Merge commit '7c278d2a'

* commit '7c278d2a':
  alacenc: support 24-bit encoding
  pcmdec: use planar sample format for pcm_s16le_planar
  vorbisdec: use float planar sample format

Conflicts:
	libavcodec/pcm.c
Merged-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/alacenc.c

@@ -27,7 +27,6 @@
 #include "mathops.h"
 
 #define DEFAULT_FRAME_SIZE        4096
-#define DEFAULT_SAMPLE_SIZE       16
 #define MAX_CHANNELS              8
 #define ALAC_EXTRADATA_SIZE       36
 #define ALAC_FRAME_HEADER_SIZE    55
@@ -66,6 +65,7 @@ typedef struct AlacEncodeContext {
     int max_prediction_order;
     int max_coded_frame_size;
     int write_sample_size;
+    int extra_bits;
     int32_t sample_buf[MAX_CHANNELS][DEFAULT_FRAME_SIZE];
     int32_t predictor_buf[DEFAULT_FRAME_SIZE];
     int interlacing_shift;
@@ -78,16 +78,26 @@ typedef struct AlacEncodeContext {
 } AlacEncodeContext;
 
 
-static void init_sample_buffers(AlacEncodeContext *s, int16_t **input_samples)
+static void init_sample_buffers(AlacEncodeContext *s,
+                                uint8_t * const *samples)
 {
     int ch, i;
-
-    for (ch = 0; ch < s->avctx->channels; ch++) {
-        int32_t       *bptr = s->sample_buf[ch];
-        const int16_t *sptr = input_samples[ch];
-        for (i = 0; i < s->frame_size; i++)
-            bptr[i] = sptr[i];
-    }
+    int shift = av_get_bytes_per_sample(s->avctx->sample_fmt) * 8 -
+                s->avctx->bits_per_raw_sample;
+
+#define COPY_SAMPLES(type) do {                             \
+        for (ch = 0; ch < s->avctx->channels; ch++) {       \
+            int32_t       *bptr = s->sample_buf[ch];        \
+            const type *sptr = (const type *)samples[ch];   \
+            for (i = 0; i < s->frame_size; i++)             \
+                bptr[i] = sptr[i] >> shift;                 \
+        }                                                   \
+    } while (0)
+
+    if (s->avctx->sample_fmt == AV_SAMPLE_FMT_S32P)
+        COPY_SAMPLES(int32_t);
+    else
+        COPY_SAMPLES(int16_t);
 }
 
 static void encode_scalar(AlacEncodeContext *s, int x,
@@ -128,7 +138,7 @@ static void write_frame_header(AlacEncodeContext *s)
     put_bits(&s->pbctx, 3,  s->avctx->channels-1);  // No. of channels -1
     put_bits(&s->pbctx, 16, 0);                     // Seems to be zero
     put_bits(&s->pbctx, 1,  encode_fs);             // Sample count is in the header
-    put_bits(&s->pbctx, 2,  0);                     // FIXME: Wasted bytes field
+    put_bits(&s->pbctx, 2,  s->extra_bits >> 3);    // Extra bytes (for 24-bit)
     put_bits(&s->pbctx, 1,  s->verbatim);           // Audio block is verbatim
     if (encode_fs)
         put_bits32(&s->pbctx, s->frame_size);       // No. of samples in the frame
@@ -345,7 +355,8 @@ static void alac_entropy_coder(AlacEncodeContext *s)
     }
 }
 
