Merge branch 'eosd' into my_master

7 files changed, 504 insertions, 346 deletions

diff --git a/Makefile b/Makefile
index 55d69f9fa1..bdd340a097 100644
--- a/Makefile
+++ b/Makefile
@@ -403,6 +403,7 @@ SRCS_COMMON = asxparser.c \
               libmpdemux/yuv4mpeg.c \
               libmpdemux/yuv4mpeg_ratio.c \
               libvo/osd.c \
+              libvo/eosd_packer.c \
               osdep/numcores.c \
               osdep/$(GETCH) \
               osdep/$(TIMER) \
diff --git a/libvo/eosd_packer.c b/libvo/eosd_packer.c
new file mode 100644
index 0000000000..103648b7c4
--- /dev/null
+++ b/libvo/eosd_packer.c
@@ -0,0 +1,254 @@
+/*
+ * Common code for packing EOSD images into larger surfaces.
+ *
+ * This file is part of mplayer2.
+ *
+ * mplayer2 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.
+ *
+ * mplayer2 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 mplayer2; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#include <libavutil/common.h>
+#include "talloc.h"
+#include "mp_msg.h"
+#include "eosd_packer.h"
+
+// Initial size of EOSD surface in pixels (x*x)
+#define EOSD_SURFACE_INITIAL_SIZE 256
+
+// Allocate an eosd_packer, which can be used to layout and cache the list of
+// EOSD images contained in a mp_eosd_images_t into a flat surface.
+// It can be free'd with talloc_free().
+// Don't forget to call eosd_init() before using it.
+struct eosd_packer *eosd_packer_create(void *talloc_ctx) {
+    return talloc_zero(talloc_ctx, struct eosd_packer);
+}
+
+// Call this when you need to completely reinitialize the EOSD state, e.g. when
+// when your EOSD surface was deleted.
+// max_width and max_height are the maximum surface sizes that should be
+// allowed.
+void eosd_packer_reinit(struct eosd_packer *state, uint32_t max_width,
+                        uint32_t max_height)
+{
+    state->max_surface_width = max_width;
+    state->max_surface_height = max_height;
+    state->surface.w = 0;
+    state->surface.h = 0;
+    state->targets_count = 0;
+}
+
+#define HEIGHT_SORT_BITS 4
+static int size_index(struct eosd_target *r)
+{
+    unsigned int h = r->source.y1;
+    int n = av_log2_16bit(h);
+    return (n << HEIGHT_SORT_BITS)
+        + (- 1 - (h << HEIGHT_SORT_BITS >> n) & (1 << HEIGHT_SORT_BITS) - 1);
+}
+
+/* Pack the given rectangles into an area of size w * h.
+ * The size of each rectangle is read from .source.x1/.source.y1.
+ * The height of each rectangle must be at least 1 and less than 65536.
+ * The .source rectangle is then set corresponding to the packed position.
+ * 'scratch' must point to work memory for num_rects+16 ints.
+ * Return 0 on success, -1 if the rectangles did not fit in w*h.
+ *
+ * The rectangles are placed in rows in order approximately sorted by
+ * height (the approximate sorting is simpler than a full one would be,
+ * and allows the algorithm to work in linear time). Additionally, to
+ * reduce wasted space when there are a few tall rectangles, empty
+ * lower-right parts of rows are filled recursively when the size of
+ * rectangles in the row drops past a power-of-two threshold. So if a
+ * row starts with rectangles of size 3x50, 10x40 and 5x20 then the
+ * free rectangle with corners (13, 20)-(w, 50) is filled recursively.
+ */
+static int pack_rectangles(struct eosd_target *rects, int num_rects,
+                           int w, int h, int *scratch)
+{
+    int bins[16 << HEIGHT_SORT_BITS];
+    int sizes[16 << HEIGHT_SORT_BITS] = {};
+    for (int i = 0; i < num_rects; i++)
+        sizes[size_index(rects + i)]++;
+    int idx = 0;
+    for (int i = 0; i < 16 << HEIGHT_SORT_BITS; i += 1 << HEIGHT_SORT_BITS) {
+        for (int j = 0; j < 1 << HEIGHT_SORT_BITS; j++) {
+            bins[i + j] = idx;
+            idx += sizes[i + j];
+        }
+        scratch[idx++] = -1;
+    }
+    for (int i = 0; i < num_rects; i++)
+        scratch[bins[size_index(rects + i)]++] = i;
+    for (int i = 0; i < 16; i++)
+        bins[i] = bins[i << HEIGHT_SORT_BITS] - sizes[i << HEIGHT_SORT_BITS];
+    struct {
+        int size, x, bottom;
+    } stack[16] = {{15, 0, h}}, s = {};
