X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fatrac3.c;h=136b16c36d45eb3abfcd4c23009b72fca3f1d0a3;hb=c5063e0348db97626aecc17c42fd41718fd62f13;hp=2f964b718380533885d3a48cd0aef4394bcb74c9;hpb=d5202e4fda3d6e3d63e032328b7626f779572e76;p=ffmpeg

diff --git a/libavcodec/atrac3.c b/libavcodec/atrac3.c
index 2f964b71838..136b16c36d4 100644
--- a/libavcodec/atrac3.c
+++ b/libavcodec/atrac3.c
@@ -1,34 +1,35 @@
 /*
  * Atrac 3 compatible decoder
- * Copyright (c) 2006-2007 Maxim Poliakovski
- * Copyright (c) 2006-2007 Benjamin Larsson
+ * Copyright (c) 2006-2008 Maxim Poliakovski
+ * Copyright (c) 2006-2008 Benjamin Larsson
  *
- * This file is part of FFmpeg.
+ * This file is part of Libav.
  *
- * FFmpeg is free software; you can redistribute it and/or
+ * Libav 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,
+ * Libav 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
+ * License along with Libav; if not, write to the Free Software
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
 /**
- * @file atrac3.c
+ * @file
  * Atrac 3 compatible decoder.
- * This decoder handles RealNetworks, RealAudio atrc data.
- * Atrac 3 is identified by the codec name atrc in RealMedia files.
+ * This decoder handles Sony's ATRAC3 data.
+ *
+ * Container formats used to store atrac 3 data:
+ * RealMedia (.rm), RIFF WAV (.wav, .at3), Sony OpenMG (.oma, .aa3).
  *
  * To use this decoder, a calling application must supply the extradata
- * bytes provided from the RealMedia container: 10 bytes or 14 bytes
- * from the WAV container.
+ * bytes provided in the containers above.
  */
 
 #include <math.h>
@@ -36,15 +37,20 @@
 #include <stdio.h>
 
 #include "avcodec.h"
-#include "bitstream.h"
+#include "get_bits.h"
 #include "dsputil.h"
 #include "bytestream.h"
+#include "fft.h"
+#include "fmtconvert.h"
 
+#include "atrac.h"
 #include "atrac3data.h"
 
 #define JOINT_STEREO    0x12
 #define STEREO          0x2
 
+#define SAMPLES_PER_FRAME 1024
+#define MDCT_SIZE          512
 
 /* These structures are needed to store the parsed gain control data. */
 typedef struct {
@@ -67,12 +73,12 @@ typedef struct {
     int               bandsCoded;
     int               numComponents;
     tonal_component   components[64];
-    float             prevFrame[1024];
+    float             prevFrame[SAMPLES_PER_FRAME];
     int               gcBlkSwitch;
     gain_block        gainBlock[2];
 
-    DECLARE_ALIGNED_16(float, spectrum[1024]);
-    DECLARE_ALIGNED_16(float, IMDCT_buf[1024]);
+    DECLARE_ALIGNED(32, float, spectrum)[SAMPLES_PER_FRAME];
+    DECLARE_ALIGNED(32, float, IMDCT_buf)[SAMPLES_PER_FRAME];
 
     float             delayBuf1[46]; ///<qmf delay buffers
     float             delayBuf2[46];
@@ -80,6 +86,7 @@ typedef struct {
 } channel_unit;
 
 typedef struct {
+    AVFrame             frame;
     GetBitContext       gb;
     //@{
     /** stream data */
@@ -104,10 +111,9 @@ typedef struct {
     //@}
     //@{
     /** data buffers */
-    float               outSamples[2048];
+    float              *outSamples[2];
     uint8_t*            decoded_bytes_buffer;
     float               tempBuf[1070];
-    DECLARE_ALIGNED_16(float,mdct_tmp[512]);
     //@}
     //@{
     /** extradata */
@@ -116,71 +122,18 @@ typedef struct {
     int                 scrambled_stream;
     int                 frame_factor;
     //@}
+
+    FFTContext          mdct_ctx;
+    FmtConvertContext   fmt_conv;
 } ATRAC3Context;
 
