draw_bmp: add integer blending for 8 bit formats

Whatever it's worth. Instead of doing a pretty stupid conversion to
float, just blend it directly. This works for most RGB formats that are
8 bits per component or below (the latter because we expand packed
fringe RGB formats for simplicity). For higher bit depth RGB this would
need extra code.

2 files changed, 79 insertions, 41 deletions

diff --git a/sub/draw_bmp.c b/sub/draw_bmp.c
index 7ef0a9f3ff..ba6ea7e646 100644
--- a/sub/draw_bmp.c
+++ b/sub/draw_bmp.c
@@ -114,6 +114,16 @@ static void blend_line_f32(void *dst, void *src, void *src_a, int w)
         dst_f[x] = src_f[x] + dst_f[x] * (1.0f - src_a_f[x]);
 }
 
+static void blend_line_u8(void *dst, void *src, void *src_a, int w)
+{
+    uint8_t *dst_i = dst;
+    uint8_t *src_i = src;
+    uint8_t *src_a_i = src_a;
+
+    for (int x = 0; x < w; x++)
+        dst_i[x] = src_i[x] + dst_i[x] * (255u - src_a_i[x]) / 255u;
+}
+
 static void blend_slice(struct mp_draw_sub_cache *p)
 {
     struct mp_image *ov = p->overlay_tmp;
@@ -500,13 +510,48 @@ static bool reinit_to_video(struct mp_draw_sub_cache *p)
     bool need_premul = params->alpha != MP_ALPHA_PREMUL &&
         (mp_imgfmt_get_desc(params->imgfmt).flags & MP_IMGFLAG_ALPHA);
 
-    int rflags = REPACK_CREATE_EXPAND_8BIT | REPACK_CREATE_PLANAR_F32;
-    p->blend_line = blend_line_f32;
+    // Intermediate format for video_overlay. Requirements:
+    //  - same subsampling as video
+    //  - uses video colorspace
+    //  - has alpha
+    //  - repacker support (to the format used in p->blend_line)
+    //  - probably 8 bit per component rather than something wasteful or strange
+    struct mp_regular_imgfmt vfdesc = {0};
+
+    int rflags = REPACK_CREATE_EXPAND_8BIT;
+    bool use_shortcut = false;
 
     p->video_to_f32 = mp_repack_create_planar(params->imgfmt, false, rflags);
     talloc_steal(p, p->video_to_f32);
     if (!p->video_to_f32)
         return false;
+    mp_get_regular_imgfmt(&vfdesc, mp_repack_get_format_dst(p->video_to_f32));
+    assert(vfdesc.num_planes); // must have succeeded
+
+    if (params->color.space == MP_CSP_RGB && vfdesc.num_planes >= 3) {
+        use_shortcut = true;
+
+        if (vfdesc.component_type == MP_COMPONENT_TYPE_UINT &&
+            vfdesc.component_size == 1 && vfdesc.component_pad == 0)
+            p->blend_line = blend_line_u8;
+    }
+
+    // If no special blender is available, blend in float.
+    if (!p->blend_line) {
+        TA_FREEP(&p->video_to_f32);
+
+        rflags |= REPACK_CREATE_PLANAR_F32;
+
+        p->video_to_f32 = mp_repack_create_planar(params->imgfmt, false, rflags);
+        talloc_steal(p, p->video_to_f32);
+        if (!p->video_to_f32)
+            return false;
+
+        mp_get_regular_imgfmt(&vfdesc, mp_repack_get_format_dst(p->video_to_f32));
+        assert(vfdesc.component_type == MP_COMPONENT_TYPE_FLOAT);
+
+        p->blend_line = blend_line_f32;
+    }
 
     p->scale_in_tiles = SCALE_IN_TILES;
 
@@ -520,18 +565,8 @@ static bool reinit_to_video(struct mp_draw_sub_cache *p)
     assert(mp_repack_get_format_dst(p->video_to_f32) ==
            mp_repack_get_format_src(p->video_from_f32));
 
