Previous filter output was incorrect. New one actually follows graph in comments described on side of filter taps.
Showing the changes to libavfilter (1 of 3 files). Show all
Unified · Split
| @@ -34,9 +34,9 @@ · 33 unmodified lines | |||
| 34 | #include "audio.h" | 34 | #include "audio.h" |
| 35 | #include "formats.h" | 35 | #include "formats.h" |
| 36 | 36 | ||
| 37 | #define NUMTAPS 64 | 37 | #define NUMTAPS 32 |
| 38 | 38 | ||
| 39 | static const int8_t filt[NUMTAPS] = { | 39 | static const int8_t filt[NUMTAPS * 2] = { |
| 40 | /* 30° 330° */ | 40 | /* 30° 330° */ |
| 41 | 4, -6, /* 32 tap stereo FIR filter. */ | 41 | 4, -6, /* 32 tap stereo FIR filter. */ |
| 42 | 4, -11, /* One side filters as if the */ | 42 | 4, -11, /* One side filters as if the */ |
| @@ -72,7 +72,10 @@ static const int8_t filt[NUMTAPS] = { · 29 unmodified lines | |||
| 72 | 4, 0}; | 72 | 4, 0}; |
| 73 | 73 | ||
| 74 | typedef struct EarwaxContext { | 74 | typedef struct EarwaxContext { |
| 75 | int16_t taps[NUMTAPS * 2]; | 75 | int16_t filter[2][NUMTAPS]; |
| 76 | int16_t taps[4][NUMTAPS * 2]; | ||
| 77 | |||
| 78 | AVFrame *frame[2]; | ||
| 76 | } EarwaxContext; | 79 | } EarwaxContext; |
| 77 | 80 | ||
| 78 | static int query_formats(AVFilterContext *ctx) | 81 | static int query_formats(AVFilterContext *ctx) |
| @@ -83,7 +86,7 @@ static int query_formats(AVFilterContext *ctx) · 4 unmodified lines | |||
| 83 | AVFilterFormats *formats = NULL; | 86 | AVFilterFormats *formats = NULL; |
| 84 | AVFilterChannelLayouts *layout = NULL; | 87 | AVFilterChannelLayouts *layout = NULL; |
| 85 | 88 | ||
| 86 | if ((ret = ff_add_format (&formats, AV_SAMPLE_FMT_S16 )) < 0 || | 89 | if ((ret = ff_add_format (&formats, AV_SAMPLE_FMT_S16P )) < 0 || |
| 87 | (ret = ff_set_common_formats (ctx , formats )) < 0 || | 90 | (ret = ff_set_common_formats (ctx , formats )) < 0 || |
| 88 | (ret = ff_add_channel_layout (&layout , AV_CH_LAYOUT_STEREO )) < 0 || | 91 | (ret = ff_add_channel_layout (&layout , AV_CH_LAYOUT_STEREO )) < 0 || |
| 89 | (ret = ff_set_common_channel_layouts (ctx , layout )) < 0 || | 92 | (ret = ff_set_common_channel_layouts (ctx , layout )) < 0 || |
| @@ -94,7 +97,8 @@ static int query_formats(AVFilterContext *ctx) · 4 unmodified lines | |||
| 94 | } | 97 | } |
| 95 | 98 | ||
| 96 | //FIXME: replace with DSPContext.scalarproduct_int16 | 99 | //FIXME: replace with DSPContext.scalarproduct_int16 |
| 97 | static inline int16_t *scalarproduct(const int16_t *in, const int16_t *endin, int16_t *out) | 100 | static inline int16_t *scalarproduct(const int16_t *in, const int16_t *endin, |
| 101 | const int16_t *filt, int16_t *out) | ||
| 98 | { | 102 | { |
| 99 | int32_t sample; | 103 | int32_t sample; |
| 100 | int16_t j; | 104 | int16_t j; |
| @@ -103,7 +107,7 @@ static inline int16_t *scalarproduct(const int16_t *in, const int16_t *endin, in · 2 unmodified lines | |||
| 103 | sample = 0; | 107 | sample = 0; |
| 104 | for (j = 0; j < NUMTAPS; j++) | 108 | for (j = 0; j < NUMTAPS; j++) |
| 105 | sample += in[j] * filt[j]; | 109 | sample += in[j] * filt[j]; |
| 106 | *out = av_clip_int16(sample >> 6); | 110 | *out = av_clip_int16(sample >> 7); |