+    int stackpos = 1;
+    int y;
+    while (stackpos) {
+        y = s.bottom;
+        s = stack[--stackpos];
+        s.size++;
+        while (s.size--) {
+            int maxy = -1;
+            int obj;
+            while ((obj = scratch[bins[s.size]]) >= 0) {
+                int bottom = y + rects[obj].source.y1;
+                if (bottom > s.bottom)
+                    break;
+                int right = s.x + rects[obj].source.x1;
+                if (right > w)
+                    break;
+                bins[s.size]++;
+                rects[obj].source.x0 = s.x;
+                rects[obj].source.x1 += s.x;
+                rects[obj].source.y0 = y;
+                rects[obj].source.y1 += y;
+                num_rects--;
+                if (maxy <= 0)
+                    stack[stackpos++] = s;
+                s.x = right;
+                maxy = FFMAX(maxy, bottom);
+            }
+            if (maxy > 0)
+                s.bottom = maxy;
+        }
+    }
+    return num_rects ? -1 : 0;
+}
+
+// padding to reduce interpolation artifacts when doing scaling & filtering
+#define EOSD_PADDING 0
+
+// Release all previous images, and packs the images in imgs into state. The
+// caller must check the change variables:
+// *out_need_reposition == true: sub-image positions changed
+// *out_need_upload == true: upload all sub-images again
+// *out_need_reallocate == true: resize the EOSD texture to state->surface.w/h
+// Logical implications: need_reallocate => need_upload => need_reposition
+void eosd_packer_generate(struct eosd_packer *state, mp_eosd_images_t *imgs,
+                          bool *out_need_reposition, bool *out_need_upload,
+                          bool *out_need_reallocate)
+{
+    int i;
+    ASS_Image *img = imgs->imgs;
+    ASS_Image *p;
+    struct eosd_surface *sfc = &state->surface;
+
+    *out_need_reposition = false;
+    *out_need_upload = false;
+    *out_need_reallocate = false;
+
+    int change_state = imgs->changed;
+
+    // eosd_reinit() was probably called, force full reupload.
+    if (state->targets_count == 0 && img)
+        change_state = 2;
+
+    if (change_state == 0)
+        return; // Nothing changed, no need to redraw
+
+    state->targets_count = 0;
+
+    *out_need_reposition = true;
+
+    if (!img)
+        return; // There's nothing to render!
+
+    if (change_state == 1)
+        goto eosd_skip_upload;
+
+    *out_need_upload = true;
+    while (1) {
+        for (p = img, i = 0; p; p = p->next) {
+            if (p->w <= 0 || p->h <= 0)
+                continue;
+            // Allocate new space for surface/target arrays
+            if (i >= state->targets_size) {
+                state->targets_size = FFMAX(state->targets_size * 2, 512);
+                state->targets  =
+                    talloc_realloc_size(state, state->targets,
+                                        state->targets_size
+                                        * sizeof(*state->targets));
+                state->scratch =
+                    talloc_realloc_size(state, state->scratch,
+                                        (state->targets_size + 16)
+                                        * sizeof(*state->scratch));
+            }
+            state->targets[i].source.x1 = p->w + EOSD_PADDING;
+            state->targets[i].source.y1 = p->h + EOSD_PADDING;
+            i++;
+        }
+        if (pack_rectangles(state->targets, i, sfc->w, sfc->h,
+                            state->scratch) >= 0)
+            break;
+        int w = FFMIN(FFMAX(sfc->w * 2, EOSD_SURFACE_INITIAL_SIZE),
+                      state->max_surface_width);
+        int h = FFMIN(FFMAX(sfc->h * 2, EOSD_SURFACE_INITIAL_SIZE),
+                      state->max_surface_height);
+        if (w == sfc->w && h == sfc->h) {
+            mp_msg(MSGT_VO, MSGL_ERR, "[eosd] EOSD bitmaps do not fit on "
+                   "a surface with the maximum supported size\n");
+            return;
+        }
+        sfc->w = w;
+        sfc->h = h;
+        *out_need_reallocate = true;
+    }
+    if (*out_need_reallocate) {
