diff --git a/libavcodec/8svx.c b/libavcodec/8svx.c
index f41b19fd58..0da48986f0 100644
--- a/libavcodec/8svx.c
+++ b/libavcodec/8svx.c
@@ -58,25 +58,6 @@ static const int8_t exponential[16] = { -128, -64, -32, -16, -8, -4, -2, -1, 0,
 
 #define MAX_FRAME_SIZE 2048
 
-/**
- * Interleave samples in buffer containing all left channel samples
- * at the beginning, and right channel samples at the end.
- * Each sample is assumed to be in signed 8-bit format.
- *
- * @param size the size in bytes of the dst and src buffer
- */
-static void interleave_stereo(uint8_t *dst, const uint8_t *src, int size)
-{
-    uint8_t *dst_end = dst + size;
-    size = size>>1;
-
-    while (dst < dst_end) {
-        *dst++ = *src;
-        *dst++ = *(src+size);
-        src++;
-    }
-}
-
 /**
  * Delta decode the compressed values in src, and put the resulting
  * decoded n samples in dst.
@@ -107,7 +88,8 @@ static int eightsvx_decode_frame(AVCodecContext *avctx, void *data,
                                  int *got_frame_ptr, AVPacket *avpkt)
 {
     EightSvxContext *esc = avctx->priv_data;
-    int n, out_data_size, ret;
+    int n, out_data_size;
+    int ch, ret;
     uint8_t *src, *dst;
 
     /* decode and interleave the first packet */
@@ -152,10 +134,7 @@ static int eightsvx_decode_frame(AVCodecContext *avctx, void *data,
             deinterleaved_samples = avpkt->data;
         }
 
-        if (avctx->channels == 2)
-            interleave_stereo(esc->samples, deinterleaved_samples, esc->samples_size);
-        else
-            memcpy(esc->samples, deinterleaved_samples, esc->samples_size);
+        memcpy(esc->samples, deinterleaved_samples, esc->samples_size);
         av_freep(&p);
     }
 
@@ -170,11 +149,14 @@ static int eightsvx_decode_frame(AVCodecContext *avctx, void *data,
     *got_frame_ptr   = 1;
     *(AVFrame *)data = esc->frame;
 
-    dst = esc->frame.data[0];
-    src = esc->samples + esc->samples_idx;
-    out_data_size = esc->frame.nb_samples * avctx->channels;
-    for (n = out_data_size; n > 0; n--)
-        *dst++ = *src++ + 128;
+    out_data_size = esc->frame.nb_samples;
+    for (ch = 0; ch<avctx->channels; ch++) {
+        dst = esc->frame.data[ch];
+        src = esc->samples + esc->samples_idx / avctx->channels + ch * esc->samples_size / avctx->channels;
+        for (n = out_data_size; n > 0; n--)
+            *dst++ = *src++ + 128;
+    }
+    out_data_size *= avctx->channels;
     esc->samples_idx += out_data_size;
 
     return esc->table ?
@@ -200,7 +182,7 @@ static av_cold int eightsvx_decode_init(AVCodecContext *avctx)
         av_log(avctx, AV_LOG_ERROR, "Invalid codec id %d.\n", avctx->codec->id);
         return AVERROR_INVALIDDATA;
     }
-    avctx->sample_fmt = AV_SAMPLE_FMT_U8;
+    avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
 
     avcodec_get_frame_defaults(&esc->frame);
     avctx->coded_frame = &esc->frame;
@@ -230,6 +212,8 @@ AVCodec ff_eightsvx_fib_decoder = {
   .close          = eightsvx_decode_close,
   .capabilities   = CODEC_CAP_DR1,
   .long_name      = NULL_IF_CONFIG_SMALL("8SVX fibonacci"),
+  .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
+                                                    AV_SAMPLE_FMT_NONE },
 };
 #endif
 #if CONFIG_EIGHTSVX_EXP_DECODER
@@ -243,6 +227,8 @@ AVCodec ff_eightsvx_exp_decoder = {
   .close          = eightsvx_decode_close,
   .capabilities   = CODEC_CAP_DR1,
   .long_name      = NULL_IF_CONFIG_SMALL("8SVX exponential"),
+  .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
+                                                    AV_SAMPLE_FMT_NONE },
 };
 #endif
 #if CONFIG_PCM_S8_PLANAR_DECODER
@@ -256,5 +242,7 @@ AVCodec ff_pcm_s8_planar_decoder = {
     .decode         = eightsvx_decode_frame,
     .capabilities   = CODEC_CAP_DR1,
     .long_name      = NULL_IF_CONFIG_SMALL("PCM signed 8-bit planar"),
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
+                                                      AV_SAMPLE_FMT_NONE },
 };
 #endif
diff --git a/libavcodec/adpcm.c b/libavcodec/adpcm.c
index 852549638b..054fa0a73d 100644
--- a/libavcodec/adpcm.c
+++ b/libavcodec/adpcm.c
@@ -1268,7 +1268,10 @@ static int adpcm_decode_frame(AVCodecContext *avctx, void *data,
 }
 
