]> git.sesse.net Git - ffmpeg/blobdiff - libavfilter/vf_overlay.c
avdevice/avdevice: Constify avdevice_list_input_sources/output_sinks
[ffmpeg] / libavfilter / vf_overlay.c
index c6a6ac82f397aa30d1203d378c03f2f073680485..bda5970b8211e442c1b5a723a0692b32cda2f554 100644 (file)
 #include "drawutils.h"
 #include "framesync.h"
 #include "video.h"
+#include "vf_overlay.h"
+
+typedef struct ThreadData {
+    AVFrame *dst, *src;
+} ThreadData;
 
 static const char *const var_names[] = {
     "main_w",    "W", ///< width  of the main    video
@@ -55,21 +60,6 @@ static const char *const var_names[] = {
     NULL
 };
 
-enum var_name {
-    VAR_MAIN_W,    VAR_MW,
-    VAR_MAIN_H,    VAR_MH,
-    VAR_OVERLAY_W, VAR_OW,
-    VAR_OVERLAY_H, VAR_OH,
-    VAR_HSUB,
-    VAR_VSUB,
-    VAR_X,
-    VAR_Y,
-    VAR_N,
-    VAR_POS,
-    VAR_T,
-    VAR_VARS_NB
-};
-
 #define MAIN    0
 #define OVERLAY 1
 
@@ -88,45 +78,6 @@ enum EvalMode {
     EVAL_MODE_NB
 };
 
-enum OverlayFormat {
-    OVERLAY_FORMAT_YUV420,
-    OVERLAY_FORMAT_YUV422,
-    OVERLAY_FORMAT_YUV444,
-    OVERLAY_FORMAT_RGB,
-    OVERLAY_FORMAT_GBRP,
-    OVERLAY_FORMAT_AUTO,
-    OVERLAY_FORMAT_NB
-};
-
-typedef struct OverlayContext {
-    const AVClass *class;
-    int x, y;                   ///< position of overlaid picture
-
-    uint8_t main_is_packed_rgb;
-    uint8_t main_rgba_map[4];
-    uint8_t main_has_alpha;
-    uint8_t overlay_is_packed_rgb;
-    uint8_t overlay_rgba_map[4];
-    uint8_t overlay_has_alpha;
-    int format;                 ///< OverlayFormat
-    int alpha_format;
-    int eval_mode;              ///< EvalMode
-
-    FFFrameSync fs;
-
-    int main_pix_step[4];       ///< steps per pixel for each plane of the main output
-    int overlay_pix_step[4];    ///< steps per pixel for each plane of the overlay
-    int hsub, vsub;             ///< chroma subsampling values
-    const AVPixFmtDescriptor *main_desc; ///< format descriptor for main input
-
-    double var_values[VAR_VARS_NB];
-    char *x_expr, *y_expr;
-
-    AVExpr *x_pexpr, *y_pexpr;
-
-    void (*blend_image)(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y);
-} OverlayContext;
-
 static av_cold void uninit(AVFilterContext *ctx)
 {
     OverlayContext *s = ctx->priv;
@@ -203,6 +154,7 @@ static int process_command(AVFilterContext *ctx, const char *cmd, const char *ar
 
 static const enum AVPixelFormat alpha_pix_fmts[] = {
     AV_PIX_FMT_YUVA420P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_YUVA444P,
+    AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_YUVA422P10,
     AV_PIX_FMT_ARGB, AV_PIX_FMT_ABGR, AV_PIX_FMT_RGBA,
     AV_PIX_FMT_BGRA, AV_PIX_FMT_GBRAP, AV_PIX_FMT_NONE
 };
@@ -221,6 +173,14 @@ static int query_formats(AVFilterContext *ctx)
         AV_PIX_FMT_YUVA420P, AV_PIX_FMT_NONE
     };
 
+    static const enum AVPixelFormat main_pix_fmts_yuv420p10[] = {
+        AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUVA420P10,
+        AV_PIX_FMT_NONE
+    };
+    static const enum AVPixelFormat overlay_pix_fmts_yuv420p10[] = {
+        AV_PIX_FMT_YUVA420P10, AV_PIX_FMT_NONE
+    };
+
     static const enum AVPixelFormat main_pix_fmts_yuv422[] = {
         AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUVJ422P, AV_PIX_FMT_YUVA422P, AV_PIX_FMT_NONE
     };
@@ -228,6 +188,13 @@ static int query_formats(AVFilterContext *ctx)
         AV_PIX_FMT_YUVA422P, AV_PIX_FMT_NONE
     };
 
+    static const enum AVPixelFormat main_pix_fmts_yuv422p10[] = {
+        AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_NONE
+    };
+    static const enum AVPixelFormat overlay_pix_fmts_yuv422p10[] = {
+        AV_PIX_FMT_YUVA422P10, AV_PIX_FMT_NONE
+    };
+
     static const enum AVPixelFormat main_pix_fmts_yuv444[] = {
         AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVA444P, AV_PIX_FMT_NONE
     };
@@ -254,76 +221,52 @@ static int query_formats(AVFilterContext *ctx)
         AV_PIX_FMT_NONE
     };
 
-    AVFilterFormats *main_formats = NULL;
-    AVFilterFormats *overlay_formats = NULL;
+    const enum AVPixelFormat *main_formats, *overlay_formats;
+    AVFilterFormats *formats;
     int ret;
 
