avfilter/vf_convolution: add slice threading support

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

1 files changed, 158 insertions, 63 deletions

diff --git a/libavfilter/vf_convolution.c b/libavfilter/vf_convolution.c
index 7b512677f8..41e92497c3 100644
--- a/libavfilter/vf_convolution.c
+++ b/libavfilter/vf_convolution.c
@@ -40,16 +40,19 @@ typedef struct ConvolutionContext {
 
     int size[4];
     int depth;
+    int bpc;
     int bstride;
     uint8_t *buffer;
+    uint8_t **bptrs;
     int nb_planes;
+    int nb_threads;
     int planewidth[4];
     int planeheight[4];
     int matrix[4][25];
     int matrix_length[4];
     int copy[4];
 
-    void (*filter[4])(struct ConvolutionContext *s, AVFrame *in, AVFrame *out, int plane);
+    int (*filter[4])(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
 } ConvolutionContext;
 
 #define OFFSET(x) offsetof(ConvolutionContext, x)
@@ -133,27 +136,39 @@ static inline void line_copy16(uint16_t *line, const uint16_t *srcp, int width,
     }
 }
 
-static void filter16_prewitt(ConvolutionContext *s, AVFrame *in, AVFrame *out, int plane)
+typedef struct ThreadData {
+    AVFrame *in, *out;
+    int plane;
+} ThreadData;
+
+static int filter16_prewitt(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    const uint16_t *src = (const uint16_t *)in->data[plane];
-    uint16_t *dst = (uint16_t *)out->data[plane];
+    ConvolutionContext *s = ctx->priv;
+    ThreadData *td = arg;
+    AVFrame *in = td->in;
+    AVFrame *out = td->out;
+    const int plane = td->plane;
     const int peak = (1 << s->depth) - 1;
     const int stride = in->linesize[plane] / 2;
     const int bstride = s->bstride;
     const int height = s->planeheight[plane];
     const int width  = s->planewidth[plane];
+    const int slice_start = (height * jobnr) / nb_jobs;
+    const int slice_end = (height * (jobnr+1)) / nb_jobs;
+    const uint16_t *src = (const uint16_t *)in->data[plane] + slice_start * stride;
+    uint16_t *dst = (uint16_t *)out->data[plane] + slice_start * (out->linesize[plane] / 2);
     const float scale = s->scale;
     const float delta = s->delta;
-    uint16_t *p0 = (uint16_t *)s->buffer + 16;
+    uint16_t *p0 = (uint16_t *)s->bptrs[jobnr] + 16;
     uint16_t *p1 = p0 + bstride;
     uint16_t *p2 = p1 + bstride;
     uint16_t *orig = p0, *end = p2;
     int y, x;
 
-    line_copy16(p0, src + stride, width, 1);
+    line_copy16(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
     line_copy16(p1, src, width, 1);
 
-    for (y = 0; y < height; y++) {
+    for (y = slice_start; y < slice_end; y++) {
         src += stride * (y < height - 1 ? 1 : -1);
         line_copy16(p2, src, width, 1);
 
@@ -179,29 +194,38 @@ static void filter16_prewitt(ConvolutionContext *s, AVFrame *in, AVFrame *out, i
         p2 = (p2 == end) ? orig: p2 + bstride;
         dst += out->linesize[plane] / 2;
     }
+
+    return 0;
 }
 
-static void filter16_sobel(ConvolutionContext *s, AVFrame *in, AVFrame *out, int plane)
+static int filter16_sobel(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    const uint16_t *src = (const uint16_t *)in->data[plane];
-    uint16_t *dst = (uint16_t *)out->data[plane];
+    ConvolutionContext *s = ctx->priv;
+    ThreadData *td = arg;
+    AVFrame *in = td->in;
+    AVFrame *out = td->out;
+    const int plane = td->plane;
     const int peak = (1 << s->depth) - 1;
     const int stride = in->linesize[plane] / 2;
     const int bstride = s->bstride;
     const int height = s->planeheight[plane];
     const int width  = s->planewidth[plane];
+    const int slice_start = (height * jobnr) / nb_jobs;
+    const int slice_end = (height * (jobnr+1)) / nb_jobs;
+    const uint16_t *src = (const uint16_t *)in->data[plane] + slice_start * stride;
+    uint16_t *dst = (uint16_t *)out->data[plane] + slice_start * (out->linesize[plane] / 2);
     const float scale = s->scale;
     const float delta = s->delta;
-    uint16_t *p0 = (uint16_t *)s->buffer + 16;
+    uint16_t *p0 = (uint16_t *)s->bptrs[jobnr] + 16;
     uint16_t *p1 = p0 + bstride;
     uint16_t *p2 = p1 + bstride;
     uint16_t *orig = p0, *end = p2;
     int y, x;
 