 
-#define ADPCM_DECODER(id_, name_, long_name_)               \
+static const enum AVSampleFormat sample_fmts_s16[]  = { AV_SAMPLE_FMT_S16,
+                                                        AV_SAMPLE_FMT_NONE };
+
+#define ADPCM_DECODER(id_, sample_fmts_, name_, long_name_) \
 AVCodec ff_ ## name_ ## _decoder = {                        \
     .name           = #name_,                               \
     .type           = AVMEDIA_TYPE_AUDIO,                   \
@@ -1278,33 +1281,34 @@ AVCodec ff_ ## name_ ## _decoder = {                        \
     .decode         = adpcm_decode_frame,                   \
     .capabilities   = CODEC_CAP_DR1,                        \
     .long_name      = NULL_IF_CONFIG_SMALL(long_name_),     \
+    .sample_fmts    = sample_fmts_,                         \
 }
 
 /* Note: Do not forget to add new entries to the Makefile as well. */
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_4XM, adpcm_4xm, "ADPCM 4X Movie");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_CT, adpcm_ct, "ADPCM Creative Technology");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA, adpcm_ea, "ADPCM Electronic Arts");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_MAXIS_XA, adpcm_ea_maxis_xa, "ADPCM Electronic Arts Maxis CDROM XA");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_R1, adpcm_ea_r1, "ADPCM Electronic Arts R1");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_R2, adpcm_ea_r2, "ADPCM Electronic Arts R2");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_R3, adpcm_ea_r3, "ADPCM Electronic Arts R3");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_XAS, adpcm_ea_xas, "ADPCM Electronic Arts XAS");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_AMV, adpcm_ima_amv, "ADPCM IMA AMV");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_APC, adpcm_ima_apc, "ADPCM IMA CRYO APC");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_DK3, adpcm_ima_dk3, "ADPCM IMA Duck DK3");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_DK4, adpcm_ima_dk4, "ADPCM IMA Duck DK4");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_EA_EACS, adpcm_ima_ea_eacs, "ADPCM IMA Electronic Arts EACS");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_EA_SEAD, adpcm_ima_ea_sead, "ADPCM IMA Electronic Arts SEAD");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_ISS, adpcm_ima_iss, "ADPCM IMA Funcom ISS");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_QT, adpcm_ima_qt, "ADPCM IMA QuickTime");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_SMJPEG, adpcm_ima_smjpeg, "ADPCM IMA Loki SDL MJPEG");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_WAV, adpcm_ima_wav, "ADPCM IMA WAV");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_WS, adpcm_ima_ws, "ADPCM IMA Westwood");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_MS, adpcm_ms, "ADPCM Microsoft");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_SBPRO_2, adpcm_sbpro_2, "ADPCM Sound Blaster Pro 2-bit");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_SBPRO_3, adpcm_sbpro_3, "ADPCM Sound Blaster Pro 2.6-bit");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_SBPRO_4, adpcm_sbpro_4, "ADPCM Sound Blaster Pro 4-bit");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_SWF, adpcm_swf, "ADPCM Shockwave Flash");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_THP, adpcm_thp, "ADPCM Nintendo Gamecube THP");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_XA, adpcm_xa, "ADPCM CDROM XA");
