sum[2 * n] += t[2 * n] * c[2 * n];
}
-static int fir_channel(AVFilterContext *ctx, void *arg, int ch, int nb_jobs)
+static int fir_quantum(AVFilterContext *ctx, AVFrame *out, int ch, int offset)
{
AudioFIRContext *s = ctx->priv;
- const float *src = (const float *)s->in[0]->extended_data[ch];
- float *sum = (float *)s->seg.sum->extended_data[ch];
- AVFrame *out = arg;
- float *block, *dst, *ptr;
+ const float *in = (const float *)s->in[0]->extended_data[ch] + offset;
+ float *block, *buf, *ptr = (float *)out->extended_data[ch] + offset;
+ const int nb_samples = FFMIN(s->min_part_size, out->nb_samples - offset);
int n, i, j;
- memset(sum, 0, sizeof(*sum) * s->seg.fft_length);
- block = (float *)s->seg.block->extended_data[ch] + s->seg.part_index * s->seg.block_size;
- memset(block, 0, sizeof(*block) * s->seg.fft_length);
+ for (int segment = 0; segment < s->nb_segments; segment++) {
+ AudioFIRSegment *seg = &s->seg[segment];
+ float *src = (float *)seg->input->extended_data[ch];
+ float *dst = (float *)seg->output->extended_data[ch];
+ float *sum = (float *)seg->sum->extended_data[ch];
- s->fdsp->vector_fmul_scalar(block, src, s->dry_gain, FFALIGN(out->nb_samples, 4));
- emms_c();
+ s->fdsp->vector_fmul_scalar(src + seg->input_offset, in, s->dry_gain, FFALIGN(nb_samples, 4));
+ emms_c();
+
+ seg->output_offset[ch] += s->min_part_size;
+ if (seg->output_offset[ch] == seg->part_size) {
+ seg->output_offset[ch] = 0;
+ } else {
+ memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
- av_rdft_calc(s->seg.rdft[ch], block);
- block[2 * s->seg.part_size] = block[1];
- block[1] = 0;
+ dst += seg->output_offset[ch];
+ for (n = 0; n < nb_samples; n++) {
+ ptr[n] += dst[n];
+ }
+ continue;
+ }
- j = s->seg.part_index;
+ memset(sum, 0, sizeof(*sum) * seg->fft_length);
+ block = (float *)seg->block->extended_data[ch] + seg->part_index[ch] * seg->block_size;
+ memset(block + seg->part_size, 0, sizeof(*block) * (seg->fft_length - seg->part_size));
- for (i = 0; i < s->seg.nb_partitions; i++) {
- const int coffset = i * s->seg.coeff_size;
- const float *block = (const float *)s->seg.block->extended_data[ch] + j * s->seg.block_size;
- const FFTComplex *coeff = s->seg.coeff[ch * !s->one2many] + coffset;
+ memcpy(block, src, sizeof(*src) * seg->part_size);
- s->fcmul_add(sum, block, (const float *)coeff, s->seg.part_size);
+ av_rdft_calc(seg->rdft[ch], block);
+ block[2 * seg->part_size] = block[1];
+ block[1] = 0;
- if (j == 0)
- j = s->seg.nb_partitions;
- j--;
- }
+ j = seg->part_index[ch];
+
+ for (i = 0; i < seg->nb_partitions; i++) {
+ const int coffset = j * seg->coeff_size;
+ const float *block = (const float *)seg->block->extended_data[ch] + i * seg->block_size;
+ const FFTComplex *coeff = (const FFTComplex *)seg->coeff->extended_data[ch * !s->one2many] + coffset;
+
+ s->afirdsp.fcmul_add(sum, block, (const float *)coeff, seg->part_size);
+
+ if (j == 0)
+ j = seg->nb_partitions;
+ j--;
+ }
+
+ sum[1] = sum[2 * seg->part_size];
+ av_rdft_calc(seg->irdft[ch], sum);
- sum[1] = sum[2 * s->seg.part_size];
- av_rdft_calc(s->seg.irdft[ch], sum);
+ buf = (float *)seg->buffer->extended_data[ch];
+ for (n = 0; n < seg->part_size; n++) {
