avfilter/drawutils: >8 bit support

Signed-off-by: Paul B Mahol <onemda@gmail.com>

3 files changed, 239 insertions, 43 deletions

diff --git a/libavfilter/drawutils.c b/libavfilter/drawutils.c
index 23a0eb5d87..7edaa4ac6c 100644
--- a/libavfilter/drawutils.c
+++ b/libavfilter/drawutils.c
@@ -24,6 +24,7 @@
 #include "libavutil/avassert.h"
 #include "libavutil/avutil.h"
 #include "libavutil/colorspace.h"
+#include "libavutil/intreadwrite.h"
 #include "libavutil/mem.h"
 #include "libavutil/pixdesc.h"
 #include "drawutils.h"
@@ -175,8 +176,10 @@ int ff_draw_init(FFDrawContext *draw, enum AVPixelFormat format, unsigned flags)
         return AVERROR(ENOSYS);
     for (i = 0; i < desc->nb_components; i++) {
         c = &desc->comp[i];
-        /* for now, only 8-bits formats */
-        if (c->depth != 8)
+        /* for now, only 8-16 bits formats */
+        if (c->depth < 8 || c->depth > 16)
+            return AVERROR(ENOSYS);
+        if (desc->flags & AV_PIX_FMT_FLAG_BE)
             return AVERROR(ENOSYS);
         if (c->plane >= MAX_PLANES)
             return AVERROR(ENOSYS);
@@ -184,6 +187,9 @@ int ff_draw_init(FFDrawContext *draw, enum AVPixelFormat format, unsigned flags)
         if (pixelstep[c->plane] != 0 &&
             pixelstep[c->plane] != c->step)
             return AVERROR(ENOSYS);
+        if (pixelstep[c->plane] == 6 &&
+            c->depth == 16)
+            return AVERROR(ENOSYS);
         pixelstep[c->plane] = c->step;
         if (pixelstep[c->plane] >= 8)
             return AVERROR(ENOSYS);
@@ -214,11 +220,18 @@ void ff_draw_color(FFDrawContext *draw, FFDrawColor *color, const uint8_t rgba[4
     if ((draw->desc->flags & AV_PIX_FMT_FLAG_RGB) &&
         ff_fill_rgba_map(rgba_map, draw->format) >= 0) {
         if (draw->nb_planes == 1) {
-            for (i = 0; i < 4; i++)
+            for (i = 0; i < 4; i++) {
                 color->comp[0].u8[rgba_map[i]] = rgba[i];
+                if (draw->desc->comp[rgba_map[i]].depth > 8) {
+                    color->comp[0].u16[rgba_map[i]] = color->comp[0].u8[rgba_map[i]] << 8;
+                }
+            }
         } else {
-            for (i = 0; i < 4; i++)
+            for (i = 0; i < 4; i++) {
                 color->comp[rgba_map[i]].u8[0] = rgba[i];
+                if (draw->desc->comp[rgba_map[i]].depth > 8)
+                    color->comp[rgba_map[i]].u16[0] = color->comp[rgba_map[i]].u8[0] << (draw->desc->comp[rgba_map[i]].depth - 8);
+            }
         }
     } else if (draw->nb_planes == 3 || draw->nb_planes == 4) {
         /* assume YUV */
@@ -226,9 +239,22 @@ void ff_draw_color(FFDrawContext *draw, FFDrawColor *color, const uint8_t rgba[4
         color->comp[1].u8[0] = RGB_TO_U_CCIR(rgba[0], rgba[1], rgba[2], 0);
         color->comp[2].u8[0] = RGB_TO_V_CCIR(rgba[0], rgba[1], rgba[2], 0);
         color->comp[3].u8[0] = rgba[3];
+        if (draw->desc->comp[0].depth > 8)
+            color->comp[0].u16[0] = color->comp[0].u8[0] << (draw->desc->comp[0].depth - 8);
+        if (draw->desc->comp[1].depth > 8)
+            color->comp[1].u16[0] = color->comp[1].u8[0] << (draw->desc->comp[1].depth - 8);
+        if (draw->desc->comp[2].depth > 8)
+            color->comp[2].u16[0] = color->comp[2].u8[0] << (draw->desc->comp[2].depth - 8);
+        if (draw->desc->comp[3].depth > 8)
