]> git.sesse.net Git - ffmpeg/blobdiff - libavfilter/af_aiir.c
avfilter/af_afir: draw IR frequency response
[ffmpeg] / libavfilter / af_aiir.c
index 6ff48e45aa1351a3e1d0314be49765984d53533d..1f2a568c1d906713358e58dfda45c3ecf9808eee 100644 (file)
 #include "avfilter.h"
 #include "internal.h"
 
+typedef struct ThreadData {
+    AVFrame *in, *out;
+} ThreadData;
+
+typedef struct Pair {
+    int a, b;
+} Pair;
+
+typedef struct BiquadContext {
+    double a0, a1, a2;
+    double b0, b1, b2;
+    double i1, i2;
+    double o1, o2;
+} BiquadContext;
+
+typedef struct IIRChannel {
+    int nb_ab[2];
+    double *ab[2];
+    double g;
+    double *cache[2];
+    BiquadContext *biquads;
+    int clippings;
+} IIRChannel;
+
 typedef struct AudioIIRContext {
     const AVClass *class;
     char *a_str, *b_str, *g_str;
     double dry_gain, wet_gain;
     int format;
+    int process;
+    int precision;
 
-    int *nb_a, *nb_b;
-    double **a, **b;
-    double *g;
-    double **input, **output;
-    int clippings;
+    IIRChannel *iir;
     int channels;
+    enum AVSampleFormat sample_format;
 
-    void (*iir_frame)(AVFilterContext *ctx, AVFrame *in, AVFrame *out);
+    int (*iir_channel)(AVFilterContext *ctx, void *arg, int ch, int nb_jobs);
 } AudioIIRContext;
 
 static int query_formats(AVFilterContext *ctx)
 {
+    AudioIIRContext *s = ctx->priv;
     AVFilterFormats *formats;
     AVFilterChannelLayouts *layouts;
-    static const enum AVSampleFormat sample_fmts[] = {
+    enum AVSampleFormat sample_fmts[] = {
         AV_SAMPLE_FMT_DBLP,
-        AV_SAMPLE_FMT_FLTP,
-        AV_SAMPLE_FMT_S32P,
-        AV_SAMPLE_FMT_S16P,
         AV_SAMPLE_FMT_NONE
     };
     int ret;
@@ -63,6 +84,7 @@ static int query_formats(AVFilterContext *ctx)
     if (ret < 0)
         return ret;
 
+    sample_fmts[0] = s->sample_format;
     formats = ff_make_format_list(sample_fmts);
     if (!formats)
         return AVERROR(ENOMEM);
@@ -76,56 +98,118 @@ static int query_formats(AVFilterContext *ctx)
     return ff_set_common_samplerates(ctx, formats);
 }
 