-static DECLARE_ALIGNED_16(float,mdct_window[512]);
-static float            qmf_window[48];
+static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
 static VLC              spectral_coeff_tab[7];
-static float            SFTable[64];
 static float            gain_tab1[16];
 static float            gain_tab2[31];
-static MDCTContext      mdct_ctx;
 static DSPContext       dsp;
 
 
-/* quadrature mirror synthesis filter */
-
-/**
- * Quadrature mirror synthesis filter.
- *
- * @param inlo      lower part of spectrum
- * @param inhi      higher part of spectrum
- * @param nIn       size of spectrum buffer
- * @param pOut      out buffer
- * @param delayBuf  delayBuf buffer
- * @param temp      temp buffer
- */
-
-
-static void iqmf (float *inlo, float *inhi, unsigned int nIn, float *pOut, float *delayBuf, float *temp)
-{
-    int   i, j;
-    float   *p1, *p3;
-
-    memcpy(temp, delayBuf, 46*sizeof(float));
-
-    p3 = temp + 46;
-
-    /* loop1 */
-    for(i=0; i<nIn; i+=2){
-        p3[2*i+0] = inlo[i  ] + inhi[i  ];
-        p3[2*i+1] = inlo[i  ] - inhi[i  ];
-        p3[2*i+2] = inlo[i+1] + inhi[i+1];
-        p3[2*i+3] = inlo[i+1] - inhi[i+1];
-    }
-
-    /* loop2 */
-    p1 = temp;
-    for (j = nIn; j != 0; j--) {
-        float s1 = 0.0;
-        float s2 = 0.0;
-
-        for (i = 0; i < 48; i += 2) {
-            s1 += p1[i] * qmf_window[i];
-            s2 += p1[i+1] * qmf_window[i+1];
-        }
-
-        pOut[0] = s2;
-        pOut[1] = s1;
-
-        p1 += 2;
-        pOut += 2;
-    }
-
-    /* Update the delay buffer. */
-    memcpy(delayBuf, temp + nIn*2, 46*sizeof(float));
-}
-
 /**
  * Regular 512 points IMDCT without overlapping, with the exception of the swapping of odd bands
  * caused by the reverse spectra of the QMF.
@@ -188,10 +141,9 @@ static void iqmf (float *inlo, float *inhi, unsigned int nIn, float *pOut, float
  * @param pInput    float input
  * @param pOutput   float output
  * @param odd_band  1 if the band is an odd band
- * @param mdct_tmp  aligned temporary buffer for the mdct
  */
 
-static void IMLT(float *pInput, float *pOutput, int odd_band, float* mdct_tmp)
+static void IMLT(ATRAC3Context *q, float *pInput, float *pOutput, int odd_band)
 {
     int     i;
 
@@ -199,7 +151,7 @@ static void IMLT(float *pInput, float *pOutput, int odd_band, float* mdct_tmp)
         /**
         * Reverse the odd bands before IMDCT, this is an effect of the QMF transform
         * or it gives better compression to do it this way.
-        * FIXME: It should be possible to handle this in ff_imdct_calc
+        * FIXME: It should be possible to handle this in imdct_calc
         * for that to happen a modification of the prerotation step of
         * all SIMD code and C code is needed.
         * Or fix the functions before so they generate a pre reversed spectrum.
@@ -209,10 +161,10 @@ static void IMLT(float *pInput, float *pOutput, int odd_band, float* mdct_tmp)
             FFSWAP(float, pInput[i], pInput[255-i]);
     }
 
-    mdct_ctx.fft.imdct_calc(&mdct_ctx,pOutput,pInput,mdct_tmp);
+    q->mdct_ctx.imdct_calc(&q->mdct_ctx,pOutput,pInput);
 
     /* Perform windowing on the output. */
-    dsp.vector_fmul(pOutput,mdct_window,512);
+    dsp.vector_fmul(pOutput, pOutput, mdct_window, MDCT_SIZE);
 