+            color->comp[3].u16[0] = color->comp[3].u8[0] << (draw->desc->comp[3].depth - 8);
     } else if (draw->format == AV_PIX_FMT_GRAY8 || draw->format == AV_PIX_FMT_GRAY8A) {
         color->comp[0].u8[0] = RGB_TO_Y_CCIR(rgba[0], rgba[1], rgba[2]);
         color->comp[1].u8[0] = rgba[3];
+    } else if (draw->format == AV_PIX_FMT_GRAY16 || draw->format == AV_PIX_FMT_YA16) {
+        color->comp[0].u8[0] = RGB_TO_Y_CCIR(rgba[0], rgba[1], rgba[2]);
+        color->comp[0].u16[0] = color->comp[0].u8[0] << 8;
+        color->comp[1].u8[0] = rgba[3];
+        color->comp[1].u16[0] = color->comp[1].u8[0] << 8;
     } else {
         av_log(NULL, AV_LOG_WARNING,
                "Color conversion not implemented for %s\n", draw->desc->name);
@@ -361,6 +387,31 @@ static void blend_line(uint8_t *dst, unsigned src, unsigned alpha,
     }
 }
 
+static void blend_line16(uint8_t *dst, unsigned src, unsigned alpha,
+                         int dx, int w, unsigned hsub, int left, int right)
+{
+    unsigned asrc = alpha * src;
+    unsigned tau = 0x10001 - alpha;
+    int x;
+
+    if (left) {
+        unsigned suba = (left * alpha) >> hsub;
+        uint16_t value = AV_RL16(dst);
+        AV_WL16(dst, (value * (0x10001 - suba) + src * suba) >> 16);
+        dst += dx;
+    }
+    for (x = 0; x < w; x++) {
+        uint16_t value = AV_RL16(dst);
+        AV_WL16(dst, (value * tau + asrc) >> 16);
+        dst += dx;
+    }
+    if (right) {
+        unsigned suba = (right * alpha) >> hsub;
+        uint16_t value = AV_RL16(dst);
+        AV_WL16(dst, (value * (0x10001 - suba) + src * suba) >> 16);
+    }
+}
+
 void ff_blend_rectangle(FFDrawContext *draw, FFDrawColor *color,
                         uint8_t *dst[], int dst_linesize[],
                         int dst_w, int dst_h,
@@ -375,8 +426,13 @@ void ff_blend_rectangle(FFDrawContext *draw, FFDrawColor *color,
     clip_interval(dst_h, &y0, &h, NULL);
     if (w <= 0 || h <= 0 || !color->rgba[3])
         return;
-    /* 0x10203 * alpha + 2 is in the [ 2 ; 0x1010101 - 2 ] range */
-    alpha = 0x10203 * color->rgba[3] + 0x2;
+    if (draw->desc->comp[0].depth <= 8) {
+        /* 0x10203 * alpha + 2 is in the [ 2 ; 0x1010101 - 2 ] range */
+        alpha = 0x10203 * color->rgba[3] + 0x2;
+    } else {
+        /* 0x101 * alpha is in the [ 2 ; 0x1001] range */
+        alpha = 0x101 * color->rgba[3] + 0x2;
+    }
     nb_planes = (draw->nb_planes - 1) | 1; /* eliminate alpha */
     for (plane = 0; plane < nb_planes; plane++) {
         nb_comp = draw->pixelstep[plane];
@@ -388,29 +444,77 @@ void ff_blend_rectangle(FFDrawContext *draw, FFDrawColor *color,
         subsampling_bounds(draw->hsub[plane], &x_sub, &w_sub, &left, &right);
         subsampling_bounds(draw->vsub[plane], &y_sub, &h_sub, &top, &bottom);
         for (comp = 0; comp < nb_comp; comp++) {
+            const int depth = draw->desc->comp[comp].depth;
+
             if (!component_used(draw, plane, comp))
                 continue;
             p = p0 + comp;
             if (top) {
-                blend_line(p, color->comp[plane].u8[comp], alpha >> 1,
-                           draw->pixelstep[plane], w_sub,
-                           draw->hsub[plane], left, right);
+                if (depth <= 8) {
+                    blend_line(p, color->comp[plane].u8[comp], alpha >> 1,