-static int write_frame(AlacEncodeContext *s, AVPacket *avpkt, int16_t **samples)
+static int write_frame(AlacEncodeContext *s, AVPacket *avpkt,
+                       uint8_t * const *samples)
 {
     int i, j;
     int prediction_type = 0;
@@ -356,9 +367,20 @@ static int write_frame(AlacEncodeContext *s, AVPacket *avpkt, int16_t **samples)
     if (s->verbatim) {
         write_frame_header(s);
         /* samples are channel-interleaved in verbatim mode */
+        if (s->avctx->sample_fmt == AV_SAMPLE_FMT_S32P) {
+            int shift = 32 - s->avctx->bits_per_raw_sample;
+            int32_t * const *samples_s32 = (int32_t * const *)samples;
+            for (i = 0; i < s->frame_size; i++)
+                for (j = 0; j < s->avctx->channels; j++)
+                    put_sbits(pb, s->avctx->bits_per_raw_sample,
+                              samples_s32[j][i] >> shift);
+        } else {
+            int16_t * const *samples_s16 = (int16_t * const *)samples;
             for (i = 0; i < s->frame_size; i++)
                 for (j = 0; j < s->avctx->channels; j++)
-                put_sbits(pb, 16, samples[j][i]);
+                    put_sbits(pb, s->avctx->bits_per_raw_sample,
+                              samples_s16[j][i]);
+        }
     } else {
         init_sample_buffers(s, samples);
         write_frame_header(s);
@@ -381,6 +403,17 @@ static int write_frame(AlacEncodeContext *s, AVPacket *avpkt, int16_t **samples)
                 put_sbits(pb, 16, s->lpc[i].lpc_coeff[j]);
         }
 
+        // write extra bits if needed
+        if (s->extra_bits) {
+            uint32_t mask = (1 << s->extra_bits) - 1;
+            for (i = 0; i < s->frame_size; i++) {
+                for (j = 0; j < s->avctx->channels; j++) {
+                    put_bits(pb, s->extra_bits, s->sample_buf[j][i] & mask);
+                    s->sample_buf[j][i] >>= s->extra_bits;
+                }
+            }
+        }
+
         // apply lpc and entropy coding to audio samples
 
         for (i = 0; i < s->avctx->channels; i++) {
@@ -433,6 +466,15 @@ static av_cold int alac_encode_init(AVCodecContext *avctx)
         return AVERROR_PATCHWELCOME;
     }
 
+    if (avctx->sample_fmt == AV_SAMPLE_FMT_S32P) {
+        if (avctx->bits_per_raw_sample != 24)
+            av_log(avctx, AV_LOG_WARNING, "encoding as 24 bits-per-sample\n");
+        avctx->bits_per_raw_sample = 24;
+    } else {
+        avctx->bits_per_raw_sample = 16;
+        s->extra_bits              = 0;
+    }
+
     // Set default compression level
     if (avctx->compression_level == FF_COMPRESSION_DEFAULT)
         s->compression_level = 2;
@@ -447,10 +489,7 @@ static av_cold int alac_encode_init(AVCodecContext *avctx)
 
     s->max_coded_frame_size = get_max_frame_size(avctx->frame_size,
                                                  avctx->channels,
-                                                 DEFAULT_SAMPLE_SIZE);
-
-    // FIXME: consider wasted_bytes
-    s->write_sample_size  = DEFAULT_SAMPLE_SIZE + avctx->channels - 1;
+                                                 avctx->bits_per_raw_sample);
 
     avctx->extradata = av_mallocz(ALAC_EXTRADATA_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
     if (!avctx->extradata) {
@@ -463,11 +502,11 @@ static av_cold int alac_encode_init(AVCodecContext *avctx)
     AV_WB32(alac_extradata,    ALAC_EXTRADATA_SIZE);
     AV_WB32(alac_extradata+4,  MKBETAG('a','l','a','c'));
     AV_WB32(alac_extradata+12, avctx->frame_size);
-    AV_WB8 (alac_extradata+17, DEFAULT_SAMPLE_SIZE);
+    AV_WB8 (alac_extradata+17, avctx->bits_per_raw_sample);
     AV_WB8 (alac_extradata+21, avctx->channels);
     AV_WB32(alac_extradata+24, s->max_coded_frame_size);
     AV_WB32(alac_extradata+28,
-            avctx->sample_rate * avctx->channels * DEFAULT_SAMPLE_SIZE); // average bitrate
+            avctx->sample_rate * avctx->channels * avctx->bits_per_raw_sample); // average bitrate
     AV_WB32(alac_extradata+32, avctx->sample_rate);
 
     // Set relevant extradata fields
@@ -536,13 +575,12 @@ static int alac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
 {
     AlacEncodeContext *s = avctx->priv_data;
     int out_bytes, max_frame_size, ret;
-    int16_t **samples = (int16_t **)frame->extended_data;
 
     s->frame_size = frame->nb_samples;
 
     if (frame->nb_samples < DEFAULT_FRAME_SIZE)
         max_frame_size = get_max_frame_size(s->frame_size, avctx->channels,
-                                            DEFAULT_SAMPLE_SIZE);
+                                            avctx->bits_per_raw_sample);
     else
         max_frame_size = s->max_coded_frame_size;
 
