avfilter/af_anlms: add double sample format support

2 files changed, 180 insertions, 80 deletions

diff --git a/libavfilter/af_anlms.c b/libavfilter/af_anlms.c
index 3191ed1b31..9d3c44575b 100644
--- a/libavfilter/af_anlms.c
+++ b/libavfilter/af_anlms.c
@@ -26,6 +26,7 @@
 #include "audio.h"
 #include "avfilter.h"
 #include "filters.h"
+#include "formats.h"
 #include "internal.h"
 
 enum OutModes {
@@ -45,6 +46,7 @@ typedef struct AudioNLMSContext {
     float eps;
     float leakage;
     int output_mode;
+    int precision;
 
     int kernel_size;
     AVFrame *offset;
@@ -56,6 +58,8 @@ typedef struct AudioNLMSContext {
 
     int anlmf;
 
+    int (*filter_channels)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs);
+
     AVFloatDSPContext *fdsp;
 } AudioNLMSContext;
 
@@ -74,93 +78,32 @@ static const AVOption anlms_options[] = {
     {  "o", "output",                0,          AV_OPT_TYPE_CONST,    {.i64=OUT_MODE},     0, 0, AT, "mode" },
     {  "n", "noise",                 0,          AV_OPT_TYPE_CONST,    {.i64=NOISE_MODE},   0, 0, AT, "mode" },
     {  "e", "error",                 0,          AV_OPT_TYPE_CONST,    {.i64=ERROR_MODE},   0, 0, AT, "mode" },
+    { "precision", "set processing precision", OFFSET(precision),  AV_OPT_TYPE_INT,    {.i64=0},   0, 2, A, "precision" },
+    {   "auto",  "set auto processing precision",                  0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, A, "precision" },
+    {   "float", "set single-floating point processing precision", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, A, "precision" },
+    {   "double","set double-floating point processing precision", 0, AV_OPT_TYPE_CONST, {.i64=2}, 0, 0, A, "precision" },
     { NULL }
 };
 
 AVFILTER_DEFINE_CLASS_EXT(anlms, "anlm(f|s)", anlms_options);
 
-static float fir_sample(AudioNLMSContext *s, float sample, float *delay,
-                        float *coeffs, float *tmp, int *offset)
+static int query_formats(AVFilterContext *ctx)
 {
-    const int order = s->order;
-    float output;
-
-    delay[*offset] = sample;
-
-    memcpy(tmp, coeffs + order - *offset, order * sizeof(float));
-
-    output = s->fdsp->scalarproduct_float(delay, tmp, s->kernel_size);
-
-    if (--(*offset) < 0)
-        *offset = order - 1;
-
-    return output;
-}
-
-static float process_sample(AudioNLMSContext *s, float input, float desired,
-                            float *delay, float *coeffs, float *tmp, int *offsetp)
-{
-    const int order = s->order;
-    const float leakage = s->leakage;
-    const float mu = s->mu;
-    const float a = 1.f - leakage;
-    float sum, output, e, norm, b;
-    int offset = *offsetp;
-
-    delay[offset + order] = input;
-
-    output = fir_sample(s, input, delay, coeffs, tmp, offsetp);
-    e = desired - output;
-
-    sum = s->fdsp->scalarproduct_float(delay, delay, s->kernel_size);
-
-    norm = s->eps + sum;
-    b = mu * e / norm;
-    if (s->anlmf)
-        b *= e * e;
-
-    memcpy(tmp, delay + offset, order * sizeof(float));
-
-    s->fdsp->vector_fmul_scalar(coeffs, coeffs, a, s->kernel_size);
+    AudioNLMSContext *s = ctx->priv;
+    static const enum AVSampleFormat sample_fmts[3][3] = {
+        { AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_DBLP, AV_SAMPLE_FMT_NONE },
+        { AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE },
+        { AV_SAMPLE_FMT_DBLP, AV_SAMPLE_FMT_NONE },
+    };
+    int ret;
 
-    s->fdsp->vector_fmac_scalar(coeffs, tmp, b, s->kernel_size);
+    if ((ret = ff_set_common_all_channel_counts(ctx)) < 0)
+        return ret;
 