 }
 
@@ -220,9 +172,9 @@ static void IMLT(float *pInput, float *pOutput, int odd_band, float* mdct_tmp)
 /**
  * Atrac 3 indata descrambling, only used for data coming from the rm container
  *
- * @param in        pointer to 8 bit array of indata
- * @param bits      amount of bits
+ * @param inbuffer  pointer to 8 bit array of indata
  * @param out       pointer to 8 bit array of outdata
+ * @param bytes     amount of bytes
  */
 
 static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
@@ -231,23 +183,22 @@ static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
     const uint32_t* buf;
     uint32_t* obuf = (uint32_t*) out;
 
-    off = (int)((long)inbuffer & 3);
+    off = (intptr_t)inbuffer & 3;
     buf = (const uint32_t*) (inbuffer - off);
-    c = be2me_32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
+    c = av_be2ne32((0x537F6103 >> (off*8)) | (0x537F6103 << (32-(off*8))));
     bytes += 3 + off;
     for (i = 0; i < bytes/4; i++)
         obuf[i] = c ^ buf[i];
 
     if (off)
-        av_log(NULL,AV_LOG_DEBUG,"Offset of %d not handled, post sample on ffmpeg-dev.\n",off);
+        av_log_ask_for_sample(NULL, "Offset of %d not handled.\n", off);
 
     return off;
 }
 
 
-static void init_atrac3_transforms(ATRAC3Context *q) {
+static av_cold int init_atrac3_transforms(ATRAC3Context *q, int is_float) {
     float enc_window[256];
-    float s;
     int i;
 
     /* Generate the mdct window, for details see
@@ -261,27 +212,23 @@ static void init_atrac3_transforms(ATRAC3Context *q) {
             mdct_window[511-i] = mdct_window[i];
         }
 
-    /* Generate the QMF window. */
-    for (i=0 ; i<24; i++) {
-        s = qmf_48tap_half[i] * 2.0;
-        qmf_window[i] = s;
-        qmf_window[47 - i] = s;
-    }
-
     /* Initialize the MDCT transform. */
-    ff_mdct_init(&mdct_ctx, 9, 1);
+    return ff_mdct_init(&q->mdct_ctx, 9, 1, is_float ? 1.0 / 32768 : 1.0);
 }
 
 /**
  * Atrac3 uninit, free all allocated memory
  */
 
-static int atrac3_decode_close(AVCodecContext *avctx)
+static av_cold int atrac3_decode_close(AVCodecContext *avctx)
 {
     ATRAC3Context *q = avctx->priv_data;
 
     av_free(q->pUnits);
     av_free(q->decoded_bytes_buffer);
+    av_freep(&q->outSamples[0]);
+
+    ff_mdct_end(&q->mdct_ctx);
 
     return 0;
 }
@@ -305,7 +252,6 @@ static void readQuantSpectralCoeffs (GetBitContext *gb, int selector, int coding
 
     if (codingFlag != 0) {
         /* constant length coding (CLC) */
-        //FIXME we don't have any samples coded in CLC mode
         numBits = CLCLengthTab[selector];
 
         if (selector > 1) {
@@ -388,7 +334,7 @@ static int decodeSpectrum (GetBitContext *gb, float *pOut)
             readQuantSpectralCoeffs (gb, subband_vlc_index[cnt], codingMode, mantissas, subbWidth);
 
             /* Decode the scale factor for this subband. */
-            SF = SFTable[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
+            SF = ff_atrac_sf_table[SF_idxs[cnt]] * iMaxQuant[subband_vlc_index[cnt]];
 
             /* Inverse quantize the coefficients. */
             for (pIn=mantissas ; first<last; first++, pIn++)
@@ -401,7 +347,7 @@ static int decodeSpectrum (GetBitContext *gb, float *pOut)
 
     /* Clear the subbands that were not coded. */
     first = subbandTab[cnt];
-    memset(pOut+first, 0, (1024 - first) * sizeof(float));
+    memset(pOut+first, 0, (SAMPLES_PER_FRAME - first) * sizeof(float));
     return numSubbands;
 }
 
@@ -431,7 +377,7 @@ static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent
 
     coding_mode_selector = get_bits(gb,2);
     if (coding_mode_selector == 2)
-        return -1;
+        return AVERROR_INVALIDDATA;
 
     coding_mode = coding_mode_selector & 1;
 
