RepoMirrors
/
ffmpeg
mirror of https://git.ffmpeg.org/ffmpeg.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

avfilter/af_afir: fix overhead for small partitions

This commit is contained in:
Paul B Mahol 2018-12-31 18:04:59 +01:00
parent c4276a7f30
commit 86555a2fbf
1 changed files with 23 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
libavfilter/af_afir.c
View File

@ -56,12 +56,12 @@ static void fcmul_add_c(float *sum, const float *t, const float *c, ptrdiff_t le
sum[2 * n] += t[2 * n] * c[2 * n]; sum[2 * n] += t[2 * n] * c[2 * n];
} }
static int fir_channel(AVFilterContext *ctx, void *arg, int ch) static int fir_quantum(AVFilterContext *ctx, AVFrame *out, int ch, int offset)
{ {
AudioFIRContext *s = ctx->priv; AudioFIRContext *s = ctx->priv;
const float *in = (const float *)s->in[0]->extended_data[ch]; const float *in = (const float *)s->in[0]->extended_data[ch] + offset;
AVFrame *out = arg; float *block, *buf, *ptr = (float *)out->extended_data[ch] + offset;
float *block, *buf, *ptr = (float *)out->extended_data[ch]; const int nb_samples = FFMIN(s->min_part_size, out->nb_samples - offset);
int n, i, j; int n, i, j;
for (int segment = 0; segment < s->nb_segments; segment++) { for (int segment = 0; segment < s->nb_segments; segment++) {
@ -70,7 +70,7 @@ static int fir_channel(AVFilterContext *ctx, void *arg, int ch)
float *dst = (float *)seg->output->extended_data[ch]; float *dst = (float *)seg->output->extended_data[ch];
float *sum = (float *)seg->sum->extended_data[ch]; float *sum = (float *)seg->sum->extended_data[ch];
s->fdsp->vector_fmul_scalar(src + seg->input_offset, in, s->dry_gain, FFALIGN(out->nb_samples, 4)); s->fdsp->vector_fmul_scalar(src + seg->input_offset, in, s->dry_gain, FFALIGN(nb_samples, 4));
emms_c(); emms_c();
seg->output_offset[ch] += s->min_part_size; seg->output_offset[ch] += s->min_part_size;
@ -80,7 +80,7 @@ static int fir_channel(AVFilterContext *ctx, void *arg, int ch)
memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src)); memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
dst += seg->output_offset[ch]; dst += seg->output_offset[ch];
for (n = 0; n < out->nb_samples; n++) { for (n = 0; n < nb_samples; n++) {
ptr[n] += dst[n]; ptr[n] += dst[n];
} }
continue; continue;
@ -127,17 +127,28 @@ static int fir_channel(AVFilterContext *ctx, void *arg, int ch)
memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src)); memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
for (n = 0; n < out->nb_samples; n++) { for (n = 0; n < nb_samples; n++) {
ptr[n] += dst[n]; ptr[n] += dst[n];
} }
} }
s->fdsp->vector_fmul_scalar(ptr, ptr, s->wet_gain, FFALIGN(out->nb_samples, 4)); s->fdsp->vector_fmul_scalar(ptr, ptr, s->wet_gain, FFALIGN(nb_samples, 4));
emms_c(); emms_c();
return 0; return 0;
} }
static int fir_channel(AVFilterContext *ctx, AVFrame *out, int ch)
{
AudioFIRContext *s = ctx->priv;
for (int offset = 0; offset < out->nb_samples; offset += s->min_part_size) {
fir_quantum(ctx, out, ch, offset);
}
return 0;
}
static int fir_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) static int fir_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
{ {
AVFrame *out = arg; AVFrame *out = arg;
@ -525,8 +536,8 @@ static int activate(AVFilterContext *ctx)
{ {
AudioFIRContext *s = ctx->priv; AudioFIRContext *s = ctx->priv;
AVFilterLink *outlink = ctx->outputs[0]; AVFilterLink *outlink = ctx->outputs[0];
int ret, status, available, wanted;
AVFrame *in = NULL; AVFrame *in = NULL;
int ret, status;
int64_t pts; int64_t pts;
FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx); FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
@ -557,7 +568,9 @@ static int activate(AVFilterContext *ctx)
return ret; return ret;
} }
ret = ff_inlink_consume_samples(ctx->inputs[0], s->min_part_size, s->min_part_size, &in); available = ff_inlink_queued_samples(ctx->inputs[0]);
wanted = FFMAX(s->min_part_size, (available / s->min_part_size) * s->min_part_size);
ret = ff_inlink_consume_samples(ctx->inputs[0], wanted, wanted, &in);
if (ret > 0) if (ret > 0)
ret = fir_frame(s, in, outlink); ret = fir_frame(s, in, outlink);