-ADPCM_DECODER(AV_CODEC_ID_ADPCM_YAMAHA, adpcm_yamaha, "ADPCM Yamaha");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_4XM,         sample_fmts_s16,  adpcm_4xm,         "ADPCM 4X Movie");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_CT,          sample_fmts_s16,  adpcm_ct,          "ADPCM Creative Technology");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA,          sample_fmts_s16,  adpcm_ea,          "ADPCM Electronic Arts");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_MAXIS_XA, sample_fmts_s16,  adpcm_ea_maxis_xa, "ADPCM Electronic Arts Maxis CDROM XA");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_R1,       sample_fmts_s16,  adpcm_ea_r1,       "ADPCM Electronic Arts R1");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_R2,       sample_fmts_s16,  adpcm_ea_r2,       "ADPCM Electronic Arts R2");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_R3,       sample_fmts_s16,  adpcm_ea_r3,       "ADPCM Electronic Arts R3");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_EA_XAS,      sample_fmts_s16,  adpcm_ea_xas,      "ADPCM Electronic Arts XAS");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_AMV,     sample_fmts_s16,  adpcm_ima_amv,     "ADPCM IMA AMV");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_APC,     sample_fmts_s16,  adpcm_ima_apc,     "ADPCM IMA CRYO APC");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_DK3,     sample_fmts_s16,  adpcm_ima_dk3,     "ADPCM IMA Duck DK3");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_DK4,     sample_fmts_s16,  adpcm_ima_dk4,     "ADPCM IMA Duck DK4");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_EA_EACS, sample_fmts_s16,  adpcm_ima_ea_eacs, "ADPCM IMA Electronic Arts EACS");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_EA_SEAD, sample_fmts_s16,  adpcm_ima_ea_sead, "ADPCM IMA Electronic Arts SEAD");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_ISS,     sample_fmts_s16,  adpcm_ima_iss,     "ADPCM IMA Funcom ISS");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_QT,      sample_fmts_s16,  adpcm_ima_qt,      "ADPCM IMA QuickTime");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_SMJPEG,  sample_fmts_s16,  adpcm_ima_smjpeg,  "ADPCM IMA Loki SDL MJPEG");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_WAV,     sample_fmts_s16,  adpcm_ima_wav,     "ADPCM IMA WAV");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_IMA_WS,      sample_fmts_s16,  adpcm_ima_ws,      "ADPCM IMA Westwood");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_MS,          sample_fmts_s16,  adpcm_ms,          "ADPCM Microsoft");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_SBPRO_2,     sample_fmts_s16,  adpcm_sbpro_2,     "ADPCM Sound Blaster Pro 2-bit");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_SBPRO_3,     sample_fmts_s16,  adpcm_sbpro_3,     "ADPCM Sound Blaster Pro 2.6-bit");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_SBPRO_4,     sample_fmts_s16,  adpcm_sbpro_4,     "ADPCM Sound Blaster Pro 4-bit");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_SWF,         sample_fmts_s16,  adpcm_swf,         "ADPCM Shockwave Flash");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_THP,         sample_fmts_s16,  adpcm_thp,         "ADPCM Nintendo Gamecube THP");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_XA,          sample_fmts_s16,  adpcm_xa,          "ADPCM CDROM XA");
+ADPCM_DECODER(AV_CODEC_ID_ADPCM_YAMAHA,      sample_fmts_s16,  adpcm_yamaha,      "ADPCM Yamaha");
diff --git a/libavcodec/apedec.c b/libavcodec/apedec.c
index 21080902ee..cba3dc0354 100644
--- a/libavcodec/apedec.c
+++ b/libavcodec/apedec.c
@@ -196,13 +196,13 @@ static av_cold int ape_decode_init(AVCodecContext *avctx)
     s->bps = avctx->bits_per_coded_sample;
     switch (s->bps) {
     case 8:
-        avctx->sample_fmt = AV_SAMPLE_FMT_U8;
+        avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
         break;
     case 16:
-        avctx->sample_fmt = AV_SAMPLE_FMT_S16;
+        avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
         break;
     case 24:
-        avctx->sample_fmt = AV_SAMPLE_FMT_S32;
+        avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
         break;
     default:
         av_log_ask_for_sample(avctx, "Unsupported bits per coded sample %d\n",
@@ -830,7 +830,7 @@ static int ape_decode_frame(AVCodecContext *avctx, void *data,
     uint8_t *sample8;
     int16_t *sample16;
     int32_t *sample24;
-    int i, ret;
+    int i, ch, ret;
     int blockstodecode;
     int bytes_used = 0;
 