+ buf[n] += sum[n];
+ }
+
+ memcpy(dst, buf, seg->part_size * sizeof(*dst));
+
+ buf = (float *)seg->buffer->extended_data[ch];
+ memcpy(buf, sum + seg->part_size, seg->part_size * sizeof(*buf));
- dst = (float *)s->seg.buffer->extended_data[ch];
- for (n = 0; n < s->seg.part_size; n++) {
- dst[n] += sum[n];
+ seg->part_index[ch] = (seg->part_index[ch] + 1) % seg->nb_partitions;
+
+ memmove(src, src + s->min_part_size, (seg->input_size - s->min_part_size) * sizeof(*src));
+
+ for (n = 0; n < nb_samples; n++) {
+ ptr[n] += dst[n];
+ }
}
- ptr = (float *)out->extended_data[ch];
- s->fdsp->vector_fmul_scalar(ptr, dst, s->wet_gain, FFALIGN(out->nb_samples, 4));
+ s->fdsp->vector_fmul_scalar(ptr, ptr, s->wet_gain, FFALIGN(nb_samples, 4));
emms_c();
- dst = (float *)s->seg.buffer->extended_data[ch];
- memcpy(dst, sum + s->seg.part_size, s->seg.part_size * sizeof(*dst));
+ 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)
+{
+ AVFrame *out = arg;
+ const int start = (out->channels * jobnr) / nb_jobs;
+ const int end = (out->channels * (jobnr+1)) / nb_jobs;
+
+ for (int ch = start; ch < end; ch++) {
+ fir_channel(ctx, out, ch);
+ }
return 0;
}
if (s->pts == AV_NOPTS_VALUE)
s->pts = in->pts;
s->in[0] = in;
- ctx->internal->execute(ctx, fir_channel, out, NULL, outlink->channels);
-
- s->seg.part_index = (s->seg.part_index + 1) % s->seg.nb_partitions;
+ ctx->internal->execute(ctx, fir_channels, out, NULL, FFMIN(outlink->channels,
+ ff_filter_get_nb_threads(ctx)));
out->pts = s->pts;
if (s->pts != AV_NOPTS_VALUE)
av_free(mag);
}
+static int init_segment(AVFilterContext *ctx, AudioFIRSegment *seg,
+ int offset, int nb_partitions, int part_size)
+{
+ AudioFIRContext *s = ctx->priv;
+
+ seg->rdft = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->rdft));
+ seg->irdft = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->irdft));
+ if (!seg->rdft || !seg->irdft)
+ return AVERROR(ENOMEM);
+
+ seg->fft_length = part_size * 2 + 1;
+ seg->part_size = part_size;
+ seg->block_size = FFALIGN(seg->fft_length, 32);
+ seg->coeff_size = FFALIGN(seg->part_size + 1, 32);
+ seg->nb_partitions = nb_partitions;
+ seg->input_size = offset + s->min_part_size;
+ seg->input_offset = offset;
+
+ seg->part_index = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->part_index));
+ seg->output_offset = av_calloc(ctx->inputs[0]->channels, sizeof(*seg->output_offset));
+ if (!seg->part_index || !seg->output_offset)
+ return AVERROR(ENOMEM);
+
+ for (int ch = 0; ch < ctx->inputs[0]->channels; ch++) {
+ seg->rdft[ch] = av_rdft_init(av_log2(2 * part_size), DFT_R2C);
+ seg->irdft[ch] = av_rdft_init(av_log2(2 * part_size), IDFT_C2R);
+ if (!seg->rdft[ch] || !seg->irdft[ch])
+ return AVERROR(ENOMEM);
+ }
+
+ seg->sum = ff_get_audio_buffer(ctx->inputs[0], seg->fft_length);
+ seg->block = ff_get_audio_buffer(ctx->inputs[0], seg->nb_partitions * seg->block_size);
+ seg->buffer = ff_get_audio_buffer(ctx->inputs[0], seg->part_size);
+ seg->coeff = ff_get_audio_buffer(ctx->inputs[1], seg->nb_partitions * seg->coeff_size * 2);
+ seg->input = ff_get_audio_buffer(ctx->inputs[0], seg->input_size);
+ seg->output = ff_get_audio_buffer(ctx->inputs[0], seg->part_size);
