mirror of
https://github.com/xenia-project/FFmpeg.git
synced 2025-02-26 07:05:59 +00:00
avfilter/vf_w3fdif: implement slice threading
Signed-off-by: Paul B Mahol <onemda@gmail.com>
This commit is contained in:
parent
7b07e72abc
commit
885afff01c
@ -40,7 +40,8 @@ typedef struct W3FDIFContext {
|
|||||||
int eof;
|
int eof;
|
||||||
int nb_planes;
|
int nb_planes;
|
||||||
AVFrame *prev, *cur, *next; ///< previous, current, next frames
|
AVFrame *prev, *cur, *next; ///< previous, current, next frames
|
||||||
int32_t *work_line; ///< line we are calculating
|
int32_t **work_line; ///< lines we are calculating
|
||||||
|
int nb_threads;
|
||||||
} W3FDIFContext;
|
} W3FDIFContext;
|
||||||
|
|
||||||
#define OFFSET(x) offsetof(W3FDIFContext, x)
|
#define OFFSET(x) offsetof(W3FDIFContext, x)
|
||||||
@ -82,9 +83,10 @@ static int query_formats(AVFilterContext *ctx)
|
|||||||
|
|
||||||
static int config_input(AVFilterLink *inlink)
|
static int config_input(AVFilterLink *inlink)
|
||||||
{
|
{
|
||||||
W3FDIFContext *s = inlink->dst->priv;
|
AVFilterContext *ctx = inlink->dst;
|
||||||
|
W3FDIFContext *s = ctx->priv;
|
||||||
const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
|
const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
|
||||||
int ret;
|
int ret, i;
|
||||||
|
|
||||||
if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
|
if ((ret = av_image_fill_linesizes(s->linesize, inlink->format, inlink->w)) < 0)
|
||||||
return ret;
|
return ret;
|
||||||
@ -93,10 +95,17 @@ static int config_input(AVFilterLink *inlink)
|
|||||||
s->planeheight[0] = s->planeheight[3] = inlink->h;
|
s->planeheight[0] = s->planeheight[3] = inlink->h;
|
||||||
|
|
||||||
s->nb_planes = av_pix_fmt_count_planes(inlink->format);
|
s->nb_planes = av_pix_fmt_count_planes(inlink->format);
|
||||||
s->work_line = av_calloc(s->linesize[0], sizeof(*s->work_line));
|
s->nb_threads = ctx->graph->nb_threads;
|
||||||
|
s->work_line = av_calloc(s->nb_threads, sizeof(*s->work_line));
|
||||||
if (!s->work_line)
|
if (!s->work_line)
|
||||||
return AVERROR(ENOMEM);
|
return AVERROR(ENOMEM);
|
||||||
|
|
||||||
|
for (i = 0; i < s->nb_threads; i++) {
|
||||||
|
s->work_line[i] = av_calloc(s->linesize[0], sizeof(*s->work_line[0]));
|
||||||
|
if (!s->work_line[i])
|
||||||
|
return AVERROR(ENOMEM);
|
||||||
|
}
|
||||||
|
|
||||||
return 0;
|
return 0;
|
||||||
}
|
}
|
||||||
|
|
||||||
@ -128,11 +137,20 @@ static const int8_t n_coef_hf[2] = { 3, 5 };
|
|||||||
static const int32_t coef_hf[2][5] = {{ -4096, 8192, -4096, 0, 0},
|
static const int32_t coef_hf[2][5] = {{ -4096, 8192, -4096, 0, 0},
|
||||||
{ 2032, -7602, 11140, -7602, 2032}};
|
{ 2032, -7602, 11140, -7602, 2032}};
|
||||||
|
|
||||||
static void deinterlace_plane(AVFilterContext *ctx, AVFrame *out,
|
typedef struct ThreadData {
|
||||||
const AVFrame *cur, const AVFrame *adj,
|
AVFrame *out, *cur, *adj;
|
||||||
const int filter, const int plane)
|
int plane;
|
||||||
|
} ThreadData;
|
||||||
|
|
||||||
|
static int deinterlace_slice(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
|
||||||
{
|
{
|
||||||
W3FDIFContext *s = ctx->priv;
|
W3FDIFContext *s = ctx->priv;
|
||||||
|
ThreadData *td = arg;
|
||||||
|
AVFrame *out = td->out;
|
||||||
|
AVFrame *cur = td->cur;
|
||||||
|
AVFrame *adj = td->adj;
|
||||||
|
const int plane = td->plane;
|
||||||
|
const int filter = s->filter;