@@ -443,7 +389,7 @@ static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent
 
         quant_step_index = get_bits(gb,3);
         if (quant_step_index <= 1)
-            return -1;
+            return AVERROR_INVALIDDATA;
 
         if (coding_mode_selector == 3)
             coding_mode = get_bits1(gb);
@@ -456,19 +402,21 @@ static int decodeTonalComponents (GetBitContext *gb, tonal_component *pComponent
 
             for (k=0; k<coded_components; k++) {
                 sfIndx = get_bits(gb,6);
+                if (component_count >= 64)
+                    return AVERROR_INVALIDDATA;
                 pComponent[component_count].pos = j * 64 + (get_bits(gb,6));
-                max_coded_values = 1024 - pComponent[component_count].pos;
+                max_coded_values = SAMPLES_PER_FRAME - pComponent[component_count].pos;
                 coded_values = coded_values_per_component + 1;
                 coded_values = FFMIN(max_coded_values,coded_values);
 
-                scalefactor = SFTable[sfIndx] * iMaxQuant[quant_step_index];
+                scalefactor = ff_atrac_sf_table[sfIndx] * iMaxQuant[quant_step_index];
 
                 readQuantSpectralCoeffs(gb, quant_step_index, coding_mode, mantissa, coded_values);
 
                 pComponent[component_count].numCoefs = coded_values;
 
                 /* inverse quant */
-                pCoef = pComponent[k].coef;
+                pCoef = pComponent[component_count].coef;
                 for (cnt = 0; cnt < coded_values; cnt++)
                     pCoef[cnt] = mantissa[cnt] * scalefactor;
 
@@ -506,7 +454,7 @@ static int decodeGainControl (GetBitContext *gb, gain_block *pGb, int numBands)
             pLevel[cf]= get_bits(gb,4);
             pLoc  [cf]= get_bits(gb,5);
             if(cf && pLoc[cf] <= pLoc[cf-1])
-                return -1;
+                return AVERROR_INVALIDDATA;
         }
     }
 
@@ -577,24 +525,28 @@ static void gainCompensateAndOverlap (float *pIn, float *pPrev, float *pOut, gai
 
 /**
  * Combine the tonal band spectrum and regular band spectrum
+ * Return position of the last tonal coefficient
  *
  * @param pSpectrum     output spectrum buffer
  * @param numComponents amount of tonal components
  * @param pComponent    tonal components for this band
  */
 
-static void addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
+static int addTonalComponents (float *pSpectrum, int numComponents, tonal_component *pComponent)
 {
-    int   cnt, i;
+    int   cnt, i, lastPos = -1;
     float   *pIn, *pOut;
 
     for (cnt = 0; cnt < numComponents; cnt++){
+        lastPos = FFMAX(pComponent[cnt].pos + pComponent[cnt].numCoefs, lastPos);
         pIn = pComponent[cnt].coef;
         pOut = &(pSpectrum[pComponent[cnt].pos]);
 
         for (i=0 ; i<pComponent[cnt].numCoefs ; i++)
             pOut[i] += pIn[i];
     }
+
+    return lastPos;
 }
 
 
@@ -714,17 +666,17 @@ static void channelWeighting (float *su1, float *su2, int *p3)
 
 static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_unit *pSnd, float *pOut, int channelNum, int codingMode)
 {
-    int   band, result=0, numSubbands, numBands;
+    int   band, result=0, numSubbands, lastTonal, numBands;
 
     if (codingMode == JOINT_STEREO && channelNum == 1) {
         if (get_bits(gb,2) != 3) {
             av_log(NULL,AV_LOG_ERROR,"JS mono Sound Unit id != 3.\n");
-            return -1;
+            return AVERROR_INVALIDDATA;
         }
     } else {
         if (get_bits(gb,6) != 0x28) {
             av_log(NULL,AV_LOG_ERROR,"Sound Unit id != 0x28.\n");
-            return -1;
+            return AVERROR_INVALIDDATA;
         }
     }
 
@@ -740,25 +692,28 @@ static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_
     numSubbands = decodeSpectrum (gb, pSnd->spectrum);
 
     /* Merge the decoded spectrum and tonal components. */
-    addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
+    lastTonal = addTonalComponents (pSnd->spectrum, pSnd->numComponents, pSnd->components);
 
