#include "libavutil/pixdesc.h"
#include "unsharp.h"
-static void apply_unsharp( uint8_t *dst, int dst_stride,
- const uint8_t *src, int src_stride,
- int width, int height, UnsharpFilterParam *fp)
-{
- uint32_t **sc = fp->sc;
- uint32_t sr[MAX_MATRIX_SIZE - 1], tmp1, tmp2;
-
- int32_t res;
- int x, y, z;
- const uint8_t *src2 = NULL; //silence a warning
- const int amount = fp->amount;
- const int steps_x = fp->steps_x;
- const int steps_y = fp->steps_y;
- const int scalebits = fp->scalebits;
- const int32_t halfscale = fp->halfscale;
-
- if (!amount) {
- av_image_copy_plane(dst, dst_stride, src, src_stride, width, height);
- return;
- }
-
- for (y = 0; y < 2 * steps_y; y++)
- memset(sc[y], 0, sizeof(sc[y][0]) * (width + 2 * steps_x));
-
- for (y = -steps_y; y < height + steps_y; y++) {
- if (y < height)
- src2 = src;
-
- memset(sr, 0, sizeof(sr[0]) * (2 * steps_x - 1));
- for (x = -steps_x; x < width + steps_x; x++) {
- tmp1 = x <= 0 ? src2[0] : x >= width ? src2[width-1] : src2[x];
- for (z = 0; z < steps_x * 2; z += 2) {
- tmp2 = sr[z + 0] + tmp1; sr[z + 0] = tmp1;
- tmp1 = sr[z + 1] + tmp2; sr[z + 1] = tmp2;
- }
- for (z = 0; z < steps_y * 2; z += 2) {
- tmp2 = sc[z + 0][x + steps_x] + tmp1; sc[z + 0][x + steps_x] = tmp1;
- tmp1 = sc[z + 1][x + steps_x] + tmp2; sc[z + 1][x + steps_x] = tmp2;
- }
- if (x >= steps_x && y >= steps_y) {
- const uint8_t *srx = src - steps_y * src_stride + x - steps_x;
- uint8_t *dsx = dst - steps_y * dst_stride + x - steps_x;
-
- res = (int32_t)*srx + ((((int32_t) * srx - (int32_t)((tmp1 + halfscale) >> scalebits)) * amount) >> 16);
- *dsx = av_clip_uint8(res);
- }
- }
- if (y >= 0) {
- dst += dst_stride;
- src += src_stride;
- }
- }
+typedef struct TheadData {
+ UnsharpFilterParam *fp;
+ uint8_t *dst;
+ const uint8_t *src;
+ int dst_stride;
+ int src_stride;
+ int width;
+ int height;
+} ThreadData;
+
+#define DEF_UNSHARP_SLICE_FUNC(name, nbits) \
+static int name##_##nbits(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \
+{ \
+ ThreadData *td = arg; \
+ UnsharpFilterParam *fp = td->fp; \
+ UnsharpContext *s = ctx->priv; \
+ uint32_t **sc = fp->sc; \
+ uint32_t *sr = fp->sr; \
+ const uint##nbits##_t *src2 = NULL; \
+ const int amount = fp->amount; \
+ const int steps_x = fp->steps_x; \
+ const int steps_y = fp->steps_y; \
+ const int scalebits = fp->scalebits; \
+ const int32_t halfscale = fp->halfscale; \
+ \
+ uint##nbits##_t *dst = (uint##nbits##_t*)td->dst; \
+ const uint##nbits##_t *src = (const uint##nbits##_t *)td->src; \
+ int dst_stride = td->dst_stride; \
+ int src_stride = td->src_stride; \
+ const int width = td->width; \
+ const int height = td->height; \
+ const int sc_offset = jobnr * 2 * steps_y; \
+ const int sr_offset = jobnr * (MAX_MATRIX_SIZE - 1); \
+ const int slice_start = (height * jobnr) / nb_jobs; \
+ const int slice_end = (height * (jobnr+1)) / nb_jobs; \
+ \
+ int32_t res; \
+ int x, y, z; \
+ uint32_t tmp1, tmp2; \
+ \
+ if (!amount) { \
+ av_image_copy_plane(td->dst + slice_start * dst_stride, dst_stride, \
+ td->src + slice_start * src_stride, src_stride, \
+ width * s->bps, slice_end - slice_start); \
+ return 0; \
+ } \
+ \
+ for (y = 0; y < 2 * steps_y; y++) \