-    line_copy16(p0, src + stride, width, 1);
+    line_copy16(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
     line_copy16(p1, src, width, 1);
 
-    for (y = 0; y < height; y++) {
+    for (y = slice_start; y < slice_end; y++) {
         src += stride * (y < height - 1 ? 1 : -1);
         line_copy16(p2, src, width, 1);
 
@@ -227,28 +251,37 @@ static void filter16_sobel(ConvolutionContext *s, AVFrame *in, AVFrame *out, int
         p2 = (p2 == end) ? orig: p2 + bstride;
         dst += out->linesize[plane] / 2;
     }
+
+    return 0;
 }
 
-static void filter_prewitt(ConvolutionContext *s, AVFrame *in, AVFrame *out, int plane)
+static int filter_prewitt(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    const uint8_t *src = in->data[plane];
-    uint8_t *dst = out->data[plane];
+    ConvolutionContext *s = ctx->priv;
+    ThreadData *td = arg;
+    AVFrame *in = td->in;
+    AVFrame *out = td->out;
+    const int plane = td->plane;
     const int stride = in->linesize[plane];
     const int bstride = s->bstride;
     const int height = s->planeheight[plane];
     const int width  = s->planewidth[plane];
+    const int slice_start = (height * jobnr) / nb_jobs;
+    const int slice_end = (height * (jobnr+1)) / nb_jobs;
+    const uint8_t *src = in->data[plane] + slice_start * stride;
+    uint8_t *dst = out->data[plane] + slice_start * out->linesize[plane];
     const float scale = s->scale;
     const float delta = s->delta;
-    uint8_t *p0 = s->buffer + 16;
+    uint8_t *p0 = s->bptrs[jobnr] + 16;
     uint8_t *p1 = p0 + bstride;
     uint8_t *p2 = p1 + bstride;
     uint8_t *orig = p0, *end = p2;
     int y, x;
 
-    line_copy8(p0, src + stride, width, 1);
+    line_copy8(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
     line_copy8(p1, src, width, 1);
 
-    for (y = 0; y < height; y++) {
+    for (y = slice_start; y < slice_end; y++) {
         src += stride * (y < height - 1 ? 1 : -1);
         line_copy8(p2, src, width, 1);
 
@@ -274,28 +307,37 @@ static void filter_prewitt(ConvolutionContext *s, AVFrame *in, AVFrame *out, int
         p2 = (p2 == end) ? orig: p2 + bstride;
         dst += out->linesize[plane];
     }
+
+    return 0;
 }
 
-static void filter_sobel(ConvolutionContext *s, AVFrame *in, AVFrame *out, int plane)
+static int filter_sobel(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    const uint8_t *src = in->data[plane];
-    uint8_t *dst = out->data[plane];
+    ConvolutionContext *s = ctx->priv;
+    ThreadData *td = arg;
+    AVFrame *in = td->in;
+    AVFrame *out = td->out;
+    const int plane = td->plane;
     const int stride = in->linesize[plane];
     const int bstride = s->bstride;
     const int height = s->planeheight[plane];
     const int width  = s->planewidth[plane];
+    const int slice_start = (height * jobnr) / nb_jobs;
+    const int slice_end = (height * (jobnr+1)) / nb_jobs;
+    const uint8_t *src = in->data[plane] + slice_start * stride;
+    uint8_t *dst = out->data[plane] + slice_start * out->linesize[plane];
     const float scale = s->scale;
     const float delta = s->delta;
-    uint8_t *p0 = s->buffer + 16;
+    uint8_t *p0 = s->bptrs[jobnr] + 16;
     uint8_t *p1 = p0 + bstride;
     uint8_t *p2 = p1 + bstride;
     uint8_t *orig = p0, *end = p2;
     int y, x;
 
-    line_copy8(p0, src + stride, width, 1);
+    line_copy8(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
     line_copy8(p1, src, width, 1);
 
-    for (y = 0; y < height; y++) {
+    for (y = slice_start; y < slice_end; y++) {
         src += stride * (y < height - 1 ? 1 : -1);
         line_copy8(p2, src, width, 1);
 
@@ -321,18 +363,27 @@ static void filter_sobel(ConvolutionContext *s, AVFrame *in, AVFrame *out, int p
         p2 = (p2 == end) ? orig: p2 + bstride;
         dst += out->linesize[plane];
     }
+
+    return 0;
 }
 