-#define IIR_FRAME(name, type, min, max, need_clipping)                  \
-static void iir_frame_## name(AVFilterContext *ctx, AVFrame *in, AVFrame *out)  \
+#define IIR_CH(name, type, min, max, need_clipping)                     \
+static int iir_ch_## name(AVFilterContext *ctx, void *arg, int ch, int nb_jobs)  \
 {                                                                       \
     AudioIIRContext *s = ctx->priv;                                     \
     const double ig = s->dry_gain;                                      \
     const double og = s->wet_gain;                                      \
-    int ch, n;                                                          \
+    ThreadData *td = arg;                                               \
+    AVFrame *in = td->in, *out = td->out;                               \
+    const type *src = (const type *)in->extended_data[ch];              \
+    double *ic = (double *)s->iir[ch].cache[0];                         \
+    double *oc = (double *)s->iir[ch].cache[1];                         \
+    const int nb_a = s->iir[ch].nb_ab[0];                               \
+    const int nb_b = s->iir[ch].nb_ab[1];                               \
+    const double *a = s->iir[ch].ab[0];                                 \
+    const double *b = s->iir[ch].ab[1];                                 \
+    int *clippings = &s->iir[ch].clippings;                             \
+    type *dst = (type *)out->extended_data[ch];                         \
+    int n;                                                              \
                                                                         \
-    for (ch = 0; ch < out->channels; ch++) {                            \
-        const type *src = (const type *)in->extended_data[ch];          \
-        double *ic = (double *)s->input[ch];                            \
-        double *oc = (double *)s->output[ch];                           \
-        const int nb_a = s->nb_a[ch];                                   \
-        const int nb_b = s->nb_b[ch];                                   \
-        const double *a = s->a[ch];                                     \
-        const double *b = s->b[ch];                                     \
-        type *dst = (type *)out->extended_data[ch];                     \
+    for (n = 0; n < in->nb_samples; n++) {                              \
+        double sample = 0.;                                             \
+        int x;                                                          \
                                                                         \
-        for (n = 0; n < in->nb_samples; n++) {                          \
-            double sample = 0.;                                         \
-            int x;                                                      \
+        memmove(&ic[1], &ic[0], (nb_b - 1) * sizeof(*ic));              \
+        memmove(&oc[1], &oc[0], (nb_a - 1) * sizeof(*oc));              \
+        ic[0] = src[n] * ig;                                            \
+        for (x = 0; x < nb_b; x++)                                      \
+            sample += b[x] * ic[x];                                     \
                                                                         \
-            memmove(&ic[1], &ic[0], (nb_b - 1) * sizeof(*ic));          \
-            memmove(&oc[1], &oc[0], (nb_a - 1) * sizeof(*oc));          \
-            ic[0] = src[n] * ig;                                        \
-            for (x = 0; x < nb_b; x++)                                  \
-                sample += b[x] * ic[x];                                 \
+        for (x = 1; x < nb_a; x++)                                      \
+            sample -= a[x] * oc[x];                                     \
                                                                         \
-            for (x = 1; x < nb_a; x++)                                  \
-                sample -= a[x] * oc[x];                                 \
+        oc[0] = sample;                                                 \
+        sample *= og;                                                   \
+        if (need_clipping && sample < min) {                            \
+            (*clippings)++;                                             \
+            dst[n] = min;                                               \
+        } else if (need_clipping && sample > max) {                     \
+            (*clippings)++;                                             \
+            dst[n] = max;                                               \
+        } else {                                                        \
+            dst[n] = sample;                                            \
+        }                                                               \
+    }                                                                   \
                                                                         \
-            oc[0] = sample;                                             \
-            sample *= og;                                               \
-            if (need_clipping && sample < min) {                        \
-                s->clippings++;                                         \
+    return 0;                                                           \
+}
+
+IIR_CH(s16p, int16_t, INT16_MIN, INT16_MAX, 1)
+IIR_CH(s32p, int32_t, INT32_MIN, INT32_MAX, 1)
+IIR_CH(fltp, float,         -1.,        1., 0)
+IIR_CH(dblp, double,        -1.,        1., 0)
+
+#define SERIAL_IIR_CH(name, type, min, max, need_clipping)                  \
+static int iir_ch_serial_## name(AVFilterContext *ctx, void *arg, int ch, int nb_jobs)  \
+{                                                                       \
+    AudioIIRContext *s = ctx->priv;                                     \
+    const double ig = s->dry_gain;                                      \
+    const double og = s->wet_gain;                                      \
+    ThreadData *td = arg;                                               \
+    AVFrame *in = td->in, *out = td->out;                               \
+    const type *src = (const type *)in->extended_data[ch];              \
+    type *dst = (type *)out->extended_data[ch];                         \
+    IIRChannel *iir = &s->iir[ch];                                      \
+    int *clippings = &iir->clippings;                                   \
+    int nb_biquads = (FFMAX(iir->nb_ab[0], iir->nb_ab[1]) + 1) / 2;     \
+    int n, i;                                                           \
+                                                                        \
+    for (i = 0; i < nb_biquads; i++) {                                  \
+        const double a1 = -iir->biquads[i].a1;                          \
+        const double a2 = -iir->biquads[i].a2;                          \
+        const double b0 = iir->biquads[i].b0;                           \
+        const double b1 = iir->biquads[i].b1;                           \
+        const double b2 = iir->biquads[i].b2;                           \
+        double i1 = iir->biquads[i].i1;                                 \
+        double i2 = iir->biquads[i].i2;                                 \
+        double o1 = iir->biquads[i].o1;                                 \
+        double o2 = iir->biquads[i].o2;                                 \
+                                                                        \
+        for (n = 0; n < in->nb_samples; n++) {                          \
+            double sample = ig * (i ? dst[n] : src[n]);                 \
+            double o0 = sample * b0 + i1 * b1 + i2 * b2 + o1 * a1 + o2 * a2; \
+                                                                        \
+            i2 = i1;                                                    \
+            i1 = src[n];                                                \
+            o2 = o1;                                                    \
+            o1 = o0;                                                    \
+            o0 *= og;                                                   \
+                                                                        \
+            if (need_clipping && o0 < min) {                            \
+                (*clippings)++;                                         \
                 dst[n] = min;                                           \
-            } else if (need_clipping && sample > max) {                 \
-                s->clippings++;                                         \
+            } else if (need_clipping && o0 > max) {                     \
+                (*clippings)++;                                         \
                 dst[n] = max;                                           \
             } else {                                                    \
-                dst[n] = sample;                                        \
+                dst[n] = o0;                                            \
             }                                                           \
         }                                                               \
+        iir->biquads[i].i1 = i1;                                        \
+        iir->biquads[i].i2 = i2;                                        \
+        iir->biquads[i].o1 = o1;                                        \
+        iir->biquads[i].o2 = o2;                                        \
     }                                                                   \
+                                                                        \
+    return 0;                                                           \
 }
 