+ if (!seg->buffer || !seg->sum || !seg->block || !seg->coeff || !seg->input || !seg->output)
+ return AVERROR(ENOMEM);
+
+ return 0;
+}
+
static int convert_coeffs(AVFilterContext *ctx)
{
AudioFIRContext *s = ctx->priv;
- int ret, i, ch, n, N;
+ int left, offset = 0, part_size, max_part_size;
+ int ret, i, ch, n;
float power = 0;
s->nb_taps = ff_inlink_queued_samples(ctx->inputs[1]);
if (s->nb_taps <= 0)
return AVERROR(EINVAL);
- for (n = av_log2(s->minp); (1 << n) < s->nb_taps; n++);
- N = FFMIN(n, av_log2(s->maxp));
- s->seg.fft_length = (1 << (N + 1)) + 1;
- s->seg.part_size = 1 << (N - 1);
- s->seg.block_size = FFALIGN(s->seg.fft_length, 32);
- s->seg.coeff_size = FFALIGN(s->seg.part_size + 1, 32);
- s->seg.nb_partitions = (s->nb_taps + s->seg.part_size - 1) / s->seg.part_size;
-
- for (ch = 0; ch < ctx->inputs[1]->channels; ch++) {
- s->seg.coeff[ch] = av_calloc(s->seg.nb_partitions * s->seg.coeff_size, sizeof(**s->seg.coeff));
- if (!s->seg.coeff[ch])
- return AVERROR(ENOMEM);
+ if (s->minp > s->maxp) {
+ s->maxp = s->minp;
}
- for (ch = 0; ch < ctx->inputs[0]->channels; ch++) {
- s->seg.rdft[ch] = av_rdft_init(N, DFT_R2C);
- s->seg.irdft[ch] = av_rdft_init(N, IDFT_C2R);
- if (!s->seg.rdft[ch] || !s->seg.irdft[ch])
- return AVERROR(ENOMEM);
- }
+ left = s->nb_taps;
+ part_size = 1 << av_log2(s->minp);
+ max_part_size = 1 << av_log2(s->maxp);
- s->seg.sum = ff_get_audio_buffer(ctx->inputs[0], s->seg.fft_length);
- s->seg.block = ff_get_audio_buffer(ctx->inputs[0], s->seg.nb_partitions * s->seg.block_size);
- s->seg.buffer = ff_get_audio_buffer(ctx->inputs[0], s->seg.part_size);
- if (!s->seg.buffer || !s->seg.sum || !s->seg.block)
- return AVERROR(ENOMEM);
+ s->min_part_size = part_size;
+
+ for (i = 0; left > 0; i++) {
+ int step = part_size == max_part_size ? INT_MAX : 1 + (i == 0);
+ int nb_partitions = FFMIN(step, (left + part_size - 1) / part_size);
+
+ s->nb_segments = i + 1;
+ ret = init_segment(ctx, &s->seg[i], offset, nb_partitions, part_size);
+ if (ret < 0)
+ return ret;
+ offset += nb_partitions * part_size;
+ left -= nb_partitions * part_size;
+ part_size *= 2;
+ part_size = FFMIN(part_size, max_part_size);
+ }
ret = ff_inlink_consume_samples(ctx->inputs[1], s->nb_taps, s->nb_taps, &s->in[1]);
if (ret < 0)
s->fdsp->vector_fmul_scalar(time, time, s->gain, FFALIGN(s->nb_taps, 4));
}
+ av_log(ctx, AV_LOG_DEBUG, "nb_taps: %d\n", s->nb_taps);
+ av_log(ctx, AV_LOG_DEBUG, "nb_segments: %d\n", s->nb_segments);
+
for (ch = 0; ch < ctx->inputs[1]->channels; ch++) {
float *time = (float *)s->in[1]->extended_data[!s->one2many * ch];
- float *block = (float *)s->seg.block->extended_data[ch];
- FFTComplex *coeff = s->seg.coeff[ch];
+ int toffset = 0;
for (i = FFMAX(1, s->length * s->nb_taps); i < s->nb_taps; i++)
time[i] = 0;
- for (i = 0; i < s->seg.nb_partitions; i++) {
- const float scale = 1.f / s->seg.part_size;
- const int toffset = i * s->seg.part_size;
- const int coffset = i * s->seg.coeff_size;
- const int remaining = s->nb_taps - (i * s->seg.part_size);
- const int size = remaining >= s->seg.part_size ? s->seg.part_size : remaining;
+ av_log(ctx, AV_LOG_DEBUG, "channel: %d\n", ch);