@@ -550,14 +588,24 @@ static int alac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
         return ret;
 
     /* use verbatim mode for compression_level 0 */
-    s->verbatim = !s->compression_level;
+    if (s->compression_level) {
+        s->verbatim   = 0;
+        s->extra_bits = avctx->bits_per_raw_sample - 16;
+    } else {
+        s->verbatim   = 1;
+        s->extra_bits = 0;
+    }
+    s->write_sample_size = avctx->bits_per_raw_sample - s->extra_bits +
+                           avctx->channels - 1;
 
-    out_bytes = write_frame(s, avpkt, samples);
+    out_bytes = write_frame(s, avpkt, frame->extended_data);
 
     if (out_bytes > max_frame_size) {
         /* frame too large. use verbatim mode */
         s->verbatim = 1;
-        out_bytes = write_frame(s, avpkt, samples);
+        s->extra_bits = 0;
+        s->write_sample_size = avctx->bits_per_raw_sample + avctx->channels - 1;
+        out_bytes = write_frame(s, avpkt, frame->extended_data);
     }
 
     avpkt->size = out_bytes;
@@ -574,7 +622,8 @@ AVCodec ff_alac_encoder = {
     .encode2        = alac_encode_frame,
     .close          = alac_encode_close,
     .capabilities   = CODEC_CAP_SMALL_LAST_FRAME,
-    .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16P,
+    .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S32P,
+                                                     AV_SAMPLE_FMT_S16P,
                                                      AV_SAMPLE_FMT_NONE },
     .long_name      = NULL_IF_CONFIG_SMALL("ALAC (Apple Lossless Audio Codec)"),
 };

libavcodec/pcm.c

@@ -358,7 +358,16 @@ static int pcm_decode_frame(AVCodecContext *avctx, void *data,
         DECODE_PLANAR(16, be16, src, samples, n, 0, 0);
         break;
     case AV_CODEC_ID_PCM_S16LE_PLANAR:
-        DECODE_PLANAR(16, le16, src, samples, n, 0, 0);
+        n /= avctx->channels;
+        for (c = 0; c < avctx->channels; c++) {
+            samples = s->frame.extended_data[c];
+#if HAVE_BIGENDIAN
+            DECODE(16, le16, src, samples, n, 0, 0)
+#else
+            memcpy(samples, src, n * 2);
+#endif
+            src += n * 2;
+        }
         break;
     case AV_CODEC_ID_PCM_S24LE_PLANAR:
         DECODE_PLANAR(32, le24, src, samples, n, 8, 0);

libavcodec/vorbisdec.c

@@ -153,9 +153,7 @@ typedef struct vorbis_context_s {
     uint8_t       mode_number; // mode number for the current packet
     uint8_t       previous_window;
     float        *channel_residues;
-    float        *channel_floors;
     float        *saved;
-    float         scale_bias; // for float->int conversion
 } vorbis_context;
 
 /* Helper functions */
@@ -194,7 +192,6 @@ static void vorbis_free(vorbis_context *vc)
     int i;
 
     av_freep(&vc->channel_residues);
-    av_freep(&vc->channel_floors);
     av_freep(&vc->saved);
 
     for (i = 0; i < vc->residue_count; i++)
@@ -953,12 +950,11 @@ static int vorbis_parse_id_hdr(vorbis_context *vc)
     }
 
     vc->channel_residues =  av_malloc((vc->blocksize[1]  / 2) * vc->audio_channels * sizeof(*vc->channel_residues));
-    vc->channel_floors   =  av_malloc((vc->blocksize[1]  / 2) * vc->audio_channels * sizeof(*vc->channel_floors));
     vc->saved            =  av_mallocz((vc->blocksize[1] / 4) * vc->audio_channels * sizeof(*vc->saved));
     vc->previous_window  = 0;
 