     switch (s->format) {
     case OVERLAY_FORMAT_YUV420:
-        if (!(main_formats    = ff_make_format_list(main_pix_fmts_yuv420)) ||
-            !(overlay_formats = ff_make_format_list(overlay_pix_fmts_yuv420))) {
-                ret = AVERROR(ENOMEM);
-                goto fail;
-            }
+        main_formats    = main_pix_fmts_yuv420;
+        overlay_formats = overlay_pix_fmts_yuv420;
+        break;
+    case OVERLAY_FORMAT_YUV420P10:
+        main_formats    = main_pix_fmts_yuv420p10;
+        overlay_formats = overlay_pix_fmts_yuv420p10;
         break;
     case OVERLAY_FORMAT_YUV422:
-        if (!(main_formats    = ff_make_format_list(main_pix_fmts_yuv422)) ||
-            !(overlay_formats = ff_make_format_list(overlay_pix_fmts_yuv422))) {
-                ret = AVERROR(ENOMEM);
-                goto fail;
-            }
+        main_formats    = main_pix_fmts_yuv422;
+        overlay_formats = overlay_pix_fmts_yuv422;
+        break;
+    case OVERLAY_FORMAT_YUV422P10:
+        main_formats    = main_pix_fmts_yuv422p10;
+        overlay_formats = overlay_pix_fmts_yuv422p10;
         break;
     case OVERLAY_FORMAT_YUV444:
-        if (!(main_formats    = ff_make_format_list(main_pix_fmts_yuv444)) ||
-            !(overlay_formats = ff_make_format_list(overlay_pix_fmts_yuv444))) {
-                ret = AVERROR(ENOMEM);
-                goto fail;
-            }
+        main_formats    = main_pix_fmts_yuv444;
+        overlay_formats = overlay_pix_fmts_yuv444;
         break;
     case OVERLAY_FORMAT_RGB:
-        if (!(main_formats    = ff_make_format_list(main_pix_fmts_rgb)) ||
-            !(overlay_formats = ff_make_format_list(overlay_pix_fmts_rgb))) {
-                ret = AVERROR(ENOMEM);
-                goto fail;
-            }
+        main_formats    = main_pix_fmts_rgb;
+        overlay_formats = overlay_pix_fmts_rgb;
         break;
     case OVERLAY_FORMAT_GBRP:
-        if (!(main_formats    = ff_make_format_list(main_pix_fmts_gbrp)) ||
-            !(overlay_formats = ff_make_format_list(overlay_pix_fmts_gbrp))) {
-                ret = AVERROR(ENOMEM);
-                goto fail;
-            }
+        main_formats    = main_pix_fmts_gbrp;
+        overlay_formats = overlay_pix_fmts_gbrp;
         break;
     case OVERLAY_FORMAT_AUTO:
-        if (!(main_formats    = ff_make_format_list(alpha_pix_fmts))) {
-                ret = AVERROR(ENOMEM);
-                goto fail;
-            }
-        break;
+        return ff_set_common_formats(ctx, ff_make_format_list(alpha_pix_fmts));
     default:
         av_assert0(0);
     }
 
-    if (s->format == OVERLAY_FORMAT_AUTO) {
-        ret = ff_set_common_formats(ctx, main_formats);
-        if (ret < 0)
-            goto fail;
-    } else {
-        if ((ret = ff_formats_ref(main_formats   , &ctx->inputs[MAIN]->out_formats   )) < 0 ||
-            (ret = ff_formats_ref(overlay_formats, &ctx->inputs[OVERLAY]->out_formats)) < 0 ||
-            (ret = ff_formats_ref(main_formats   , &ctx->outputs[MAIN]->in_formats   )) < 0)
-                goto fail;
-    }
+    formats = ff_make_format_list(main_formats);
+    if ((ret = ff_formats_ref(formats, &ctx->inputs[MAIN]->outcfg.formats)) < 0 ||
+        (ret = ff_formats_ref(formats, &ctx->outputs[MAIN]->incfg.formats)) < 0)
+        return ret;
 
-    return 0;
-fail:
-    if (main_formats)
-        av_freep(&main_formats->formats);
-    av_freep(&main_formats);
-    if (overlay_formats)
-        av_freep(&overlay_formats->formats);
-    av_freep(&overlay_formats);
-    return ret;
+    return ff_formats_ref(ff_make_format_list(overlay_formats),
+                          &ctx->inputs[OVERLAY]->outcfg.formats);
 }
 
 static int config_input_overlay(AVFilterLink *inlink)
@@ -403,10 +346,10 @@ static int config_output(AVFilterLink *outlink)
  * Blend image in src to destination buffer dst at position (x, y).
  */
 
-static av_always_inline void blend_image_packed_rgb(AVFilterContext *ctx,
+static av_always_inline void blend_slice_packed_rgb(AVFilterContext *ctx,
                                    AVFrame *dst, const AVFrame *src,
                                    int main_has_alpha, int x, int y,
-                                   int is_straight)
+                                   int is_straight, int jobnr, int nb_jobs)
 {
     OverlayContext *s = ctx->priv;
     int i, imax, j, jmax;
@@ -425,13 +368,19 @@ static av_always_inline void blend_image_packed_rgb(AVFilterContext *ctx,
     const int sb = s->overlay_rgba_map[B];
     const int sa = s->overlay_rgba_map[A];
     const int sstep = s->overlay_pix_step[0];
+    int slice_start, slice_end;
     uint8_t *S, *sp, *d, *dp;
 
     i = FFMAX(-y, 0);
-    sp = src->data[0] + i     * src->linesize[0];
-    dp = dst->data[0] + (y+i) * dst->linesize[0];
+    imax = FFMIN3(-y + dst_h, FFMIN(src_h, dst_h), y + src_h);
+
+    slice_start = i + (imax * jobnr) / nb_jobs;
+    slice_end = i + (imax * (jobnr+1)) / nb_jobs;
 
-    for (imax = FFMIN(-y + dst_h, src_h); i < imax; i++) {
+    sp = src->data[0] + (slice_start)     * src->linesize[0];
+    dp = dst->data[0] + (y + slice_start) * dst->linesize[0];
+
+    for (i = slice_start; i < slice_end; i++) {
         j = FFMAX(-x, 0);
         S = sp + j     * sstep;
         d = dp + (x+j) * dstep;
@@ -484,144 +433,225 @@ static av_always_inline void blend_image_packed_rgb(AVFilterContext *ctx,
     }
 }
 