-    memcpy(coeffs + order, coeffs, order * sizeof(float));
+    if ((ret = ff_set_common_formats_from_list(ctx, sample_fmts[s->precision])) < 0)
+        return ret;
 
-    switch (s->output_mode) {
-    case IN_MODE:       output = input;         break;
-    case DESIRED_MODE:  output = desired;       break;
-    case OUT_MODE:   output = desired - output; break;
-    case NOISE_MODE: output = input - output;   break;
-    case ERROR_MODE:                            break;
-    }
-    return output;
-}
-
-static int process_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
-{
-    AudioNLMSContext *s = ctx->priv;
-    AVFrame *out = arg;
-    const int start = (out->ch_layout.nb_channels * jobnr) / nb_jobs;
-    const int end = (out->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
-
-    for (int c = start; c < end; c++) {
-        const float *input = (const float *)s->frame[0]->extended_data[c];
-        const float *desired = (const float *)s->frame[1]->extended_data[c];
-        float *delay = (float *)s->delay->extended_data[c];
-        float *coeffs = (float *)s->coeffs->extended_data[c];
-        float *tmp = (float *)s->tmp->extended_data[c];
-        int *offset = (int *)s->offset->extended_data[c];
-        float *output = (float *)out->extended_data[c];
-
-        for (int n = 0; n < out->nb_samples; n++) {
-            output[n] = process_sample(s, input[n], desired[n], delay, coeffs, tmp, offset);
-            if (ctx->is_disabled)
-                output[n] = input[n];
-        }
-    }
-
-    return 0;
+    return ff_set_common_all_samplerates(ctx);
 }
 
 static int activate(AVFilterContext *ctx)
@@ -195,7 +138,7 @@ static int activate(AVFilterContext *ctx)
             return AVERROR(ENOMEM);
         }
 
-        ff_filter_execute(ctx, process_channels, out, NULL,
+        ff_filter_execute(ctx, s->filter_channels, out, NULL,
                           FFMIN(ctx->outputs[0]->ch_layout.nb_channels, ff_filter_get_nb_threads(ctx)));
 
         out->pts = s->frame[0]->pts;
@@ -228,6 +171,13 @@ static int activate(AVFilterContext *ctx)
     return 0;
 }
 
+#define DEPTH 32
+#include "anlms_template.c"
+
+#undef DEPTH
+#define DEPTH 64
+#include "anlms_template.c"
+
 static int config_output(AVFilterLink *outlink)
 {
     AVFilterContext *ctx = outlink->src;
@@ -247,6 +197,15 @@ static int config_output(AVFilterLink *outlink)
     if (!s->delay || !s->coeffs || !s->offset || !s->tmp)
         return AVERROR(ENOMEM);
 
+    switch (outlink->format) {
+    case AV_SAMPLE_FMT_DBLP:
+        s->filter_channels = filter_channels_double;
+        break;
+    case AV_SAMPLE_FMT_FLTP:
+        s->filter_channels = filter_channels_float;
+        break;
+    }
+
     return 0;
 }
 