+                               draw->pixelstep[plane], w_sub,
+                               draw->hsub[plane], left, right);
+                } else {
+                    blend_line16(p, color->comp[plane].u16[comp], alpha >> 1,
+                                 draw->pixelstep[plane], w_sub,
+                                 draw->hsub[plane], left, right);
+                }
                 p += dst_linesize[plane];
             }
-            for (y = 0; y < h_sub; y++) {
-                blend_line(p, color->comp[plane].u8[comp], alpha,
-                           draw->pixelstep[plane], w_sub,
-                           draw->hsub[plane], left, right);
-                p += dst_linesize[plane];
+            if (depth <= 8) {
+                for (y = 0; y < h_sub; y++) {
+                    blend_line(p, color->comp[plane].u8[comp], alpha,
+                               draw->pixelstep[plane], w_sub,
+                               draw->hsub[plane], left, right);
+                    p += dst_linesize[plane];
+                }
+            } else {
+                for (y = 0; y < h_sub; y++) {
+                    blend_line16(p, color->comp[plane].u16[comp], alpha,
+                                 draw->pixelstep[plane], w_sub,
+                                 draw->hsub[plane], left, right);
+                    p += dst_linesize[plane];
+                }
+            }
+            if (bottom) {
+                if (depth <= 8) {
+                    blend_line(p, color->comp[plane].u8[comp], alpha >> 1,
+                               draw->pixelstep[plane], w_sub,
+                               draw->hsub[plane], left, right);
+                } else {
+                    blend_line16(p, color->comp[plane].u16[comp], alpha >> 1,
+                                 draw->pixelstep[plane], w_sub,
+                                 draw->hsub[plane], left, right);
+                }
             }
-            if (bottom)
-                blend_line(p, color->comp[plane].u8[comp], alpha >> 1,
-                           draw->pixelstep[plane], w_sub,
-                           draw->hsub[plane], left, right);
         }
     }
 }
 
+static void blend_pixel16(uint8_t *dst, unsigned src, unsigned alpha,
+                          const uint8_t *mask, int mask_linesize, int l2depth,
+                          unsigned w, unsigned h, unsigned shift, unsigned xm0)
+{
+    unsigned xm, x, y, t = 0;
+    unsigned xmshf = 3 - l2depth;
+    unsigned xmmod = 7 >> l2depth;
+    unsigned mbits = (1 << (1 << l2depth)) - 1;
+    unsigned mmult = 255 / mbits;
+    uint16_t value = AV_RL16(dst);
+
+    for (y = 0; y < h; y++) {
+        xm = xm0;
+        for (x = 0; x < w; x++) {
+            t += ((mask[xm >> xmshf] >> ((~xm & xmmod) << l2depth)) & mbits)
+                 * mmult;
+            xm++;
+        }
+        mask += mask_linesize;
+    }
+    alpha = (t >> shift) * alpha;
+    AV_WL16(dst, ((0x10001 - alpha) * value + alpha * src) >> 16);
+}
+
 static void blend_pixel(uint8_t *dst, unsigned src, unsigned alpha,
                         const uint8_t *mask, int mask_linesize, int l2depth,
                         unsigned w, unsigned h, unsigned shift, unsigned xm0)
@@ -434,6 +538,31 @@ static void blend_pixel(uint8_t *dst, unsigned src, unsigned alpha,
