avfilter/vf_amplify: improve performance

This commit is contained in:
Paul B Mahol 2022-02-27 23:40:57 +01:00
parent a2dbd17788
commit 567cab3bd8
1 changed files with 9 additions and 13 deletions

View File

@ -96,8 +96,7 @@ typedef struct ThreadData {
} ThreadData;
#define AMPLIFY_SLICE(type, stype, clip) \
const stype llimit = s->llimit; \
const stype hlimit = s->hlimit; \
const stype limit[2] = { s->llimit, s->hlimit }; \
\
for (int p = 0; p < s->nb_planes; p++) { \
const int slice_start = (s->height[p] * jobnr) / nb_jobs; \
@ -116,23 +115,19 @@ typedef struct ThreadData {
for (int y = slice_start; y < slice_end; y++) { \
for (int x = 0; x < s->linesize[p] / sizeof(type); x++) { \
stype src = *(type *)(in[radius]->data[p] + y * in[radius]->linesize[p] + x * sizeof(type));\
float diff, avg; \
float diff, abs_diff, avg; \
stype sum = 0; \
\
for (int i = 0; i < nb_inputs; i++) { \
sum += *(type *)(in[i]->data[p] + y * in[i]->linesize[p] + x * sizeof(type));\
} \
\
avg = sum / (float)nb_inputs; \
avg = sum * scale; \
diff = src - avg; \
abs_diff = fabsf(diff); \
\
if (fabsf(diff) < threshold && fabsf(diff) > tolerance) { \
stype amp; \
if (diff < 0) { \
amp = -FFMIN(FFABS(diff * factor), llimit); \
} else { \
amp = FFMIN(FFABS(diff * factor), hlimit); \
} \
if (abs_diff < threshold && abs_diff > tolerance) { \
float amp = copysignf(fminf(abs_diff * factor, limit[diff >= 0]), diff); \
dst[x] = clip(src + amp, depth); \
} else { \
dst[x] = src; \
@ -143,8 +138,8 @@ typedef struct ThreadData {
} \
}
#define CLIP8(x, depth) av_clip_uint8(x)
#define CLIP16(x, depth) av_clip_uintp2_c(x, depth)
#define CLIP8(x, depth) av_clip_uint8(lrintf(x))
#define CLIP16(x, depth) av_clip_uintp2_c(lrintf(x), depth)
#define NOP(x, depth) (x)
static int amplify_frame(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
@ -157,6 +152,7 @@ static int amplify_frame(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs
const int nb_inputs = s->nb_inputs;
const float threshold = s->threshold;
const float tolerance = s->tolerance;
const float scale = 1.f / nb_inputs;
const float factor = s->factor;
const int depth = s->depth;