@@ -317,7 +276,7 @@ const AVFilter ff_af_anlmf = {
     .activate       = activate,
     FILTER_INPUTS(inputs),
     FILTER_OUTPUTS(outputs),
-    FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_FLTP),
+    FILTER_QUERY_FUNC(query_formats),
     .flags          = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL |
                       AVFILTER_FLAG_SLICE_THREADS,
     .process_command = ff_filter_process_command,
diff --git a/libavfilter/anlms_template.c b/libavfilter/anlms_template.c
new file mode 100644
index 0000000000..b25df4fa18
--- /dev/null
+++ b/libavfilter/anlms_template.c
@@ -0,0 +1,141 @@
+/*
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#undef ONE
+#undef ftype
+#undef SAMPLE_FORMAT
+#if DEPTH == 32
+#define SAMPLE_FORMAT float
+#define ftype float
+#define ONE 1.f
+#else
+#define SAMPLE_FORMAT double
+#define ftype double
+#define ONE 1.0
+#endif
+
+#define fn3(a,b)   a##_##b
+#define fn2(a,b)   fn3(a,b)
+#define fn(a)      fn2(a, SAMPLE_FORMAT)
+
+#if DEPTH == 64
+static double scalarproduct_double(const double *v1, const double *v2, int len)
+{
+    double p = 0.0;
+
+    for (int i = 0; i < len; i++)
+        p += v1[i] * v2[i];
+
+    return p;
+}
+#endif
+
+static ftype fn(fir_sample)(AudioNLMSContext *s, ftype sample, ftype *delay,
+                            ftype *coeffs, ftype *tmp, int *offset)
+{
+    const int order = s->order;
+    ftype output;
+
+    delay[*offset] = sample;
+
+    memcpy(tmp, coeffs + order - *offset, order * sizeof(ftype));
+
+#if DEPTH == 32
+    output = s->fdsp->scalarproduct_float(delay, tmp, s->kernel_size);
+#else
+    output = scalarproduct_double(delay, tmp, s->kernel_size);
+#endif
+
+    if (--(*offset) < 0)
+        *offset = order - 1;
+
+    return output;
+}
+
+static ftype fn(process_sample)(AudioNLMSContext *s, ftype input, ftype desired,
+                                ftype *delay, ftype *coeffs, ftype *tmp, int *offsetp)
+{
+    const int order = s->order;
+    const ftype leakage = s->leakage;
+    const ftype mu = s->mu;
+    const ftype a = ONE - leakage;
+    ftype sum, output, e, norm, b;
+    int offset = *offsetp;
+
+    delay[offset + order] = input;
+
+    output = fn(fir_sample)(s, input, delay, coeffs, tmp, offsetp);
+    e = desired - output;
+
+#if DEPTH == 32
+    sum = s->fdsp->scalarproduct_float(delay, delay, s->kernel_size);
+#else
+    sum = scalarproduct_double(delay, delay, s->kernel_size);
+#endif
+    norm = s->eps + sum;
+    b = mu * e / norm;
+    if (s->anlmf)
+        b *= e * e;
+
+    memcpy(tmp, delay + offset, order * sizeof(ftype));
+
+#if DEPTH == 32
+    s->fdsp->vector_fmul_scalar(coeffs, coeffs, a, s->kernel_size);
+    s->fdsp->vector_fmac_scalar(coeffs, tmp, b, s->kernel_size);
+#else
+    s->fdsp->vector_dmul_scalar(coeffs, coeffs, a, s->kernel_size);
+    s->fdsp->vector_dmac_scalar(coeffs, tmp, b, s->kernel_size);
+#endif
+
+    memcpy(coeffs + order, coeffs, order * sizeof(ftype));
+
+    switch (s->output_mode) {
+    case IN_MODE:       output = input;         break;
+    case DESIRED_MODE:  output = desired;       break;
+    case OUT_MODE:   output = desired - output; break;
+    case NOISE_MODE: output = input - output;   break;
+    case ERROR_MODE:                            break;
+    }
+    return output;
+}
+
+static int fn(filter_channels)(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
+{
+    AudioNLMSContext *s = ctx->priv;
+    AVFrame *out = arg;
+    const int start = (out->ch_layout.nb_channels * jobnr) / nb_jobs;
+    const int end = (out->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
+
+    for (int c = start; c < end; c++) {
+        const ftype *input = (const ftype *)s->frame[0]->extended_data[c];
+        const ftype *desired = (const ftype *)s->frame[1]->extended_data[c];
+        ftype *delay = (ftype *)s->delay->extended_data[c];
+        ftype *coeffs = (ftype *)s->coeffs->extended_data[c];
+        ftype *tmp = (ftype *)s->tmp->extended_data[c];
+        int *offset = (int *)s->offset->extended_data[c];
+        ftype *output = (ftype *)out->extended_data[c];
+
+        for (int n = 0; n < out->nb_samples; n++) {
+            output[n] = fn(process_sample)(s, input[n], desired[n], delay, coeffs, tmp, offset);
+            if (ctx->is_disabled)
+                output[n] = input[n];
+        }
+    }
+
+    return 0;
+}