-static void filter16_3x3(ConvolutionContext *s, AVFrame *in, AVFrame *out, int plane)
+static int filter16_3x3(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    const uint16_t *src = (const uint16_t *)in->data[plane];
-    uint16_t *dst = (uint16_t *)out->data[plane];
+    ConvolutionContext *s = ctx->priv;
+    ThreadData *td = arg;
+    AVFrame *in = td->in;
+    AVFrame *out = td->out;
+    const int plane = td->plane;
     const int peak = (1 << s->depth) - 1;
     const int stride = in->linesize[plane] / 2;
     const int bstride = s->bstride;
     const int height = s->planeheight[plane];
     const int width  = s->planewidth[plane];
-    uint16_t *p0 = (uint16_t *)s->buffer + 16;
+    const int slice_start = (height * jobnr) / nb_jobs;
+    const int slice_end = (height * (jobnr+1)) / nb_jobs;
+    const uint16_t *src = (const uint16_t *)in->data[plane] + slice_start * stride;
+    uint16_t *dst = (uint16_t *)out->data[plane] + slice_start * (out->linesize[plane] / 2);
+    uint16_t *p0 = (uint16_t *)s->bptrs[jobnr] + 16;
     uint16_t *p1 = p0 + bstride;
     uint16_t *p2 = p1 + bstride;
     uint16_t *orig = p0, *end = p2;
@@ -341,10 +392,10 @@ static void filter16_3x3(ConvolutionContext *s, AVFrame *in, AVFrame *out, int p
     const float bias = s->bias[plane];
     int y, x;
 
-    line_copy16(p0, src + stride, width, 1);
+    line_copy16(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
     line_copy16(p1, src, width, 1);
 
-    for (y = 0; y < height; y++) {
+    for (y = slice_start; y < slice_end; y++) {
         src += stride * (y < height - 1 ? 1 : -1);
         line_copy16(p2, src, width, 1);
 
@@ -367,18 +418,27 @@ static void filter16_3x3(ConvolutionContext *s, AVFrame *in, AVFrame *out, int p
         p2 = (p2 == end) ? orig: p2 + bstride;
         dst += out->linesize[plane] / 2;
     }
+
+    return 0;
 }
 
-static void filter16_5x5(ConvolutionContext *s, AVFrame *in, AVFrame *out, int plane)
+static int filter16_5x5(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    const uint16_t *src = (const uint16_t *)in->data[plane];
-    uint16_t *dst = (uint16_t *)out->data[plane];
+    ConvolutionContext *s = ctx->priv;
+    ThreadData *td = arg;
+    AVFrame *in = td->in;
+    AVFrame *out = td->out;
+    const int plane = td->plane;
     const int peak = (1 << s->depth) - 1;
     const int stride = in->linesize[plane] / 2;
     const int bstride = s->bstride;
     const int height = s->planeheight[plane];
     const int width  = s->planewidth[plane];
-    uint16_t *p0 = (uint16_t *)s->buffer + 16;
+    const int slice_start = (height * jobnr) / nb_jobs;
+    const int slice_end = (height * (jobnr+1)) / nb_jobs;
+    const uint16_t *src = (const uint16_t *)in->data[plane] + slice_start * stride;
+    uint16_t *dst = (uint16_t *)out->data[plane] + slice_start * (out->linesize[plane] / 2);
+    uint16_t *p0 = (uint16_t *)s->bptrs[jobnr] + 16;
     uint16_t *p1 = p0 + bstride;
     uint16_t *p2 = p1 + bstride;
     uint16_t *p3 = p2 + bstride;
@@ -389,14 +449,13 @@ static void filter16_5x5(ConvolutionContext *s, AVFrame *in, AVFrame *out, int p
     float bias = s->bias[plane];
     int y, x, i;
 
-    line_copy16(p0, src + 2 * stride, width, 2);
-    line_copy16(p1, src + stride, width, 2);
+    line_copy16(p0, src + 2 * stride * (slice_start < 2 ? 1 : -1), width, 2);
+    line_copy16(p1, src + stride * (slice_start == 0 ? 1 : -1), width, 2);
     line_copy16(p2, src, width, 2);
     src += stride;
     line_copy16(p3, src, width, 2);
 