     *dst = ((0x1010101 - alpha) * *dst + alpha * src) >> 24;
 }
 
+static void blend_line_hv16(uint8_t *dst, int dst_delta,
+                            unsigned src, unsigned alpha,
+                            const uint8_t *mask, int mask_linesize, int l2depth, int w,
+                            unsigned hsub, unsigned vsub,
+                            int xm, int left, int right, int hband)
+{
+    int x;
+
+    if (left) {
+        blend_pixel16(dst, src, alpha, mask, mask_linesize, l2depth,
+                      left, hband, hsub + vsub, xm);
+        dst += dst_delta;
+        xm += left;
+    }
+    for (x = 0; x < w; x++) {
+        blend_pixel16(dst, src, alpha, mask, mask_linesize, l2depth,
+                      1 << hsub, hband, hsub + vsub, xm);
+        dst += dst_delta;
+        xm += 1 << hsub;
+    }
+    if (right)
+        blend_pixel16(dst, src, alpha, mask, mask_linesize, l2depth,
+                      right, hband, hsub + vsub, xm);
+}
+
 static void blend_line_hv(uint8_t *dst, int dst_delta,
                           unsigned src, unsigned alpha,
                           const uint8_t *mask, int mask_linesize, int l2depth, int w,
@@ -474,9 +603,13 @@ void ff_blend_mask(FFDrawContext *draw, FFDrawColor *color,
     mask += ym0 * mask_linesize;
     if (mask_w <= 0 || mask_h <= 0 || !color->rgba[3])
         return;
-    /* alpha is in the [ 0 ; 0x10203 ] range,
-       alpha * mask is in the [ 0 ; 0x1010101 - 4 ] range */
-    alpha = (0x10307 * color->rgba[3] + 0x3) >> 8;
+    if (draw->desc->comp[0].depth <= 8) {
+        /* alpha is in the [ 0 ; 0x10203 ] range,
+           alpha * mask is in the [ 0 ; 0x1010101 - 4 ] range */
+        alpha = (0x10307 * color->rgba[3] + 0x3) >> 8;
+    } else {
+        alpha = (0x101 * color->rgba[3] + 0x2) >> 8;
+    }
     nb_planes = (draw->nb_planes - 1) | 1; /* eliminate alpha */
     for (plane = 0; plane < nb_planes; plane++) {
         nb_comp = draw->pixelstep[plane];
@@ -488,34 +621,65 @@ void ff_blend_mask(FFDrawContext *draw, FFDrawColor *color,
         subsampling_bounds(draw->hsub[plane], &x_sub, &w_sub, &left, &right);
         subsampling_bounds(draw->vsub[plane], &y_sub, &h_sub, &top, &bottom);
         for (comp = 0; comp < nb_comp; comp++) {
+            const int depth = draw->desc->comp[comp].depth;
+
             if (!component_used(draw, plane, comp))
                 continue;
             p = p0 + comp;
             m = mask;
             if (top) {
-                blend_line_hv(p, draw->pixelstep[plane],
-                              color->comp[plane].u8[comp], alpha,
-                              m, mask_linesize, l2depth, w_sub,
-                              draw->hsub[plane], draw->vsub[plane],
-                              xm0, left, right, top);
+                if (depth <= 8) {
+                    blend_line_hv(p, draw->pixelstep[plane],
+                                  color->comp[plane].u8[comp], alpha,
+                                  m, mask_linesize, l2depth, w_sub,
+                                  draw->hsub[plane], draw->vsub[plane],
+                                  xm0, left, right, top);
+                } else {
+                    blend_line_hv16(p, draw->pixelstep[plane],
+                                    color->comp[plane].u16[comp], alpha,
