[FFmpeg-cvslog] avfilter/vf_colorbalance: add slice threading
Paul B Mahol
git at videolan.org
Sat May 5 13:52:26 EEST 2018
ffmpeg | branch: master | Paul B Mahol <onemda at gmail.com> | Sat May 5 12:50:16 2018 +0200| [0ef7a4519745967473f69fd8819af9d558132d4b] | committer: Paul B Mahol
avfilter/vf_colorbalance: add slice threading
Signed-off-by: Paul B Mahol <onemda at gmail.com>
> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=0ef7a4519745967473f69fd8819af9d558132d4b
---
libavfilter/vf_colorbalance.c | 98 +++++++++++++++++++++++++++++--------------
1 file changed, 66 insertions(+), 32 deletions(-)
diff --git a/libavfilter/vf_colorbalance.c b/libavfilter/vf_colorbalance.c
index d502080d45..4e7476ae2f 100644
--- a/libavfilter/vf_colorbalance.c
+++ b/libavfilter/vf_colorbalance.c
@@ -31,6 +31,10 @@
#define B 2
#define A 3
+typedef struct ThreadData {
+ AVFrame *in, *out;
+} ThreadData;
+
typedef struct Range {
double shadows;
double midtones;
@@ -48,8 +52,7 @@ typedef struct ColorBalanceContext {
uint8_t rgba_map[4];
int step;
- int (*clip)(int x);
- void (*apply_lut)(AVFilterContext *ctx, AVFrame *in, AVFrame *out);
+ int (*apply_lut)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
} ColorBalanceContext;
#define OFFSET(x) offsetof(ColorBalanceContext, x)
@@ -93,20 +96,25 @@ static int query_formats(AVFilterContext *ctx)
return ff_set_common_formats(ctx, fmts_list);
}
-static void apply_lut8_p(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
+static int apply_lut8_p(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
{
ColorBalanceContext *s = ctx->priv;
- const uint8_t *srcg = in->data[0];
- const uint8_t *srcb = in->data[1];
- const uint8_t *srcr = in->data[2];
- const uint8_t *srca = in->data[3];
- uint8_t *dstg = out->data[0];
- uint8_t *dstb = out->data[1];
- uint8_t *dstr = out->data[2];
- uint8_t *dsta = out->data[3];
+ ThreadData *td = arg;
+ AVFrame *in = td->in;
+ AVFrame *out = td->out;
+ const int slice_start = (out->height * jobnr) / nb_jobs;
+ const int slice_end = (out->height * (jobnr+1)) / nb_jobs;
+ const uint8_t *srcg = in->data[0] + slice_start * in->linesize[0];
+ const uint8_t *srcb = in->data[1] + slice_start * in->linesize[1];
+ const uint8_t *srcr = in->data[2] + slice_start * in->linesize[2];
+ const uint8_t *srca = in->data[3] + slice_start * in->linesize[3];
+ uint8_t *dstg = out->data[0] + slice_start * out->linesize[0];
+ uint8_t *dstb = out->data[1] + slice_start * out->linesize[1];
+ uint8_t *dstr = out->data[2] + slice_start * out->linesize[2];
+ uint8_t *dsta = out->data[3] + slice_start * out->linesize[3];
int i, j;
- for (i = 0; i < out->height; i++) {
+ for (i = slice_start; i < slice_end; i++) {
for (j = 0; j < out->width; j++) {
dstg[j] = s->lut[G][srcg[j]];
dstb[j] = s->lut[B][srcb[j]];
@@ -124,22 +132,29 @@ static void apply_lut8_p(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
dstr += out->linesize[2];
dsta += out->linesize[3];
}
+
+ return 0;
}
-static void apply_lut16_p(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
+static int apply_lut16_p(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
{
ColorBalanceContext *s = ctx->priv;
- const uint16_t *srcg = (const uint16_t *)in->data[0];
- const uint16_t *srcb = (const uint16_t *)in->data[1];
- const uint16_t *srcr = (const uint16_t *)in->data[2];
- const uint16_t *srca = (const uint16_t *)in->data[3];
- uint16_t *dstg = (uint16_t *)out->data[0];
- uint16_t *dstb = (uint16_t *)out->data[1];
- uint16_t *dstr = (uint16_t *)out->data[2];
- uint16_t *dsta = (uint16_t *)out->data[3];
+ ThreadData *td = arg;
+ AVFrame *in = td->in;
+ AVFrame *out = td->out;
+ const int slice_start = (out->height * jobnr) / nb_jobs;
+ const int slice_end = (out->height * (jobnr+1)) / nb_jobs;
+ const uint16_t *srcg = (const uint16_t *)in->data[0] + slice_start * in->linesize[0] / 2;
+ const uint16_t *srcb = (const uint16_t *)in->data[1] + slice_start * in->linesize[1] / 2;
+ const uint16_t *srcr = (const uint16_t *)in->data[2] + slice_start * in->linesize[2] / 2;
+ const uint16_t *srca = (const uint16_t *)in->data[3] + slice_start * in->linesize[3] / 2;