-static av_always_inline void blend_plane(AVFilterContext *ctx,
-                                         AVFrame *dst, const AVFrame *src,
-                                         int src_w, int src_h,
-                                         int dst_w, int dst_h,
-                                         int i, int hsub, int vsub,
-                                         int x, int y,
-                                         int main_has_alpha,
-                                         int dst_plane,
-                                         int dst_offset,
-                                         int dst_step,
-                                         int straight,
-                                         int yuv)
-{
-    int src_wp = AV_CEIL_RSHIFT(src_w, hsub);
-    int src_hp = AV_CEIL_RSHIFT(src_h, vsub);
-    int dst_wp = AV_CEIL_RSHIFT(dst_w, hsub);
-    int dst_hp = AV_CEIL_RSHIFT(dst_h, vsub);
-    int yp = y>>vsub;
-    int xp = x>>hsub;
-    uint8_t *s, *sp, *d, *dp, *dap, *a, *da, *ap;
-    int jmax, j, k, kmax;
-
-    j = FFMAX(-yp, 0);
-    sp = src->data[i] + j         * src->linesize[i];
-    dp = dst->data[dst_plane]
-                      + (yp+j)    * dst->linesize[dst_plane]
-                      + dst_offset;
-    ap = src->data[3] + (j<<vsub) * src->linesize[3];
-    dap = dst->data[3] + ((yp+j) << vsub) * dst->linesize[3];
-
-    for (jmax = FFMIN(-yp + dst_hp, src_hp); j < jmax; j++) {
-        k = FFMAX(-xp, 0);
-        d = dp + (xp+k) * dst_step;
-        s = sp + k;
-        a = ap + (k<<hsub);
-        da = dap + ((xp+k) << hsub);
-
-        for (kmax = FFMIN(-xp + dst_wp, src_wp); k < kmax; k++) {
-            int alpha_v, alpha_h, alpha;
-
-            // average alpha for color components, improve quality
-            if (hsub && vsub && j+1 < src_hp && k+1 < src_wp) {
-                alpha = (a[0] + a[src->linesize[3]] +
-                         a[1] + a[src->linesize[3]+1]) >> 2;
-            } else if (hsub || vsub) {
-                alpha_h = hsub && k+1 < src_wp ?
-                    (a[0] + a[1]) >> 1 : a[0];
-                alpha_v = vsub && j+1 < src_hp ?
-                    (a[0] + a[src->linesize[3]]) >> 1 : a[0];
-                alpha = (alpha_v + alpha_h) >> 1;
-            } else
-                alpha = a[0];
-            // if the main channel has an alpha channel, alpha has to be calculated
-            // to create an un-premultiplied (straight) alpha value
-            if (main_has_alpha && alpha != 0 && alpha != 255) {
-                // average alpha for color components, improve quality
-                uint8_t alpha_d;
-                if (hsub && vsub && j+1 < src_hp && k+1 < src_wp) {
-                    alpha_d = (da[0] + da[dst->linesize[3]] +
-                               da[1] + da[dst->linesize[3]+1]) >> 2;
-                } else if (hsub || vsub) {
-                    alpha_h = hsub && k+1 < src_wp ?
-                        (da[0] + da[1]) >> 1 : da[0];
-                    alpha_v = vsub && j+1 < src_hp ?
-                        (da[0] + da[dst->linesize[3]]) >> 1 : da[0];
-                    alpha_d = (alpha_v + alpha_h) >> 1;
-                } else
-                    alpha_d = da[0];
-                alpha = UNPREMULTIPLY_ALPHA(alpha, alpha_d);
-            }
-            if (straight) {
-                *d = FAST_DIV255(*d * (255 - alpha) + *s * alpha);
-            } else {
-                if (i && yuv)
-                    *d = av_clip(FAST_DIV255((*d - 128) * (255 - alpha)) + *s - 128, -128, 128) + 128;
-                else
-                    *d = FFMIN(FAST_DIV255(*d * (255 - alpha)) + *s, 255);
-            }
-            s++;
-            d += dst_step;
-            da += 1 << hsub;
-            a += 1 << hsub;
-        }
-        dp += dst->linesize[dst_plane];
-        sp += src->linesize[i];
-        ap += (1 << vsub) * src->linesize[3];
-        dap += (1 << vsub) * dst->linesize[3];
-    }
+#define DEFINE_BLEND_PLANE(depth, nbits)                                                                   \
+static av_always_inline void blend_plane_##depth##_##nbits##bits(AVFilterContext *ctx,                     \
+                                         AVFrame *dst, const AVFrame *src,                                 \
+                                         int src_w, int src_h,                                             \
+                                         int dst_w, int dst_h,                                             \
+                                         int i, int hsub, int vsub,                                        \
+                                         int x, int y,                                                     \
+                                         int main_has_alpha,                                               \
+                                         int dst_plane,                                                    \
+                                         int dst_offset,                                                   \
+                                         int dst_step,                                                     \
+                                         int straight,                                                     \
+                                         int yuv,                                                          \
+                                         int jobnr,                                                        \
+                                         int nb_jobs)                                                      \
+{                                                                                                          \
+    OverlayContext *octx = ctx->priv;                                                                      \
+    int src_wp = AV_CEIL_RSHIFT(src_w, hsub);                                                              \
+    int src_hp = AV_CEIL_RSHIFT(src_h, vsub);                                                              \
+    int dst_wp = AV_CEIL_RSHIFT(dst_w, hsub);                                                              \
+    int dst_hp = AV_CEIL_RSHIFT(dst_h, vsub);                                                              \
+    int yp = y>>vsub;                                                                                      \
+    int xp = x>>hsub;                                                                                      \
+    uint##depth##_t *s, *sp, *d, *dp, *dap, *a, *da, *ap;                                                  \
+    int jmax, j, k, kmax;                                                                                  \
+    int slice_start, slice_end;                                                                            \
+    const uint##depth##_t max = (1 << nbits) - 1;                                                          \
+    const uint##depth##_t mid = (1 << (nbits -1)) ;                                                        \
+    int bytes = depth / 8;                                                                                 \
+                                                                                                           \
+    dst_step /= bytes;                                                                                     \
+    j = FFMAX(-yp, 0);                                                                                     \
+    jmax = FFMIN3(-yp + dst_hp, FFMIN(src_hp, dst_hp), yp + src_hp);                                       \
+                                                                                                           \
+    slice_start = j + (jmax * jobnr) / nb_jobs;                                                            \
+    slice_end = j + (jmax * (jobnr+1)) / nb_jobs;                                                          \
+                                                                                                           \
+    sp = (uint##depth##_t *)(src->data[i] + (slice_start) * src->linesize[i]);                             \
+    dp = (uint##depth##_t *)(dst->data[dst_plane]                                                          \
+                      + (yp + slice_start) * dst->linesize[dst_plane]                                      \
+                      + dst_offset);                                                                       \
+    ap = (uint##depth##_t *)(src->data[3] + (slice_start << vsub) * src->linesize[3]);                     \
+    dap = (uint##depth##_t *)(dst->data[3] + ((yp + slice_start) << vsub) * dst->linesize[3]);             \
+                                                                                                           \
+    for (j = slice_start; j < slice_end; j++) {                                                            \
+        k = FFMAX(-xp, 0);                                                                                 \
+        d = dp + (xp+k) * dst_step;                                                                        \
+        s = sp + k;                                                                                        \
+        a = ap + (k<<hsub);                                                                                \
+        da = dap + ((xp+k) << hsub);                                                                       \
+        kmax = FFMIN(-xp + dst_wp, src_wp);                                                                \
+                                                                                                           \
+        if (nbits == 8 && ((vsub && j+1 < src_hp) || !vsub) && octx->blend_row[i]) {                       \
+            int c = octx->blend_row[i]((uint8_t*)d, (uint8_t*)da, (uint8_t*)s,                             \
+                    (uint8_t*)a, kmax - k, src->linesize[3]);                                              \
+                                                                                                           \
+            s += c;                                                                                        \
+            d += dst_step * c;                                                                             \
+            da += (1 << hsub) * c;                                                                         \
+            a += (1 << hsub) * c;                                                                          \
+            k += c;                                                                                        \
+        }                                                                                                  \
+        for (; k < kmax; k++) {                                                                            \
+            int alpha_v, alpha_h, alpha;                                                                   \
+                                                                                                           \
+            /* average alpha for color components, improve quality */                                      \
+            if (hsub && vsub && j+1 < src_hp && k+1 < src_wp) {                                            \
+                alpha = (a[0] + a[src->linesize[3]] +                                                      \
+                         a[1] + a[src->linesize[3]+1]) >> 2;                                               \
+            } else if (hsub || vsub) {                                                                     \
+                alpha_h = hsub && k+1 < src_wp ?                                                           \
+                    (a[0] + a[1]) >> 1 : a[0];                                                             \
+                alpha_v = vsub && j+1 < src_hp ?                                                           \
+                    (a[0] + a[src->linesize[3]]) >> 1 : a[0];                                              \
+                alpha = (alpha_v + alpha_h) >> 1;                                                          \
+            } else                                                                                         \
+                alpha = a[0];                                                                              \
+            /* if the main channel has an alpha channel, alpha has to be calculated */                     \
+            /* to create an un-premultiplied (straight) alpha value */                                     \
+            if (main_has_alpha && alpha != 0 && alpha != max) {                                            \
+                /* average alpha for color components, improve quality */                                  \
+                uint8_t alpha_d;                                                                           \
+                if (hsub && vsub && j+1 < src_hp && k+1 < src_wp) {                                        \
+                    alpha_d = (da[0] + da[dst->linesize[3]] +                                              \
+                               da[1] + da[dst->linesize[3]+1]) >> 2;                                       \
+                } else if (hsub || vsub) {                                                                 \
+                    alpha_h = hsub && k+1 < src_wp ?                                                       \
+                        (da[0] + da[1]) >> 1 : da[0];                                                      \