-IIR_FRAME(s16p, int16_t, INT16_MIN, INT16_MAX, 1)
-IIR_FRAME(s32p, int32_t, INT32_MIN, INT32_MAX, 1)
-IIR_FRAME(fltp, float,         -1.,        1., 0)
-IIR_FRAME(dblp, double,        -1.,        1., 0)
+SERIAL_IIR_CH(s16p, int16_t, INT16_MIN, INT16_MAX, 1)
+SERIAL_IIR_CH(s32p, int32_t, INT32_MIN, INT32_MAX, 1)
+SERIAL_IIR_CH(fltp, float,         -1.,        1., 0)
+SERIAL_IIR_CH(dblp, double,        -1.,        1., 0)
 
 static void count_coefficients(char *item_str, int *nb_items)
 {
@@ -141,8 +225,9 @@ static void count_coefficients(char *item_str, int *nb_items)
     }
 }
 
-static int read_gains(AVFilterContext *ctx, char *item_str, int nb_items, double *dst)
+static int read_gains(AVFilterContext *ctx, char *item_str, int nb_items)
 {
+    AudioIIRContext *s = ctx->priv;
     char *p, *arg, *old_str, *prev_arg = NULL, *saveptr = NULL;
     int i;
 
@@ -159,7 +244,7 @@ static int read_gains(AVFilterContext *ctx, char *item_str, int nb_items, double
         }
 
         p = NULL;
-        if (sscanf(arg, "%lf", &dst[i]) != 1) {
+        if (sscanf(arg, "%lf", &s->iir[i].g) != 1) {
             av_log(ctx, AV_LOG_ERROR, "Invalid gains supplied: %s\n", arg);
             av_freep(&old_str);
             return AVERROR(EINVAL);
@@ -198,7 +283,7 @@ static int read_tf_coefficients(AVFilterContext *ctx, char *item_str, int nb_ite
     return 0;
 }
 
-static int read_zp_coefficients(AVFilterContext *ctx, char *item_str, int nb_items, double *dst)
+static int read_zp_coefficients(AVFilterContext *ctx, char *item_str, int nb_items, double *dst, const char *format)
 {
     char *p, *arg, *old_str, *saveptr = NULL;
     int i;
@@ -211,7 +296,7 @@ static int read_zp_coefficients(AVFilterContext *ctx, char *item_str, int nb_ite
             break;
 
         p = NULL;
-        if (sscanf(arg, "%lf %lfi", &dst[i*2], &dst[i*2+1]) != 2) {
+        if (sscanf(arg, format, &dst[i*2], &dst[i*2+1]) != 2) {
             av_log(ctx, AV_LOG_ERROR, "Invalid coefficients supplied: %s\n", arg);
             av_freep(&old_str);
             return AVERROR(EINVAL);
@@ -223,7 +308,9 @@ static int read_zp_coefficients(AVFilterContext *ctx, char *item_str, int nb_ite
     return 0;
 }
 
-static int read_channels(AVFilterContext *ctx, int channels, uint8_t *item_str, int *nb, double **c, double **cache)
+static const char *format[] = { "%lf", "%lf %lfi", "%lf %lfr", "%lf %lfd" };
+
+static int read_channels(AVFilterContext *ctx, int channels, uint8_t *item_str, int ab)
 {
     AudioIIRContext *s = ctx->priv;
     char *p, *arg, *old_str, *prev_arg = NULL, *saveptr = NULL;
@@ -233,6 +320,8 @@ static int read_channels(AVFilterContext *ctx, int channels, uint8_t *item_str,
     if (!p)
         return AVERROR(ENOMEM);
     for (i = 0; i < channels; i++) {
+        IIRChannel *iir = &s->iir[i];
+
         if (!(arg = av_strtok(p, "|", &saveptr)))
             arg = prev_arg;
 
@@ -241,20 +330,20 @@ static int read_channels(AVFilterContext *ctx, int channels, uint8_t *item_str,
             return AVERROR(EINVAL);
         }
 
-        count_coefficients(arg, &nb[i]);
+        count_coefficients(arg, &iir->nb_ab[ab]);
 
         p = NULL;
-        cache[i] = av_calloc(nb[i] + 1, sizeof(double));
-        c[i] = av_calloc(nb[i] * (s->format + 1), sizeof(double));
-        if (!c[i] || !cache[i]) {
+        iir->cache[ab] = av_calloc(iir->nb_ab[ab] + 1, sizeof(double));
+        iir->ab[ab] = av_calloc(iir->nb_ab[ab] * (!!s->format + 1), sizeof(double));
+        if (!iir->ab[ab] || !iir->cache[ab]) {
             av_freep(&old_str);
             return AVERROR(ENOMEM);
         }
 
