avfilter: slice processing for geq

Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>

diff --git a/libavfilter/vf_geq.c b/libavfilter/vf_geq.c
index 0bd81fd58644981db42bf840831c1be2ceb8a578..0651ed0b36aba6bb599bec00c90f0b103648e587 100644 (file)
--- a/libavfilter/vf_geq.c
+++ b/libavfilter/vf_geq.c
@@ -41,6 +41,9 @@ typedef struct GEQContext {
     AVExpr *e[4];               ///< expressions for each plane
     char *expr_str[4+3];        ///< expression strings for each plane
     AVFrame *picref;            ///< current input buffer
+    uint8_t *dst;               ///< reference pointer to the 8bits output
+    uint16_t *dst16;            ///< reference pointer to the 16bits output
+    double values[VAR_VARS_NB]; ///< expression values
     int hsub, vsub;             ///< chroma subsampling
     int planes;                 ///< number of planes
     int is_rgb;
@@ -226,8 +229,62 @@ static int geq_config_props(AVFilterLink *inlink)
 
     geq->hsub = desc->log2_chroma_w;
     geq->vsub = desc->log2_chroma_h;
+    geq->bps = desc->comp[0].depth;
     geq->planes = desc->nb_components;
-    geq->bps    = desc->comp[0].depth;
+    return 0;
+}
+
+typedef struct ThreadData {
+    int height;
+    int width;
+    int plane;
+    int linesize;
+} ThreadData;
+
+static int slice_geq_filter(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
+{
+    GEQContext *geq = ctx->priv;
+    ThreadData *td = arg;
+    const int height = td->height;
+    const int width = td->width;
+    const int plane = td->plane;
+    const int linesize = td->linesize;
+    const int slice_start = (height *  jobnr) / nb_jobs;
+    const int slice_end = (height * (jobnr+1)) / nb_jobs;
+    int x, y;
+    uint8_t *ptr;
+    uint16_t *ptr16;
+
+    double values[VAR_VARS_NB];
+    values[VAR_W] = geq->values[VAR_W];
+    values[VAR_H] = geq->values[VAR_H];
+    values[VAR_N] = geq->values[VAR_N];
+    values[VAR_SW] = geq->values[VAR_SW];
+    values[VAR_SH] = geq->values[VAR_SH];
+    values[VAR_T] = geq->values[VAR_T];
+
+    if (geq->bps == 8) {
+        for (y = slice_start; y < slice_end; y++) {
+            ptr = geq->dst + linesize * y;
+            values[VAR_Y] = y;
+
+            for (x = 0; x < width; x++) {
+                values[VAR_X] = x;
+                ptr[x] = av_expr_eval(geq->e[plane], values, geq);
+            }
+            ptr += linesize;
+        }
+    }
+    else {
+        for (y = slice_start; y < slice_end; y++) {
+            ptr16 = geq->dst16 + (linesize/2) * y;
+            values[VAR_Y] = y;
+            for (x = 0; x < width; x++) {
+                values[VAR_X] = x;
+                ptr16[x] = av_expr_eval(geq->e[plane], values, geq);
+            }
+        }
+    }
 
     return 0;
 }
@@ -235,13 +292,14 @@ static int geq_config_props(AVFilterLink *inlink)
 static int geq_filter_frame(AVFilterLink *inlink, AVFrame *in)
 {
     int plane;
-    GEQContext *geq = inlink->dst->priv;
+    AVFilterContext *ctx = inlink->dst;
+    const int nb_threads = ff_filter_get_nb_threads(ctx);
+    GEQContext *geq = ctx->priv;
     AVFilterLink *outlink = inlink->dst->outputs[0];
     AVFrame *out;
-    double values[VAR_VARS_NB] = {
-        [VAR_N] = inlink->frame_count_out,
-        [VAR_T] = in->pts == AV_NOPTS_VALUE ? NAN : in->pts * av_q2d(inlink->time_base),
-    };
+
+    geq->values[VAR_N] = inlink->frame_count_out,
+    geq->values[VAR_T] = in->pts == AV_NOPTS_VALUE ? NAN : in->pts * av_q2d(inlink->time_base),
 
     geq->picref = in;
     out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
@@ -252,34 +310,25 @@ static int geq_filter_frame(AVFilterLink *inlink, AVFrame *in)
     av_frame_copy_props(out, in);
 
     for (plane = 0; plane < geq->planes && out->data[plane]; plane++) {
-        int x, y;
-        uint8_t *dst = out->data[plane];
-        uint16_t *dst16 = (uint16_t*)out->data[plane];
-        const int linesize = out->linesize[plane];
         const int w = (plane == 1 || plane == 2) ? AV_CEIL_RSHIFT(inlink->w, geq->hsub) : inlink->w;
         const int h = (plane == 1 || plane == 2) ? AV_CEIL_RSHIFT(inlink->h, geq->vsub) : inlink->h;
+        const int linesize = out->linesize[plane];
+        ThreadData td;
 
-        values[VAR_W]  = w;
-        values[VAR_H]  = h;
-        values[VAR_SW] = w / (double)inlink->w;
-        values[VAR_SH] = h / (double)inlink->h;
+        geq->dst = out->data[plane];
+        geq->dst16 = (uint16_t*)out->data[plane];
 
-        for (y = 0; y < h; y++) {
-            values[VAR_Y] = y;
-            if (geq->bps > 8) {
-                for (x = 0; x < w; x++) {
-                    values[VAR_X] = x;
-                    dst16[x] = av_expr_eval(geq->e[plane], values, geq);
-                }
-                dst16 += linesize / 2;
-            } else {
-                for (x = 0; x < w; x++) {
-                    values[VAR_X] = x;
-                    dst[x] = av_expr_eval(geq->e[plane], values, geq);
-                }
-                dst += linesize;
-            }
-        }
+        geq->values[VAR_W]  = w;
+        geq->values[VAR_H]  = h;
+        geq->values[VAR_SW] = w / (double)inlink->w;
+        geq->values[VAR_SH] = h / (double)inlink->h;
+
+        td.width = w;
+        td.height = h;
+        td.plane = plane;
+        td.linesize = linesize;
+
+        ctx->internal->execute(ctx, slice_geq_filter, &td, NULL, FFMIN(h, nb_threads));
     }
 
     av_frame_free(&geq->picref);
@@ -323,5 +372,5 @@ AVFilter ff_vf_geq = {
     .inputs        = geq_inputs,
     .outputs       = geq_outputs,
     .priv_class    = &geq_class,
-    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
+    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
 };