-
-    for (y = 0; y < height; y++) {
+    for (y = slice_start; y < slice_end; y++) {
         uint16_t *array[] = {
             p0 - 2, p0 - 1, p0, p0 + 1, p0 + 2,
             p1 - 2, p1 - 1, p1, p1 + 1, p1 + 2,
@@ -425,17 +484,26 @@ static void filter16_5x5(ConvolutionContext *s, AVFrame *in, AVFrame *out, int p
         p4 = (p4 == end) ? orig: p4 + bstride;
         dst += out->linesize[plane] / 2;
     }
+
+    return 0;
 }
 
-static void filter_3x3(ConvolutionContext *s, AVFrame *in, AVFrame *out, int plane)
+static int filter_3x3(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    const uint8_t *src = in->data[plane];
-    uint8_t *dst = out->data[plane];
+    ConvolutionContext *s = ctx->priv;
+    ThreadData *td = arg;
+    AVFrame *in = td->in;
+    AVFrame *out = td->out;
+    const int plane = td->plane;
     const int stride = in->linesize[plane];
     const int bstride = s->bstride;
     const int height = s->planeheight[plane];
     const int width  = s->planewidth[plane];
-    uint8_t *p0 = s->buffer + 16;
+    const int slice_start = (height * jobnr) / nb_jobs;
+    const int slice_end = (height * (jobnr+1)) / nb_jobs;
+    const uint8_t *src = in->data[plane] + slice_start * stride;
+    uint8_t *dst = out->data[plane] + slice_start * out->linesize[plane];
+    uint8_t *p0 = s->bptrs[jobnr] + 16;
     uint8_t *p1 = p0 + bstride;
     uint8_t *p2 = p1 + bstride;
     uint8_t *orig = p0, *end = p2;
@@ -444,10 +512,10 @@ static void filter_3x3(ConvolutionContext *s, AVFrame *in, AVFrame *out, int pla
     const float bias = s->bias[plane];
     int y, x;
 
-    line_copy8(p0, src + stride, width, 1);
+    line_copy8(p0, src + stride * (slice_start == 0 ? 1 : -1), width, 1);
     line_copy8(p1, src, width, 1);
 
-    for (y = 0; y < height; y++) {
+    for (y = slice_start; y < slice_end; y++) {
         src += stride * (y < height - 1 ? 1 : -1);
         line_copy8(p2, src, width, 1);
 
@@ -470,17 +538,26 @@ static void filter_3x3(ConvolutionContext *s, AVFrame *in, AVFrame *out, int pla
         p2 = (p2 == end) ? orig: p2 + bstride;
         dst += out->linesize[plane];
     }
+
+    return 0;
 }
 
-static void filter_5x5(ConvolutionContext *s, AVFrame *in, AVFrame *out, int plane)
+static int filter_5x5(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
 {
-    const uint8_t *src = in->data[plane];
-    uint8_t *dst = out->data[plane];
+    ConvolutionContext *s = ctx->priv;
+    ThreadData *td = arg;
+    AVFrame *in = td->in;
+    AVFrame *out = td->out;
+    const int plane = td->plane;
     const int stride = in->linesize[plane];
     const int bstride = s->bstride;
     const int height = s->planeheight[plane];
     const int width  = s->planewidth[plane];
-    uint8_t *p0 = s->buffer + 16;
+    const int slice_start = (height * jobnr) / nb_jobs;
+    const int slice_end = (height * (jobnr+1)) / nb_jobs;
+    const uint8_t *src = in->data[plane] + slice_start * stride;
+    uint8_t *dst = out->data[plane] + slice_start * out->linesize[plane];
+    uint8_t *p0 = s->bptrs[jobnr] + 16;
     uint8_t *p1 = p0 + bstride;
     uint8_t *p2 = p1 + bstride;
     uint8_t *p3 = p2 + bstride;
@@ -491,14 +568,14 @@ static void filter_5x5(ConvolutionContext *s, AVFrame *in, AVFrame *out, int pla
     float bias = s->bias[plane];
     int y, x, i;
 
-    line_copy8(p0, src + 2 * stride, width, 2);
-    line_copy8(p1, src + stride, width, 2);
+    line_copy8(p0, src + 2 * stride * (slice_start < 2 ? 1 : -1), width, 2);
+    line_copy8(p1, src + stride * (slice_start == 0 ? 1 : -1), width, 2);
     line_copy8(p2, src, width, 2);
     src += stride;
     line_copy8(p3, src, width, 2);
 
