diracdsp: add dequantization SIMD
[ffmpeg.git] / libavcodec / resample.c
index fea8e75..4c5eb9f 100644 (file)
  */
 
 /**
- * @file libavcodec/resample.c
+ * @file
  * samplerate conversion for both audio and video
  */
 
+#include <string.h>
+
 #include "avcodec.h"
 #include "audioconvert.h"
-#include "opt.h"
+#include "libavutil/opt.h"
+#include "libavutil/mem.h"
+#include "libavutil/samplefmt.h"
+
+#if FF_API_AVCODEC_RESAMPLE
+FF_DISABLE_DEPRECATION_WARNINGS
+
+#define MAX_CHANNELS 8
 
 struct AVResampleContext;
 
@@ -36,20 +45,22 @@ static const char *context_to_name(void *ptr)
 }
 
 static const AVOption options[] = {{NULL}};
-static const AVClass audioresample_context_class = { "ReSampleContext", context_to_name, options };
+static const AVClass audioresample_context_class = {
+    "ReSampleContext", context_to_name, options, LIBAVUTIL_VERSION_INT
+};
 
 struct ReSampleContext {
     struct AVResampleContext *resample_context;
-    short *temp[2];
+    short *temp[MAX_CHANNELS];
     int temp_len;
     float ratio;
     /* channel convert */
     int input_channels, output_channels, filter_channels;
     AVAudioConvert *convert_ctx[2];
-    enum SampleFormat sample_fmt[2]; ///< input and output sample format
-    unsigned sample_size[2];         ///< size of one sample in sample_fmt
-    short *buffer[2];                ///< buffers used for conversion to S16
-    unsigned buffer_size[2];         ///< sizes of allocated buffers
+    enum AVSampleFormat sample_fmt[2]; ///< input and output sample format
+    unsigned sample_size[2];           ///< size of one sample in sample_fmt
+    short *buffer[2];                  ///< buffers used for conversion to S16
+    unsigned buffer_size[2];           ///< sizes of allocated buffers
 };
 
 /* n1: number of samples */
@@ -103,65 +114,125 @@ static void mono_to_stereo(short *output, short *input, int n1)
     }
 }
 
-/* XXX: should use more abstract 'N' channels system */
-static void stereo_split(short *output1, short *output2, short *input, int n)
+/*
+5.1 to stereo input: [fl, fr, c, lfe, rl, rr]
+- Left = front_left + rear_gain * rear_left + center_gain * center
+- Right = front_right + rear_gain * rear_right + center_gain * center
+Where rear_gain is usually around 0.5-1.0 and
+      center_gain is almost always 0.7 (-3 dB)
+*/
+static void surround_to_stereo(short **output, short *input, int channels, int samples)
 {
     int i;
+    short l, r;
+
+    for (i = 0; i < samples; i++) {
+        int fl,fr,c,rl,rr;
+        fl = input[0];
+        fr = input[1];
+        c = input[2];
+        // lfe = input[3];
+        rl = input[4];
+        rr = input[5];
+
+        l = av_clip_int16(fl + (0.5 * rl) + (0.7 * c));
+        r = av_clip_int16(fr + (0.5 * rr) + (0.7 * c));
+
+        /* output l & r. */
+        *output[0]++ = l;
+        *output[1]++ = r;
+
+        /* increment input. */
+        input += channels;
+    }
+}
+
+static void deinterleave(short **output, short *input, int channels, int samples)
+{
+    int i, j;
 
-    for(i=0;i<n;i++) {
-        *output1++ = *input++;
-        *output2++ = *input++;
+    for (i = 0; i < samples; i++) {
+        for (j = 0; j < channels; j++) {
+            *output[j]++ = *input++;
+        }
     }
 }
 
-static void stereo_mux(short *output, short *input1, short *input2, int n)
+static void interleave(short *output, short **input, int channels, int samples)
 {
-    int i;
+    int i, j;
 