         if (s->format) {
-            ret = read_zp_coefficients(ctx, arg, nb[i], c[i]);
+            ret = read_zp_coefficients(ctx, arg, iir->nb_ab[ab], iir->ab[ab], format[s->format]);
         } else {
-            ret = read_tf_coefficients(ctx, arg, nb[i], c[i]);
+            ret = read_tf_coefficients(ctx, arg, iir->nb_ab[ab], iir->ab[ab]);
         }
         if (ret < 0) {
             av_freep(&old_str);
@@ -304,7 +393,7 @@ static int expand(AVFilterContext *ctx, double *pz, int nb, double *coeffs)
         multiply(pz[2 * i], pz[2 * i + 1], nb, coeffs);
 
     for (i = 0; i < nb + 1; i++) {
-        if (fabs(coeffs[2 * i + 1]) > DBL_EPSILON) {
+        if (fabs(coeffs[2 * i + 1]) > FLT_EPSILON) {
             av_log(ctx, AV_LOG_ERROR, "coeff: %lf of z^%d is not real; poles/zeros are not complex conjugates.\n",
                    coeffs[2 * i + 1], i);
             return AVERROR(EINVAL);
@@ -317,47 +406,242 @@ static int expand(AVFilterContext *ctx, double *pz, int nb, double *coeffs)
 static int convert_zp2tf(AVFilterContext *ctx, int channels)
 {
     AudioIIRContext *s = ctx->priv;
-    int ch, i, j, ret;
+    int ch, i, j, ret = 0;
 
     for (ch = 0; ch < channels; ch++) {
+        IIRChannel *iir = &s->iir[ch];
         double *topc, *botc;
 
-        topc = av_calloc((s->nb_b[ch] + 1) * 2, sizeof(*topc));
-        botc = av_calloc((s->nb_a[ch] + 1) * 2, sizeof(*botc));
-        if (!topc || !botc)
-            return AVERROR(ENOMEM);
+        topc = av_calloc((iir->nb_ab[0] + 1) * 2, sizeof(*topc));
+        botc = av_calloc((iir->nb_ab[1] + 1) * 2, sizeof(*botc));
+        if (!topc || !botc) {
+            ret = AVERROR(ENOMEM);
+            goto fail;
+        }
 
-        ret = expand(ctx, s->a[ch], s->nb_a[ch], botc);
+        ret = expand(ctx, iir->ab[0], iir->nb_ab[0], botc);
         if (ret < 0) {
-            av_free(topc);
-            av_free(botc);
-            return ret;
+            goto fail;
         }
 
-        ret = expand(ctx, s->b[ch], s->nb_b[ch], topc);
+        ret = expand(ctx, iir->ab[1], iir->nb_ab[1], topc);
         if (ret < 0) {
-            av_free(topc);
-            av_free(botc);
-            return ret;
+            goto fail;
         }
 
-        for (j = 0, i = s->nb_b[ch]; i >= 0; j++, i--) {
-            s->b[ch][j] = topc[2 * i];
+        for (j = 0, i = iir->nb_ab[1]; i >= 0; j++, i--) {
+            iir->ab[1][j] = topc[2 * i];
         }
-        s->nb_b[ch]++;
+        iir->nb_ab[1]++;
 
-        for (j = 0, i = s->nb_a[ch]; i >= 0; j++, i--) {
-            s->a[ch][j] = botc[2 * i];
+        for (j = 0, i = iir->nb_ab[0]; i >= 0; j++, i--) {
+            iir->ab[0][j] = botc[2 * i];
         }
-        s->nb_a[ch]++;
+        iir->nb_ab[0]++;
 