-    ff_mdct_init(&vc->mdct[0], bl0, 1, -vc->scale_bias);
-    ff_mdct_init(&vc->mdct[1], bl1, 1, -vc->scale_bias);
+    ff_mdct_init(&vc->mdct[0], bl0, 1, -1.0);
+    ff_mdct_init(&vc->mdct[1], bl1, 1, -1.0);
 
     av_dlog(NULL, " vorbis version %d \n audio_channels %d \n audio_samplerate %d \n bitrate_max %d \n bitrate_nom %d \n bitrate_min %d \n blk_0 %d blk_1 %d \n ",
             vc->version, vc->audio_channels, vc->audio_samplerate, vc->bitrate_maximum, vc->bitrate_nominal, vc->bitrate_minimum, vc->blocksize[0], vc->blocksize[1]);
@@ -990,13 +986,7 @@ static av_cold int vorbis_decode_init(AVCodecContext *avccontext)
     avpriv_float_dsp_init(&vc->fdsp, avccontext->flags & CODEC_FLAG_BITEXACT);
     ff_fmt_convert_init(&vc->fmt_conv, avccontext);
 
-    if (avccontext->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
-        avccontext->sample_fmt = AV_SAMPLE_FMT_FLT;
-        vc->scale_bias = 1.0f;
-    } else {
-        avccontext->sample_fmt = AV_SAMPLE_FMT_S16;
-        vc->scale_bias = 32768.0f;
-    }
+    avccontext->sample_fmt = AV_SAMPLE_FMT_FLTP;
 
     if (!headers_len) {
         av_log(avccontext, AV_LOG_ERROR, "Extradata missing.\n");
@@ -1487,7 +1477,7 @@ void ff_vorbis_inverse_coupling(float *mag, float *ang, int blocksize)
 
 // Decode the audio packet using the functions above
 
-static int vorbis_parse_audio_packet(vorbis_context *vc)
+static int vorbis_parse_audio_packet(vorbis_context *vc, float **floor_ptr)
 {
     GetBitContext *gb = &vc->gb;
     FFTContext *mdct;
@@ -1498,7 +1488,6 @@ static int vorbis_parse_audio_packet(vorbis_context *vc)
     uint8_t do_not_decode[255];
     vorbis_mapping *mapping;
     float *ch_res_ptr   = vc->channel_residues;
-    float *ch_floor_ptr = vc->channel_floors;
     uint8_t res_chan[255];
     unsigned res_num = 0;
     int retlen  = 0;
@@ -1530,7 +1519,8 @@ static int vorbis_parse_audio_packet(vorbis_context *vc)
     }
 
     memset(ch_res_ptr,   0, sizeof(float) * vc->audio_channels * vlen); //FIXME can this be removed ?
-    memset(ch_floor_ptr, 0, sizeof(float) * vc->audio_channels * vlen); //FIXME can this be removed ?
+    for (i = 0; i < vc->audio_channels; ++i)
+        memset(floor_ptr[i], 0, vlen * sizeof(floor_ptr[0][0])); //FIXME can this be removed ?
 
 // Decode floor
 
@@ -1543,14 +1533,13 @@ static int vorbis_parse_audio_packet(vorbis_context *vc)
             floor = &vc->floors[mapping->submap_floor[0]];
         }
 
-        ret = floor->decode(vc, &floor->data, ch_floor_ptr);
+        ret = floor->decode(vc, &floor->data, floor_ptr[i]);
 
         if (ret < 0) {
             av_log(vc->avccontext, AV_LOG_ERROR, "Invalid codebook in vorbis_floor_decode.\n");
             return AVERROR_INVALIDDATA;
         }
         no_residue[i] = ret;
-        ch_floor_ptr += vlen;
     }
 
 // Nonzero vector propagate
@@ -1614,10 +1603,9 @@ static int vorbis_parse_audio_packet(vorbis_context *vc)
     mdct = &vc->mdct[blockflag];
 
     for (j = vc->audio_channels-1;j >= 0; j--) {
-        ch_floor_ptr = vc->channel_floors   + j           * blocksize / 2;
         ch_res_ptr   = vc->channel_residues + res_chan[j] * blocksize / 2;
-        vc->fdsp.vector_fmul(ch_floor_ptr, ch_floor_ptr, ch_res_ptr, blocksize / 2);
-        mdct->imdct_half(mdct, ch_res_ptr, ch_floor_ptr);
+        vc->fdsp.vector_fmul(floor_ptr[j], floor_ptr[j], ch_res_ptr, blocksize / 2);
+        mdct->imdct_half(mdct, ch_res_ptr, floor_ptr[j]);
     }
 