 
-    for (y = 0; y < height; y++) {
+    for (y = slice_start; y < slice_end; y++) {
         uint8_t *array[] = {
             p0 - 2, p0 - 1, p0, p0 + 1, p0 + 2,
             p1 - 2, p1 - 1, p1, p1 + 1, p1 + 2,
@@ -527,6 +604,8 @@ static void filter_5x5(ConvolutionContext *s, AVFrame *in, AVFrame *out, int pla
         p4 = (p4 == end) ? orig: p4 + bstride;
         dst += out->linesize[plane];
     }
+
+    return 0;
 }
 
 static int config_input(AVFilterLink *inlink)
@@ -534,22 +613,31 @@ static int config_input(AVFilterLink *inlink)
     AVFilterContext *ctx = inlink->dst;
     ConvolutionContext *s = ctx->priv;
     const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
-    int ret, p;
+    int p;
 
     s->depth = desc->comp[0].depth;
-    if ((ret = av_image_fill_linesizes(s->planewidth, inlink->format, inlink->w)) < 0)
-        return ret;
 
+    s->planewidth[1] = s->planewidth[2] = AV_CEIL_RSHIFT(inlink->w, desc->log2_chroma_w);
+    s->planewidth[0] = s->planewidth[3] = inlink->w;
     s->planeheight[1] = s->planeheight[2] = AV_CEIL_RSHIFT(inlink->h, desc->log2_chroma_h);
     s->planeheight[0] = s->planeheight[3] = inlink->h;
 
     s->nb_planes = av_pix_fmt_count_planes(inlink->format);
+    s->nb_threads = ff_filter_get_nb_threads(ctx);
+    s->bptrs = av_calloc(s->nb_threads, sizeof(*s->bptrs));
+    if (!s->bptrs)
+        return AVERROR(ENOMEM);
 
     s->bstride = s->planewidth[0] + 32;
-    s->buffer = av_malloc_array(5 * s->bstride, (s->depth + 7) / 8);
+    s->bpc = (s->depth + 7) / 8;
+    s->buffer = av_malloc_array(5 * s->bstride * s->nb_threads, s->bpc);
     if (!s->buffer)
         return AVERROR(ENOMEM);
 
+    for (p = 0; p < s->nb_threads; p++) {
+        s->bptrs[p] = s->buffer + 5 * s->bstride * s->bpc * p;
+    }
+
     if (!strcmp(ctx->filter->name, "convolution")) {
         if (s->depth > 8) {
             for (p = 0; p < s->nb_planes; p++) {
@@ -574,8 +662,9 @@ static int config_input(AVFilterLink *inlink)
 
 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
 {
-    ConvolutionContext *s = inlink->dst->priv;
-    AVFilterLink *outlink = inlink->dst->outputs[0];
+    AVFilterContext *ctx = inlink->dst;
+    ConvolutionContext *s = ctx->priv;
+    AVFilterLink *outlink = ctx->outputs[0];
     AVFrame *out;
     int plane;
 
@@ -587,15 +676,20 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
     av_frame_copy_props(out, in);
 
     for (plane = 0; plane < s->nb_planes; plane++) {
+        ThreadData td;
+
         if (s->copy[plane]) {
             av_image_copy_plane(out->data[plane], out->linesize[plane],
                                 in->data[plane], in->linesize[plane],
-                                s->planewidth[plane],
+                                s->planewidth[plane] * s->bpc,
                                 s->planeheight[plane]);
             continue;
         }
 
-        s->filter[plane](s, in, out, plane);
+        td.in = in;
+        td.out = out;
+        td.plane = plane;
+        ctx->internal->execute(ctx, s->filter[plane], &td, NULL, FFMIN(s->planeheight[plane], s->nb_threads));
     }
 
     av_frame_free(&in);
@@ -661,6 +755,7 @@ static av_cold void uninit(AVFilterContext *ctx)
 {
     ConvolutionContext *s = ctx->priv;
 
+    av_freep(&s->bptrs);
     av_freep(&s->buffer);
 }
 
@@ -694,7 +789,7 @@ AVFilter ff_vf_convolution = {
     .query_formats = query_formats,
     .inputs        = convolution_inputs,
     .outputs       = convolution_outputs,
-    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
+    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
 };
 
 #endif /* CONFIG_CONVOLUTION_FILTER */
@@ -720,7 +815,7 @@ AVFilter ff_vf_prewitt = {
     .query_formats = query_formats,
     .inputs        = convolution_inputs,
     .outputs       = convolution_outputs,
-    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
+    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
 };
 
 #endif /* CONFIG_PREWITT_FILTER */
@@ -746,7 +841,7 @@ AVFilter ff_vf_sobel = {
     .query_formats = query_formats,
     .inputs        = convolution_inputs,
     .outputs       = convolution_outputs,
-    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
+    .flags         = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
 };
 
 #endif /* CONFIG_SOBEL_FILTER */