+fail:
         av_free(topc);
         av_free(botc);
+        if (ret < 0)
+            break;
+    }
+
+    return ret;
+}
+
+static int decompose_zp2biquads(AVFilterContext *ctx, int channels)
+{
+    AudioIIRContext *s = ctx->priv;
+    int ch, ret;
+
+    for (ch = 0; ch < channels; ch++) {
+        IIRChannel *iir = &s->iir[ch];
+        int nb_biquads = (FFMAX(iir->nb_ab[0], iir->nb_ab[1]) + 1) / 2;
+        int current_biquad = 0;
+
+        iir->biquads = av_calloc(nb_biquads, sizeof(BiquadContext));
+        if (!iir->biquads)
+            return AVERROR(ENOMEM);
+
+        while (nb_biquads--) {
+            Pair outmost_pole = { -1, -1 };
+            Pair nearest_zero = { -1, -1 };
+            double zeros[4] = { 0 };
+            double poles[4] = { 0 };
+            double b[6] = { 0 };
+            double a[6] = { 0 };
+            double min_distance = DBL_MAX;
+            double max_mag = 0;
+            int i;
+
+            for (i = 0; i < iir->nb_ab[0]; i++) {
+                double mag;
+
+                if (isnan(iir->ab[0][2 * i]) || isnan(iir->ab[0][2 * i + 1]))
+                    continue;
+                mag = hypot(iir->ab[0][2 * i], iir->ab[0][2 * i + 1]);
+
+                if (mag > max_mag) {
+                    max_mag = mag;
+                    outmost_pole.a = i;
+                }
+            }
+
+            for (i = 0; i < iir->nb_ab[1]; i++) {
+                if (isnan(iir->ab[0][2 * i]) || isnan(iir->ab[0][2 * i + 1]))
+                    continue;
+
+                if (iir->ab[0][2 * i    ] ==  iir->ab[0][2 * outmost_pole.a    ] &&
+                    iir->ab[0][2 * i + 1] == -iir->ab[0][2 * outmost_pole.a + 1]) {
+                    outmost_pole.b = i;
+                    break;
+                }
+            }
+
+            av_log(ctx, AV_LOG_VERBOSE, "outmost_pole is %d.%d\n", outmost_pole.a, outmost_pole.b);
+
+            if (outmost_pole.a < 0 || outmost_pole.b < 0)
+                return AVERROR(EINVAL);
+
+            for (i = 0; i < iir->nb_ab[1]; i++) {
+                double distance;
+
+                if (isnan(iir->ab[1][2 * i]) || isnan(iir->ab[1][2 * i + 1]))
+                    continue;
+                distance = hypot(iir->ab[0][2 * outmost_pole.a    ] - iir->ab[1][2 * i    ],
+                                 iir->ab[0][2 * outmost_pole.a + 1] - iir->ab[1][2 * i + 1]);
+
+                if (distance < min_distance) {
+                    min_distance = distance;
+                    nearest_zero.a = i;
+                }
+            }
+
+            for (i = 0; i < iir->nb_ab[1]; i++) {
+                if (isnan(iir->ab[1][2 * i]) || isnan(iir->ab[1][2 * i + 1]))
+                    continue;
+
+                if (iir->ab[1][2 * i    ] ==  iir->ab[1][2 * nearest_zero.a    ] &&
+                    iir->ab[1][2 * i + 1] == -iir->ab[1][2 * nearest_zero.a + 1]) {
+                    nearest_zero.b = i;
+                    break;
+                }
+            }
+
+            av_log(ctx, AV_LOG_VERBOSE, "nearest_zero is %d.%d\n", nearest_zero.a, nearest_zero.b);
+
+            if (nearest_zero.a < 0 || nearest_zero.b < 0)
+                return AVERROR(EINVAL);
+
+            poles[0] = iir->ab[0][2 * outmost_pole.a    ];
+            poles[1] = iir->ab[0][2 * outmost_pole.a + 1];
+
+            zeros[0] = iir->ab[1][2 * nearest_zero.a    ];
+            zeros[1] = iir->ab[1][2 * nearest_zero.a + 1];
+
+            if (nearest_zero.a == nearest_zero.b && outmost_pole.a == outmost_pole.b) {
+                zeros[2] = 0;
+                zeros[3] = 0;
+
+                poles[2] = 0;
+                poles[3] = 0;
+            } else {
+                poles[2] = iir->ab[0][2 * outmost_pole.b    ];
+                poles[3] = iir->ab[0][2 * outmost_pole.b + 1];
+
+                zeros[2] = iir->ab[1][2 * nearest_zero.b    ];
+                zeros[3] = iir->ab[1][2 * nearest_zero.b + 1];
+            }
+
+            ret = expand(ctx, zeros, 2, b);
+            if (ret < 0)
+                return ret;
+
+            ret = expand(ctx, poles, 2, a);
+            if (ret < 0)
+                return ret;
+
+            iir->ab[0][2 * outmost_pole.a] = iir->ab[0][2 * outmost_pole.a + 1] = NAN;
+            iir->ab[0][2 * outmost_pole.b] = iir->ab[0][2 * outmost_pole.b + 1] = NAN;
+            iir->ab[1][2 * nearest_zero.a] = iir->ab[1][2 * nearest_zero.a + 1] = NAN;
+            iir->ab[1][2 * nearest_zero.b] = iir->ab[1][2 * nearest_zero.b + 1] = NAN;
+
+            iir->biquads[current_biquad].a0 = 1.0;
+            iir->biquads[current_biquad].a1 = a[2] / a[4];
+            iir->biquads[current_biquad].a2 = a[0] / a[4];
+            iir->biquads[current_biquad].b0 = b[4] / a[4] * (current_biquad ? 1.0 : iir->g);
+            iir->biquads[current_biquad].b1 = b[2] / a[4] * (current_biquad ? 1.0 : iir->g);
+            iir->biquads[current_biquad].b2 = b[0] / a[4] * (current_biquad ? 1.0 : iir->g);
+
+            av_log(ctx, AV_LOG_VERBOSE, "a=%lf %lf %lf:b=%lf %lf %lf\n",
+                   iir->biquads[current_biquad].a0,
+                   iir->biquads[current_biquad].a1,
+                   iir->biquads[current_biquad].a2,
+                   iir->biquads[current_biquad].b0,
+                   iir->biquads[current_biquad].b1,
+                   iir->biquads[current_biquad].b2);
+
+            current_biquad++;
+        }
     }
 