@@ -930,27 +930,24 @@ static int ape_decode_frame(AVCodecContext *avctx, void *data,
 
     switch (s->bps) {
     case 8:
-        sample8 = (uint8_t *)s->frame.data[0];
-        for (i = 0; i < blockstodecode; i++) {
-            *sample8++ = (s->decoded[0][i] + 0x80) & 0xff;
-            if (s->channels == 2)
-                *sample8++ = (s->decoded[1][i] + 0x80) & 0xff;
+        for (ch = 0; ch < s->channels; ch++) {
+            sample8 = (uint8_t *)s->frame.data[ch];
+            for (i = 0; i < blockstodecode; i++)
+                *sample8++ = (s->decoded[ch][i] + 0x80) & 0xff;
         }
         break;
     case 16:
-        sample16 = (int16_t *)s->frame.data[0];
-        for (i = 0; i < blockstodecode; i++) {
-            *sample16++ = s->decoded[0][i];
-            if (s->channels == 2)
-                *sample16++ = s->decoded[1][i];
+        for (ch = 0; ch < s->channels; ch++) {
+            sample16 = (int16_t *)s->frame.data[ch];
+            for (i = 0; i < blockstodecode; i++)
+                *sample16++ = s->decoded[ch][i];
         }
         break;
     case 24:
-        sample24 = (int32_t *)s->frame.data[0];
-        for (i = 0; i < blockstodecode; i++) {
-            *sample24++ = s->decoded[0][i] << 8;
-            if (s->channels == 2)
-                *sample24++ = s->decoded[1][i] << 8;
+        for (ch = 0; ch < s->channels; ch++) {
+            sample24 = (int32_t *)s->frame.data[ch];
+            for (i = 0; i < blockstodecode; i++)
+                *sample24++ = s->decoded[ch][i] << 8;
         }
         break;
     }
@@ -995,5 +992,9 @@ AVCodec ff_ape_decoder = {
     .capabilities   = CODEC_CAP_SUBFRAMES | CODEC_CAP_DELAY | CODEC_CAP_DR1,
     .flush          = ape_flush,
     .long_name      = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
+                                                      AV_SAMPLE_FMT_S16P,
+                                                      AV_SAMPLE_FMT_S32P,
+                                                      AV_SAMPLE_FMT_NONE },
     .priv_class     = &ape_decoder_class,
 };
diff --git a/libavcodec/atrac3.c b/libavcodec/atrac3.c
index 7d076bed99..b900882a7a 100644
--- a/libavcodec/atrac3.c
+++ b/libavcodec/atrac3.c
@@ -112,7 +112,6 @@ typedef struct {
     //@}
     //@{
     /** data buffers */
-    float              *outSamples[2];
     uint8_t*            decoded_bytes_buffer;
     float               tempBuf[1070];
     //@}
@@ -198,7 +197,7 @@ static int decode_bytes(const uint8_t* inbuffer, uint8_t* out, int bytes){
 }
 
 
-static av_cold int init_atrac3_transforms(ATRAC3Context *q, int is_float) {
+static av_cold int init_atrac3_transforms(ATRAC3Context *q) {
     float enc_window[256];
     int i;
 
@@ -214,7 +213,7 @@ static av_cold int init_atrac3_transforms(ATRAC3Context *q, int is_float) {
         }
 
     /* Initialize the MDCT transform. */
-    return ff_mdct_init(&q->mdct_ctx, 9, 1, is_float ? 1.0 / 32768 : 1.0);
+    return ff_mdct_init(&q->mdct_ctx, 9, 1, 1.0 / 32768);
 }
 
 /**
@@ -227,7 +226,6 @@ static av_cold int atrac3_decode_close(AVCodecContext *avctx)
 
     av_free(q->pUnits);
     av_free(q->decoded_bytes_buffer);
-    av_freep(&q->outSamples[0]);
 
     ff_mdct_end(&q->mdct_ctx);
 
@@ -838,8 +836,6 @@ static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
     ATRAC3Context *q = avctx->priv_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,
@@ -853,8 +849,6 @@ static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
         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) {
@@ -864,27 +858,13 @@ static int atrac3_decode_frame(AVCodecContext *avctx, void *data,
         databuf = buf;
     }
 
-    if (q->channels == 1 && avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
-        result = decodeFrame(q, databuf, &samples_flt);
-    else
-        result = decodeFrame(q, databuf, q->outSamples);
+    result = decodeFrame(q, databuf, (float **)q->frame.extended_data);
 
     if (result != 0) {
         av_log(NULL,AV_LOG_ERROR,"Frame decoding error!\n");
         return result;
     }
 