+                    alpha_v = vsub && j+1 < src_hp ?                                                       \
+                        (da[0] + da[dst->linesize[3]]) >> 1 : da[0];                                       \
+                    alpha_d = (alpha_v + alpha_h) >> 1;                                                    \
+                } else                                                                                     \
+                    alpha_d = da[0];                                                                       \
+                alpha = UNPREMULTIPLY_ALPHA(alpha, alpha_d);                                               \
+            }                                                                                              \
+            if (straight) {                                                                                \
+                if (nbits > 8)                                                                             \
+                   *d = (*d * (max - alpha) + *s * alpha) / max;                                           \
+                else                                                                                       \
+                    *d = FAST_DIV255(*d * (255 - alpha) + *s * alpha);                                     \
+            } else {                                                                                       \
+                if (nbits > 8) {                                                                           \
+                    if (i && yuv)                                                                          \
+                        *d = av_clip((*d * (max - alpha) + *s * alpha) / max + *s - mid, -mid, mid) + mid; \
+                    else                                                                                   \
+                        *d = FFMIN((*d * (max - alpha) + *s * alpha) / max + *s, max);                     \
+                } else {                                                                                   \
+                    if (i && yuv)                                                                          \
+                        *d = av_clip(FAST_DIV255((*d - mid) * (max - alpha)) + *s - mid, -mid, mid) + mid; \
+                    else                                                                                   \
+                        *d = FFMIN(FAST_DIV255(*d * (max - alpha)) + *s, max);                             \
+                }                                                                                          \
+            }                                                                                              \
+            s++;                                                                                           \
+            d += dst_step;                                                                                 \
+            da += 1 << hsub;                                                                               \
+            a += 1 << hsub;                                                                                \
+        }                                                                                                  \
+        dp += dst->linesize[dst_plane] / bytes;                                                            \
+        sp += src->linesize[i] / bytes;                                                                    \
+        ap += (1 << vsub) * src->linesize[3] / bytes;                                                      \
+        dap += (1 << vsub) * dst->linesize[3] / bytes;                                                     \
+    }                                                                                                      \
 }
-
-static inline void alpha_composite(const AVFrame *src, const AVFrame *dst,
-                                   int src_w, int src_h,
-                                   int dst_w, int dst_h,
-                                   int x, int y)
-{
-    uint8_t alpha;          ///< the amount of overlay to blend on to main
-    uint8_t *s, *sa, *d, *da;
-    int i, imax, j, jmax;
-
-    i = FFMAX(-y, 0);
-    sa = src->data[3] + i     * src->linesize[3];
-    da = dst->data[3] + (y+i) * dst->linesize[3];
-
-    for (imax = FFMIN(-y + dst_h, src_h); i < imax; i++) {
-        j = FFMAX(-x, 0);
-        s = sa + j;
-        d = da + x+j;
-
-        for (jmax = FFMIN(-x + dst_w, src_w); j < jmax; j++) {
-            alpha = *s;
-            if (alpha != 0 && alpha != 255) {
-                uint8_t alpha_d = *d;
-                alpha = UNPREMULTIPLY_ALPHA(alpha, alpha_d);
-            }
-            switch (alpha) {
-            case 0:
-                break;
-            case 255:
-                *d = *s;
-                break;
-            default:
-                // apply alpha compositing: main_alpha += (1-main_alpha) * overlay_alpha
-                *d += FAST_DIV255((255 - *d) * *s);
-            }
-            d += 1;
-            s += 1;
-        }
-        da += dst->linesize[3];
-        sa += src->linesize[3];
-    }
+DEFINE_BLEND_PLANE(8, 8)
+DEFINE_BLEND_PLANE(16, 10)
+
+#define DEFINE_ALPHA_COMPOSITE(depth, nbits)                                                               \
+static inline void alpha_composite_##depth##_##nbits##bits(const AVFrame *src, const AVFrame *dst,         \
+                                   int src_w, int src_h,                                                   \
+                                   int dst_w, int dst_h,                                                   \
+                                   int x, int y,                                                           \
+                                   int jobnr, int nb_jobs)                                                 \
+{                                                                                                          \
+    uint##depth##_t alpha;          /* the amount of overlay to blend on to main */                        \
+    uint##depth##_t *s, *sa, *d, *da;                                                                      \
+    int i, imax, j, jmax;                                                                                  \
+    int slice_start, slice_end;                                                                            \
+    const uint##depth##_t max = (1 << nbits) - 1;                                                          \
+    int bytes = depth / 8;                                                                                 \
+                                                                                                           \
+    imax = FFMIN(-y + dst_h, src_h);                                                                       \
+    slice_start = (imax * jobnr) / nb_jobs;                                                                \
+    slice_end = ((imax * (jobnr+1)) / nb_jobs);                                                            \
+                                                                                                           \
+    i = FFMAX(-y, 0);                                                                                      \
+    sa = (uint##depth##_t *)(src->data[3] + (i + slice_start) * src->linesize[3]);                         \
+    da = (uint##depth##_t *)(dst->data[3] + (y + i + slice_start) * dst->linesize[3]);                     \
+                                                                                                           \
+    for (i = i + slice_start; i < slice_end; i++) {                                                        \
+        j = FFMAX(-x, 0);                                                                                  \
+        s = sa + j;                                                                                        \
+        d = da + x+j;                                                                                      \
+                                                                                                           \
+        for (jmax = FFMIN(-x + dst_w, src_w); j < jmax; j++) {                                             \
+            alpha = *s;                                                                                    \
+            if (alpha != 0 && alpha != max) {                                                              \
+                uint8_t alpha_d = *d;                                                                      \
+                alpha = UNPREMULTIPLY_ALPHA(alpha, alpha_d);                                               \
+            }                                                                                              \
+            if (alpha == max)                                                                              \
+                *d = *s;                                                                                   \
+            else if (alpha > 0) {                                                                          \
+                /* apply alpha compositing: main_alpha += (1-main_alpha) * overlay_alpha */                \
+                if (nbits > 8)                                                                             \
+                    *d += (max - *d) * *s / max;                                                           \
+                else                                                                                       \
+                    *d += FAST_DIV255((max - *d) * *s);                                                    \
+            }                                                                                              \
+            d += 1;                                                                                        \
+            s += 1;                                                                                        \
+        }                                                                                                  \
+        da += dst->linesize[3] / bytes;                                                                    \
+        sa += src->linesize[3] / bytes;                                                                    \
+    }                                                                                                      \
+}
+DEFINE_ALPHA_COMPOSITE(8, 8)
+DEFINE_ALPHA_COMPOSITE(16, 10)
+
+#define DEFINE_BLEND_SLICE_YUV(depth, nbits)                                                               \
+static av_always_inline void blend_slice_yuv_##depth##_##nbits##bits(AVFilterContext *ctx,                 \
+                                             AVFrame *dst, const AVFrame *src,                             \
+                                             int hsub, int vsub,                                           \
+                                             int main_has_alpha,                                           \
+                                             int x, int y,                                                 \
+                                             int is_straight,                                              \
+                                             int jobnr, int nb_jobs)                                       \
+{                                                                                                          \
+    OverlayContext *s = ctx->priv;                                                                         \
+    const int src_w = src->width;                                                                          \
+    const int src_h = src->height;                                                                         \
+    const int dst_w = dst->width;                                                                          \
+    const int dst_h = dst->height;                                                                         \
+                                                                                                           \
+    blend_plane_##depth##_##nbits##bits(ctx, dst, src, src_w, src_h, dst_w, dst_h, 0, 0,       0,          \
+                x, y, main_has_alpha, s->main_desc->comp[0].plane, s->main_desc->comp[0].offset,           \
+                s->main_desc->comp[0].step, is_straight, 1, jobnr, nb_jobs);                               \
+    blend_plane_##depth##_##nbits##bits(ctx, dst, src, src_w, src_h, dst_w, dst_h, 1, hsub, vsub,          \
+                x, y, main_has_alpha, s->main_desc->comp[1].plane, s->main_desc->comp[1].offset,           \
+                s->main_desc->comp[1].step, is_straight, 1, jobnr, nb_jobs);                               \
+    blend_plane_##depth##_##nbits##bits(ctx, dst, src, src_w, src_h, dst_w, dst_h, 2, hsub, vsub,          \
+                x, y, main_has_alpha, s->main_desc->comp[2].plane, s->main_desc->comp[2].offset,           \
+                s->main_desc->comp[2].step, is_straight, 1, jobnr, nb_jobs);                               \
+                                                                                                           \
+    if (main_has_alpha)                                                                                    \
+        alpha_composite_##depth##_##nbits##bits(src, dst, src_w, src_h, dst_w, dst_h, x, y,                \
+                                                jobnr, nb_jobs);                                           \
 }
+DEFINE_BLEND_SLICE_YUV(8, 8)
+DEFINE_BLEND_SLICE_YUV(16, 10)
 