     return 0;
 }
 
+static void convert_pr2zp(AVFilterContext *ctx, int channels)
+{
+    AudioIIRContext *s = ctx->priv;
+    int ch;
+
+    for (ch = 0; ch < channels; ch++) {
+        IIRChannel *iir = &s->iir[ch];
+        int n;
+
+        for (n = 0; n < iir->nb_ab[0]; n++) {
+            double r = iir->ab[0][2*n];
+            double angle = iir->ab[0][2*n+1];
+
+            iir->ab[0][2*n]   = r * cos(angle);
+            iir->ab[0][2*n+1] = r * sin(angle);
+        }
+
+        for (n = 0; n < iir->nb_ab[1]; n++) {
+            double r = iir->ab[1][2*n];
+            double angle = iir->ab[1][2*n+1];
+
+            iir->ab[1][2*n]   = r * cos(angle);
+            iir->ab[1][2*n+1] = r * sin(angle);
+        }
+    }
+}
+
+static void convert_pd2zp(AVFilterContext *ctx, int channels)
+{
+    AudioIIRContext *s = ctx->priv;
+    int ch;
+
+    for (ch = 0; ch < channels; ch++) {
+        IIRChannel *iir = &s->iir[ch];
+        int n;
+
+        for (n = 0; n < iir->nb_ab[0]; n++) {
+            double r = iir->ab[0][2*n];
+            double angle = M_PI*iir->ab[0][2*n+1]/180.;
+
+            iir->ab[0][2*n]   = r * cos(angle);
+            iir->ab[0][2*n+1] = r * sin(angle);
+        }
+
+        for (n = 0; n < iir->nb_ab[1]; n++) {
+            double r = iir->ab[1][2*n];
+            double angle = M_PI*iir->ab[1][2*n+1]/180.;
+
+            iir->ab[1][2*n]   = r * cos(angle);
+            iir->ab[1][2*n+1] = r * sin(angle);
+        }
+    }
+}
+
 static int config_output(AVFilterLink *outlink)
 {
     AVFilterContext *ctx = outlink->src;
@@ -366,49 +650,66 @@ static int config_output(AVFilterLink *outlink)
     int ch, ret, i;
 
     s->channels = inlink->channels;
-    s->a = av_calloc(inlink->channels, sizeof(*s->a));
-    s->b = av_calloc(inlink->channels, sizeof(*s->b));
-    s->g = av_calloc(inlink->channels, sizeof(*s->g));
-    s->nb_a = av_calloc(inlink->channels, sizeof(*s->nb_a));
-    s->nb_b = av_calloc(inlink->channels, sizeof(*s->nb_b));
-    s->input = av_calloc(inlink->channels, sizeof(*s->input));
-    s->output = av_calloc(inlink->channels, sizeof(*s->output));
-    if (!s->a || !s->b || !s->nb_a || !s->nb_b || !s->input || !s->output)
+    s->iir = av_calloc(s->channels, sizeof(*s->iir));
+    if (!s->iir)
         return AVERROR(ENOMEM);
 
-    ret = read_gains(ctx, s->g_str, inlink->channels, s->g);
+    ret = read_gains(ctx, s->g_str, inlink->channels);
     if (ret < 0)
         return ret;
 
-    ret = read_channels(ctx, inlink->channels, s->a_str, s->nb_a, s->a, s->output);
+    ret = read_channels(ctx, inlink->channels, s->a_str, 0);
     if (ret < 0)
         return ret;
 
-    ret = read_channels(ctx, inlink->channels, s->b_str, s->nb_b, s->b, s->input);
+    ret = read_channels(ctx, inlink->channels, s->b_str, 1);
     if (ret < 0)
         return ret;
 