| 107 | out++; | 111 | out++; |
| 108 | in++; | 112 | in++; |
| 109 | } | 113 | } |
| @@ -111,40 +115,102 @@ static inline int16_t *scalarproduct(const int16_t *in, const int16_t *endin, in · 1 unmodified line | |||
| 111 | return out; | 115 | return out; |
| 112 | } | 116 | } |
| 113 | 117 | ||
| 114 | static int filter_frame(AVFilterLink *inlink, AVFrame *insamples) | 118 | static int config_input(AVFilterLink *inlink) |
| 115 | { | 119 | { |
| 116 | AVFilterLink *outlink = inlink->dst->outputs[0]; | 120 | EarwaxContext *s = inlink->dst->priv; |
| 117 | int16_t *taps, *endin, *in, *out; | ||
| 118 | AVFrame *outsamples = ff_get_audio_buffer(outlink, insamples->nb_samples); | ||
| 119 | int len; | ||
| 120 | 121 | ||
| 121 | if (!outsamples) { | 122 | for (int i = 0; i < NUMTAPS; i++) { |
| 122 | av_frame_free(&insamples); | 123 | s->filter[0][i] = filt[i * 2]; |
| 123 | return AVERROR(ENOMEM); | 124 | s->filter[1][i] = filt[i * 2 + 1]; |
| 124 | } | 125 | } |
| 125 | av_frame_copy_props(outsamples, insamples); | ||
| 126 | 126 | ||
| 127 | taps = ((EarwaxContext *)inlink->dst->priv)->taps; | 127 | return 0; |
| 128 | out = (int16_t *)outsamples->data[0]; | 128 | } |
| 129 | in = (int16_t *)insamples ->data[0]; | 129 | |
| 130 | static void convolve(AVFilterContext *ctx, AVFrame *in, | ||
| 131 | int input_ch, int output_ch, | ||
| 132 | int filter_ch, int tap_ch) | ||
| 133 | { | ||
| 134 | EarwaxContext *s = ctx->priv; | ||
| 135 | int16_t *taps, *endin, *dst, *src; | ||
| 136 | int len; | ||
| 137 | |||
| 138 | taps = s->taps[tap_ch]; | ||
| 139 | dst = (int16_t *)s->frame[input_ch]->data[output_ch]; | ||
| 140 | src = (int16_t *)in->data[input_ch]; | ||
| 130 | 141 | ||
| 131 | len = FFMIN(NUMTAPS, 2*insamples->nb_samples); | 142 | len = FFMIN(NUMTAPS, in->nb_samples); |
| 132 | // copy part of new input and process with saved input | 143 | // copy part of new input and process with saved input |
| 133 | memcpy(taps+NUMTAPS, in, len * sizeof(*taps)); | 144 | memcpy(taps+NUMTAPS, src, len * sizeof(*taps)); |
| 134 | out = scalarproduct(taps, taps + len, out); | 145 | dst = scalarproduct(taps, taps + len, s->filter[filter_ch], dst); |
| 135 | 146 | ||
| 136 | // process current input | 147 | // process current input |
| 137 | if (2*insamples->nb_samples >= NUMTAPS ){ | 148 | if (2*in->nb_samples >= NUMTAPS ){ |
| 138 | endin = in + insamples->nb_samples * 2 - NUMTAPS; | 149 | endin = src + in->nb_samples - NUMTAPS; |
| 139 | scalarproduct(in, endin, out); | 150 | scalarproduct(src, endin, s->filter[filter_ch], dst); |
| 140 | 151 | ||
| 141 | // save part of input for next round | 152 | // save part of input for next round |
| 142 | memcpy(taps, endin, NUMTAPS * sizeof(*taps)); | 153 | memcpy(taps, endin, NUMTAPS * sizeof(*taps)); |
| 143 | } else | 154 | } else { |
| 144 | memmove(taps, taps + 2*insamples->nb_samples, NUMTAPS * sizeof(*taps)); | 155 | memmove(taps, taps + in->nb_samples, NUMTAPS * sizeof(*taps)); |
| 156 | } | ||
| 157 | } | ||
| 158 | |||
| 159 | static void mix(AVFilterContext *ctx, AVFrame *out, | ||
| 160 | int output_ch, int f0, int f1, int i0, int i1) | ||
| 161 | { | ||