|
||||||
uint8_t *in_line, *in_lines_cur[5], *in_lines_adj[5];
|
uint8_t *in_line, *in_lines_cur[5], *in_lines_adj[5];
|
||||||
uint8_t *out_line, *out_pixel;
|
uint8_t *out_line, *out_pixel;
|
||||||
int32_t *work_line, *work_pixel;
|
int32_t *work_line, *work_pixel;
|
||||||
@ -144,15 +162,17 @@ static void deinterlace_plane(AVFilterContext *ctx, AVFrame *out,
|
|||||||
const int cur_line_stride = cur->linesize[plane];
|
const int cur_line_stride = cur->linesize[plane];
|
||||||
const int adj_line_stride = adj->linesize[plane];
|
const int adj_line_stride = adj->linesize[plane];
|
||||||
const int dst_line_stride = out->linesize[plane];
|
const int dst_line_stride = out->linesize[plane];
|
||||||
|
const int start = (height * jobnr) / nb_jobs;
|
||||||
|
const int end = (height * (jobnr+1)) / nb_jobs;
|
||||||
int i, j, y_in, y_out;
|
int i, j, y_in, y_out;
|
||||||
|
|
||||||
/* copy unchanged the lines of the field */
|
/* copy unchanged the lines of the field */
|
||||||
y_out = s->field == cur->top_field_first;
|
y_out = start + (s->field == cur->top_field_first) - (start & 1);
|
||||||
|
|
||||||
in_line = cur_data + (y_out * cur_line_stride);
|
in_line = cur_data + (y_out * cur_line_stride);
|
||||||
out_line = dst_data + (y_out * dst_line_stride);
|
out_line = dst_data + (y_out * dst_line_stride);
|
||||||
|
|
||||||
while (y_out < height) {
|
while (y_out < end) {
|
||||||
memcpy(out_line, in_line, linesize);
|
memcpy(out_line, in_line, linesize);
|
||||||
y_out += 2;
|
y_out += 2;
|
||||||
in_line += cur_line_stride * 2;
|
in_line += cur_line_stride * 2;
|
||||||
@ -160,13 +180,13 @@ static void deinterlace_plane(AVFilterContext *ctx, AVFrame *out,
|
|||||||
}
|
}
|
||||||
|
|
||||||
/* interpolate other lines of the field */
|
/* interpolate other lines of the field */
|
||||||
y_out = s->field != cur->top_field_first;
|
y_out = start + (s->field != cur->top_field_first) - (start & 1);
|
||||||
|
|
||||||
out_line = dst_data + (y_out * dst_line_stride);
|
out_line = dst_data + (y_out * dst_line_stride);
|
||||||
|
|
||||||
while (y_out < height) {
|
while (y_out < end) {
|
||||||
/* clear workspace */
|
/* clear workspace */
|
||||||
memset(s->work_line, 0, sizeof(*s->work_line) * linesize);
|
memset(s->work_line[jobnr], 0, sizeof(*s->work_line[jobnr]) * linesize);
|
||||||
|
|
||||||
/* get low vertical frequencies from current field */
|
/* get low vertical frequencies from current field */
|
||||||
for (j = 0; j < n_coef_lf[filter]; j++) {
|
for (j = 0; j < n_coef_lf[filter]; j++) {
|
||||||
@ -180,7 +200,7 @@ static void deinterlace_plane(AVFilterContext *ctx, AVFrame *out,
|
|||||||
in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
|
in_lines_cur[j] = cur_data + (y_in * cur_line_stride);
|
||||||
}
|
}
|
||||||
|
|
||||||
work_line = s->work_line;
|
work_line = s->work_line[jobnr];
|
||||||
switch (n_coef_lf[filter]) {
|
switch (n_coef_lf[filter]) {
|
||||||
case 2:
|
case 2:
|
||||||
for (i = 0; i < linesize; i++) {
|
for (i = 0; i < linesize; i++) {
|
||||||
@ -210,7 +230,7 @@ static void deinterlace_plane(AVFilterContext *ctx, AVFrame *out,
|
|||||||
in_lines_adj[j] = adj_data + (y_in * adj_line_stride);
|
in_lines_adj[j] = adj_data + (y_in * adj_line_stride);
|
||||||
}
|
}
|
||||||
|
|
||||||
work_line = s->work_line;
|
work_line = s->work_line[jobnr];
|
||||||
switch (n_coef_hf[filter]) {