-    if (s->format) {
+    if (s->format == 2) {
+        convert_pr2zp(ctx, inlink->channels);
+    } else if (s->format == 3) {
+        convert_pd2zp(ctx, inlink->channels);
+    }
+
+    if (s->format == 0)
+        av_log(ctx, AV_LOG_WARNING, "tf coefficients format is not recommended for too high number of zeros/poles.\n");
+
+    if (s->format > 0 && s->process == 0) {
+        av_log(ctx, AV_LOG_WARNING, "Direct processsing is not recommended for zp coefficients format.\n");
+
         ret = convert_zp2tf(ctx, inlink->channels);
         if (ret < 0)
             return ret;
+    } else if (s->format == 0 && s->process == 1) {
+        av_log(ctx, AV_LOG_ERROR, "Serial cascading is not implemented for transfer function.\n");
+        return AVERROR_PATCHWELCOME;
+    } else if (s->format > 0 && s->process == 1) {
+        if (inlink->format == AV_SAMPLE_FMT_S16P)
+            av_log(ctx, AV_LOG_WARNING, "Serial cascading is not recommended for i16 precision.\n");
+
+        ret = decompose_zp2biquads(ctx, inlink->channels);
+        if (ret < 0)
+            return ret;
     }
 
     for (ch = 0; ch < inlink->channels; ch++) {
-        for (i = 1; i < s->nb_a[ch]; i++) {
-            s->a[ch][i] /= s->a[ch][0];
+        IIRChannel *iir = &s->iir[ch];
+
+        for (i = 1; i < iir->nb_ab[0]; i++) {
+            iir->ab[0][i] /= iir->ab[0][0];
         }
 
-        for (i = 0; i < s->nb_b[ch]; i++) {
-            s->b[ch][i] *= s->g[ch] / s->a[ch][0];
+        for (i = 0; i < iir->nb_ab[1]; i++) {
+            iir->ab[1][i] *= iir->g / iir->ab[0][0];
         }
     }
 
     switch (inlink->format) {
-    case AV_SAMPLE_FMT_DBLP: s->iir_frame = iir_frame_dblp; break;
-    case AV_SAMPLE_FMT_FLTP: s->iir_frame = iir_frame_fltp; break;
-    case AV_SAMPLE_FMT_S32P: s->iir_frame = iir_frame_s32p; break;
-    case AV_SAMPLE_FMT_S16P: s->iir_frame = iir_frame_s16p; break;
+    case AV_SAMPLE_FMT_DBLP: s->iir_channel = s->process == 1 ? iir_ch_serial_dblp : iir_ch_dblp; break;
+    case AV_SAMPLE_FMT_FLTP: s->iir_channel = s->process == 1 ? iir_ch_serial_fltp : iir_ch_fltp; break;
+    case AV_SAMPLE_FMT_S32P: s->iir_channel = s->process == 1 ? iir_ch_serial_s32p : iir_ch_s32p; break;
+    case AV_SAMPLE_FMT_S16P: s->iir_channel = s->process == 1 ? iir_ch_serial_s16p : iir_ch_s16p; break;
     }
 
     return 0;
@@ -419,7 +720,9 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
     AVFilterContext *ctx = inlink->dst;
     AudioIIRContext *s = ctx->priv;
     AVFilterLink *outlink = ctx->outputs[0];
+    ThreadData td;
     AVFrame *out;
+    int ch;
 
     if (av_frame_is_writable(in)) {
         out = in;
@@ -432,11 +735,16 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in)
         av_frame_copy_props(out, in);
     }
 
-    s->iir_frame(ctx, in, out);
+    td.in  = in;
+    td.out = out;
+    ctx->internal->execute(ctx, s->iir_channel, &td, NULL, outlink->channels);
 
-    if (s->clippings > 0)
-        av_log(ctx, AV_LOG_WARNING, "clipping %d times. Please reduce gain.\n", s->clippings);
-    s->clippings = 0;
+    for (ch = 0; ch < outlink->channels; ch++) {
+        if (s->iir[ch].clippings > 0)
+            av_log(ctx, AV_LOG_WARNING, "Channel %d clipping %d times. Please reduce gain.\n",
+                   ch, s->iir[ch].clippings);
+        s->iir[ch].clippings = 0;
+    }
 
     if (in != out)
         av_frame_free(&in);
@@ -453,6 +761,14 @@ static av_cold int init(AVFilterContext *ctx)
         return AVERROR(EINVAL);
     }
 
+    switch (s->precision) {
+    case 0: s->sample_format = AV_SAMPLE_FMT_DBLP; break;
+    case 1: s->sample_format = AV_SAMPLE_FMT_FLTP; break;
+    case 2: s->sample_format = AV_SAMPLE_FMT_S32P; break;
+    case 3: s->sample_format = AV_SAMPLE_FMT_S16P; break;
+    default: return AVERROR_BUG;
+    }
+
     return 0;
 }
 
@@ -461,27 +777,17 @@ static av_cold void uninit(AVFilterContext *ctx)
     AudioIIRContext *s = ctx->priv;
     int ch;
 
