int eof_hrirs;
int ir_len;
+ int air_len;
int mapping[64];
float *data_ir[2];
float *temp_src[2];
FFTComplex *temp_fft[2];
+ FFTComplex *temp_afft[2];
FFTContext *fft[2], *ifft[2];
FFTComplex *data_hrtf[2];
p = NULL;
if (parse_channel_name(s, s->nb_irs, &arg, &out_ch_id, buf)) {
- av_log(ctx, AV_LOG_WARNING, "Failed to parse \'%s\' as channel name.\n", buf);
+ av_log(ctx, AV_LOG_WARNING, "Failed to parse \'%s\' as channel name.\n", arg);
continue;
}
s->mapping[s->nb_irs] = out_ch_id;
float **ringbuffer;
float **temp_src;
FFTComplex **temp_fft;
+ FFTComplex **temp_afft;
} ThreadData;
static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
float *ringbuffer = td->ringbuffer[jobnr];
float *temp_src = td->temp_src[jobnr];
const int ir_len = s->ir_len;
+ const int air_len = s->air_len;
const float *src = (const float *)in->data[0];
float *dst = (float *)out->data[0];
const int in_channels = in->channels;
if (l == s->lfe_channel) {
*dst += *(buffer[s->lfe_channel] + wr) * s->gain_lfe;
- temp_ir += FFALIGN(ir_len, 16);
+ temp_ir += air_len;
continue;
}
if (read + ir_len < buffer_length) {
memcpy(temp_src, bptr + read, ir_len * sizeof(*temp_src));
} else {
- int len = FFMIN(ir_len - (read % ir_len), buffer_length - read);
+ int len = FFMIN(air_len - (read % ir_len), buffer_length - read);
memcpy(temp_src, bptr + read, len * sizeof(*temp_src));
- memcpy(temp_src + len, bptr, (ir_len - len) * sizeof(*temp_src));
+ memcpy(temp_src + len, bptr, (air_len - len) * sizeof(*temp_src));
}
- dst[0] += s->fdsp->scalarproduct_float(temp_ir, temp_src, ir_len);
- temp_ir += FFALIGN(ir_len, 16);
+ dst[0] += s->fdsp->scalarproduct_float(temp_ir, temp_src, FFALIGN(ir_len, 32));
+ temp_ir += air_len;
}
- if (fabs(*dst) > 1)
- *n_clippings += 1;
+ if (fabsf(dst[0]) > 1)
+ n_clippings[0]++;
dst += 2;
src += in_channels;
const int buffer_length = s->buffer_length;
const uint32_t modulo = (uint32_t)buffer_length - 1;
FFTComplex *fft_in = s->temp_fft[jobnr];
+ FFTComplex *fft_acc = s->temp_afft[jobnr];
FFTContext *ifft = s->ifft[jobnr];
FFTContext *fft = s->fft[jobnr];
const int n_fft = s->n_fft;
dst += offset;
- n_read = FFMIN(s->ir_len, in->nb_samples);
+ n_read = FFMIN(ir_len, in->nb_samples);
for (j = 0; j < n_read; j++) {
dst[2 * j] = ringbuffer[wr];
ringbuffer[wr] = 0.0;
dst[2 * j] = 0;
}
+ memset(fft_acc, 0, sizeof(FFTComplex) * n_fft);
+
for (i = 0; i < in_channels; i++) {
if (i == s->lfe_channel) {
for (j = 0; j < in->nb_samples; j++) {
const float re = fft_in[j].re;
const float im = fft_in[j].im;
- fft_in[j].re = re * hcomplex->re - im * hcomplex->im;
- fft_in[j].im = re * hcomplex->im + im * hcomplex->re;
+ fft_acc[j].re += re * hcomplex->re - im * hcomplex->im;
+ fft_acc[j].im += re * hcomplex->im + im * hcomplex->re;
}
+ }
- av_fft_permute(ifft, fft_in);
- av_fft_calc(ifft, fft_in);
+ av_fft_permute(ifft, fft_acc);
+ av_fft_calc(ifft, fft_acc);
- for (j = 0; j < in->nb_samples; j++) {
- dst[2 * j] += fft_in[j].re * fft_scale;
- }
+ for (j = 0; j < in->nb_samples; j++) {
+ dst[2 * j] += fft_acc[j].re * fft_scale;
+ }
- for (j = 0; j < ir_len - 1; j++) {
- int write_pos = (wr + j) & modulo;
+ for (j = 0; j < ir_len - 1; j++) {
+ int write_pos = (wr + j) & modulo;
- *(ringbuffer + write_pos) += fft_in[in->nb_samples + j].re * fft_scale;