+                                    m, mask_linesize, l2depth, w_sub,
+                                    draw->hsub[plane], draw->vsub[plane],
+                                    xm0, left, right, top);
+                }
                 p += dst_linesize[plane];
                 m += top * mask_linesize;
             }
-            for (y = 0; y < h_sub; y++) {
-                blend_line_hv(p, draw->pixelstep[plane],
-                              color->comp[plane].u8[comp], alpha,
-                              m, mask_linesize, l2depth, w_sub,
-                              draw->hsub[plane], draw->vsub[plane],
-                              xm0, left, right, 1 << draw->vsub[plane]);
-                p += dst_linesize[plane];
-                m += mask_linesize << draw->vsub[plane];
+            if (depth <= 8) {
+                for (y = 0; y < h_sub; y++) {
+                    blend_line_hv(p, draw->pixelstep[plane],
+                                  color->comp[plane].u8[comp], alpha,
+                                  m, mask_linesize, l2depth, w_sub,
+                                  draw->hsub[plane], draw->vsub[plane],
+                                  xm0, left, right, 1 << draw->vsub[plane]);
+                    p += dst_linesize[plane];
+                    m += mask_linesize << draw->vsub[plane];
+                }
+            } else {
+                for (y = 0; y < h_sub; y++) {
+                    blend_line_hv16(p, draw->pixelstep[plane],
+                                    color->comp[plane].u16[comp], alpha,
+                                    m, mask_linesize, l2depth, w_sub,
+                                    draw->hsub[plane], draw->vsub[plane],
+                                    xm0, left, right, 1 << draw->vsub[plane]);
+                    p += dst_linesize[plane];
+                    m += mask_linesize << draw->vsub[plane];
+                }
+            }
+            if (bottom) {
+                if (depth <= 8) {
+                    blend_line_hv(p, draw->pixelstep[plane],
+                                  color->comp[plane].u8[comp], alpha,
+                                  m, mask_linesize, l2depth, w_sub,
+                                  draw->hsub[plane], draw->vsub[plane],
+                                  xm0, left, right, bottom);
+                } else {
+                    blend_line_hv16(p, draw->pixelstep[plane],
+                                    color->comp[plane].u16[comp], alpha,
+                                    m, mask_linesize, l2depth, w_sub,
+                                    draw->hsub[plane], draw->vsub[plane],
+                                    xm0, left, right, bottom);
+                }
             }
-            if (bottom)
-                blend_line_hv(p, draw->pixelstep[plane],
-                              color->comp[plane].u8[comp], alpha,
-                              m, mask_linesize, l2depth, w_sub,
-                              draw->hsub[plane], draw->vsub[plane],
-                              xm0, left, right, bottom);
         }
     }
 }
diff --git a/libavfilter/drawutils.h b/libavfilter/drawutils.h
index e247dd6440..1fb3e4f256 100644
--- a/libavfilter/drawutils.h
+++ b/libavfilter/drawutils.h
@@ -60,9 +60,9 @@ typedef struct FFDrawContext {
