[FFmpeg-devel] [PATCH] lavfi/noise: support slice threading

[Paul B Mahol]

Signed-off-by: Paul B Mahol <onemda at gmail.com>
---
 libavfilter/vf_noise.c | 56 +++++++++++++++++++++++++++++++++++---------------
 1 file changed, 39 insertions(+), 17 deletions(-)

diff --git a/libavfilter/vf_noise.c b/libavfilter/vf_noise.c
index 12aca24..4d59d58 100644
--- a/libavfilter/vf_noise.c
+++ b/libavfilter/vf_noise.c
@@ -67,6 +67,13 @@ typedef struct {
     void (*line_noise_avg)(uint8_t *dst, const uint8_t *src, int len, int8_t **shift);
 } NoiseContext;
 
+typedef struct ThreadData {
+    const uint8_t *src;
+    uint8_t *dst;
+    int dst_linesize, src_linesize;
+    int comp;
+} ThreadData;
+
 #define OFFSET(x) offsetof(NoiseContext, x)
 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
 
@@ -328,22 +335,30 @@ static inline void line_noise_avg_mmx(uint8_t *dst, const uint8_t *src,
 #endif
 }
 
-static void noise(uint8_t *dst, const uint8_t *src,
-                  int dst_linesize, int src_linesize,
-                  int width, int height, NoiseContext *n, int comp)
+static int noise(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
+    NoiseContext *n = ctx->priv;
+    ThreadData *td = arg;
+    const int comp = td->comp;
+    int height = n->height[comp];
+    int start = (height *  jobnr   ) / nb_jobs;
+    int end   = (height * (jobnr+1)) / nb_jobs;
     int8_t *noise = n->param[comp].noise;
     int flags = n->param[comp].flags;
     AVLFG *lfg = &n->param[comp].lfg;
+    uint8_t *dst = td->dst + start * td->dst_linesize;
+    const uint8_t *src = td->src + start * td->src_linesize;
+    const int width = n->linesize[comp];
     int shift, y;
 
     if (!noise) {
         if (dst != src)
-            av_image_copy_plane(dst, dst_linesize, src, src_linesize, width, height);
-        return;
+            av_image_copy_plane(dst, td->dst_linesize, src, td->src_linesize,
+                                width, end - start);
+        return 0;
     }
 
-    for (y = 0; y < height; y++) {
+    for (y = start; y < end; y++) {
         if (flags & NOISE_TEMPORAL)
             shift = av_lfg_get(lfg) & (MAX_SHIFT - 1);
         else
@@ -355,19 +370,18 @@ static void noise(uint8_t *dst, const uint8_t *src,
         } else {
             n->line_noise(dst, src, noise, width, shift);
         }
-        dst += dst_linesize;
-        src += src_linesize;
+        dst += td->dst_linesize;
+        src += td->src_linesize;
     }
 
-    n->param[comp].shiftptr++;
-    if (n->param[comp].shiftptr == 3)
-        n->param[comp].shiftptr = 0;
+    return 0;
 }
 
 static int filter_frame(AVFilterLink *inlink, AVFrame *inpicref)
 {
-    NoiseContext *n = inlink->dst->priv;
-    AVFilterLink *outlink = inlink->dst->outputs[0];
+    AVFilterContext *ctx = inlink->dst;
+    AVFilterLink *outlink = ctx->outputs[0];
+    NoiseContext *n = ctx->priv;
     AVFrame *out;
     int i;
 
@@ -382,9 +396,17 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *inpicref)
         av_frame_copy_props(out, inpicref);
     }
 
-    for (i = 0; i < n->nb_planes; i++)
-        noise(out->data[i], inpicref->data[i], out->linesize[i],
-              inpicref->linesize[i], n->linesize[i], n->height[i], n, i);
+    for (i = 0; i < n->nb_planes; i++) {
+        ThreadData td = { .src = inpicref->data[i], .dst = out->data[i],
+                          .src_linesize = inpicref->linesize[i],
+                          .dst_linesize = out->linesize[i], .comp = i };
+
+        ctx->internal->execute(ctx, noise, &td, NULL, FFMIN(n->height[i], ctx->graph->nb_threads));
+
+        n->param[i].shiftptr++;
+        if (n->param[i].shiftptr == 3)
+            n->param[i].shiftptr = 0;
+    }
     emms_c();
 
     if (inpicref != out)
@@ -463,5 +485,5 @@ AVFilter avfilter_vf_noise = {
     .inputs        = noise_inputs,
     .outputs       = noise_outputs,
     .priv_class    = &noise_class,
-    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
+    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
 };
-- 
1.7.11.2