 
-    /* Convert number of subbands into number of MLT/QMF bands */
+    /* calculate number of used MLT/QMF bands according to the amount of coded spectral lines */
     numBands = (subbandTab[numSubbands] - 1) >> 8;
+    if (lastTonal >= 0)
+        numBands = FFMAX((lastTonal + 256) >> 8, numBands);
 
 
     /* Reconstruct time domain samples. */
     for (band=0; band<4; band++) {
         /* Perform the IMDCT step without overlapping. */
         if (band <= numBands) {
-            IMLT(&(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1,q->mdct_tmp);
+            IMLT(q, &(pSnd->spectrum[band*256]), pSnd->IMDCT_buf, band&1);
         } else
             memset(pSnd->IMDCT_buf, 0, 512 * sizeof(float));
 
         /* gain compensation and overlapping */
-        gainCompensateAndOverlap (pSnd->IMDCT_buf, &(pSnd->prevFrame[band*256]), &(pOut[band*256]),
-                                    &((pSnd->gainBlock[1 - (pSnd->gcBlkSwitch)]).gBlock[band]),
-                                    &((pSnd->gainBlock[pSnd->gcBlkSwitch]).gBlock[band]));
+        gainCompensateAndOverlap(pSnd->IMDCT_buf, &pSnd->prevFrame[band * 256],
+                                 &pOut[band * 256],
+                                 &pSnd->gainBlock[1 - pSnd->gcBlkSwitch].gBlock[band],
+                                 &pSnd->gainBlock[    pSnd->gcBlkSwitch].gBlock[band]);
     }
 
     /* Swap the gain control buffers for the next frame. */
@@ -774,11 +729,12 @@ static int decodeChannelSoundUnit (ATRAC3Context *q, GetBitContext *gb, channel_
  * @param databuf       the input data
  */
 
-static int decodeFrame(ATRAC3Context *q, uint8_t* databuf)
+static int decodeFrame(ATRAC3Context *q, const uint8_t* databuf,
+                       float **out_samples)
 {
     int   result, i;
     float   *p1, *p2, *p3, *p4;
-    uint8_t    *ptr1, *ptr2;
+    uint8_t *ptr1;
 
     if (q->codingMode == JOINT_STEREO) {
 
@@ -786,23 +742,29 @@ static int decodeFrame(ATRAC3Context *q, uint8_t* databuf)
         /* decode Sound Unit 1 */
         init_get_bits(&q->gb,databuf,q->bits_per_frame);
 
-        result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, q->outSamples, 0, JOINT_STEREO);
+        result = decodeChannelSoundUnit(q,&q->gb, q->pUnits, out_samples[0], 0, JOINT_STEREO);
         if (result != 0)
-            return (result);
+            return result;
 
         /* Framedata of the su2 in the joint-stereo mode is encoded in
          * reverse byte order so we need to swap it first. */
-        ptr1 = databuf;
-        ptr2 = databuf+q->bytes_per_frame-1;
-        for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
-            FFSWAP(uint8_t,*ptr1,*ptr2);
+        if (databuf == q->decoded_bytes_buffer) {
+            uint8_t *ptr2 = q->decoded_bytes_buffer+q->bytes_per_frame-1;
+            ptr1 = q->decoded_bytes_buffer;
+            for (i = 0; i < (q->bytes_per_frame/2); i++, ptr1++, ptr2--) {
+                FFSWAP(uint8_t,*ptr1,*ptr2);
+            }
+        } else {
+            const uint8_t *ptr2 = databuf+q->bytes_per_frame-1;
+            for (i = 0; i < q->bytes_per_frame; i++)
+                q->decoded_bytes_buffer[i] = *ptr2--;
         }
 
         /* Skip the sync codes (0xF8). */
-        ptr1 = databuf;
+        ptr1 = q->decoded_bytes_buffer;
         for (i = 4; *ptr1 == 0xF8; i++, ptr1++) {
             if (i >= q->bytes_per_frame)
-                return -1;
+                return AVERROR_INVALIDDATA;
         }
 
 
@@ -821,14 +783,14 @@ static int decodeFrame(ATRAC3Context *q, uint8_t* databuf)
         }
 