+        mp_msg(MSGT_VO, MSGL_V, "[eosd] Allocate a %dx%d surface for "
+               "EOSD bitmaps.\n", sfc->w, sfc->h);
+    }
+
+eosd_skip_upload:
+    for (p = img; p; p = p->next) {
+        if (p->w <= 0 || p->h <= 0)
+            continue;
+        struct eosd_target *target = &state->targets[state->targets_count];
+        target->source.x1 -= EOSD_PADDING;
+        target->source.y1 -= EOSD_PADDING;
+        target->dest.x0 = p->dst_x;
+        target->dest.y0 = p->dst_y;
+        target->dest.x1 = p->w + p->dst_x;
+        target->dest.y1 = p->h + p->dst_y;
+        target->color = p->color;
+        target->ass_img = p;
+        state->targets_count++;
+    }
+}
+
+// Calculate the bounding box of all sub-rectangles in the EOSD surface that
+// will be used for EOSD rendering.
+// If the bounding box is empty, return false.
+bool eosd_packer_calculate_source_bb(struct eosd_packer *state,
+                                     struct eosd_rect *out_bb)
+{
+    struct eosd_rect bb = { state->surface.w, state->surface.h, 0, 0 };
+
+    for (int n = 0; n < state->targets_count; n++) {
+        struct eosd_rect s = state->targets[n].source;
+        bb.x0 = FFMIN(bb.x0, s.x0);
+        bb.y0 = FFMIN(bb.y0, s.y0);
+        bb.x1 = FFMAX(bb.x1, s.x1);
+        bb.y1 = FFMAX(bb.y1, s.y1);
+    }
+
+    // avoid degenerate bounding box if empty
+    bb.x0 = FFMIN(bb.x0, bb.x1);
+    bb.y0 = FFMIN(bb.y0, bb.y1);
+
+    *out_bb = bb;
+    return state->targets_count > 0;
+}
diff --git a/libvo/eosd_packer.h b/libvo/eosd_packer.h
new file mode 100644
index 0000000000..e207a4a2dd
--- /dev/null
+++ b/libvo/eosd_packer.h
@@ -0,0 +1,71 @@
+/*
+ * This file is part of mplayer2.
+ *
+ * mplayer2 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.
+ *
+ * mplayer2 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 mplayer2; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#ifndef MPLAYER_EOSD_PACKER_H
+#define MPLAYER_EOSD_PACKER_H
+
+#include <inttypes.h>
+#include <stdbool.h>
+
+#include "sub/ass_mp.h"
+
+// Pool of surfaces
+struct eosd_surface {
+    //void *native_surface;
+    int w;
+    int h;
+};
+
+struct eosd_rect {
+    int x0, y0, x1, y1;
+};
+
+// List of surfaces to be rendered
+struct eosd_target {
+    struct eosd_rect source;    // position in EOSD surface
+    struct eosd_rect dest;      // position on screen
+    uint32_t color;             // libass-style color of the image
+    // NOTE: This must not be accessed after you return from your VO's
+    //       VOCTRL_DRAW_EOSD call - libass will free or reuse the associated
+    //       memory. Feel free to set this to NULL to make erroneous accesses to
+    //       this member fail early.
+    ASS_Image *ass_img;
+};
+
+struct eosd_packer {
+    struct eosd_surface surface;
+    struct eosd_target *targets;
+    int targets_count;  // number of valid elements in targets
+    int targets_size;   // number of allocated elements in targets
+
+    uint32_t max_surface_width;
+    uint32_t max_surface_height;
+
+    int *scratch;
+};
+
+struct eosd_packer *eosd_packer_create(void *talloc_ctx);
+void eosd_packer_reinit(struct eosd_packer *state, uint32_t max_width,
+                        uint32_t max_height);
+void eosd_packer_generate(struct eosd_packer *state, mp_eosd_images_t *imgs,
+                          bool *out_need_reposition, bool *out_need_upload,
+                          bool *out_need_reallocate);
+bool eosd_packer_calculate_source_bb(struct eosd_packer *state,
+                                     struct eosd_rect *out_bb);
+
+#endif /* MPLAYER_EOSD_PACKER_H */
diff --git a/libvo/gl_common.c b/libvo/gl_common.c
index 7cca800d40..aa34521802 100644
--- a/libvo/gl_common.c
+++ b/libvo/gl_common.c
@@ -360,6 +360,12 @@ static const extfunc_desc_t extfuncs[] = {
     DEF_FUNC_DESC(ColorMask),
     DEF_FUNC_DESC(ReadPixels),
     DEF_FUNC_DESC(ReadBuffer),
+    DEF_FUNC_DESC(VertexPointer),
+    DEF_FUNC_DESC(ColorPointer),
+    DEF_FUNC_DESC(TexCoordPointer),
+    DEF_FUNC_DESC(DrawArrays),
+    DEF_FUNC_DESC(EnableClientState),
+    DEF_FUNC_DESC(DisableClientState),
 