- memset(block, 0, sizeof(*block) * s->seg.fft_length);
- memcpy(block, time + toffset, size * sizeof(*block));
+ for (int segment = 0; segment < s->nb_segments; segment++) {
+ AudioFIRSegment *seg = &s->seg[segment];
+ float *block = (float *)seg->block->extended_data[ch];
+ FFTComplex *coeff = (FFTComplex *)seg->coeff->extended_data[ch];
- av_rdft_calc(s->seg.rdft[0], block);
+ av_log(ctx, AV_LOG_DEBUG, "segment: %d\n", segment);
- coeff[coffset].re = block[0] * scale;
- coeff[coffset].im = 0;
- for (n = 1; n < s->seg.part_size; n++) {
- coeff[coffset + n].re = block[2 * n] * scale;
- coeff[coffset + n].im = block[2 * n + 1] * scale;
+ for (i = 0; i < seg->nb_partitions; i++) {
+ const float scale = 1.f / seg->part_size;
+ const int coffset = i * seg->coeff_size;
+ const int remaining = s->nb_taps - toffset;
+ const int size = remaining >= seg->part_size ? seg->part_size : remaining;
+
+ memset(block, 0, sizeof(*block) * seg->fft_length);
+ memcpy(block, time + toffset, size * sizeof(*block));
+
+ av_rdft_calc(seg->rdft[0], block);
+
+ coeff[coffset].re = block[0] * scale;
+ coeff[coffset].im = 0;
+ for (n = 1; n < seg->part_size; n++) {
+ coeff[coffset + n].re = block[2 * n] * scale;
+ coeff[coffset + n].im = block[2 * n + 1] * scale;
+ }
+ coeff[coffset + seg->part_size].re = block[1] * scale;
+ coeff[coffset + seg->part_size].im = 0;
+
+ toffset += size;
}
- coeff[coffset + s->seg.part_size].re = block[1] * scale;
- coeff[coffset + s->seg.part_size].im = 0;
+
+ av_log(ctx, AV_LOG_DEBUG, "nb_partitions: %d\n", seg->nb_partitions);
+ av_log(ctx, AV_LOG_DEBUG, "partition size: %d\n", seg->part_size);
+ av_log(ctx, AV_LOG_DEBUG, "block size: %d\n", seg->block_size);
+ av_log(ctx, AV_LOG_DEBUG, "fft_length: %d\n", seg->fft_length);
+ av_log(ctx, AV_LOG_DEBUG, "coeff_size: %d\n", seg->coeff_size);
+ av_log(ctx, AV_LOG_DEBUG, "input_size: %d\n", seg->input_size);
+ av_log(ctx, AV_LOG_DEBUG, "input_offset: %d\n", seg->input_offset);
}
}
av_frame_free(&s->in[1]);
- av_log(ctx, AV_LOG_DEBUG, "nb_taps: %d\n", s->nb_taps);
- av_log(ctx, AV_LOG_DEBUG, "nb_partitions: %d\n", s->seg.nb_partitions);
- av_log(ctx, AV_LOG_DEBUG, "partition size: %d\n", s->seg.part_size);
- av_log(ctx, AV_LOG_DEBUG, "fft_length: %d\n", s->seg.fft_length);
-
s->have_coeffs = 1;
return 0;
{
AudioFIRContext *s = ctx->priv;
AVFilterLink *outlink = ctx->outputs[0];
+ int ret, status, available, wanted;
AVFrame *in = NULL;
- int ret, status;
int64_t pts;
FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
return ret;
}
- ret = ff_inlink_consume_samples(ctx->inputs[0], s->seg.part_size, s->seg.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)
ret = fir_frame(s, in, outlink);
}
}
- if (ff_inlink_queued_samples(ctx->inputs[0]) >= s->seg.part_size) {
+ if (ff_inlink_queued_samples(ctx->inputs[0]) >= s->min_part_size) {
ff_filter_set_ready(ctx, 10);
return 0;
}
outlink->channel_layout = ctx->inputs[0]->channel_layout;
outlink->channels = ctx->inputs[0]->channels;
- s->seg.coeff = av_calloc(ctx->inputs[1]->channels, sizeof(*s->seg.coeff));
- s->seg.rdft = av_calloc(outlink->channels, sizeof(*s->seg.rdft));