-    if (s->a) {
+    if (s->iir) {
         for (ch = 0; ch < s->channels; ch++) {
-            av_freep(&s->a[ch]);
-            av_freep(&s->output[ch]);
+            IIRChannel *iir = &s->iir[ch];
+            av_freep(&iir->ab[0]);
+            av_freep(&iir->ab[1]);
+            av_freep(&iir->cache[0]);
+            av_freep(&iir->cache[1]);
+            av_freep(&iir->biquads);
         }
     }
-    av_freep(&s->a);
-
-    if (s->b) {
-        for (ch = 0; ch < s->channels; ch++) {
-            av_freep(&s->b[ch]);
-            av_freep(&s->input[ch]);
-        }
-    }
-    av_freep(&s->b);
-
-    av_freep(&s->input);
-    av_freep(&s->output);
-
-    av_freep(&s->nb_a);
-    av_freep(&s->nb_b);
+    av_freep(&s->iir);
 }
 
 static const AVFilterPad inputs[] = {
@@ -506,14 +812,24 @@ static const AVFilterPad outputs[] = {
 #define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
 
 static const AVOption aiir_options[] = {
-    { "z", "set B/numerator/zeros coefficients",   OFFSET(b_str),    AV_OPT_TYPE_STRING, {.str="1 1"}, 0, 0, AF },
-    { "p", "set A/denominator/poles coefficients", OFFSET(a_str),    AV_OPT_TYPE_STRING, {.str="1 1"}, 0, 0, AF },
+    { "z", "set B/numerator/zeros coefficients",   OFFSET(b_str),    AV_OPT_TYPE_STRING, {.str="1+0i 1-0i"}, 0, 0, AF },
+    { "p", "set A/denominator/poles coefficients", OFFSET(a_str),    AV_OPT_TYPE_STRING, {.str="1+0i 1-0i"}, 0, 0, AF },
     { "k", "set channels gains",                   OFFSET(g_str),    AV_OPT_TYPE_STRING, {.str="1|1"}, 0, 0, AF },
     { "dry", "set dry gain",                       OFFSET(dry_gain), AV_OPT_TYPE_DOUBLE, {.dbl=1},     0, 1, AF },
     { "wet", "set wet gain",                       OFFSET(wet_gain), AV_OPT_TYPE_DOUBLE, {.dbl=1},     0, 1, AF },
-    { "f", "set coefficients format",              OFFSET(format),   AV_OPT_TYPE_INT,    {.i64=0},     0, 1, AF, "format" },
+    { "f", "set coefficients format",              OFFSET(format),   AV_OPT_TYPE_INT,    {.i64=1},     0, 3, AF, "format" },
     { "tf", "transfer function",                   0,                AV_OPT_TYPE_CONST,  {.i64=0},     0, 0, AF, "format" },
     { "zp", "Z-plane zeros/poles",                 0,                AV_OPT_TYPE_CONST,  {.i64=1},     0, 0, AF, "format" },
+    { "pr", "Z-plane zeros/poles (polar radians)", 0,                AV_OPT_TYPE_CONST,  {.i64=2},     0, 0, AF, "format" },
+    { "pd", "Z-plane zeros/poles (polar degrees)", 0,                AV_OPT_TYPE_CONST,  {.i64=3},     0, 0, AF, "format" },
+    { "r", "set kind of processing",               OFFSET(process),  AV_OPT_TYPE_INT,    {.i64=1},     0, 1, AF, "process" },
+    { "d", "direct",                               0,                AV_OPT_TYPE_CONST,  {.i64=0},     0, 0, AF, "process" },
+    { "s", "serial cascading",                     0,                AV_OPT_TYPE_CONST,  {.i64=1},     0, 0, AF, "process" },
+    { "e", "set precision",                        OFFSET(precision),AV_OPT_TYPE_INT,    {.i64=0},     0, 3, AF, "precision" },
+    { "dbl", "double-precision floating-point",    0,                AV_OPT_TYPE_CONST,  {.i64=0},     0, 0, AF, "precision" },
+    { "flt", "single-precision floating-point",    0,                AV_OPT_TYPE_CONST,  {.i64=1},     0, 0, AF, "precision" },
+    { "i32", "32-bit integers",                    0,                AV_OPT_TYPE_CONST,  {.i64=2},     0, 0, AF, "precision" },
+    { "i16", "16-bit integers",                    0,                AV_OPT_TYPE_CONST,  {.i64=3},     0, 0, AF, "precision" },
     { NULL },
 };
 
@@ -523,10 +839,11 @@ AVFilter ff_af_aiir = {
     .name          = "aiir",
     .description   = NULL_IF_CONFIG_SMALL("Apply Infinite Impulse Response filter with supplied coefficients."),
     .priv_size     = sizeof(AudioIIRContext),
+    .priv_class    = &aiir_class,
     .init          = init,
     .uninit        = uninit,
     .query_formats = query_formats,
     .inputs        = inputs,
     .outputs       = outputs,
-    .priv_class    = &aiir_class,
+    .flags         = AVFILTER_FLAG_SLICE_THREADS,
 };