-    // Find a reasonable intermediate format for video_overlay. Requirements:
-    //  - same subsampling
-    //  - has alpha
-    //  - uses video colorspace
-    //  - REPACK_CREATE_PLANAR_F32 support
-    //  - probably not using float (vaguely wastes memory)
-    struct mp_regular_imgfmt vfdesc = {0};
-    mp_get_regular_imgfmt(&vfdesc, mp_repack_get_format_dst(p->video_to_f32));
-    assert(vfdesc.component_type == MP_COMPONENT_TYPE_FLOAT);
-
     int overlay_fmt = 0;
-    if (params->color.space == MP_CSP_RGB && vfdesc.num_planes >= 3) {
+    if (use_shortcut) {
         // No point in doing anything fancy.
         overlay_fmt = IMGFMT_BGRA;
         p->scale_in_tiles = false;
@@ -643,6 +678,9 @@ static bool reinit_to_video(struct mp_draw_sub_cache *p)
         int xs = p->video_overlay->fmt.chroma_xs;
         int ys = p->video_overlay->fmt.chroma_ys;
         if (xs || ys) {
+            // Require float so format selection becomes simpler (maybe).
+            assert(rflags & REPACK_CREATE_PLANAR_F32);
+
             // For extracting the alpha plane, construct a gray format that is
             // compatible with the alpha one.
             struct mp_regular_imgfmt odesc = {0};
diff --git a/test/ref/draw_bmp.txt b/test/ref/draw_bmp.txt
index ccbe86e5a7..aa0c3a0a80 100644
--- a/test/ref/draw_bmp.txt
+++ b/test/ref/draw_bmp.txt
@@ -1,7 +1,7 @@
-0bgr        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-0rgb        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-abgr        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
-argb        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
+0bgr        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+0rgb        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+abgr        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrap, a=unknown, ca=unknown, ca_f=unknown
+argb        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrap, a=unknown, ca=unknown, ca_f=unknown
 ayuv64      = align=1:1 ov=yuva444p, ov_f=yuva444pf, v_f=yuva444pf, a=unknown, ca=unknown, ca_f=unknown
 ayuv64be    = align=1:1 ov=yuva444p, ov_f=yuva444pf, v_f=yuva444pf, a=unknown, ca=unknown, ca_f=unknown
 bayer_bggr16= no
@@ -16,20 +16,20 @@ bayer_grbg8 = no
 bayer_rggb16= no
 bayer_rggb16be= no
 bayer_rggb8 = no
-bgr0        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-bgr24       = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
+bgr0        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+bgr24       = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
 bgr4        = no
-bgr444      = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-bgr444be    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
+bgr444      = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+bgr444be    = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
 bgr48       = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
 bgr48be     = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-bgr4_byte   = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-bgr555      = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-bgr555be    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-bgr565      = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-bgr565be    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-bgr8        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-bgra        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
+bgr4_byte   = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+bgr555      = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+bgr555be    = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+bgr565      = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+bgr565be    = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+bgr8        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+bgra        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrap, a=unknown, ca=unknown, ca_f=unknown
 bgra64      = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
 bgra64be    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
 cuda        = no
@@ -37,7 +37,7 @@ d3d11       = no
 d3d11va_vld = no
 drm_prime   = no
 dxva2_vld   = no
-gbrap       = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
+gbrap       = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrap, a=unknown, ca=unknown, ca_f=unknown
 gbrap10     = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
 gbrap10be   = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
 gbrap12     = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
@@ -46,7 +46,7 @@ gbrap16     = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=u
 gbrap16be   = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
 gbrapf32    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
 gbrapf32be  = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
-gbrp        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
+gbrp        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
 gbrp1       = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
 gbrp10      = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
 gbrp10be    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
@@ -97,21 +97,21 @@ p016        = align=2:2 ov=yuva420p, ov_f=yuva420pf, v_f=yuv420pf, a=gray, ca=gr
 p016be      = align=2:2 ov=yuva420p, ov_f=yuva420pf, v_f=yuv420pf, a=gray, ca=gray, ca_f=grayf32
 pal8        = no
 qsv         = no
-rgb0        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-rgb24       = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
+rgb0        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+rgb24       = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
 rgb30       = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
 rgb4        = no
-rgb444      = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-rgb444be    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
+rgb444      = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+rgb444be    = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
 rgb48       = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
 rgb48be     = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-rgb4_byte   = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-rgb555      = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-rgb555be    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-rgb565      = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-rgb565be    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-rgb8        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrpf32, a=unknown, ca=unknown, ca_f=unknown
-rgba        = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
+rgb4_byte   = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+rgb555      = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+rgb555be    = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+rgb565      = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+rgb565be    = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+rgb8        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrp, a=unknown, ca=unknown, ca_f=unknown
+rgba        = align=1:1 ov=unknown, ov_f=gbrap, v_f=gbrap, a=unknown, ca=unknown, ca_f=unknown
 rgba64      = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
 rgba64be    = align=1:1 ov=unknown, ov_f=gbrapf32, v_f=gbrapf32, a=unknown, ca=unknown, ca_f=unknown
 uyvy422     = align=2:1 ov=yuva422p, ov_f=yuva422pf, v_f=yuv422pf, a=gray, ca=gray, ca_f=grayf32