchar *logstr = talloc_zero_size(NULL, log_length + 1); - gl->GetProgramInfoLog(program, log_length, NULL, logstr); - MP_MSG(p, pri, "shader link log (status=%d): %s\n", status, logstr); - talloc_free(logstr); - } -} - -#define PRELUDE_END "// -- prelude end\n" - -static GLuint create_program(struct gl_video *p, const char *name, - const char *header, const char *vertex, - const char *frag, struct gl_vao *vao) -{ - GL *gl = p->gl; - MP_VERBOSE(p, "compiling shader program '%s', header:\n", name); - const char *real_header = strstr(header, PRELUDE_END); - real_header = real_header ? real_header + strlen(PRELUDE_END) : header; - mp_log_source(p->log, MSGL_V, real_header); - GLuint prog = gl->CreateProgram(); - prog_create_shader(p, prog, GL_VERTEX_SHADER, header, vertex); - prog_create_shader(p, prog, GL_FRAGMENT_SHADER, header, frag); - gl_vao_bind_attribs(vao, prog); - link_shader(p, prog); - return prog; -} - -static void shader_def(char **shader, const char *name, - const char *value) -{ - *shader = talloc_asprintf_append(*shader, "#define %s %s\n", name, value); -} - -static void shader_def_opt(char **shader, const char *name, bool b) -{ - if (b) - shader_def(shader, name, "1"); -} - -#define APPENDF(s_ptr, ...) \ - *(s_ptr) = talloc_asprintf_append(*(s_ptr), __VA_ARGS__) - -static void shader_setup_scaler(char **shader, struct scaler *scaler, int pass) -{ - int unit = scaler->index; - const char *target = unit == 0 ? "SAMPLE" : "SAMPLE_C"; - if (!scaler->kernel) { - APPENDF(shader, "#define %s(p0, p1, p2) " - "sample_%s(p0, p1, p2, filter_param1_%c)\n", - target, scaler->name, "lc"[unit]); - } else { - int size = scaler->kernel->size; - const char *lut_tex = scaler->lut_name; - char name[40]; - snprintf(name, sizeof(name), "sample_scaler%d", unit); - APPENDF(shader, "#define DEF_SCALER%d \\\n ", unit); - char lut_fn[40]; - if (scaler->kernel->polar) { - double radius = scaler->kernel->radius; - int bound = (int)ceil(radius); - // SAMPLE_CONVOLUTION_POLAR_R(NAME, R, LUT, WEIGHTS_FN, ANTIRING) - APPENDF(shader, "SAMPLE_CONVOLUTION_POLAR_R(%s, %f, %s, WEIGHTS%d, %f)\n", - name, radius, lut_tex, unit, scaler->antiring); - - // Pre-compute unrolled weights matrix - APPENDF(shader, "#define WEIGHTS%d(LUT) \\\n ", unit); - for (int y = 1-bound; y <= bound; y++) { - for (int x = 1-bound; x <= bound; 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); - - if (d < radius - 1) { - // Samples definitely inside the main ring - APPENDF(shader, "SAMPLE_POLAR_%s(LUT, %f, %d, %d) \\\n ", - // The center 4 coefficients are the primary - // contributors, used to clamp the result for - // anti-ringing - (x >= 0 && y >= 0 && x <= 1 && y <= 1) - ? "PRIMARY" : "HELPER", - radius, x, y); - } else if (d < radius) { - // Samples on the edge, these are potential values - APPENDF(shader, "SAMPLE_POLAR_POTENTIAL(LUT, %f, %d, %d) \\\n ", - radius, x, y); - } - } - } - APPENDF(shader, "\n"); - } else { - if (size == 2 || size == 6) { - snprintf(lut_fn, sizeof(lut_fn), "weights%d", size); - } else { - snprintf(lut_fn, sizeof(lut_fn), "weights_scaler%d", unit); - APPENDF(shader, "WEIGHTS_N(%s, %d) \\\n ", lut_fn, size); - } - if (pass != -1) { - // The direction/pass assignment is rather arbitrary, but fixed in - // other parts of the code (like FBO setup). - const char *direction = pass == 0 ? "0, 1" : "1, 0"; - // SAMPLE_CONVOLUTION_SEP_N(NAME, DIR, N, LUT, WEIGHTS_FUNC, ANTIRING) - APPENDF(shader, "SAMPLE_CONVOLUTION_SEP_N(%s, vec2(%s), %d, %s, %s, %f)\n", - name, direction, size, lut_tex, lut_fn, scaler->antiring); - } else { - // SAMPLE_CONVOLUTION_N(NAME, N, LUT, WEIGHTS_FUNC) - APPENDF(shader, "SAMPLE_CONVOLUTION_N(%s, %d, %s, %s)\n", - name, size, lut_tex, lut_fn); - } - } - APPENDF(shader, "#define %s %s\n", target, name); - } -} - -static void compile_shaders(struct gl_video *p) -{ - GL *gl = p->gl; - - debug_check_gl(p, "before shaders"); - - delete_shaders(p); - - void *tmp = talloc_new(NULL); - - struct bstr src = bstr0(vo_opengl_shaders); - char *vertex_shader = get_section(tmp, src, "vertex_all"); - char *shader_prelude = get_section(tmp, src, "prelude"); - char *s_video = get_section(tmp, src, "frag_video"); - - bool rg = gl->mpgl_caps & MPGL_CAP_TEX_RG; - bool tex1d = gl->mpgl_caps & MPGL_CAP_1D_TEX; - bool tex3d = gl->mpgl_caps & MPGL_CAP_3D_TEX; - bool arrays = gl->mpgl_caps & MPGL_CAP_1ST_CLASS_ARRAYS; - char *header = - talloc_asprintf(tmp, "#version %d%s\n" - "#define HAVE_RG %d\n" - "#define HAVE_1DTEX %d\n" - "#define HAVE_3DTEX %d\n" - "#define HAVE_ARRAYS %d\n" - "%s%s", - gl->glsl_version, gl->es >= 300 ? " es" : "", - rg, tex1d, tex3d, arrays, shader_prelude, PRELUDE_END); - - bool use_cms = p->opts.srgb || p->use_lut_3d; - // 3DLUT overrides sRGB - bool use_srgb = p->opts.srgb && !p->use_lut_3d; - - float input_gamma = 1.0; - float conv_gamma = 1.0; - - bool is_xyz = p->image_desc.flags & MP_IMGFLAG_XYZ; - if (is_xyz) { - input_gamma *= 2.6; - // Note that this results in linear light, so we make sure to enable - // use_linear_light for XYZ inputs as well. - } - - p->input_gamma = input_gamma; - p->conv_gamma = conv_gamma; - - bool use_input_gamma = p->input_gamma != 1.0; - bool use_conv_gamma = p->conv_gamma != 1.0; - bool use_const_luma = p->image_params.colorspace == MP_CSP_BT_2020_C; - enum mp_csp_trc gamma_fun = p->image_params.gamma; - - // If either color correction option (3dlut or srgb) is enabled, or if - // sigmoidal upscaling is requested, or if the source is linear XYZ, we - // always scale in linear light - bool use_linear_light = p->opts.linear_scaling || p->opts.sigmoid_upscaling - || use_cms || is_xyz; - - // The inverse of the above transformation is normally handled by - // the CMS cases, but if CMS is disabled we need to go back manually - bool use_inv_bt1886 = false; - if (use_linear_light && !use_cms) { - if (gamma_fun == MP_CSP_TRC_SRGB) { - use_srgb = true; - } else { - use_inv_bt1886 = true; - } - } - - // Optionally transform to sigmoidal color space if requested. - p->sigmoid_enabled = p->opts.sigmoid_upscaling; - bool use_sigmoid = p->sigmoid_enabled && p->upscaling; - - // Figure out the right color spaces we need to convert, if any - enum mp_csp_prim prim_src = p->image_params.primaries, prim_dest; - if (use_cms) { - // sRGB mode wants sRGB aka BT.709 primaries, but the 3DLUT is - // always built against BT.2020. - prim_dest = p->opts.srgb ? MP_CSP_PRIM_BT_709 : MP_CSP_PRIM_BT_2020; - } else { - // If no CMS is being done we just want to output stuff as-is, - // in the native colorspace of the source. - prim_dest = prim_src; - } - - // XYZ input has no defined input color space, so we can directly convert - // it to whatever output space we actually need. - if (p->image_desc.flags & MP_IMGFLAG_XYZ) - prim_src = prim_dest; - - // Set the colorspace primaries and figure out whether we need to perform - // an extra conversion. - p->csp_src = mp_get_csp_primaries(prim_src); - p->csp_dest = mp_get_csp_primaries(prim_dest); - - bool use_cms_matrix = prim_src != prim_dest; - - if (p->gl_target == GL_TEXTURE_RECTANGLE) { - shader_def(&header, "VIDEO_SAMPLER", "sampler2DRect"); - shader_def_opt(&header, "USE_RECTANGLE", true); - } else { - shader_def(&header, "VIDEO_SAMPLER", "sampler2D"); - } - - // Need to pass alpha through the whole chain. (Not needed for OSD shaders.) - if (p->opts.alpha_mode == 1) - shader_def_opt(&header, "USE_ALPHA", p->has_alpha); - - char *header_osd = talloc_strdup(tmp, header); - shader_def_opt(&header_osd, "USE_OSD_LINEAR_CONV_BT1886", - use_cms && gamma_fun == MP_CSP_TRC_BT_1886); - shader_def_opt(&header_osd, "USE_OSD_LINEAR_CONV_SRGB", - use_cms && gamma_fun == MP_CSP_TRC_SRGB); - shader_def_opt(&header_osd, "USE_OSD_CMS_MATRIX", use_cms_matrix); - shader_def_opt(&header_osd, "USE_OSD_3DLUT", p->use_lut_3d); - shader_def_opt(&header_osd, "USE_OSD_SRGB", use_cms && use_srgb); - - for (int n = 0; n < SUBBITMAP_COUNT; n++) { - const char *name = osd_shaders[n]; - if (name) { - char *s_osd = get_section(tmp, src, name); - p->osd_programs[n] = create_program(p, name, header_osd, - vertex_shader, s_osd, - &p->osd->vao); - } - } - - struct gl_vao *v = &p->vao; // VAO to use to draw primitives - - char *header_conv = talloc_strdup(tmp, ""); - char *header_final = talloc_strdup(tmp, ""); - char *header_inter = talloc_strdup(tmp, ""); - char *header_sep = NULL; - - if (p->image_desc.id == IMGFMT_NV12 || p->image_desc.id == IMGFMT_NV21) { - shader_def(&header_conv, "USE_CONV", "CONV_NV12"); - } else if (p->plane_count > 1) { - shader_def(&header_conv, "USE_CONV", "CONV_PLANAR"); - } - - if (p->color_swizzle[0]) - shader_def(&header_conv, "USE_COLOR_SWIZZLE", p->color_swizzle); - shader_def_opt(&header_conv, "USE_INPUT_GAMMA", use_input_gamma); - shader_def_opt(&header_conv, "USE_COLORMATRIX", !p->is_rgb); - shader_def_opt(&header_conv, "USE_CONV_GAMMA", use_conv_gamma); - shader_def_opt(&header_conv, "USE_CONST_LUMA", use_const_luma); - shader_def_opt(&header_conv, "USE_LINEAR_LIGHT_BT1886", - use_linear_light && gamma_fun == MP_CSP_TRC_BT_1886); - shader_def_opt(&header_conv, "USE_LINEAR_LIGHT_SRGB", - use_linear_light && gamma_fun == MP_CSP_TRC_SRGB); - shader_def_opt(&header_conv, "USE_SIGMOID", use_sigmoid); - if (p->opts.alpha_mode > 0 && p->has_alpha && p->plane_count > 3) - shader_def(&header_conv, "USE_ALPHA_PLANE", "3"); - if (p->opts.alpha_mode == 2 && p->has_alpha) - shader_def(&header_conv, "USE_ALPHA_BLEND", "1"); - shader_def_opt(&header_conv, "USE_CHROMA_FIX", - p->chroma_fix[0] != 1.0f || p->chroma_fix[1] != 1.0f); - - shader_def_opt(&header_final, "USE_SIGMOID_INV", use_sigmoid); - shader_def_opt(&header_final, "USE_INV_GAMMA", p->user_gamma_enabled); - shader_def_opt(&header_final, "USE_CMS_MATRIX", use_cms_matrix); - shader_def_opt(&header_final, "USE_3DLUT", p->use_lut_3d); - shader_def_opt(&header_final, "USE_SRGB", use_srgb); - shader_def_opt(&header_final, "USE_INV_BT1886", use_inv_bt1886); - shader_def_opt(&header_final, "USE_DITHER", p->dither_texture != 0); - shader_def_opt(&header_final, "USE_TEMPORAL_DITHER", p->opts.temporal_dither); - - if (p->scalers[0].kernel && !p->scalers[0].kernel->polar) { - header_sep = talloc_strdup(tmp, ""); - shader_def_opt(&header_sep, "FIXED_SCALE", true); - shader_setup_scaler(&header_sep, &p->scalers[0], 0); - shader_setup_scaler(&header_inter, &p->scalers[0], 1); - } else { - shader_setup_scaler(&header_inter, &p->scalers[0], -1); - } - - bool use_interpolation = p->opts.smoothmotion; - - if (use_interpolation) { - shader_def_opt(&header_inter, "FIXED_SCALE", true); - shader_def_opt(&header_final, "USE_LINEAR_INTERP