+ memset(sc[sc_offset + y], 0, sizeof(sc[y][0]) * (width + 2 * steps_x)); \
+ \
+ dst_stride = dst_stride / s->bps; \
+ src_stride = src_stride / s->bps; \
+ /* if this is not the first tile, we start from (slice_start - steps_y) */ \
+ /* so we can get smooth result at slice boundary */ \
+ if (slice_start > steps_y) { \
+ src += (slice_start - steps_y) * src_stride; \
+ dst += (slice_start - steps_y) * dst_stride; \
+ } \
+ \
+ for (y = -steps_y + slice_start; y < steps_y + slice_end; y++) { \
+ if (y < height) \
+ src2 = src; \
+ \
+ memset(sr + sr_offset, 0, sizeof(sr[0]) * (2 * steps_x - 1)); \
+ for (x = -steps_x; x < width + steps_x; x++) { \
+ tmp1 = x <= 0 ? src2[0] : x >= width ? src2[width-1] : src2[x]; \
+ for (z = 0; z < steps_x * 2; z += 2) { \
+ tmp2 = sr[sr_offset + z + 0] + tmp1; sr[sr_offset + z + 0] = tmp1; \
+ tmp1 = sr[sr_offset + z + 1] + tmp2; sr[sr_offset + z + 1] = tmp2; \
+ } \
+ for (z = 0; z < steps_y * 2; z += 2) { \
+ tmp2 = sc[sc_offset + z + 0][x + steps_x] + tmp1; \
+ sc[sc_offset + z + 0][x + steps_x] = tmp1; \
+ tmp1 = sc[sc_offset + z + 1][x + steps_x] + tmp2; \
+ sc[sc_offset + z + 1][x + steps_x] = tmp2; \
+ } \
+ if (x >= steps_x && y >= (steps_y + slice_start)) { \
+ const uint##nbits##_t *srx = src - steps_y * src_stride + x - steps_x; \
+ uint##nbits##_t *dsx = dst - steps_y * dst_stride + x - steps_x; \
+ \
+ res = (int32_t)*srx + ((((int32_t) * srx - \
+ (int32_t)((tmp1 + halfscale) >> scalebits)) * amount) >> (8+nbits)); \
+ *dsx = av_clip_uint##nbits(res); \
+ } \
+ } \
+ if (y >= 0) { \
+ dst += dst_stride; \
+ src += src_stride; \
+ } \
+ } \
+ return 0; \
}
+DEF_UNSHARP_SLICE_FUNC(unsharp_slice, 16)
+DEF_UNSHARP_SLICE_FUNC(unsharp_slice, 8)
static int apply_unsharp_c(AVFilterContext *ctx, AVFrame *in, AVFrame *out)
{
UnsharpContext *s = ctx->priv;
int i, plane_w[3], plane_h[3];
UnsharpFilterParam *fp[3];
+ ThreadData td;
+
plane_w[0] = inlink->w;
plane_w[1] = plane_w[2] = AV_CEIL_RSHIFT(inlink->w, s->hsub);
plane_h[0] = inlink->h;
fp[0] = &s->luma;
fp[1] = fp[2] = &s->chroma;
for (i = 0; i < 3; i++) {
- apply_unsharp(out->data[i], out->linesize[i], in->data[i], in->linesize[i], plane_w[i], plane_h[i], fp[i]);
+ td.fp = fp[i];
+ td.dst = out->data[i];
+ td.src = in->data[i];
+ td.width = plane_w[i];
+ td.height = plane_h[i];
+ td.dst_stride = out->linesize[i];
+ td.src_stride = in->linesize[i];
+ ctx->internal->execute(ctx, s->unsharp_slice, &td, NULL, FFMIN(plane_h[i], s->nb_threads));
}
return 0;
}
static const enum AVPixelFormat pix_fmts[] = {
AV_PIX_FMT_YUV420P, AV_PIX_FMT_YUV422P, AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUV410P,
AV_PIX_FMT_YUV411P, AV_PIX_FMT_YUV440P, AV_PIX_FMT_YUVJ420P, AV_PIX_FMT_YUVJ422P,
+ AV_PIX_FMT_YUV420P9, AV_PIX_FMT_YUV422P9, AV_PIX_FMT_YUV444P9,
+ AV_PIX_FMT_YUV420P10, AV_PIX_FMT_YUV422P10, AV_PIX_FMT_YUV444P10, AV_PIX_FMT_YUV440P10,
+ AV_PIX_FMT_YUV420P12, AV_PIX_FMT_YUV422P12, AV_PIX_FMT_YUV444P12, AV_PIX_FMT_YUV440P12,
+ AV_PIX_FMT_YUV420P16, AV_PIX_FMT_YUV422P16, AV_PIX_FMT_YUV444P16,
AV_PIX_FMT_YUVJ444P, AV_PIX_FMT_YUVJ440P, AV_PIX_FMT_NONE
};
static int init_filter_param(AVFilterContext *ctx, UnsharpFilterParam *fp, const char *effect_type, int width)