- }
+ *(ringbuffer + write_pos) += fft_acc[in->nb_samples + j].re * fft_scale;
}
for (i = 0; i < out->nb_samples; i++) {
- if (fabs(*dst) > 1) {
+ if (fabsf(dst[0]) > 1) {
n_clippings[0]++;
}
td.delay = s->delay; td.ir = s->data_ir; td.n_clippings = n_clippings;
td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;
td.temp_fft = s->temp_fft;
+ td.temp_afft = s->temp_afft;
if (s->type == TIME_DOMAIN) {
ctx->internal->execute(ctx, headphone_convolute, &td, NULL, 2);
int n_fft;
int i, j, k;
- s->buffer_length = 1 << (32 - ff_clz(s->ir_len));
- s->n_fft = n_fft = 1 << (32 - ff_clz(s->ir_len + s->size));
+ s->air_len = 1 << (32 - ff_clz(ir_len));
+ s->buffer_length = 1 << (32 - ff_clz(s->air_len));
+ s->n_fft = n_fft = 1 << (32 - ff_clz(ir_len + s->size));
if (s->type == FREQUENCY_DOMAIN) {
fft_in_l = av_calloc(n_fft, sizeof(*fft_in_l));
av_fft_end(s->fft[0]);
av_fft_end(s->fft[1]);
- s->fft[0] = av_fft_init(log2(s->n_fft), 0);
- s->fft[1] = av_fft_init(log2(s->n_fft), 0);
+ s->fft[0] = av_fft_init(av_log2(s->n_fft), 0);
+ s->fft[1] = av_fft_init(av_log2(s->n_fft), 0);
av_fft_end(s->ifft[0]);
av_fft_end(s->ifft[1]);
- s->ifft[0] = av_fft_init(log2(s->n_fft), 1);
- s->ifft[1] = av_fft_init(log2(s->n_fft), 1);
+ s->ifft[0] = av_fft_init(av_log2(s->n_fft), 1);
+ s->ifft[1] = av_fft_init(av_log2(s->n_fft), 1);
if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) {
av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft);
}
}
- s->data_ir[0] = av_calloc(FFALIGN(s->ir_len, 16), sizeof(float) * s->nb_irs);
- s->data_ir[1] = av_calloc(FFALIGN(s->ir_len, 16), sizeof(float) * s->nb_irs);
+ s->data_ir[0] = av_calloc(s->air_len, sizeof(float) * s->nb_irs);
+ s->data_ir[1] = av_calloc(s->air_len, sizeof(float) * s->nb_irs);
s->delay[0] = av_calloc(s->nb_irs, sizeof(float));
s->delay[1] = av_calloc(s->nb_irs, sizeof(float));
s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float));
s->temp_fft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
s->temp_fft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
- if (!s->temp_fft[0] || !s->temp_fft[1]) {
+ s->temp_afft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
+ s->temp_afft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
+ if (!s->temp_fft[0] || !s->temp_fft[1] ||
+ !s->temp_afft[0] || !s->temp_afft[1]) {
ret = AVERROR(ENOMEM);
goto fail;
}
}
if (s->type == TIME_DOMAIN) {
- s->temp_src[0] = av_calloc(FFALIGN(ir_len, 16), sizeof(float));
- s->temp_src[1] = av_calloc(FFALIGN(ir_len, 16), sizeof(float));
+ s->temp_src[0] = av_calloc(s->air_len, sizeof(float));
+ s->temp_src[1] = av_calloc(s->air_len, sizeof(float));
- data_ir_l = av_calloc(nb_irs * FFALIGN(ir_len, 16), sizeof(*data_ir_l));
- data_ir_r = av_calloc(nb_irs * FFALIGN(ir_len, 16), sizeof(*data_ir_r));
+ data_ir_l = av_calloc(nb_irs * s->air_len, sizeof(*data_ir_l));
+ data_ir_r = av_calloc(nb_irs * s->air_len, sizeof(*data_ir_r));
if (!data_ir_r || !data_ir_l || !s->temp_src[0] || !s->temp_src[1]) {
ret = AVERROR(ENOMEM);
goto fail;
ret = ff_inlink_consume_samples(ctx->inputs[i + 1], len, len, &s->in[i + 1].frame);
if (ret < 0)
- return ret;
+ goto fail;
ptr = (float *)s->in[i + 1].frame->extended_data[0];
if (s->hrir_fmt == HRIR_STEREO) {
if (idx == -1)
continue;
if (s->type == TIME_DOMAIN) {