| 162 | EarwaxContext *s = ctx->priv; | ||
| 163 | const int16_t *srcl = (const int16_t *)s->frame[f0]->data[i0]; | ||
| 164 | const int16_t *srcr = (const int16_t *)s->frame[f1]->data[i1]; | ||
| 165 | int16_t *dst = (int16_t *)out->data[output_ch]; | ||
| 166 | |||
| 167 | for (int n = 0; n < out->nb_samples; n++) | ||
| 168 | dst[n] = av_clip_int16(srcl[n] + srcr[n]); | ||
| 169 | } | ||
| 170 | |||
| 171 | static int filter_frame(AVFilterLink *inlink, AVFrame *in) | ||
| 172 | { | ||
| 173 | AVFilterContext *ctx = inlink->dst; | ||
| 174 | EarwaxContext *s = ctx->priv; | ||
| 175 | AVFilterLink *outlink = ctx->outputs[0]; | ||
| 176 | AVFrame *out = ff_get_audio_buffer(outlink, in->nb_samples); | ||
| 177 | |||
| 178 | for (int ch = 0; ch < 2; ch++) { | ||
| 179 | if (!s->frame[ch] || s->frame[ch]->nb_samples < in->nb_samples) { | ||
| 180 | av_frame_free(&s->frame[ch]); | ||
| 181 | s->frame[ch] = ff_get_audio_buffer(outlink, in->nb_samples); | ||
| 182 | if (!s->frame[ch]) { | ||
| 183 | av_frame_free(&in); | ||
| 184 | av_frame_free(&out); | ||
| 185 | return AVERROR(ENOMEM); | ||
| 186 | } | ||
| 187 | } | ||
| 188 | } | ||
| 189 | |||
| 190 | if (!out) { | ||
| 191 | av_frame_free(&in); | ||
| 192 | return AVERROR(ENOMEM); | ||
| 193 | } | ||
| 194 | av_frame_copy_props(out, in); | ||
| 195 | |||
| 196 | convolve(ctx, in, 0, 0, 0, 0); | ||
| 197 | convolve(ctx, in, 0, 1, 1, 1); | ||
| 198 | convolve(ctx, in, 1, 0, 0, 2); | ||
| 199 | convolve(ctx, in, 1, 1, 1, 3); | ||
| 200 | |||
| 201 | mix(ctx, out, 0, 0, 1, 1, 0); | ||
| 202 | mix(ctx, out, 1, 0, 1, 0, 1); | ||
| 203 | |||
| 204 | av_frame_free(&in); | ||
| 205 | return ff_filter_frame(outlink, out); | ||
| 206 | } | ||
| 207 | |||
| 208 | static av_cold void uninit(AVFilterContext *ctx) | ||
| 209 | { | ||
| 210 | EarwaxContext *s = ctx->priv; | ||
| 145 | 211 | ||
| 146 | av_frame_free(&insamples); | 212 | av_frame_free(&s->frame[0]); |
| 147 | return ff_filter_frame(outlink, outsamples); | 213 | av_frame_free(&s->frame[1]); |
| 148 | } | 214 | } |
| 149 | 215 | ||
| 150 | static const AVFilterPad earwax_inputs[] = { | 216 | static const AVFilterPad earwax_inputs[] = { |
| @@ -152,6 +218,7 @@ static const AVFilterPad earwax_inputs[] = { · 1 unmodified line | |||
| 152 | .name = "default", | 218 | .name = "default", |
| 153 | .type = AVMEDIA_TYPE_AUDIO, | 219 | .type = AVMEDIA_TYPE_AUDIO, |
| 154 | .filter_frame = filter_frame, | 220 | .filter_frame = filter_frame, |
| 221 | .config_props = config_input, | ||
| 155 | }, | 222 | }, |
| 156 | { NULL } | 223 | { NULL } |
| 157 | }; | 224 | }; |
| @@ -169,6 +236,7 @@ AVFilter ff_af_earwax = { · 11 unmodified lines | |||
| 169 | .description = NULL_IF_CONFIG_SMALL("Widen the stereo image."), | 236 | .description = NULL_IF_CONFIG_SMALL("Widen the stereo image."), |
| 170 | .query_formats = query_formats, | 237 | .query_formats = query_formats, |
| 171 | .priv_size = sizeof(EarwaxContext), | 238 | .priv_size = sizeof(EarwaxContext), |
| 239 | .uninit = uninit, | ||
| 172 | .inputs = earwax_inputs, | 240 | .inputs = earwax_inputs, |
| 173 | .outputs = earwax_outputs, | 241 | .outputs = earwax_outputs, |
| 174 | }; | 242 | }; |