 typedef struct FFDrawColor {
     uint8_t rgba[4];
     union {
-        uint32_t u32;
-        uint16_t u16;
-        uint8_t  u8[4];
+        uint32_t u32[4];
+        uint16_t u16[8];
+        uint8_t  u8[16];
     } comp[MAX_PLANES];
 } FFDrawColor;
 
diff --git a/tests/ref/fate/filter-pixfmts-pad b/tests/ref/fate/filter-pixfmts-pad
index 9e474b9e99..a734d312bc 100644
--- a/tests/ref/fate/filter-pixfmts-pad
+++ b/tests/ref/fate/filter-pixfmts-pad
@@ -7,20 +7,52 @@ bgr24               f8b65ad845905c7d0c93ca28dfbb826f
 bgra                929aac15e848038e367c250037575f9f
 gbrap               6712984b4a068ffa534f0cb35b2adc6f
 gbrp                3c94d39256db2409015df913fd330a90
+gbrp10le            f206db2a32ad484a59cf3e596bdc2866
+gbrp12le            be6463331985e2c8ed7bd910093ca4a2
+gbrp14le            9ae804cf217bec0a737c36c20573cbe5
+gbrp9le             9a86dab5661c213ce2b7e00ae48b4d1f
 gray                ddc663a0491df3959d9c5795dceaa72e
+gray16le            468bda6155bdc7a7a20c34d6e599fd16
 rgb0                78d500c8361ab6423a4826a00268c908
 rgb24               17f9e2e0c609009acaf2175c42d4a2a5
 rgba                b157c90191463d34fb3ce77b36c96386
+xyz12le             85abf80b77a9236a76ba0b00fcbdea2d
 ya8                 5fc0f471207ddf7aa01b07027d56b672
 yuv410p             cb871dcc1e84a7ef1d21f9237b88cf6e
 yuv411p             aec2c1740de9a62db0d41f4dda9121b0
 yuv420p             4398e408fc35436ce4b20468946f58b6
+yuv420p10le         74518a7d68457c54da3300e80c683e9c
+yuv420p12le         32086c64c814315aa9253580708a192e
+yuv420p14le         265e9fefb8b92f3cae678eb3cf4e00b4
+yuv420p16le         1ca89e47164d5f2481e39d4ac6eab2ed
+yuv420p9le          59f0209084d1d1dd2861e058d1df63f4
 yuv422p             e43d68568d9f782908ba56bf1e09d5d5
+yuv422p10le         18667cf7f87d8ffe1c4cc3db2100c559
+yuv422p12le         71e32220fa1bbdcac95fb57ba0dbd4f0
+yuv422p14le         a0064be4d29079a811e5a315341da09a
+yuv422p16le         836057e9999c763697c66c21869492b8
+yuv422p9le          3213bed797e7b4cefa4f174dc33bf246
 yuv440p             a7e34de74c96b0224fe1381ec1db2ba7
+yuv440p10le         f91a481e27be7ea4a38a878e5e4330a3
+yuv440p12le         0e9953a09ac09fe5f1523b350cf7cb72
 yuv444p             6bfd89286dc36f2789b77d747ed8fa22
+yuv444p10le         b10afb5fda970dbce8aac5aab981a8cb
+yuv444p12le         8f1b4b0f91d309fdaaec8a1ba6c7107f
+yuv444p14le         2a87a84fa01237de9c08dd7e9ffbf78d
+yuv444p16le         28508867fe7470c3539f8a84e8fb8271
+yuv444p9le          99d39ef897c6037d6c7aa6e7a0398f84
 yuva420p            842c27169ecdcf6de79f2b787367b51c
+yuva420p10le        336fcb42c5b665c2028661e73325b359
+yuva420p16le        ff45de790e7bdd3c25d8aad51289aba9
+yuva420p9le         8ef1f3b3e01b5ce222e4caeec3dec396
 yuva422p            91dcecc4bfdff1f0db9ef8b9b5b9ac2a
+yuva422p10le        1ba292c74c8646fd077a6116142b1bc8
+yuva422p16le        383226550fe9c93d6e8bf0d45d1423d1
+yuva422p9le         0fb76788c905c6d448143aa3c5eae116
 yuva444p            fb60941a57596b277417a3c7c00aa194
+yuva444p10le        251ea4ead8300d752eb355a08cbb0352
+yuva444p16le        5b65287e1862d2d9f1ad2cfdcde94661
+yuva444p9le         e6946c10b94c271e7ea24b3bcff314e1
 yuvj411p            ca967e68759a4956729dd366adc7e7fa
 yuvj420p            c00611cd5f1558047d579d8a7d30e381
 yuvj422p            b3acdf07147a7598836065836ad8420b