         /* Decode Sound Unit 2. */
-        result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], &q->outSamples[1024], 1, JOINT_STEREO);
+        result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[1], out_samples[1], 1, JOINT_STEREO);
         if (result != 0)
-            return (result);
+            return result;
 
         /* Reconstruct the channel coefficients. */
-        reverseMatrixing(q->outSamples, &q->outSamples[1024], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
+        reverseMatrixing(out_samples[0], out_samples[1], q->matrix_coeff_index_prev, q->matrix_coeff_index_now);
 
-        channelWeighting(q->outSamples, &q->outSamples[1024], q->weighting_delay);
+        channelWeighting(out_samples[0], out_samples[1], q->weighting_delay);
 
     } else {
         /* normal stereo mode or mono */
@@ -836,24 +798,25 @@ static int decodeFrame(ATRAC3Context *q, uint8_t* databuf)
         for (i=0 ; i<q->channels ; i++) {
 
             /* Set the bitstream reader at the start of a channel sound unit. */
-            init_get_bits(&q->gb, databuf+((i*q->bytes_per_frame)/q->channels), (q->bits_per_frame)/q->channels);
+            init_get_bits(&q->gb,
+                          databuf + i * q->bytes_per_frame / q->channels,
+                          q->bits_per_frame / q->channels);
 
-            result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], &q->outSamples[i*1024], i, q->codingMode);
+            result = decodeChannelSoundUnit(q,&q->gb, &q->pUnits[i], out_samples[i], i, q->codingMode);
             if (result != 0)
-                return (result);
+                return result;
         }
     }
 
     /* Apply the iQMF synthesis filter. */
-    p1= q->outSamples;
     for (i=0 ; i<q->channels ; i++) {
+        p1 = out_samples[i];
         p2= p1+256;
         p3= p2+256;
         p4= p3+256;
-        iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
-        iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
-        iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
-        p1 +=1024;
+        ff_atrac_iqmf (p1, p2, 256, p1, q->pUnits[i].delayBuf1, q->tempBuf);
+        ff_atrac_iqmf (p4, p3, 256, p3, q->pUnits[i].delayBuf2, q->tempBuf);
+        ff_atrac_iqmf (p1, p3, 512, p1, q->pUnits[i].delayBuf3, q->tempBuf);
     }
 
     return 0;
@@ -866,16 +829,31 @@ static int decodeFrame(ATRAC3Context *q, uint8_t* databuf)
  * @param avctx     pointer to the AVCodecContext
  */
 
-static int atrac3_decode_frame(AVCodecContext *avctx,
-            void *data, int *data_size,
-            const uint8_t *buf, int buf_size) {
+static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
+                               int *got_frame_ptr, AVPacket *avpkt)
+{
+    const uint8_t *buf = avpkt->data;
+    int buf_size = avpkt->size;
     ATRAC3Context *q = avctx->priv_data;
-    int result = 0, i;
-    uint8_t* databuf;
-    int16_t* samples = data;
+    int result;
+    const uint8_t* databuf;
+    float   *samples_flt;
+    int16_t *samples_s16;
+
+    if (buf_size < avctx->block_align) {
+        av_log(avctx, AV_LOG_ERROR,
+               "Frame too small (%d bytes). Truncated file?\n", buf_size);
+        return AVERROR_INVALIDDATA;
+    }
 
-    if (buf_size < avctx->block_align)
-        return buf_size;
+    /* get output buffer */
+    q->frame.nb_samples = SAMPLES_PER_FRAME;
+    if ((result = avctx->get_buffer(avctx, &q->frame)) < 0) {
+        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
+        return result;
+    }
+    samples_flt = (float   *)q->frame.data[0];
+    samples_s16 = (int16_t *)q->frame.data[0];
 
     /* Check if we need to descramble and what buffer to pass on. */
     if (q->scrambled_stream) {
@@ -885,27 +863,30 @@ static int atrac3_decode_frame(AVCodecContext *avctx,
         databuf = buf;
     }
 
-    result = decodeFrame(q, databuf);
+    if (q->channels == 1 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
+        result = decodeFrame(q, databuf, &samples_flt);
+    else
+        result = decodeFrame(q, databuf, q->outSamples);
 
     if (result != 0) {
         av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
-        return -1;
+        return result;
     }
 