|
switch (n_coef_hf[filter]) {
|
||||||
case 3:
|
case 3:
|
||||||
for (i = 0; i < linesize; i++) {
|
for (i = 0; i < linesize; i++) {
|
||||||
@ -238,7 +258,7 @@ static void deinterlace_plane(AVFilterContext *ctx, AVFrame *out,
|
|||||||
}
|
}
|
||||||
|
|
||||||
/* save scaled result to the output frame, scaling down by 256 * 256 */
|
/* save scaled result to the output frame, scaling down by 256 * 256 */
|
||||||
work_pixel = s->work_line;
|
work_pixel = s->work_line[jobnr];
|
||||||
out_pixel = out_line;
|
out_pixel = out_line;
|
||||||
|
|
||||||
for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
|
for (j = 0; j < linesize; j++, out_pixel++, work_pixel++)
|
||||||
@ -248,6 +268,8 @@ static void deinterlace_plane(AVFilterContext *ctx, AVFrame *out,
|
|||||||
y_out += 2;
|
y_out += 2;
|
||||||
out_line += dst_line_stride * 2;
|
out_line += dst_line_stride * 2;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
return 0;
|
||||||
}
|
}
|
||||||
|
|
||||||
static int filter(AVFilterContext *ctx, int is_second)
|
static int filter(AVFilterContext *ctx, int is_second)
|
||||||
@ -255,6 +277,7 @@ static int filter(AVFilterContext *ctx, int is_second)
|
|||||||
W3FDIFContext *s = ctx->priv;
|
W3FDIFContext *s = ctx->priv;
|
||||||
AVFilterLink *outlink = ctx->outputs[0];
|
AVFilterLink *outlink = ctx->outputs[0];
|
||||||
AVFrame *out, *adj;
|
AVFrame *out, *adj;
|
||||||
|
ThreadData td;
|
||||||
int plane;
|
int plane;
|
||||||
|
|
||||||
out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
|
out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
|
||||||
@ -278,8 +301,11 @@ static int filter(AVFilterContext *ctx, int is_second)
|
|||||||
}
|
}
|
||||||
|
|
||||||
adj = s->field ? s->next : s->prev;
|
adj = s->field ? s->next : s->prev;
|
||||||
for (plane = 0; plane < s->nb_planes; plane++)
|
td.out = out; td.cur = s->cur; td.adj = adj;
|
||||||
deinterlace_plane(ctx, out, s->cur, adj, s->filter, plane);
|
for (plane = 0; plane < s->nb_planes; plane++) {
|
||||||
|
td.plane = plane;
|
||||||
|
ctx->internal->execute(ctx, deinterlace_slice, &td, NULL, FFMIN(s->planeheight[plane], s->nb_threads));
|
||||||
|
}
|
||||||
|
|
||||||
s->field = !s->field;
|
s->field = !s->field;
|
||||||
|
|
||||||
@ -355,10 +381,15 @@ static int request_frame(AVFilterLink *outlink)
|
|||||||
static av_cold void uninit(AVFilterContext *ctx)
|
static av_cold void uninit(AVFilterContext *ctx)
|
||||||
{
|
{
|
||||||
W3FDIFContext *s = ctx->priv;
|
W3FDIFContext *s = ctx->priv;
|
||||||
|
int i;
|
||||||
|
|
||||||
av_frame_free(&s->prev);
|
av_frame_free(&s->prev);
|
||||||
av_frame_free(&s->cur );
|
av_frame_free(&s->cur );
|
||||||
av_frame_free(&s->next);
|
av_frame_free(&s->next);
|
||||||
|
|
||||||
|
for (i = 0; i < s->nb_threads; i++)
|
||||||
|
av_freep(&s->work_line[i]);
|
||||||
|
|
||||||
av_freep(&s->work_line);
|
av_freep(&s->work_line);
|
||||||
}
|
}
|
||||||
|
|
||||||
@ -391,5 +422,5 @@ AVFilter ff_vf_w3fdif = {
|
|||||||
.query_formats = query_formats,
|
.query_formats = query_formats,
|
||||||
.inputs = w3fdif_inputs,
|
.inputs = w3fdif_inputs,
|
||||||
.outputs = w3fdif_outputs,
|
.outputs = w3fdif_outputs,
|
||||||
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
|
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
|
||||||
};
|
};
|
||||||
|
Loading…
x
Reference in New Issue
Block a user