-    for(i=0;i<n;i++) {
-        *output++ = *input1++;
-        *output++ = *input2++;
+    for (i = 0; i < samples; i++) {
+        for (j = 0; j < channels; j++) {
+            *output++ = *input[j]++;
+        }
     }
 }
 
 static void ac3_5p1_mux(short *output, short *input1, short *input2, int n)
 {
     int i;
-    short l,r;
-
-    for(i=0;i<n;i++) {
-      l=*input1++;
-      r=*input2++;
-      *output++ = l;           /* left */
-      *output++ = (l/2)+(r/2); /* center */
-      *output++ = r;           /* right */
-      *output++ = 0;           /* left surround */
-      *output++ = 0;           /* right surroud */
-      *output++ = 0;           /* low freq */
+    short l, r;
+
+    for (i = 0; i < n; i++) {
+        l = *input1++;
+        r = *input2++;
+        *output++ = l;                  /* left */
+        *output++ = (l / 2) + (r / 2);  /* center */
+        *output++ = r;                  /* right */
+        *output++ = 0;                  /* left surround */
+        *output++ = 0;                  /* right surroud */
+        *output++ = 0;                  /* low freq */
     }
 }
 
+#define SUPPORT_RESAMPLE(ch1, ch2, ch3, ch4, ch5, ch6, ch7, ch8) \
+    ch8<<7 | ch7<<6 | ch6<<5 | ch5<<4 | ch4<<3 | ch3<<2 | ch2<<1 | ch1<<0
+
+static const uint8_t supported_resampling[MAX_CHANNELS] = {
+    // output ch:    1  2  3  4  5  6  7  8
+    SUPPORT_RESAMPLE(1, 1, 0, 0, 0, 0, 0, 0), // 1 input channel
+    SUPPORT_RESAMPLE(1, 1, 0, 0, 0, 1, 0, 0), // 2 input channels
+    SUPPORT_RESAMPLE(0, 0, 1, 0, 0, 0, 0, 0), // 3 input channels
+    SUPPORT_RESAMPLE(0, 0, 0, 1, 0, 0, 0, 0), // 4 input channels
+    SUPPORT_RESAMPLE(0, 0, 0, 0, 1, 0, 0, 0), // 5 input channels
+    SUPPORT_RESAMPLE(0, 1, 0, 0, 0, 1, 0, 0), // 6 input channels
+    SUPPORT_RESAMPLE(0, 0, 0, 0, 0, 0, 1, 0), // 7 input channels
+    SUPPORT_RESAMPLE(0, 0, 0, 0, 0, 0, 0, 1), // 8 input channels
+};
+
 ReSampleContext *av_audio_resample_init(int output_channels, int input_channels,
                                         int output_rate, int input_rate,
-                                        enum SampleFormat sample_fmt_out,
-                                        enum SampleFormat sample_fmt_in,
+                                        enum AVSampleFormat sample_fmt_out,
+                                        enum AVSampleFormat sample_fmt_in,
                                         int filter_length, int log2_phase_count,
                                         int linear, double cutoff)
 {
     ReSampleContext *s;
 
-    if ( input_channels > 2)
-      {
-        av_log(NULL, AV_LOG_ERROR, "Resampling with input channels greater than 2 unsupported.\n");
+    if (input_channels > MAX_CHANNELS) {
+        av_log(NULL, AV_LOG_ERROR,
+               "Resampling with input channels greater than %d is unsupported.\n",
+               MAX_CHANNELS);
         return NULL;
-      }
+    }
+    if (!(supported_resampling[input_channels-1] & (1<<(output_channels-1)))) {
+        int i;
+        av_log(NULL, AV_LOG_ERROR, "Unsupported audio resampling. Allowed "
+               "output channels for %d input channel%s", input_channels,
+               input_channels > 1 ? "s:" : ":");
+        for (i = 0; i < MAX_CHANNELS; i++)
+            if (supported_resampling[input_channels-1] & (1<<i))
+                av_log(NULL, AV_LOG_ERROR, " %d", i + 1);
+        av_log(NULL, AV_LOG_ERROR, "\n");
+        return NULL;
+    }
 
     s = av_mallocz(sizeof(ReSampleContext));
-    if (!s)
-      {
+    if (!s) {
         av_log(NULL, AV_LOG_ERROR, "Can't allocate memory for resample context.\n");
         return NULL;
-      }
+    }
 
     s->ratio = (float)output_rate / (float)input_rate;
 
@@ -172,85 +243,60 @@ ReSampleContext *av_audio_resample_init(int output_channels, int input_channels,
     if (s->output_channels < s->filter_channels)
         s->filter_channels = s->output_channels;
 