 // Overlap/add, save data for next overlapping
@@ -1628,7 +1616,7 @@ static int vorbis_parse_audio_packet(vorbis_context *vc)
         unsigned bs1 = vc->blocksize[1];
         float *residue    = vc->channel_residues + res_chan[j] * blocksize / 2;
         float *saved      = vc->saved + j * bs1 / 4;
-        float *ret        = vc->channel_floors + j * retlen;
+        float *ret        = floor_ptr[j];
         float *buf        = residue;
         const float *win  = vc->win[blockflag & previous_window];
 
@@ -1657,14 +1645,31 @@ static int vorbis_decode_frame(AVCodecContext *avccontext, void *data,
     int buf_size       = avpkt->size;
     vorbis_context *vc = avccontext->priv_data;
     GetBitContext *gb = &vc->gb;
-    const float *channel_ptrs[255];
+    float *channel_ptrs[255];
     int i, len, ret;
 
     av_dlog(NULL, "packet length %d \n", buf_size);
 
+    /* get output buffer */
+    vc->frame.nb_samples = vc->blocksize[1] / 2;
+    if ((ret = avccontext->get_buffer(avccontext, &vc->frame)) < 0) {
+        av_log(avccontext, AV_LOG_ERROR, "get_buffer() failed\n");
+        return ret;
+    }
+
+    if (vc->audio_channels > 8) {
+        for (i = 0; i < vc->audio_channels; i++)
+            channel_ptrs[i] = (float *)vc->frame.extended_data[i];
+    } else {
+        for (i = 0; i < vc->audio_channels; i++) {
+            int ch = ff_vorbis_channel_layout_offsets[vc->audio_channels - 1][i];
+            channel_ptrs[ch] = (float *)vc->frame.extended_data[i];
+        }
+    }
+
     init_get_bits(gb, buf, buf_size*8);
 
-    if ((len = vorbis_parse_audio_packet(vc)) <= 0)
+    if ((len = vorbis_parse_audio_packet(vc, channel_ptrs)) <= 0)
         return len;
 
     if (!vc->first_frame) {
@@ -1676,30 +1681,7 @@ static int vorbis_decode_frame(AVCodecContext *avccontext, void *data,
     av_dlog(NULL, "parsed %d bytes %d bits, returned %d samples (*ch*bits) \n",
             get_bits_count(gb) / 8, get_bits_count(gb) % 8, len);
 
-    /* get output buffer */
     vc->frame.nb_samples = len;
-    if ((ret = avccontext->get_buffer(avccontext, &vc->frame)) < 0) {
-        av_log(avccontext, AV_LOG_ERROR, "get_buffer() failed\n");
-        return ret;
-    }
-
-    if (vc->audio_channels > 8) {
-        for (i = 0; i < vc->audio_channels; i++)
-            channel_ptrs[i] = vc->channel_floors + i * len;
-    } else {
-        for (i = 0; i < vc->audio_channels; i++)
-            channel_ptrs[i] = vc->channel_floors +
-                              len * ff_vorbis_channel_layout_offsets[vc->audio_channels - 1][i];
-    }
-
-    if (avccontext->sample_fmt == AV_SAMPLE_FMT_FLT)
-        vc->fmt_conv.float_interleave((float *)vc->frame.data[0], channel_ptrs,
-                                      len, vc->audio_channels);
-    else
-        vc->fmt_conv.float_to_int16_interleave((int16_t *)vc->frame.data[0],
-                                               channel_ptrs, len,
-                                               vc->audio_channels);
-
     *got_frame_ptr   = 1;
     *(AVFrame *)data = vc->frame;
 
@@ -1740,7 +1722,6 @@ AVCodec ff_vorbis_decoder = {
     .capabilities    = CODEC_CAP_DR1,
     .long_name       = NULL_IF_CONFIG_SMALL("Vorbis"),
     .channel_layouts = ff_vorbis_channel_layouts,
-    .sample_fmts     = (const enum AVSampleFormat[]) {
-        AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_NONE
-    },
+    .sample_fmts     = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
+                                                       AV_SAMPLE_FMT_NONE },
 };