-static av_always_inline void blend_image_yuv(AVFilterContext *ctx,
-                                             AVFrame *dst, const AVFrame *src,
-                                             int hsub, int vsub,
-                                             int main_has_alpha,
-                                             int x, int y,
-                                             int is_straight)
+static av_always_inline void blend_slice_planar_rgb(AVFilterContext *ctx,
+                                                    AVFrame *dst, const AVFrame *src,
+                                                    int hsub, int vsub,
+                                                    int main_has_alpha,
+                                                    int x, int y,
+                                                    int is_straight,
+                                                    int jobnr,
+                                                    int nb_jobs)
 {
     OverlayContext *s = ctx->priv;
     const int src_w = src->width;
@@ -629,139 +659,210 @@ static av_always_inline void blend_image_yuv(AVFilterContext *ctx,
     const int dst_w = dst->width;
     const int dst_h = dst->height;
 
-    blend_plane(ctx, dst, src, src_w, src_h, dst_w, dst_h, 0, 0,       0, x, y, main_has_alpha,
-                s->main_desc->comp[0].plane, s->main_desc->comp[0].offset, s->main_desc->comp[0].step, is_straight, 1);
-    blend_plane(ctx, dst, src, src_w, src_h, dst_w, dst_h, 1, hsub, vsub, x, y, main_has_alpha,
-                s->main_desc->comp[1].plane, s->main_desc->comp[1].offset, s->main_desc->comp[1].step, is_straight, 1);
-    blend_plane(ctx, dst, src, src_w, src_h, dst_w, dst_h, 2, hsub, vsub, x, y, main_has_alpha,
-                s->main_desc->comp[2].plane, s->main_desc->comp[2].offset, s->main_desc->comp[2].step, is_straight, 1);
+    blend_plane_8_8bits(ctx, dst, src, src_w, src_h, dst_w, dst_h, 0, 0,   0, x, y, main_has_alpha,
+                s->main_desc->comp[1].plane, s->main_desc->comp[1].offset, s->main_desc->comp[1].step, is_straight, 0,
+                jobnr, nb_jobs);
+    blend_plane_8_8bits(ctx, dst, src, src_w, src_h, dst_w, dst_h, 1, hsub, vsub, x, y, main_has_alpha,
+                s->main_desc->comp[2].plane, s->main_desc->comp[2].offset, s->main_desc->comp[2].step, is_straight, 0,
+                jobnr, nb_jobs);
+    blend_plane_8_8bits(ctx, dst, src, src_w, src_h, dst_w, dst_h, 2, hsub, vsub, x, y, main_has_alpha,
+                s->main_desc->comp[0].plane, s->main_desc->comp[0].offset, s->main_desc->comp[0].step, is_straight, 0,
+                jobnr, nb_jobs);
 