     DEF_EXT_DESC(GenBuffers, NULL,
                  ("glGenBuffers", "glGenBuffersARB")),
diff --git a/libvo/gl_common.h b/libvo/gl_common.h
index 8e3294f441..5a0fd9b8a8 100644
--- a/libvo/gl_common.h
+++ b/libvo/gl_common.h
@@ -463,6 +463,13 @@ struct GL {
     void (GLAPIENTRY *ReadPixels)(GLint, GLint, GLsizei, GLsizei, GLenum,
                                   GLenum, GLvoid *);
     void (GLAPIENTRY *ReadBuffer)(GLenum);
+    void (GLAPIENTRY *VertexPointer)(GLint, GLenum, GLsizei, const GLvoid *);
+    void (GLAPIENTRY *ColorPointer)(GLint, GLenum, GLsizei, const GLvoid *);
+    void (GLAPIENTRY *TexCoordPointer)(GLint, GLenum, GLsizei, const GLvoid *);
+    void (GLAPIENTRY *DrawArrays)(GLenum, GLint, GLsizei);
+    void (GLAPIENTRY *EnableClientState)(GLenum);
+    void (GLAPIENTRY *DisableClientState)(GLenum);
+
 
     // OpenGL extension functions
     void (GLAPIENTRY *GenBuffers)(GLsizei, GLuint *);
diff --git a/libvo/vo_gl.c b/libvo/vo_gl.c
index e68465f009..7bf5cf2ec1 100644
--- a/libvo/vo_gl.c
+++ b/libvo/vo_gl.c
@@ -38,6 +38,7 @@
 #include "osd.h"
 #include "sub/font_load.h"
 #include "sub/sub.h"
+#include "eosd_packer.h"
 
 #include "gl_common.h"
 #include "aspect.h"
@@ -49,19 +50,17 @@ static int preinit_nosw(struct vo *vo, const char *arg);
 //! How many parts the OSD may consist of at most
 #define MAX_OSD_PARTS 20
 
-#define LARGE_EOSD_TEX_SIZE 512
-#define TINYTEX_SIZE 16
-#define TINYTEX_COLS (LARGE_EOSD_TEX_SIZE / TINYTEX_SIZE)
-#define TINYTEX_MAX (TINYTEX_COLS * TINYTEX_COLS)
-#define SMALLTEX_SIZE 32
-#define SMALLTEX_COLS (LARGE_EOSD_TEX_SIZE / SMALLTEX_SIZE)
-#define SMALLTEX_MAX (SMALLTEX_COLS * SMALLTEX_COLS)
-
 //for gl_priv.use_yuv
 #define MASK_ALL_YUV (~(1 << YUV_CONVERSION_NONE))
 #define MASK_NOT_COMBINERS (~((1 << YUV_CONVERSION_NONE) | (1 << YUV_CONVERSION_COMBINERS)))
 #define MASK_GAMMA_SUPPORT (MASK_NOT_COMBINERS & ~(1 << YUV_CONVERSION_FRAGMENT))
 
+struct vertex_eosd {
+    float x, y;
+    uint8_t color[4];
+    float u, v;
+};
+
 struct gl_priv {
     MPGLContext *glctx;
     GL *gl;
@@ -74,17 +73,17 @@ struct gl_priv {
     //! Alpha textures for OSD
     GLuint osdatex[MAX_OSD_PARTS];
 #endif
-    GLuint *eosdtex;
-    GLuint largeeosdtex[2];
+    GLuint eosd_texture;
+    int eosd_texture_width, eosd_texture_height;
+    struct eosd_packer *eosd;
+    struct vertex_eosd *eosd_va;
     //! Display lists that draw the OSD parts
     GLuint osdDispList[MAX_OSD_PARTS];
 #ifndef FAST_OSD
     GLuint osdaDispList[MAX_OSD_PARTS];
 #endif
-    GLuint eosdDispList;
     //! How many parts the OSD currently consists of
     int osdtexCnt;
-    int eosdtexCnt;
     int osd_color;
 
     int use_ycbcr;
@@ -285,150 +284,111 @@ static void clearOSD(struct vo *vo)
 }
 
 /**
- * \brief remove textures, display list and free memory used by EOSD
+ * \brief construct display list from ass image list
+ * \param img image list to create OSD from.
  */
-static void clearEOSD(struct vo *vo)
+static void genEOSD(struct vo *vo, mp_eosd_images_t *imgs)
 {
     struct gl_priv *p = vo->priv;
     GL *gl = p->gl;
 
-    if (p->eosdDispList)
-        gl->DeleteLists(p->eosdDispList, 1);
-    p->eosdDispList = 0;
-    if (p->eosdtexCnt)
-        gl->DeleteTextures(p->eosdtexCnt, p->eosdtex);
-    p->eosdtexCnt = 0;
-    free(p->eosdtex);
-    p->eosdtex = NULL;
-}
+    bool need_repos, need_upload, need_allocate;
+    eosd_packer_generate(p->eosd, imgs, &need_repos, &need_upload,
+                         &need_allocate);
 
-static inline int is_tinytex(ASS_Image *i, int tinytexcur)
-{
-    return i->w < TINYTEX_SIZE && i->h < TINYTEX_SIZE
-           && tinytexcur < TINYTEX_MAX;
-}
+    if (!need_repos)
+        return;
 
-static inline int is_smalltex(ASS_Image *i, int smalltexcur)
-{
-    return i->w < SMALLTEX_SIZE && i->h < SMALLTEX_SIZE
-           && smalltexcur < SMALLTEX_MAX;
-}
+    if (!p->eosd_texture)
+        gl->GenTextures(1, &p->eosd_texture);
 