-    s->sample_fmt [0] = sample_fmt_in;
-    s->sample_fmt [1] = sample_fmt_out;
-    s->sample_size[0] = av_get_bits_per_sample_format(s->sample_fmt[0])>>3;
-    s->sample_size[1] = av_get_bits_per_sample_format(s->sample_fmt[1])>>3;
+    s->sample_fmt[0]  = sample_fmt_in;
+    s->sample_fmt[1]  = sample_fmt_out;
+    s->sample_size[0] = av_get_bytes_per_sample(s->sample_fmt[0]);
+    s->sample_size[1] = av_get_bytes_per_sample(s->sample_fmt[1]);
 
-    if (s->sample_fmt[0] != SAMPLE_FMT_S16) {
-        if (!(s->convert_ctx[0] = av_audio_convert_alloc(SAMPLE_FMT_S16, 1,
+    if (s->sample_fmt[0] != AV_SAMPLE_FMT_S16) {
+        if (!(s->convert_ctx[0] = av_audio_convert_alloc(AV_SAMPLE_FMT_S16, 1,
                                                          s->sample_fmt[0], 1, NULL, 0))) {
             av_log(s, AV_LOG_ERROR,
                    "Cannot convert %s sample format to s16 sample format\n",
-                   avcodec_get_sample_fmt_name(s->sample_fmt[0]));
+                   av_get_sample_fmt_name(s->sample_fmt[0]));
             av_free(s);
             return NULL;
         }
     }
 
-    if (s->sample_fmt[1] != SAMPLE_FMT_S16) {
+    if (s->sample_fmt[1] != AV_SAMPLE_FMT_S16) {
         if (!(s->convert_ctx[1] = av_audio_convert_alloc(s->sample_fmt[1], 1,
-                                                         SAMPLE_FMT_S16, 1, NULL, 0))) {
+                                                         AV_SAMPLE_FMT_S16, 1, NULL, 0))) {
             av_log(s, AV_LOG_ERROR,
                    "Cannot convert s16 sample format to %s sample format\n",
-                   avcodec_get_sample_fmt_name(s->sample_fmt[1]));
+                   av_get_sample_fmt_name(s->sample_fmt[1]));
             av_audio_convert_free(s->convert_ctx[0]);
             av_free(s);
             return NULL;
         }
     }
 
-/*
- * AC-3 output is the only case where filter_channels could be greater than 2.
- * input channels can't be greater than 2, so resample the 2 channels and then
- * expand to 6 channels after the resampling.
- */
-    if(s->filter_channels>2)
-      s->filter_channels = 2;
-
-#define TAPS 16
-    s->resample_context= av_resample_init(output_rate, input_rate,
-                         filter_length, log2_phase_count, linear, cutoff);
+    s->resample_context = av_resample_init(output_rate, input_rate,
+                                           filter_length, log2_phase_count,
+                                           linear, cutoff);
 
-    *(AVClass**)s->resample_context = &audioresample_context_class;
+    *(const AVClass**)s->resample_context = &audioresample_context_class;
 
     return s;
 }
 
-#if LIBAVCODEC_VERSION_MAJOR < 53
-ReSampleContext *audio_resample_init(int output_channels, int input_channels,
-                                     int output_rate, int input_rate)
-{
-    return av_audio_resample_init(output_channels, input_channels,
-                                  output_rate, input_rate,
-                                  SAMPLE_FMT_S16, SAMPLE_FMT_S16,
-                                  TAPS, 10, 0, 0.8);
-}
-#endif
-
 /* resample audio. 'nb_samples' is the number of input samples */
 /* XXX: optimize it ! */
 int audio_resample(ReSampleContext *s, short *output, short *input, int nb_samples)
 {
     int i, nb_samples1;
-    short *bufin[2];
-    short *bufout[2];
-    short *buftmp2[2], *buftmp3[2];
+    short *bufin[MAX_CHANNELS];
+    short *bufout[MAX_CHANNELS];
+    short *buftmp2[MAX_CHANNELS], *buftmp3[MAX_CHANNELS];
     short *output_bak = NULL;
     int lenout;
 
