sub: move RGBA scaling to vo_vaapi

vo_vaapi is the only thing which can't scale RGBA on the GPU. (Other
cases of RGBA scaling are handled in draw_bmp.c for some reason.)

Move this code and get rid of the osd_conv_cache thing.

Functionally, nothing changes.

sub/img_convert.c

@@ -30,17 +30,6 @@
 #include "video/mp_image.h"
 #include "video/sws_utils.h"
 
-struct osd_conv_cache {
-    struct sub_bitmap part[MP_SUB_BB_LIST_MAX];
-    struct sub_bitmap *parts;
-    void *scratch;
-};
-
-struct osd_conv_cache *osd_conv_cache_new(void)
-{
-    return talloc_zero(NULL, struct osd_conv_cache);
-}
-
 void mp_blur_rgba_sub_bitmap(struct sub_bitmap *d, double gblur)
 {
     struct mp_image *tmp1 = mp_image_alloc(IMGFMT_BGRA, d->w, d->h);
@@ -58,57 +47,6 @@ void mp_blur_rgba_sub_bitmap(struct sub_bitmap *d, double gblur)
     talloc_free(tmp1);
 }
 
-// If RGBA parts need scaling, scale them.
-bool osd_scale_rgba(struct osd_conv_cache *c, struct sub_bitmaps *imgs)
-{
-    struct sub_bitmaps src = *imgs;
-    if (src.format != SUBBITMAP_RGBA)
-        return false;
-
-    bool need_scale = false;
-    for (int n = 0; n < src.num_parts; n++) {
-        struct sub_bitmap *sb = &src.parts[n];
-        if (sb->w != sb->dw || sb->h != sb->dh)
-            need_scale = true;
-    }
-    if (!need_scale)
-        return false;
-
-    talloc_free(c->parts);
-    imgs->parts = c->parts = talloc_array(c, struct sub_bitmap, src.num_parts);
-    imgs->packed = NULL;
-
-    // Note: we scale all parts, since most likely all need scaling anyway, and
-    //       to get a proper copy of all data in the imgs list.
-    for (int n = 0; n < src.num_parts; n++) {
-        struct sub_bitmap *d = &imgs->parts[n];
-        struct sub_bitmap *s = &src.parts[n];
-
-        struct mp_image src_image = {0};
-        mp_image_setfmt(&src_image, IMGFMT_BGRA);
-        mp_image_set_size(&src_image, s->w, s->h);
-        src_image.planes[0] = s->bitmap;
-        src_image.stride[0] = s->stride;
-
-        d->x = s->x;
-        d->y = s->y;
-        d->w = d->dw = s->dw;
-        d->h = d->dh = s->dh;
-        struct mp_image *image = mp_image_alloc(IMGFMT_BGRA, d->w, d->h);
-        talloc_steal(c->parts, image);
-        if (image) {
-            d->stride = image->stride[0];
-            d->bitmap = image->planes[0];
-
-            mp_image_swscale(image, &src_image, mp_sws_fast_flags);
-        } else {
-            // on OOM, skip the region; just don't scale it
-            *d = *s;
-        }
-    }
-    return true;
-}
-
 bool mp_sub_bitmaps_bb(struct sub_bitmaps *imgs, struct mp_rect *out_bb)
 {
     struct mp_rect bb = {INT_MAX, INT_MAX, INT_MIN, INT_MIN};

sub/img_convert.h

@@ -3,16 +3,12 @@
 
 #include <stdbool.h>
 
-struct osd_conv_cache;
 struct sub_bitmaps;
 struct sub_bitmap;
 struct mp_rect;
 
-struct osd_conv_cache *osd_conv_cache_new(void);
-
 // Sub postprocessing
 void mp_blur_rgba_sub_bitmap(struct sub_bitmap *d, double gblur);
-bool osd_scale_rgba(struct osd_conv_cache *c, struct sub_bitmaps *imgs);
 
 bool mp_sub_bitmaps_bb(struct sub_bitmaps *imgs, struct mp_rect *out_bb);
 

video/out/vo_vaapi.c

@@ -32,8 +32,9 @@
 #include "common/msg.h"
 #include "video/out/vo.h"
 #include "video/mp_image_pool.h"
-#include "sub/osd.h"
+#include "video/sws_utils.h"
 #include "sub/img_convert.h"
+#include "sub/osd.h"
 #include "x11_common.h"
 
 #include "video/mp_image.h"
@@ -58,7 +59,6 @@ struct vaapi_osd_part {
     int change_id;
     struct vaapi_osd_image image;
     struct vaapi_subpic subpic;
-    struct osd_conv_cache *conv_cache;
 };
 
 #define MAX_OUTPUT_SURFACES 2
@@ -336,8 +336,6 @@ static void draw_osd_cb(void *pctx, struct sub_bitmaps *imgs)
     if (imgs->change_id != part->change_id) {
         part->change_id = imgs->change_id;
 
-        osd_scale_rgba(part->conv_cache, imgs);
-
         struct mp_rect bb;
         if (!mp_sub_bitmaps_bb(imgs, &bb))
             goto error;
@@ -365,6 +363,25 @@ static void draw_osd_cb(void *pctx, struct sub_bitmaps *imgs)
         for (int n = 0; n < imgs->num_parts; n++) {
             struct sub_bitmap *sub = &imgs->parts[n];
 
+            struct mp_image src = {0};
+            mp_image_setfmt(&src, IMGFMT_BGRA);
+            mp_image_set_size(&src, sub->w, sub->h);
+            src.planes[0] = sub->bitmap;
+            src.stride[0] = sub->stride;
+
+            struct mp_image *bmp = &src;
+
+            struct mp_image *tmp = NULL;
+            if (sub->dw != sub->w || sub->dh != sub->h) {
+                tmp = mp_image_alloc(IMGFMT_BGRA, sub->dw, sub->dh);
+                if (!tmp)
+                    goto error;
+
+                mp_image_swscale(tmp, &src, mp_sws_fast_flags);
+
+                bmp = tmp;
+            }
+
             // Note: nothing guarantees that the sub-bitmaps don't overlap.
             //       But in all currently existing cases, they don't.
             //       We simply hope that this won't change, and nobody will
@@ -373,8 +390,10 @@ static void draw_osd_cb(void *pctx, struct sub_bitmaps *imgs)
             size_t dst = (sub->y - bb.y0) * vaimg.stride[0] +
                          (sub->x - bb.x0) * 4;
 
-            memcpy_pic(vaimg.planes[0] + dst, sub->bitmap, sub->w * 4, sub->h,
+            memcpy_pic(vaimg.planes[0] + dst, bmp->planes[0], sub->dw * 4,
-                       vaimg.stride[0], sub->stride);
+                       sub->dh, vaimg.stride[0], bmp->stride[0]);
+
+            talloc_free(tmp);
         }
 
         if (!va_image_unmap(p->mpvaapi, &img->image))
@@ -630,7 +649,6 @@ static int preinit(struct vo *vo)
         struct vaapi_osd_part *part = &p->osd_parts[n];
         part->image.image.image_id = VA_INVALID_ID;
         part->image.subpic_id = VA_INVALID_ID;
-        part->conv_cache = talloc_steal(vo, osd_conv_cache_new());
     }
 
     int max_display_attrs = vaMaxNumDisplayAttributes(p->display);