-static inline void tinytex_pos(int tinytexcur, int *x, int *y)
-{
-    *x = (tinytexcur % TINYTEX_COLS) * TINYTEX_SIZE;
-    *y = (tinytexcur / TINYTEX_COLS) * TINYTEX_SIZE;
-}
+    gl->BindTexture(p->target, p->eosd_texture);
 
-static inline void smalltex_pos(int smalltexcur, int *x, int *y)
-{
-    *x = (smalltexcur % SMALLTEX_COLS) * SMALLTEX_SIZE;
-    *y = (smalltexcur / SMALLTEX_COLS) * SMALLTEX_SIZE;
+    if (need_allocate) {
+        texSize(vo, p->eosd->surface.w, p->eosd->surface.h,
+                &p->eosd_texture_width, &p->eosd_texture_height);
+        // xxx it doesn't need to be cleared, that's a waste of time
+        glCreateClearTex(gl, p->target, GL_ALPHA, GL_ALPHA, GL_UNSIGNED_BYTE,
+                         GL_NEAREST, p->eosd_texture_width,
+                         p->eosd_texture_height, 0);
+    }
+
+    // 2 triangles primitives per quad = 6 vertices per quad
+    // not using GL_QUADS, as it is deprecated in OpenGL 3.x and later
+    p->eosd_va = talloc_realloc_size(p->eosd, p->eosd_va,
+                                     p->eosd->targets_count
+                                     * sizeof(struct vertex_eosd) * 6);
+
+    float eosd_w = p->eosd_texture_width;
+    float eosd_h = p->eosd_texture_height;
+
+    if (p->use_rectangle == 1)
+        eosd_w = eosd_h = 1.0f;
+
+    for (int n = 0; n < p->eosd->targets_count; n++) {
+        struct eosd_target *target = &p->eosd->targets[n];
+        ASS_Image *i = target->ass_img;
+
+        if (need_upload) {
+            glUploadTex(gl, p->target, GL_ALPHA, GL_UNSIGNED_BYTE, i->bitmap,
+                        i->stride, target->source.x0, target->source.y0,
+                        i->w, i->h, 0);
+        }
+
+        uint8_t color[4] = { i->color >> 24, (i->color >> 16) & 0xff,
+                            (i->color >> 8) & 0xff, 255 - (i->color & 0xff) };
+
+        float x0 = target->dest.x0;
+        float y0 = target->dest.y0;
+        float x1 = target->dest.x1;
+        float y1 = target->dest.y1;
+        float tx0 = target->source.x0 / eosd_w;
+        float ty0 = target->source.y0 / eosd_h;
+        float tx1 = target->source.x1 / eosd_w;
+        float ty1 = target->source.y1 / eosd_h;
+
+#define COLOR_INIT {color[0], color[1], color[2], color[3]}
+        struct vertex_eosd *va = &p->eosd_va[n * 6];
+        va[0] = (struct vertex_eosd) { x0, y0, COLOR_INIT, tx0, ty0 };
+        va[1] = (struct vertex_eosd) { x0, y1, COLOR_INIT, tx0, ty1 };
+        va[2] = (struct vertex_eosd) { x1, y0, COLOR_INIT, tx1, ty0 };
+        va[3] = (struct vertex_eosd) { x1, y1, COLOR_INIT, tx1, ty1 };
+        va[4] = va[2];
+        va[5] = va[1];
+#undef COLOR_INIT
+    }
+
+    gl->BindTexture(p->target, 0);
 }
 
-/**
- * \brief construct display list from ass image list
- * \param img image list to create OSD from.
- *            A value of NULL has the same effect as clearEOSD()
- */
-static void genEOSD(struct vo *vo, mp_eosd_images_t *imgs)
+// Note: relies on state being setup, like projection matrix and blending
+static void drawEOSD(struct vo *vo)
 {
     struct gl_priv *p = vo->priv;
     GL *gl = p->gl;
 
-    int sx, sy;
-    int tinytexcur = 0;
-    int smalltexcur = 0;
-    GLuint *curtex;
-    GLint scale_type = p->scaled_osd ? GL_LINEAR : GL_NEAREST;
-    ASS_Image *img = imgs->imgs;
-    ASS_Image *i;
-
-    if (imgs->changed == 0) // there are elements, but they are unchanged
+    if (p->eosd->targets_count == 0)
         return;
-    if (img && imgs->changed == 1) // there are elements, but they just moved
-        goto skip_upload;
 