     if (main_has_alpha)
-        alpha_composite(src, dst, src_w, src_h, dst_w, dst_h, x, y);
+        alpha_composite_8_8bits(src, dst, src_w, src_h, dst_w, dst_h, x, y, jobnr, nb_jobs);
 }
 
-static av_always_inline void blend_image_planar_rgb(AVFilterContext *ctx,
-                                                    AVFrame *dst, const AVFrame *src,
-                                                    int hsub, int vsub,
-                                                    int main_has_alpha,
-                                                    int x, int y,
-                                                    int is_straight)
+static int blend_slice_yuv420(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
     OverlayContext *s = ctx->priv;
-    const int src_w = src->width;
-    const int src_h = src->height;
-    const int dst_w = dst->width;
-    const int dst_h = dst->height;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 1, 1, 0, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
+}
 
-    blend_plane(ctx, dst, src, src_w, src_h, dst_w, dst_h, 0, 0,       0, x, y, main_has_alpha,
-                s->main_desc->comp[1].plane, s->main_desc->comp[1].offset, s->main_desc->comp[1].step, is_straight, 0);
-    blend_plane(ctx, dst, src, src_w, src_h, dst_w, dst_h, 1, hsub, vsub, x, y, main_has_alpha,
-                s->main_desc->comp[2].plane, s->main_desc->comp[2].offset, s->main_desc->comp[2].step, is_straight, 0);
-    blend_plane(ctx, dst, src, src_w, src_h, dst_w, dst_h, 2, hsub, vsub, x, y, main_has_alpha,
-                s->main_desc->comp[0].plane, s->main_desc->comp[0].offset, s->main_desc->comp[0].step, is_straight, 0);
+static int blend_slice_yuva420(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
+{
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 1, 1, 1, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
+}
 