-    if (q->channels == 1) {
-        /* mono */
-        for (i = 0; i<1024; i++)
-            samples[i] = av_clip_int16(round(q->outSamples[i]));
-        *data_size = 1024 * sizeof(int16_t);
-    } else {
-        /* stereo */
-        for (i = 0; i < 1024; i++) {
-            samples[i*2] = av_clip_int16(round(q->outSamples[i]));
-            samples[i*2+1] = av_clip_int16(round(q->outSamples[1024+i]));
-        }
-        *data_size = 2048 * sizeof(int16_t);
+    /* interleave */
+    if (q->channels == 2 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT) {
+        q->fmt_conv.float_interleave(samples_flt,
+                                     (const float **)q->outSamples,
+                                     SAMPLES_PER_FRAME, 2);
+    } else if (avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
+        q->fmt_conv.float_to_int16_interleave(samples_s16,
+                                              (const float **)q->outSamples,
+                                              SAMPLES_PER_FRAME, q->channels);
     }
 
+    *got_frame_ptr   = 1;
+    *(AVFrame *)data = q->frame;
+
     return avctx->block_align;
 }
 
@@ -916,11 +897,13 @@ static int atrac3_decode_frame(AVCodecContext *avctx,
  * @param avctx     pointer to the AVCodecContext
  */
 
-static int atrac3_decode_init(AVCodecContext *avctx)
+static av_cold int atrac3_decode_init(AVCodecContext *avctx)
 {
-    int i;
+    int i, ret;
     const uint8_t *edata_ptr = avctx->extradata;
     ATRAC3Context *q = avctx->priv_data;
+    static VLC_TYPE atrac3_vlc_table[4096][2];
+    static int vlcs_initialized = 0;
 
     /* Take data from the AVCodecContext (RM container). */
     q->sample_rate = avctx->sample_rate;
@@ -940,7 +923,7 @@ static int atrac3_decode_init(AVCodecContext *avctx)
         av_log(avctx,AV_LOG_DEBUG,"[12-13] %d\n",bytestream_get_le16(&edata_ptr));  //Unknown always 0
 
         /* setup */
-        q->samples_per_frame = 1024 * q->channels;
+        q->samples_per_frame = SAMPLES_PER_FRAME * q->channels;
         q->atrac3version = 4;
         q->delay = 0x88E;
         if (q->codingMode)
@@ -953,7 +936,7 @@ static int atrac3_decode_init(AVCodecContext *avctx)
         if ((q->bytes_per_frame == 96*q->channels*q->frame_factor) || (q->bytes_per_frame == 152*q->channels*q->frame_factor) || (q->bytes_per_frame == 192*q->channels*q->frame_factor)) {
         } else {
             av_log(avctx,AV_LOG_ERROR,"Unknown frame/channel/frame_factor configuration %d/%d/%d\n", q->bytes_per_frame, q->channels, q->frame_factor);
-            return -1;
+            return AVERROR_INVALIDDATA;
         }
 
     } else if (avctx->extradata_size == 10) {
@@ -973,17 +956,17 @@ static int atrac3_decode_init(AVCodecContext *avctx)
 
     if (q->atrac3version != 4) {
         av_log(avctx,AV_LOG_ERROR,"Version %d != 4.\n",q->atrac3version);
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
 
-    if (q->samples_per_frame != 1024 && q->samples_per_frame != 2048) {
+    if (q->samples_per_frame != SAMPLES_PER_FRAME && q->samples_per_frame != SAMPLES_PER_FRAME*2) {
         av_log(avctx,AV_LOG_ERROR,"Unknown amount of samples per frame %d.\n",q->samples_per_frame);
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
 
     if (q->delay != 0x88E) {
         av_log(avctx,AV_LOG_ERROR,"Unknown amount of delay %x != 0x88E.\n",q->delay);
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
 
     if (q->codingMode == STEREO) {
@@ -992,17 +975,17 @@ static int atrac3_decode_init(AVCodecContext *avctx)
         av_log(avctx,AV_LOG_DEBUG,"Joint stereo detected.\n");
     } else {
         av_log(avctx,AV_LOG_ERROR,"Unknown channel coding mode %x!\n",q->codingMode);
-        return -1;
+        return AVERROR_INVALIDDATA;
     }
 
     if (avctx->channels <= 0 || avctx->channels > 2 /*|| ((avctx->channels * 1024) != q->samples_per_frame)*/) {
         av_log(avctx,AV_LOG_ERROR,"Channel configuration error!\n");
-        return -1;
+        return AVERROR(EINVAL);
     }
 
 
     if(avctx->block_align >= UINT_MAX/2)
-        return -1;
+        return AVERROR(EINVAL);
 
