X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavfilter%2Faf_headphone.c;h=10638f9e7b57f9c3cd3e6da8a83c053960683af8;hb=7549f0ac1baffabfa964962c0c0067e8da692982;hp=6b210e1436de7de154b3dba95b09e6e95e39eeed;hpb=ce265b0bf5d0c77a092a1f5fbeb652c7cdea5fc7;p=ffmpeg

diff --git a/libavfilter/af_headphone.c b/libavfilter/af_headphone.c
index 6b210e1436d..10638f9e7b5 100644
--- a/libavfilter/af_headphone.c
+++ b/libavfilter/af_headphone.c
@@ -20,7 +20,6 @@
 
 #include <math.h>
 
-#include "libavutil/audio_fifo.h"
 #include "libavutil/avstring.h"
 #include "libavutil/channel_layout.h"
 #include "libavutil/float_dsp.h"
@@ -51,6 +50,7 @@ typedef struct HeadphoneContext {
     int eof_hrirs;
 
     int ir_len;
+    int air_len;
 
     int mapping[64];
 
@@ -73,13 +73,13 @@ typedef struct HeadphoneContext {
     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];
 
     AVFloatDSPContext *fdsp;
     struct headphone_inputs {
-        AVAudioFifo *fifo;
         AVFrame     *frame;
         int          ir_len;
         int          delay_l;
@@ -159,6 +159,7 @@ typedef struct ThreadData {
     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)
@@ -174,6 +175,7 @@ static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int n
     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;
@@ -202,7 +204,7 @@ static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int n
 
             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;
             }
 
@@ -211,18 +213,18 @@ static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int n
             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;
@@ -251,6 +253,7 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
     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;
@@ -262,7 +265,7 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
 
     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;
@@ -273,6 +276,8 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
         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++) {
@@ -297,26 +302,26 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
             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]++;
         }
 
@@ -328,20 +333,13 @@ static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr,
     return 0;
 }
 
-static int read_ir(AVFilterLink *inlink, int input_number, AVFrame *frame)
+static int check_ir(AVFilterLink *inlink, int input_number)
 {
     AVFilterContext *ctx = inlink->dst;
     HeadphoneContext *s = ctx->priv;
-    int ir_len, max_ir_len, ret;
-
-    ret = av_audio_fifo_write(s->in[input_number].fifo, (void **)frame->extended_data,
-                             frame->nb_samples);
-    av_frame_free(&frame);
+    int ir_len, max_ir_len;
 
-    if (ret < 0)
-        return ret;
-
-    ir_len = av_audio_fifo_size(s->in[input_number].fifo);
+    ir_len = ff_inlink_queued_samples(inlink);
     max_ir_len = 65536;
     if (ir_len > max_ir_len) {
         av_log(ctx, AV_LOG_ERROR, "Too big length of IRs: %d > %d.\n", ir_len, max_ir_len);
@@ -371,6 +369,7 @@ static int headphone_frame(HeadphoneContext *s, AVFrame *in, AVFilterLink *outli
     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);
@@ -405,8 +404,9 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
     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));
@@ -418,12 +418,12 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
 
         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);
@@ -432,8 +432,8 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
         }
     }
 
-    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));
 
@@ -445,7 +445,10 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
         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;
         }
@@ -457,20 +460,12 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
         goto fail;
     }
 
-    for (i = 0; i < s->nb_inputs - 1; i++) {
-        s->in[i + 1].frame = ff_get_audio_buffer(ctx->inputs[i + 1], s->ir_len);
-        if (!s->in[i + 1].frame) {
-            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;
@@ -490,7 +485,9 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
         int delay_r = s->in[i + 1].delay_r;
         float *ptr;
 
-        av_audio_fifo_read(s->in[i + 1].fifo, (void **)s->in[i + 1].frame->extended_data, len);
+        ret = ff_inlink_consume_samples(ctx->inputs[i + 1], len, len, &s->in[i + 1].frame);
+        if (ret < 0)
+            goto fail;
         ptr = (float *)s->in[i + 1].frame->extended_data[0];
 
         if (s->hrir_fmt == HRIR_STEREO) {
@@ -510,7 +507,7 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
             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;
@@ -553,7 +550,7 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
 
                 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;
@@ -577,11 +574,13 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
                 }
             }
         }
+
+        av_frame_free(&s->in[i + 1].frame);
     }
 
     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));
@@ -600,6 +599,9 @@ static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
 
 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);
 
@@ -623,27 +625,15 @@ static int activate(AVFilterContext *ctx)
     FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
     if (!s->eof_hrirs) {
         for (i = 1; i < s->nb_inputs; i++) {
-            AVFrame *ir = NULL;
-            int64_t pts;
-            int status;
-
             if (s->in[i].eof)
                 continue;
 
-            if ((ret = ff_inlink_consume_frame(ctx->inputs[i], &ir)) > 0) {
-                ret = read_ir(ctx->inputs[i], i, ir);
-                if (ret < 0)
-                    return ret;
-            }
-            if (ret < 0)
+            if ((ret = check_ir(ctx->inputs[i], i)) < 0)
                 return ret;
 
             if (!s->in[i].eof) {
-                if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
-                    if (status == AVERROR_EOF) {
-                        s->in[i].eof = 1;
-                    }
-                }
+                if (ff_outlink_get_status(ctx->inputs[i]) == AVERROR_EOF)
+                    s->in[i].eof = 1;
             }
         }
 
@@ -659,6 +649,7 @@ static int activate(AVFilterContext *ctx)
                         ff_inlink_request_frame(ctx->inputs[i]);
                 }
             }
+
             return 0;
         } else {
             s->eof_hrirs = 1;
@@ -803,7 +794,6 @@ static int config_output(AVFilterLink *outlink)
     AVFilterContext *ctx = outlink->src;
     HeadphoneContext *s = ctx->priv;
     AVFilterLink *inlink = ctx->inputs[0];
-    int i;
 
     if (s->hrir_fmt == HRIR_MULTI) {
         AVFilterLink *hrir_link = ctx->inputs[1];
@@ -814,12 +804,7 @@ static int config_output(AVFilterLink *outlink)
         }
     }
 
-    for (i = 0; i < s->nb_inputs; i++) {
-        s->in[i].fifo = av_audio_fifo_alloc(ctx->inputs[i]->format, ctx->inputs[i]->channels, 1024);
-        if (!s->in[i].fifo)
-            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;
 }
@@ -843,13 +828,13 @@ static av_cold void uninit(AVFilterContext *ctx)
     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++) {
-        av_frame_free(&s->in[i].frame);
-        av_audio_fifo_free(s->in[i].fifo);
         if (ctx->input_pads && i)
             av_freep(&ctx->input_pads[i].name);
     }