- s->seg.irdft = av_calloc(outlink->channels, sizeof(*s->seg.irdft));
- if (!s->seg.coeff || !s->seg.rdft || !s->seg.irdft)
- return AVERROR(ENOMEM);
-
s->nb_channels = outlink->channels;
s->nb_coef_channels = ctx->inputs[1]->channels;
s->pts = AV_NOPTS_VALUE;
return 0;
}
-static av_cold void uninit(AVFilterContext *ctx)
+static void uninit_segment(AVFilterContext *ctx, AudioFIRSegment *seg)
{
AudioFIRContext *s = ctx->priv;
- int ch;
- if (s->seg.coeff) {
- for (ch = 0; ch < s->nb_coef_channels; ch++) {
- av_freep(&s->seg.coeff[ch]);
+ if (seg->rdft) {
+ for (int ch = 0; ch < s->nb_channels; ch++) {
+ av_rdft_end(seg->rdft[ch]);
}
}
- av_freep(&s->seg.coeff);
+ av_freep(&seg->rdft);
- if (s->seg.rdft) {
- for (ch = 0; ch < s->nb_channels; ch++) {
- av_rdft_end(s->seg.rdft[ch]);
+ if (seg->irdft) {
+ for (int ch = 0; ch < s->nb_channels; ch++) {
+ av_rdft_end(seg->irdft[ch]);
}
}
- av_freep(&s->seg.rdft);
-
- if (s->seg.irdft) {
- for (ch = 0; ch < s->nb_channels; ch++) {
- av_rdft_end(s->seg.irdft[ch]);
- }
- }
- av_freep(&s->seg.irdft);
+ av_freep(&seg->irdft);
+
+ av_freep(&seg->output_offset);
+ av_freep(&seg->part_index);
+
+ av_frame_free(&seg->block);
+ av_frame_free(&seg->sum);
+ av_frame_free(&seg->buffer);
+ av_frame_free(&seg->coeff);
+ av_frame_free(&seg->input);
+ av_frame_free(&seg->output);
+ seg->input_size = 0;
+}
- av_frame_free(&s->in[1]);
+static av_cold void uninit(AVFilterContext *ctx)
+{
+ AudioFIRContext *s = ctx->priv;
- av_frame_free(&s->seg.block);
- av_frame_free(&s->seg.sum);
- av_frame_free(&s->seg.buffer);
+ for (int i = 0; i < s->nb_segments; i++) {
+ uninit_segment(ctx, &s->seg[i]);
+ }
av_freep(&s->fdsp);
+ av_frame_free(&s->in[1]);
for (int i = 0; i < ctx->nb_outputs; i++)
av_freep(&ctx->output_pads[i].name);
return 0;
}
+void ff_afir_init(AudioFIRDSPContext *dsp)
+{
+ dsp->fcmul_add = fcmul_add_c;
+
+ if (ARCH_X86)
+ ff_afir_init_x86(dsp);
+}
+
static av_cold int init(AVFilterContext *ctx)
{
AudioFIRContext *s = ctx->priv;
}
}
- s->fcmul_add = fcmul_add_c;
-
s->fdsp = avpriv_float_dsp_alloc(0);
if (!s->fdsp)
return AVERROR(ENOMEM);
- if (ARCH_X86)
- ff_afir_init_x86(s);
+ ff_afir_init(&s->afirdsp);
return 0;
}
static const AVFilterPad afir_inputs[] = {
{
- .name = "main",
- .type = AVMEDIA_TYPE_AUDIO,
+ .name = "main",
+ .type = AVMEDIA_TYPE_AUDIO,
},{
- .name = "ir",
- .type = AVMEDIA_TYPE_AUDIO,
+ .name = "ir",
+ .type = AVMEDIA_TYPE_AUDIO,
},
{ NULL }
};
{ "channel", "set IR channel to display frequency response", OFFSET(ir_channel), AV_OPT_TYPE_INT, {.i64=0}, 0, 1024, VF },
{ "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "hd720"}, 0, 0, VF },
{ "rate", "set video rate", OFFSET(frame_rate), AV_OPT_TYPE_VIDEO_RATE, {.str = "25"}, 0, INT32_MAX, VF },
- { "minp", "set min partition size", OFFSET(minp), AV_OPT_TYPE_INT, {.i64=16}, 16, 65536, AF },
- { "maxp", "set max partition size", OFFSET(maxp), AV_OPT_TYPE_INT, {.i64=65536}, 16, 65536, AF },
+ { "minp", "set min partition size", OFFSET(minp), AV_OPT_TYPE_INT, {.i64=8192}, 8, 32768, AF },
+ { "maxp", "set max partition size", OFFSET(maxp), AV_OPT_TYPE_INT, {.i64=8192}, 8, 32768, AF },
{ NULL }
};
projects / ffmpeg / blobdiff