-    clearEOSD(vo);
-    if (!img)
-        return;
-    if (!p->largeeosdtex[0]) {
-        gl->GenTextures(2, p->largeeosdtex);
-        for (int n = 0; n < 2; n++) {
-            gl->BindTexture(p->target, p->largeeosdtex[n]);
-            glCreateClearTex(gl, p->target, GL_ALPHA, GL_ALPHA,
-                             GL_UNSIGNED_BYTE, scale_type,
-                             LARGE_EOSD_TEX_SIZE, LARGE_EOSD_TEX_SIZE, 0);
-        }
-    }
-    for (i = img; i; i = i->next) {
-        if (i->w <= 0 || i->h <= 0 || i->stride < i->w)
-            continue;
-        if (is_tinytex(i, tinytexcur))
-            tinytexcur++;
-        else if (is_smalltex(i, smalltexcur))
-            smalltexcur++;
-        else
-            p->eosdtexCnt++;
-    }
-    mp_msg(MSGT_VO, MSGL_DBG2, "EOSD counts (tiny, small, all): %i, %i, %i\n",
-           tinytexcur, smalltexcur, p->eosdtexCnt);
-    if (p->eosdtexCnt) {
-        p->eosdtex = calloc(p->eosdtexCnt, sizeof(GLuint));
-        gl->GenTextures(p->eosdtexCnt, p->eosdtex);
-    }
-    tinytexcur = smalltexcur = 0;
-    for (i = img, curtex = p->eosdtex; i; i = i->next) {
-        int x = 0, y = 0;
-        if (i->w <= 0 || i->h <= 0 || i->stride < i->w) {
-            mp_msg(MSGT_VO, MSGL_V, "Invalid dimensions OSD for part!\n");
-            continue;
-        }
-        if (is_tinytex(i, tinytexcur)) {
-            tinytex_pos(tinytexcur, &x, &y);
-            gl->BindTexture(p->target, p->largeeosdtex[0]);
-            tinytexcur++;
-        } else if (is_smalltex(i, smalltexcur)) {
-            smalltex_pos(smalltexcur, &x, &y);
-            gl->BindTexture(p->target, p->largeeosdtex[1]);
-            smalltexcur++;
-        } else {
-            texSize(vo, i->w, i->h, &sx, &sy);
-            gl->BindTexture(p->target, *curtex++);
-            glCreateClearTex(gl, p->target, GL_ALPHA, GL_ALPHA,
-                             GL_UNSIGNED_BYTE, scale_type, sx, sy, 0);
-        }
-        glUploadTex(gl, p->target, GL_ALPHA, GL_UNSIGNED_BYTE, i->bitmap,
-                    i->stride, x, y, i->w, i->h, 0);
-    }
-    p->eosdDispList = gl->GenLists(1);
-skip_upload:
-    gl->NewList(p->eosdDispList, GL_COMPILE);
-    tinytexcur = smalltexcur = 0;
-    for (i = img, curtex = p->eosdtex; i; i = i->next) {
-        int x = 0, y = 0;
-        if (i->w <= 0 || i->h <= 0 || i->stride < i->w)
-            continue;
-        gl->Color4ub(i->color >> 24, (i->color >> 16) & 0xff,
-                     (i->color >> 8) & 0xff, 255 - (i->color & 0xff));
-        if (is_tinytex(i, tinytexcur)) {
-            tinytex_pos(tinytexcur, &x, &y);
-            sx = sy = LARGE_EOSD_TEX_SIZE;
-            gl->BindTexture(p->target, p->largeeosdtex[0]);
-            tinytexcur++;
-        } else if (is_smalltex(i, smalltexcur)) {
-            smalltex_pos(smalltexcur, &x, &y);
-            sx = sy = LARGE_EOSD_TEX_SIZE;
-            gl->BindTexture(p->target, p->largeeosdtex[1]);
-            smalltexcur++;
-        } else {
-            texSize(vo, i->w, i->h, &sx, &sy);
-            gl->BindTexture(p->target, *curtex++);
-        }
-        glDrawTex(gl, i->dst_x, i->dst_y, i->w, i->h, x, y, i->w, i->h, sx, sy,
-                  p->use_rectangle == 1, 0, 0);
-    }
-    gl->EndList();
+    gl->BindTexture(p->target, p->eosd_texture);
+
+    struct vertex_eosd *va = p->eosd_va;
+    size_t stride = sizeof(struct vertex_eosd);
+
+    gl->VertexPointer(2, GL_FLOAT, stride, &va[0].x);
+    gl->ColorPointer(4, GL_UNSIGNED_BYTE, stride, &va[0].color[0]);
+    gl->TexCoordPointer(2, GL_FLOAT, stride, &va[0].u);
+
+    gl->EnableClientState(GL_VERTEX_ARRAY);
+    gl->EnableClientState(GL_TEXTURE_COORD_ARRAY);
+    gl->EnableClientState(GL_COLOR_ARRAY);
+
+    gl->DrawArrays(GL_TRIANGLES, 0, p->eosd->targets_count * 6);
+
+    gl->DisableClientState(GL_VERTEX_ARRAY);
+    gl->DisableClientState(GL_TEXTURE_COORD_ARRAY);
+    gl->DisableClientState(GL_COLOR_ARRAY);
+
     gl->BindTexture(p->target, 0);
 }
 