     /* Pad the data buffer with FF_INPUT_BUFFER_PADDING_SIZE,
      * this is for the bitstream reader. */
@@ -1011,17 +994,29 @@ static int atrac3_decode_init(AVCodecContext *avctx)
 
 
     /* Initialize the VLC tables. */
-    for (i=0 ; i<7 ; i++) {
-        init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
-            huff_bits[i], 1, 1,
-            huff_codes[i], 1, 1, INIT_VLC_USE_STATIC);
+    if (!vlcs_initialized) {
+        for (i=0 ; i<7 ; i++) {
+            spectral_coeff_tab[i].table = &atrac3_vlc_table[atrac3_vlc_offs[i]];
+            spectral_coeff_tab[i].table_allocated = atrac3_vlc_offs[i + 1] - atrac3_vlc_offs[i];
+            init_vlc (&spectral_coeff_tab[i], 9, huff_tab_sizes[i],
+                huff_bits[i], 1, 1,
+                huff_codes[i], 1, 1, INIT_VLC_USE_NEW_STATIC);
+        }
+        vlcs_initialized = 1;
     }
 
-    init_atrac3_transforms(q);
+    if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT)
+        avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
+    else
+        avctx->sample_fmt = AV_SAMPLE_FMT_S16;
 
-    /* Generate the scale factors. */
-    for (i=0 ; i<64 ; i++)
-        SFTable[i] = pow(2.0, (i - 15) / 3.0);
+    if ((ret = init_atrac3_transforms(q, avctx->sample_fmt == AV_SAMPLE_FMT_FLT))) {
+        av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
+        av_freep(&q->decoded_bytes_buffer);
+        return ret;
+    }
+
+    ff_atrac_generate_tables();
 
     /* Generate gain tables. */
     for (i=0 ; i<16 ; i++)
@@ -1044,26 +1039,40 @@ static int atrac3_decode_init(AVCodecContext *avctx)
         q->matrix_coeff_index_next[i] = 3;
     }
 
-    dsputil_init(&dsp, avctx);
+    ff_dsputil_init(&dsp, avctx);
+    ff_fmt_convert_init(&q->fmt_conv, avctx);
 
     q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
     if (!q->pUnits) {
-        av_free(q->decoded_bytes_buffer);
+        atrac3_decode_close(avctx);
         return AVERROR(ENOMEM);
     }
 
+    if (avctx->channels > 1 || avctx->sample_fmt == AV_SAMPLE_FMT_S16) {
+        q->outSamples[0] = av_mallocz(SAMPLES_PER_FRAME * avctx->channels * sizeof(*q->outSamples[0]));
+        q->outSamples[1] = q->outSamples[0] + SAMPLES_PER_FRAME;
+        if (!q->outSamples[0]) {
+            atrac3_decode_close(avctx);
+            return AVERROR(ENOMEM);
+        }
+    }
+
+    avcodec_get_frame_defaults(&q->frame);
+    avctx->coded_frame = &q->frame;
+
     return 0;
 }
 
 
-AVCodec atrac3_decoder =
+AVCodec ff_atrac3_decoder =
 {
-    .name = "atrac 3",
-    .type = CODEC_TYPE_AUDIO,
+    .name = "atrac3",
+    .type = AVMEDIA_TYPE_AUDIO,
     .id = CODEC_ID_ATRAC3,
     .priv_data_size = sizeof(ATRAC3Context),
     .init = atrac3_decode_init,
     .close = atrac3_decode_close,
     .decode = atrac3_decode_frame,
-    .long_name = "Atrac 3 (Adaptive TRansform Acoustic Coding 3)",
+    .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
+    .long_name = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
 };