{
int z;
+ UnsharpContext *s = ctx->priv;
const char *effect = fp->amount == 0 ? "none" : fp->amount < 0 ? "blur" : "sharpen";
if (!(fp->msize_x & fp->msize_y & 1)) {
av_log(ctx, AV_LOG_VERBOSE, "effect:%s type:%s msize_x:%d msize_y:%d amount:%0.2f\n",
effect, effect_type, fp->msize_x, fp->msize_y, fp->amount / 65535.0);
- for (z = 0; z < 2 * fp->steps_y; z++)
+ fp->sr = av_malloc_array((MAX_MATRIX_SIZE - 1) * s->nb_threads, sizeof(uint32_t));
+ fp->sc = av_mallocz_array(2 * fp->steps_y * s->nb_threads, sizeof(uint32_t *));
+ if (!fp->sr || !fp->sc)
+ return AVERROR(ENOMEM);
+
+ for (z = 0; z < 2 * fp->steps_y * s->nb_threads; z++)
if (!(fp->sc[z] = av_malloc_array(width + 2 * fp->steps_x,
sizeof(*(fp->sc[z])))))
return AVERROR(ENOMEM);
return 0;
}
-static int config_props(AVFilterLink *link)
+static int config_input(AVFilterLink *inlink)
{
- UnsharpContext *s = link->dst->priv;
- const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(link->format);
+ UnsharpContext *s = inlink->dst->priv;
+ const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format);
int ret;
s->hsub = desc->log2_chroma_w;
s->vsub = desc->log2_chroma_h;
+ s->bitdepth = desc->comp[0].depth;
+ s->bps = s->bitdepth > 8 ? 2 : 1;
+ s->unsharp_slice = s->bitdepth > 8 ? unsharp_slice_16 : unsharp_slice_8;
- ret = init_filter_param(link->dst, &s->luma, "luma", link->w);
+ // ensure (height / nb_threads) > 4 * steps_y,
+ // so that we don't have too much overlap between two threads
+ s->nb_threads = FFMIN(ff_filter_get_nb_threads(inlink->dst),
+ inlink->h / (4 * s->luma.steps_y));
+
+ ret = init_filter_param(inlink->dst, &s->luma, "luma", inlink->w);
if (ret < 0)
return ret;
- ret = init_filter_param(link->dst, &s->chroma, "chroma", AV_CEIL_RSHIFT(link->w, s->hsub));
+ ret = init_filter_param(inlink->dst, &s->chroma, "chroma", AV_CEIL_RSHIFT(inlink->w, s->hsub));
if (ret < 0)
return ret;
return 0;
}
-static void free_filter_param(UnsharpFilterParam *fp)
+static void free_filter_param(UnsharpFilterParam *fp, int nb_threads)
{
int z;
- for (z = 0; z < 2 * fp->steps_y; z++)
- av_freep(&fp->sc[z]);
+ if (fp->sc) {
+ for (z = 0; z < 2 * fp->steps_y * nb_threads; z++)
+ av_freep(&fp->sc[z]);
+ av_freep(&fp->sc);
+ }
+ av_freep(&fp->sr);
}
static av_cold void uninit(AVFilterContext *ctx)
{
UnsharpContext *s = ctx->priv;
- free_filter_param(&s->luma);
- free_filter_param(&s->chroma);
+ free_filter_param(&s->luma, s->nb_threads);
+ free_filter_param(&s->chroma, s->nb_threads);
}
static int filter_frame(AVFilterLink *link, AVFrame *in)
.name = "default",
.type = AVMEDIA_TYPE_VIDEO,
.filter_frame = filter_frame,
- .config_props = config_props,
+ .config_props = config_input,
},
{ NULL }
};
{ NULL }
};
-AVFilter ff_vf_unsharp = {
+const AVFilter ff_vf_unsharp = {
.name = "unsharp",
.description = NULL_IF_CONFIG_SMALL("Sharpen or blur the input video."),
.priv_size = sizeof(UnsharpContext),
.query_formats = query_formats,
.inputs = avfilter_vf_unsharp_inputs,
.outputs = avfilter_vf_unsharp_outputs,
- .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC,
+ .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
};
projects / ffmpeg / blobdiff