@@ -450,10 +410,10 @@ static void uninitGl(struct vo *vo)
         gl->DeleteTextures(i, p->default_texs);
     p->default_texs[0] = 0;
     clearOSD(vo);
-    clearEOSD(vo);
-    if (p->largeeosdtex[0])
-        gl->DeleteTextures(2, p->largeeosdtex);
-    p->largeeosdtex[0] = 0;
+    if (p->eosd_texture)
+        gl->DeleteTextures(1, &p->eosd_texture);
+    eosd_packer_reinit(p->eosd, 0, 0);
+    p->eosd_texture = 0;
     if (gl->DeleteBuffers && p->buffer)
         gl->DeleteBuffers(1, &p->buffer);
     p->buffer = 0;
@@ -633,6 +593,10 @@ static int initGl(struct vo *vo, uint32_t d_width, uint32_t d_height)
         update_yuvconv(vo);
     }
 
+    GLint max_texture_size;
+    gl->GetIntegerv(GL_MAX_TEXTURE_SIZE, &max_texture_size);
+    eosd_packer_reinit(p->eosd, max_texture_size, max_texture_size);
+
     resize(vo, d_width, d_height);
 
     gl->ClearColor(0.0f, 0.0f, 0.0f, 0.0f);
@@ -786,7 +750,7 @@ static void do_render_osd(struct vo *vo, int type)
     GL *gl = p->gl;
 
     int draw_osd  = (type & RENDER_OSD) && p->osdtexCnt > 0;
-    int draw_eosd = (type & RENDER_EOSD) && p->eosdDispList;
+    int draw_eosd = (type & RENDER_EOSD);
     if (!draw_osd && !draw_eosd)
         return;
     // set special rendering parameters
@@ -799,7 +763,7 @@ static void do_render_osd(struct vo *vo, int type)
     gl->Enable(GL_BLEND);
     if (draw_eosd) {
         gl->BlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA);
-        gl->CallList(p->eosdDispList);
+        drawEOSD(vo);
     }
     if (draw_osd) {
         gl->Color4ub((p->osd_color >> 16) & 0xff, (p->osd_color >> 8) & 0xff,
@@ -1239,6 +1203,8 @@ static int preinit_internal(struct vo *vo, const char *arg, int allow_sw,
         .osd_color = 0xffffff,
     };
 
+    p->eosd = eosd_packer_create(vo);
+
     //essentially unused; for legacy warnings only
     int user_colorspace = 0;
     int levelconv = -1;
diff --git a/libvo/vo_vdpau.c b/libvo/vo_vdpau.c
index f8b70a4255..7b2f974ac2 100644
--- a/libvo/vo_vdpau.c
+++ b/libvo/vo_vdpau.c
@@ -56,6 +56,7 @@
 #include "libavutil/mathematics.h"
 
 #include "sub/ass_mp.h"
+#include "eosd_packer.h"
 
 #define WRAP_ADD(x, a, m) ((a) < 0 \
                            ? ((x)+(a)+(m) < (m) ? (x)+(a)+(m) : (x)+(a)) \
@@ -85,9 +86,6 @@
 /* number of palette entries */
 #define PALETTE_SIZE 256
 
-/* Initial size of EOSD surface in pixels (x*x) */
-#define EOSD_SURFACE_INITIAL_SIZE 256
-
 /* Pixelformat used for output surfaces */
 #define OUTPUT_RGBA_FORMAT VDP_RGBA_FORMAT_B8G8R8A8
 
@@ -173,24 +171,8 @@ struct vdpctx {
 
     // EOSD
     // Pool of surfaces
-    struct eosd_bitmap_surface {
-        VdpBitmapSurface surface;
-        int w;
-        int h;
-        uint32_t max_width;
-        uint32_t max_height;
-    } eosd_surface;
-
-    // List of surfaces to be rendered
-    struct eosd_target {
-        VdpRect source;
-        VdpRect dest;
-        VdpColor color;
-    } *eosd_targets;
-    int eosd_targets_size;
-    int *eosd_scratch;
-
-    int eosd_render_count;
+    VdpBitmapSurface eosd_surface;
+    struct eosd_packer *eosd_packer;
 
     // Video equalizer
     struct mp_csp_equalizer video_eq;
@@ -796,13 +778,16 @@ static int initialize_vdpau_objects(struct vo *vo)
     if (create_vdp_mixer(vo, vc->vdp_chroma_type) < 0)
         return -1;
 