-    if (s->input_channels == s->output_channels && s->ratio == 1.0 && 0) {
-        /* nothing to do */
-        memcpy(output, input, nb_samples * s->input_channels * sizeof(short));
-        return nb_samples;
-    }
-
-    if (s->sample_fmt[0] != SAMPLE_FMT_S16) {
+    if (s->sample_fmt[0] != AV_SAMPLE_FMT_S16) {
         int istride[1] = { s->sample_size[0] };
         int ostride[1] = { 2 };
         const void *ibuf[1] = { input };
         void       *obuf[1];
-        unsigned input_size = nb_samples*s->input_channels*2;
+        unsigned input_size = nb_samples * s->input_channels * 2;
 
         if (!s->buffer_size[0] || s->buffer_size[0] < input_size) {
             av_free(s->buffer[0]);
@@ -265,22 +311,25 @@ int audio_resample(ReSampleContext *s, short *output, short *input, int nb_sampl
         obuf[0] = s->buffer[0];
 
         if (av_audio_convert(s->convert_ctx[0], obuf, ostride,
-                             ibuf, istride, nb_samples*s->input_channels) < 0) {
-            av_log(s->resample_context, AV_LOG_ERROR, "Audio sample format conversion failed\n");
+                             ibuf, istride, nb_samples * s->input_channels) < 0) {
+            av_log(s->resample_context, AV_LOG_ERROR,
+                   "Audio sample format conversion failed\n");
             return 0;
         }
 
-        input  = s->buffer[0];
+        input = s->buffer[0];
     }
 
-    lenout= 4*nb_samples * s->ratio + 16;
+    lenout= 2*s->output_channels*nb_samples * s->ratio + 16;
 
-    if (s->sample_fmt[1] != SAMPLE_FMT_S16) {
+    if (s->sample_fmt[1] != AV_SAMPLE_FMT_S16) {
+        int out_size = lenout * av_get_bytes_per_sample(s->sample_fmt[1]) *
+                       s->output_channels;
         output_bak = output;
 
-        if (!s->buffer_size[1] || s->buffer_size[1] < lenout) {
+        if (!s->buffer_size[1] || s->buffer_size[1] < out_size) {
             av_free(s->buffer[1]);
-            s->buffer_size[1] = lenout;
+            s->buffer_size[1] = out_size;
             s->buffer[1] = av_malloc(s->buffer_size[1]);
             if (!s->buffer[1]) {
                 av_log(s->resample_context, AV_LOG_ERROR, "Could not allocate buffer\n");
@@ -292,83 +341,99 @@ int audio_resample(ReSampleContext *s, short *output, short *input, int nb_sampl
     }
 
     /* XXX: move those malloc to resample init code */
-    for(i=0; i<s->filter_channels; i++){
-        bufin[i]= av_malloc( (nb_samples + s->temp_len) * sizeof(short) );
+    for (i = 0; i < s->filter_channels; i++) {
+        bufin[i] = av_malloc_array((nb_samples + s->temp_len), sizeof(short));
+        bufout[i] = av_malloc_array(lenout, sizeof(short));
+
+        if (!bufin[i] || !bufout[i]) {
+            av_log(s->resample_context, AV_LOG_ERROR, "Could not allocate buffer\n");
+            nb_samples1 = 0;
+            goto fail;
+        }
+
         memcpy(bufin[i], s->temp[i], s->temp_len * sizeof(short));
         buftmp2[i] = bufin[i] + s->temp_len;
     }
 
-    /* make some zoom to avoid round pb */
-    bufout[0]= av_malloc( lenout * sizeof(short) );
-    bufout[1]= av_malloc( lenout * sizeof(short) );
-
-    if (s->input_channels == 2 &&
-        s->output_channels == 1) {
+    if (s->input_channels == 2 && s->output_channels == 1) {
         buftmp3[0] = output;
         stereo_to_mono(buftmp2[0], input, nb_samples);
     } else if (s->output_channels >= 2 && s->input_channels == 1) {
         buftmp3[0] = bufout[0];
-        memcpy(buftmp2[0], input, nb_samples*sizeof(short));
-    } else if (s->output_channels >= 2) {
+        memcpy(buftmp2[0], input, nb_samples * sizeof(short));
+    } else if (s->input_channels == 6 && s->output_channels ==2) {
         buftmp3[0] = bufout[0];
         buftmp3[1] = bufout[1];
-        stereo_split(buftmp2[0], buftmp2[1], input, nb_samples);
+        surround_to_stereo(buftmp2, input, s->input_channels, nb_samples);
+    } else if (s->output_channels >= s->input_channels && s->input_channels >= 2) {
+        for (i = 0; i < s->input_channels; i++) {
+            buftmp3[i] = bufout[i];
+        }
+        deinterleave(buftmp2, input, s->input_channels, nb_samples);
     } else {
         buftmp3[0] = output;
-        memcpy(buftmp2[0], input, nb_samples*sizeof(short));
+        memcpy(buftmp2[0], input, nb_samples * sizeof(short));
     }
 