+ uint16_t *dstg = (uint16_t *)out->data[0] + slice_start * out->linesize[0] / 2;
+ uint16_t *dstb = (uint16_t *)out->data[1] + slice_start * out->linesize[1] / 2;
+ uint16_t *dstr = (uint16_t *)out->data[2] + slice_start * out->linesize[2] / 2;
+ uint16_t *dsta = (uint16_t *)out->data[3] + slice_start * out->linesize[3] / 2;
int i, j;
- for (i = 0; i < out->height; i++) {
+ for (i = slice_start; i < slice_end; i++) {
for (j = 0; j < out->width; j++) {
dstg[j] = s->lut[G][srcg[j]];
dstb[j] = s->lut[B][srcb[j]];
@@ -157,13 +172,20 @@ static void apply_lut16_p(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
dstr += out->linesize[2] / 2;
dsta += out->linesize[3] / 2;
}
+
+ return 0;
}
-static void apply_lut8(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
+static int apply_lut8(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
{
ColorBalanceContext *s = ctx->priv;
+ ThreadData *td = arg;
+ AVFrame *in = td->in;
+ AVFrame *out = td->out;
AVFilterLink *outlink = ctx->outputs[0];
- const uint8_t *srcrow = in->data[0];
+ const int slice_start = (out->height * jobnr) / nb_jobs;
+ const int slice_end = (out->height * (jobnr+1)) / nb_jobs;
+ const uint8_t *srcrow = in->data[0] + slice_start * in->linesize[0];
const uint8_t roffset = s->rgba_map[R];
const uint8_t goffset = s->rgba_map[G];
const uint8_t boffset = s->rgba_map[B];
@@ -172,8 +194,8 @@ static void apply_lut8(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
uint8_t *dstrow;
int i, j;
- dstrow = out->data[0];
- for (i = 0; i < outlink->h; i++) {
+ dstrow = out->data[0] + slice_start * out->linesize[0];
+ for (i = slice_start; i < slice_end; i++) {
const uint8_t *src = srcrow;
uint8_t *dst = dstrow;
@@ -188,13 +210,20 @@ static void apply_lut8(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
srcrow += in->linesize[0];
dstrow += out->linesize[0];
}
+
+ return 0;
}
-static void apply_lut16(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
+static int apply_lut16(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
{
ColorBalanceContext *s = ctx->priv;
+ ThreadData *td = arg;
+ AVFrame *in = td->in;
+ AVFrame *out = td->out;
AVFilterLink *outlink = ctx->outputs[0];
- const uint16_t *srcrow = (const uint16_t *)in->data[0];
+ const int slice_start = (out->height * jobnr) / nb_jobs;
+ const int slice_end = (out->height * (jobnr+1)) / nb_jobs;
+ const uint16_t *srcrow = (const uint16_t *)in->data[0] + slice_start * in->linesize[0] / 2;
const uint8_t roffset = s->rgba_map[R];
const uint8_t goffset = s->rgba_map[G];
const uint8_t boffset = s->rgba_map[B];
@@ -203,8 +232,8 @@ static void apply_lut16(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
uint16_t *dstrow;
int i, j;
- dstrow = (uint16_t *)out->data[0];
- for (i = 0; i < outlink->h; i++) {
+ dstrow = (uint16_t *)out->data[0] + slice_start * out->linesize[0] / 2;
+ for (i = slice_start; i < slice_end; i++) {
const uint16_t *src = srcrow;
uint16_t *dst = dstrow;
@@ -219,6 +248,8 @@ static void apply_lut16(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
srcrow += in->linesize[0] / 2;
dstrow += out->linesize[0] / 2;
}
+
+ return 0;
}
static int config_output(AVFilterLink *outlink)
@@ -296,6 +327,7 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
AVFilterContext *ctx = inlink->dst;
ColorBalanceContext *s = ctx->priv;
AVFilterLink *outlink = ctx->outputs[0];
+ ThreadData td;
AVFrame *out;
if (av_frame_is_writable(in)) {
@@ -309,7 +341,9 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
av_frame_copy_props(out, in);
}
- s->apply_lut(ctx, in, out);
+ td.in = in;
+ td.out = out;
+ ctx->internal->execute(ctx, s->apply_lut, &td, NULL, FFMIN(outlink->h, ff_filter_get_nb_threads(ctx)));
if (in != out)
av_frame_free(&in);
@@ -342,5 +376,5 @@ AVFilter ff_vf_colorbalance = {
.query_formats = query_formats,
.inputs = colorbalance_inputs,
.outputs = colorbalance_outputs,
- .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
+ .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
};
More information about the ffmpeg-cvslog
mailing list