+    int max_width = 0, max_height = 0;
     vdp_st = vdp->
         bitmap_surface_query_capabilities(vc->vdp_device,
                                           VDP_RGBA_FORMAT_A8,
                                           &(VdpBool){0},
-                                          &vc->eosd_surface.max_width,
-                                          &vc->eosd_surface.max_height);
+                                          &max_width,
+                                          &max_height);
     CHECK_ST_WARNING("Query to get max EOSD surface size failed");
+    eosd_packer_reinit(vc->eosd_packer, max_width, max_height);
+
     forget_frames(vo);
     resize(vo);
     return 0;
@@ -822,11 +807,9 @@ static void mark_vdpau_objects_uninitialized(struct vo *vo)
     for (int i = 0; i <= MAX_OUTPUT_SURFACES; i++)
         vc->output_surfaces[i] = VDP_INVALID_HANDLE;
     vc->vdp_device = VDP_INVALID_HANDLE;
-    vc->eosd_surface = (struct eosd_bitmap_surface){
-        .surface = VDP_INVALID_HANDLE,
-    };
+    vc->eosd_surface = VDP_INVALID_HANDLE;
+    eosd_packer_reinit(vc->eosd_packer, 0, 0);
     vc->output_surface_width = vc->output_surface_height = -1;
-    vc->eosd_render_count = 0;
 }
 
 static int handle_preemption(struct vo *vo)
@@ -1022,6 +1005,8 @@ static void draw_osd_I8A8(void *ctx, int x0, int y0, int w, int h,
                      "vdp_output_surface_render_output_surface");
 }
 
+#define EOSD_VDP_RC(r) &(VdpRect){r.x0, r.y0, r.x1, r.y1}
+
 static void draw_eosd(struct vo *vo)
 {
     struct vdpctx *vc = vo->priv;
@@ -1044,200 +1029,66 @@ static void draw_eosd(struct vo *vo)
         .blend_equation_alpha = VDP_OUTPUT_SURFACE_RENDER_BLEND_EQUATION_ADD,
     };
 
-    for (i = 0; i < vc->eosd_render_count; i++) {
+    for (i = 0; i < vc->eosd_packer->targets_count; i++) {
+        struct eosd_target *target = &vc->eosd_packer->targets[i];
+        VdpColor color = {
+            .alpha = 1.0 - ((target->color >> 0) & 0xff) / 255.0,
+            .blue  = ((target->color >>  8) & 0xff) / 255.0,
+            .green = ((target->color >> 16) & 0xff) / 255.0,
+            .red   = ((target->color >> 24) & 0xff) / 255.0,
+        };
         vdp_st = vdp->
             output_surface_render_bitmap_surface(output_surface,
-                                                 &vc->eosd_targets[i].dest,
-                                                 vc->eosd_surface.surface,
-                                                 &vc->eosd_targets[i].source,
-                                                 &vc->eosd_targets[i].color,
+                                                 EOSD_VDP_RC(target->dest),
+                                                 vc->eosd_surface,
+                                                 EOSD_VDP_RC(target->source),
+                                                 &color,
                                                  &blend_state,
                                                  VDP_OUTPUT_SURFACE_RENDER_ROTATE_0);
         CHECK_ST_WARNING("EOSD: Error when rendering");
     }
 }
 
-#define HEIGHT_SORT_BITS 4
-static int size_index(struct eosd_target *r)
-{
-    unsigned int h = r->source.y1;
-    int n = av_log2_16bit(h);
-    return (n << HEIGHT_SORT_BITS)
-        + (- 1 - (h << HEIGHT_SORT_BITS >> n) & (1 << HEIGHT_SORT_BITS) - 1);
-}
-
-/* Pack the given rectangles into an area of size w * h.
- * The size of each rectangle is read from .source.x1/.source.y1.
- * The height of each rectangle must be at least 1 and less than 65536.
- * The .source rectangle is then set corresponding to the packed position.
- * 'scratch' must point to work memory for num_rects+16 ints.
- * Return 0 on success, -1 if the rectangles did not fit in w*h.
- *
- * The rectangles are placed in rows in order approximately sorted by
- * height (the approximate sorting is simpler than a full one would be,
- * and allows the algorithm to work in linear time). Additionally, to
- * reduce wasted space when there are a few tall rectangles, empty
- * lower-right parts of rows are filled recursively when the size of
- * rectangles in the row drops past a power-of-two threshold. So if a
- * row starts with rectangles of size 3x50, 10x40 and 5x20 then the
- * free rectangle with corners (13, 20)-(w, 50) is filled recursively.
- */
-static int pack_rectangles(struct eosd_target *rects, int num_rects,
-                           int w, int h, int *scratch)
-{
-    int bins[16 << HEIGHT_SORT_BITS];
-    int sizes[16 << HEIGHT_SORT_BITS] = {};
-    for (int i = 0; i < num_rects; i++)
-        sizes[size_index(rects + i)]++;
-    int idx = 0;
-    for (int i = 0; i < 16 << HEIGHT_SORT_BITS; i += 1 << HEIGHT_SORT_BITS) {
-        for (int j = 0; j < 1 << HEIGHT_SORT_BITS; j++) {
-            bins[i + j] = idx;
-            idx += sizes[i + j];
-        }
-        scratch[idx++] = -1;
-    }
-    for (int i = 0; i < num_rects; i++)
-        scratch[bins[size_index(rects + i)]++] = i;