     nb_samples += s->temp_len;
 
     /* resample each channel */
     nb_samples1 = 0; /* avoid warning */
-    for(i=0;i<s->filter_channels;i++) {
+    for (i = 0; i < s->filter_channels; i++) {
         int consumed;
-        int is_last= i+1 == s->filter_channels;
+        int is_last = i + 1 == s->filter_channels;
 
-        nb_samples1 = av_resample(s->resample_context, buftmp3[i], bufin[i], &consumed, nb_samples, lenout, is_last);
-        s->temp_len= nb_samples - consumed;
-        s->temp[i]= av_realloc(s->temp[i], s->temp_len*sizeof(short));
-        memcpy(s->temp[i], bufin[i] + consumed, s->temp_len*sizeof(short));
+        nb_samples1 = av_resample(s->resample_context, buftmp3[i], bufin[i],
+                                  &consumed, nb_samples, lenout, is_last);
+        s->temp_len = nb_samples - consumed;
+        s->temp[i] = av_realloc_array(s->temp[i], s->temp_len, sizeof(short));
+        memcpy(s->temp[i], bufin[i] + consumed, s->temp_len * sizeof(short));
     }
 
     if (s->output_channels == 2 && s->input_channels == 1) {
         mono_to_stereo(output, buftmp3[0], nb_samples1);
-    } else if (s->output_channels == 2) {
-        stereo_mux(output, buftmp3[0], buftmp3[1], nb_samples1);
-    } else if (s->output_channels == 6) {
+    } else if (s->output_channels == 6 && s->input_channels == 2) {
         ac3_5p1_mux(output, buftmp3[0], buftmp3[1], nb_samples1);
+    } else if ((s->output_channels == s->input_channels && s->input_channels >= 2) ||
+               (s->output_channels == 2 && s->input_channels == 6)) {
+        interleave(output, buftmp3, s->output_channels, nb_samples1);
     }
 
-    if (s->sample_fmt[1] != SAMPLE_FMT_S16) {
+    if (s->sample_fmt[1] != AV_SAMPLE_FMT_S16) {
         int istride[1] = { 2 };
         int ostride[1] = { s->sample_size[1] };
         const void *ibuf[1] = { output };
         void       *obuf[1] = { output_bak };
 
         if (av_audio_convert(s->convert_ctx[1], obuf, ostride,
-                             ibuf, istride, nb_samples1*s->output_channels) < 0) {
-            av_log(s->resample_context, AV_LOG_ERROR, "Audio sample format convertion failed\n");
+                             ibuf, istride, nb_samples1 * s->output_channels) < 0) {
+            av_log(s->resample_context, AV_LOG_ERROR,
+                   "Audio sample format conversion failed\n");
             return 0;
         }
     }
 
-    for(i=0; i<s->filter_channels; i++)
+fail:
+    for (i = 0; i < s->filter_channels; i++) {
         av_free(bufin[i]);
+        av_free(bufout[i]);
+    }
 
-    av_free(bufout[0]);
-    av_free(bufout[1]);
     return nb_samples1;
 }
 
 void audio_resample_close(ReSampleContext *s)
 {
+    int i;
     av_resample_close(s->resample_context);
-    av_freep(&s->temp[0]);
-    av_freep(&s->temp[1]);
+    for (i = 0; i < s->filter_channels; i++)
+        av_freep(&s->temp[i]);
     av_freep(&s->buffer[0]);
     av_freep(&s->buffer[1]);
     av_audio_convert_free(s->convert_ctx[0]);
     av_audio_convert_free(s->convert_ctx[1]);
     av_free(s);
 }
+
+FF_ENABLE_DEPRECATION_WARNINGS
+#endif