-    if (main_has_alpha)
-        alpha_composite(src, dst, src_w, src_h, dst_w, dst_h, x, y);
+static int blend_slice_yuv420p10(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
+{
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_16_10bits(ctx, td->dst, td->src, 1, 1, 0, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuv420(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuva420p10(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 1, 1, 0, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_16_10bits(ctx, td->dst, td->src, 1, 1, 1, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuva420(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuv422p10(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 1, 1, 1, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_16_10bits(ctx, td->dst, td->src, 1, 0, 0, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuv422(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuva422p10(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 1, 0, 0, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_16_10bits(ctx, td->dst, td->src, 1, 0, 1, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuva422(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuv422(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 1, 0, 1, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 1, 0, 0, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuv444(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuva422(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 0, 0, 0, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 1, 0, 1, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuva444(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuv444(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 0, 0, 1, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 0, 0, 0, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_gbrp(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuva444(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_planar_rgb(ctx, dst, src, 0, 0, 0, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 0, 0, 1, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_gbrap(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_gbrp(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_planar_rgb(ctx, dst, src, 0, 0, 1, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_planar_rgb(ctx, td->dst, td->src, 0, 0, 0, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuv420_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_gbrap(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 1, 1, 0, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_planar_rgb(ctx, td->dst, td->src, 0, 0, 1, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuva420_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuv420_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 1, 1, 1, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 1, 1, 0, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuv422_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuva420_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 1, 0, 0, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 1, 1, 1, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuva422_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuv422_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 1, 0, 1, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 1, 0, 0, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuv444_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuva422_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 0, 0, 0, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 1, 0, 1, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_yuva444_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuv444_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_yuv(ctx, dst, src, 0, 0, 1, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 0, 0, 0, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_gbrp_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_yuva444_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_planar_rgb(ctx, dst, src, 0, 0, 0, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_yuv_8_8bits(ctx, td->dst, td->src, 0, 0, 1, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
+}
+
+static int blend_slice_gbrp_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
+{
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_planar_rgb(ctx, td->dst, td->src, 0, 0, 0, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_gbrap_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_gbrap_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_planar_rgb(ctx, dst, src, 0, 0, 1, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_planar_rgb(ctx, td->dst, td->src, 0, 0, 1, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_rgb(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_rgb(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_packed_rgb(ctx, dst, src, 0, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_packed_rgb(ctx, td->dst, td->src, 0, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_rgba(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_rgba(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_packed_rgb(ctx, dst, src, 1, x, y, 1);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_packed_rgb(ctx, td->dst, td->src, 1, s->x, s->y, 1, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_rgb_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_rgb_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_packed_rgb(ctx, dst, src, 0, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_packed_rgb(ctx, td->dst, td->src, 0, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
 }
 
-static void blend_image_rgba_pm(AVFilterContext *ctx, AVFrame *dst, const AVFrame *src, int x, int y)
+static int blend_slice_rgba_pm(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    blend_image_packed_rgb(ctx, dst, src, 1, x, y, 0);
+    OverlayContext *s = ctx->priv;
+    ThreadData *td = arg;
+    blend_slice_packed_rgb(ctx, td->dst, td->src, 1, s->x, s->y, 0, jobnr, nb_jobs);
+    return 0;
 }
 
 static int config_input_main(AVFilterLink *inlink)
@@ -781,39 +882,51 @@ static int config_input_main(AVFilterLink *inlink)
     s->main_has_alpha = ff_fmt_is_in(inlink->format, alpha_pix_fmts);
     switch (s->format) {
     case OVERLAY_FORMAT_YUV420:
-        s->blend_image = s->main_has_alpha ? blend_image_yuva420 : blend_image_yuv420;
+        s->blend_slice = s->main_has_alpha ? blend_slice_yuva420 : blend_slice_yuv420;
+        break;
+    case OVERLAY_FORMAT_YUV420P10:
+        s->blend_slice = s->main_has_alpha ? blend_slice_yuva420p10 : blend_slice_yuv420p10;
         break;
     case OVERLAY_FORMAT_YUV422:
-        s->blend_image = s->main_has_alpha ? blend_image_yuva422 : blend_image_yuv422;
+        s->blend_slice = s->main_has_alpha ? blend_slice_yuva422 : blend_slice_yuv422;
+        break;
+    case OVERLAY_FORMAT_YUV422P10:
+        s->blend_slice = s->main_has_alpha ? blend_slice_yuva422p10 : blend_slice_yuv422p10;
         break;
     case OVERLAY_FORMAT_YUV444:
-        s->blend_image = s->main_has_alpha ? blend_image_yuva444 : blend_image_yuv444;
+        s->blend_slice = s->main_has_alpha ? blend_slice_yuva444 : blend_slice_yuv444;
         break;
     case OVERLAY_FORMAT_RGB:
-        s->blend_image = s->main_has_alpha ? blend_image_rgba : blend_image_rgb;
+        s->blend_slice = s->main_has_alpha ? blend_slice_rgba : blend_slice_rgb;
         break;
     case OVERLAY_FORMAT_GBRP:
-        s->blend_image = s->main_has_alpha ? blend_image_gbrap : blend_image_gbrp;
+        s->blend_slice = s->main_has_alpha ? blend_slice_gbrap : blend_slice_gbrp;
         break;
     case OVERLAY_FORMAT_AUTO:
         switch (inlink->format) {
         case AV_PIX_FMT_YUVA420P:
-            s->blend_image = blend_image_yuva420;
+            s->blend_slice = blend_slice_yuva420;
+            break;
+        case AV_PIX_FMT_YUVA420P10:
+            s->blend_slice = blend_slice_yuva420p10;
             break;
         case AV_PIX_FMT_YUVA422P:
-            s->blend_image = blend_image_yuva422;
+            s->blend_slice = blend_slice_yuva422;
+            break;
+        case AV_PIX_FMT_YUVA422P10:
+            s->blend_slice = blend_slice_yuva422p10;
             break;
         case AV_PIX_FMT_YUVA444P:
-            s->blend_image = blend_image_yuva444;
+            s->blend_slice = blend_slice_yuva444;
             break;
         case AV_PIX_FMT_ARGB:
         case AV_PIX_FMT_RGBA:
         case AV_PIX_FMT_BGRA:
         case AV_PIX_FMT_ABGR:
-            s->blend_image = blend_image_rgba;
+            s->blend_slice = blend_slice_rgba;
             break;
         case AV_PIX_FMT_GBRAP:
-            s->blend_image = blend_image_gbrap;
+            s->blend_slice = blend_slice_gbrap;
             break;
         default:
             av_assert0(0);
@@ -823,43 +936,43 @@ static int config_input_main(AVFilterLink *inlink)
     }
 
     if (!s->alpha_format)
-        return 0;
+        goto end;
 