- offset = idx * FFALIGN(len, 16);
+ offset = idx * s->air_len;
for (j = 0; j < len; j++) {
data_ir_l[offset + j] = ptr[len * 2 - j * 2 - 2] * gain_lin;
data_ir_r[offset + j] = ptr[len * 2 - j * 2 - 1] * gain_lin;
I = idx * 2;
if (s->type == TIME_DOMAIN) {
- offset = idx * FFALIGN(len, 16);
+ offset = idx * s->air_len;
for (j = 0; j < len; j++) {
data_ir_l[offset + j] = ptr[len * N - j * N - N + I ] * gain_lin;
data_ir_r[offset + j] = ptr[len * N - j * N - N + I + 1] * gain_lin;
}
if (s->type == TIME_DOMAIN) {
- memcpy(s->data_ir[0], data_ir_l, sizeof(float) * nb_irs * FFALIGN(ir_len, 16));
- memcpy(s->data_ir[1], data_ir_r, sizeof(float) * nb_irs * FFALIGN(ir_len, 16));
+ memcpy(s->data_ir[0], data_ir_l, sizeof(float) * nb_irs * s->air_len);
+ memcpy(s->data_ir[1], data_ir_r, sizeof(float) * nb_irs * s->air_len);
} else {
s->data_hrtf[0] = av_calloc(n_fft * s->nb_irs, sizeof(FFTComplex));
s->data_hrtf[1] = av_calloc(n_fft * s->nb_irs, sizeof(FFTComplex));
fail:
+ for (i = 0; i < s->nb_inputs - 1; i++)
+ av_frame_free(&s->in[i + 1].frame);
+
av_freep(&data_ir_l);
av_freep(&data_ir_r);
if ((ret = check_ir(ctx->inputs[i], i)) < 0)
return ret;
- if (!s->in[i].eof) {
if (ff_outlink_get_status(ctx->inputs[i]) == AVERROR_EOF)
s->in[i].eof = 1;
- }
}
for (i = 1; i < s->nb_inputs; i++) {
if (!layouts)
return AVERROR(ENOMEM);
- ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->out_channel_layouts);
+ ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->outcfg.channel_layouts);
if (ret)
return ret;
ret = ff_add_channel_layout(&stereo_layout, AV_CH_LAYOUT_STEREO);
+ if (ret)
+ return ret;
+ ret = ff_channel_layouts_ref(stereo_layout, &ctx->outputs[0]->incfg.channel_layouts);
if (ret)
return ret;
if (s->hrir_fmt == HRIR_MULTI) {
hrir_layouts = ff_all_channel_counts();
if (!hrir_layouts)
- ret = AVERROR(ENOMEM);
- ret = ff_channel_layouts_ref(hrir_layouts, &ctx->inputs[1]->out_channel_layouts);
+ return AVERROR(ENOMEM);
+ ret = ff_channel_layouts_ref(hrir_layouts, &ctx->inputs[1]->outcfg.channel_layouts);
if (ret)
return ret;
} else {
for (i = 1; i < s->nb_inputs; i++) {
- ret = ff_channel_layouts_ref(stereo_layout, &ctx->inputs[i]->out_channel_layouts);
+ ret = ff_channel_layouts_ref(stereo_layout, &ctx->inputs[i]->outcfg.channel_layouts);
if (ret)
return ret;
}
}
- ret = ff_channel_layouts_ref(stereo_layout, &ctx->outputs[0]->in_channel_layouts);
- if (ret)
- return ret;
-
formats = ff_all_samplerates();
if (!formats)
return AVERROR(ENOMEM);
}
}
- s->gain_lfe = expf((s->gain - 3 * inlink->channels - 6 + s->lfe_gain) / 20 * M_LN10);
+ s->gain_lfe = expf((s->gain - 3 * inlink->channels + s->lfe_gain) / 20 * M_LN10);
return 0;
}
static av_cold void uninit(AVFilterContext *ctx)
{
HeadphoneContext *s = ctx->priv;
- int i;
av_fft_end(s->ifft[0]);
av_fft_end(s->ifft[1]);
av_freep(&s->temp_src[1]);
av_freep(&s->temp_fft[0]);
av_freep(&s->temp_fft[1]);
+ av_freep(&s->temp_afft[0]);
+ av_freep(&s->temp_afft[1]);
av_freep(&s->data_hrtf[0]);
av_freep(&s->data_hrtf[1]);
av_freep(&s->fdsp);
- for (i = 0; i < s->nb_inputs; i++) {
- if (ctx->input_pads && i)
- av_freep(&ctx->input_pads[i].name);
- }
av_freep(&s->in);
+ for (unsigned i = 1; i < ctx->nb_inputs; i++)
+ av_freep(&ctx->input_pads[i].name);
}
#define OFFSET(x) offsetof(HeadphoneContext, x)