     switch (s->format) {
     case OVERLAY_FORMAT_YUV420:
-        s->blend_image = s->main_has_alpha ? blend_image_yuva420_pm : blend_image_yuv420_pm;
+        s->blend_slice = s->main_has_alpha ? blend_slice_yuva420_pm : blend_slice_yuv420_pm;
         break;
     case OVERLAY_FORMAT_YUV422:
-        s->blend_image = s->main_has_alpha ? blend_image_yuva422_pm : blend_image_yuv422_pm;
+        s->blend_slice = s->main_has_alpha ? blend_slice_yuva422_pm : blend_slice_yuv422_pm;
         break;
     case OVERLAY_FORMAT_YUV444:
-        s->blend_image = s->main_has_alpha ? blend_image_yuva444_pm : blend_image_yuv444_pm;
+        s->blend_slice = s->main_has_alpha ? blend_slice_yuva444_pm : blend_slice_yuv444_pm;
         break;
     case OVERLAY_FORMAT_RGB:
-        s->blend_image = s->main_has_alpha ? blend_image_rgba_pm : blend_image_rgb_pm;
+        s->blend_slice = s->main_has_alpha ? blend_slice_rgba_pm : blend_slice_rgb_pm;
         break;
     case OVERLAY_FORMAT_GBRP:
-        s->blend_image = s->main_has_alpha ? blend_image_gbrap_pm : blend_image_gbrp_pm;
+        s->blend_slice = s->main_has_alpha ? blend_slice_gbrap_pm : blend_slice_gbrp_pm;
         break;
     case OVERLAY_FORMAT_AUTO:
         switch (inlink->format) {
         case AV_PIX_FMT_YUVA420P:
-            s->blend_image = blend_image_yuva420_pm;
+            s->blend_slice = blend_slice_yuva420_pm;
             break;
         case AV_PIX_FMT_YUVA422P:
-            s->blend_image = blend_image_yuva422_pm;
+            s->blend_slice = blend_slice_yuva422_pm;
             break;
         case AV_PIX_FMT_YUVA444P:
-            s->blend_image = blend_image_yuva444_pm;
+            s->blend_slice = blend_slice_yuva444_pm;
             break;
         case AV_PIX_FMT_ARGB:
         case AV_PIX_FMT_RGBA:
         case AV_PIX_FMT_BGRA:
         case AV_PIX_FMT_ABGR:
-            s->blend_image = blend_image_rgba_pm;
+            s->blend_slice = blend_slice_rgba_pm;
             break;
         case AV_PIX_FMT_GBRAP:
-            s->blend_image = blend_image_gbrap_pm;
+            s->blend_slice = blend_slice_gbrap_pm;
             break;
         default:
             av_assert0(0);
@@ -867,6 +980,12 @@ static int config_input_main(AVFilterLink *inlink)
         }
         break;
     }
+
+end:
+    if (ARCH_X86)
+        ff_overlay_init_x86(s, s->format, inlink->format,
+                            s->alpha_format, s->main_has_alpha);
+
     return 0;
 }
 
@@ -904,9 +1023,15 @@ static int do_blend(FFFrameSync *fs)
                s->var_values[VAR_Y], s->y);
     }
 
-    if (s->x < mainpic->width  && s->x + second->width  >= 0 ||
-        s->y < mainpic->height && s->y + second->height >= 0)
-        s->blend_image(ctx, mainpic, second, s->x, s->y);
+    if (s->x < mainpic->width  && s->x + second->width  >= 0 &&
+        s->y < mainpic->height && s->y + second->height >= 0) {
+        ThreadData td;
+
+        td.dst = mainpic;
+        td.src = second;
+        ctx->internal->execute(ctx, s->blend_slice, &td, NULL, FFMIN(FFMAX(1, FFMIN3(s->y + second->height, FFMIN(second->height, mainpic->height), mainpic->height - s->y)),
+                                                                     ff_filter_get_nb_threads(ctx)));
+    }
     return ff_filter_frame(ctx->outputs[0], mainpic);
 }
 
@@ -928,8 +1053,8 @@ static int activate(AVFilterContext *ctx)
 #define FLAGS AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
 
 static const AVOption overlay_options[] = {
-    { "x", "set the x expression", OFFSET(x_expr), AV_OPT_TYPE_STRING, {.str = "0"}, CHAR_MIN, CHAR_MAX, FLAGS },
-    { "y", "set the y expression", OFFSET(y_expr), AV_OPT_TYPE_STRING, {.str = "0"}, CHAR_MIN, CHAR_MAX, FLAGS },
+    { "x", "set the x expression", OFFSET(x_expr), AV_OPT_TYPE_STRING, {.str = "0"}, 0, 0, FLAGS },
+    { "y", "set the y expression", OFFSET(y_expr), AV_OPT_TYPE_STRING, {.str = "0"}, 0, 0, FLAGS },
     { "eof_action", "Action to take when encountering EOF from secondary input ",
         OFFSET(fs.opt_eof_action), AV_OPT_TYPE_INT, { .i64 = EOF_ACTION_REPEAT },
         EOF_ACTION_REPEAT, EOF_ACTION_PASS, .flags = FLAGS, "eof_action" },
@@ -942,7 +1067,9 @@ static const AVOption overlay_options[] = {
     { "shortest", "force termination when the shortest input terminates", OFFSET(fs.opt_shortest), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
     { "format", "set output format", OFFSET(format), AV_OPT_TYPE_INT, {.i64=OVERLAY_FORMAT_YUV420}, 0, OVERLAY_FORMAT_NB-1, FLAGS, "format" },
         { "yuv420", "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_YUV420}, .flags = FLAGS, .unit = "format" },
+        { "yuv420p10", "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_YUV420P10}, .flags = FLAGS, .unit = "format" },
         { "yuv422", "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_YUV422}, .flags = FLAGS, .unit = "format" },
+        { "yuv422p10", "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_YUV422P10}, .flags = FLAGS, .unit = "format" },
         { "yuv444", "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_YUV444}, .flags = FLAGS, .unit = "format" },
         { "rgb",    "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_RGB},    .flags = FLAGS, .unit = "format" },
         { "gbrp",   "", 0, AV_OPT_TYPE_CONST, {.i64=OVERLAY_FORMAT_GBRP},   .flags = FLAGS, .unit = "format" },
@@ -992,5 +1119,6 @@ AVFilter ff_vf_overlay = {
     .process_command = process_command,
     .inputs        = avfilter_vf_overlay_inputs,
     .outputs       = avfilter_vf_overlay_outputs,
-    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
+    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL |
+                     AVFILTER_FLAG_SLICE_THREADS,
 };