-    /* 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;
 
@@ -1006,12 +986,9 @@ static av_cold int atrac3_decode_init(AVCodecContext *avctx)
         vlcs_initialized = 1;
     }
 
-    if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT)
-        avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
-    else
-        avctx->sample_fmt = AV_SAMPLE_FMT_S16;
+    avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
 
-    if ((ret = init_atrac3_transforms(q, avctx->sample_fmt == AV_SAMPLE_FMT_FLT))) {
+    if ((ret = init_atrac3_transforms(q))) {
         av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");
         av_freep(&q->decoded_bytes_buffer);
         return ret;
@@ -1049,15 +1026,6 @@ static av_cold int atrac3_decode_init(AVCodecContext *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;
 
@@ -1076,4 +1044,6 @@ AVCodec ff_atrac3_decoder =
     .decode         = atrac3_decode_frame,
     .capabilities   = CODEC_CAP_SUBFRAMES | CODEC_CAP_DR1,
     .long_name      = NULL_IF_CONFIG_SMALL("Atrac 3 (Adaptive TRansform Acoustic Coding 3)"),
+    .sample_fmts     = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
+                                                       AV_SAMPLE_FMT_NONE },
 };
diff --git a/libavcodec/cook.c b/libavcodec/cook.c
index cca97e3046..b4d1a37ca1 100644
--- a/libavcodec/cook.c
+++ b/libavcodec/cook.c
@@ -119,9 +119,10 @@ typedef struct cook {
     void (*interpolate)(struct cook *q, float *buffer,
                         int gain_index, int gain_index_next);
 
-    void (*saturate_output)(struct cook *q, int chan, float *out);
+    void (*saturate_output)(struct cook *q, float *out);
 
     AVCodecContext*     avctx;
+    DSPContext          dsp;
     AVFrame             frame;
     GetBitContext       gb;
     /* stream data */
@@ -887,18 +888,15 @@ static inline void decode_bytes_and_gain(COOKContext *q, COOKSubpacket *p,
  * Saturate the output signal and interleave.
  *
  * @param q                 pointer to the COOKContext
- * @param chan              channel to saturate
  * @param out               pointer to the output vector
  */
-static void saturate_output_float(COOKContext *q, int chan, float *out)
+static void saturate_output_float(COOKContext *q, float *out)
 {
-    int j;
-    float *output = q->mono_mdct_output + q->samples_per_channel;
-    for (j = 0; j < q->samples_per_channel; j++) {
-        out[chan + q->nb_channels * j] = av_clipf(output[j], -1.0, 1.0);
-    }
+    q->dsp.vector_clipf(out, q->mono_mdct_output + q->samples_per_channel,
+                        -1.0f, 1.0f, FFALIGN(q->samples_per_channel, 8));
 }
 
+
 /**
  * Final part of subpacket decoding:
  *  Apply modulated lapped transform, gain compensation,
@@ -909,15 +907,14 @@ static void saturate_output_float(COOKContext *q, int chan, float *out)
  * @param gains_ptr         array of current/prev gain pointers
  * @param previous_buffer   pointer to the previous buffer to be used for overlapping
  * @param out               pointer to the output buffer
- * @param chan              0: left or single channel, 1: right channel
  */
 static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
                                          cook_gains *gains_ptr, float *previous_buffer,
-                                         float *out, int chan)
+                                         float *out)
 {
     imlt_gain(q, decode_buffer, gains_ptr, previous_buffer);
     if (out)
-        q->saturate_output(q, chan, out);
+        q->saturate_output(q, out);
 }
 
 
@@ -930,7 +927,7 @@ static inline void mlt_compensate_output(COOKContext *q, float *decode_buffer,
  * @param outbuffer         pointer to the outbuffer
  */
 static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
-                            const uint8_t *inbuffer, float *outbuffer)
+                            const uint8_t *inbuffer, float **outbuffer)
 {
     int sub_packet_size = p->size;
     int res;
@@ -953,15 +950,18 @@ static int decode_subpacket(COOKContext *q, COOKSubpacket *p,
     }
 
     mlt_compensate_output(q, q->decode_buffer_1, &p->gains1,
-                          p->mono_previous_buffer1, outbuffer, p->ch_idx);
+                          p->mono_previous_buffer1,
+                          outbuffer ? outbuffer[p->ch_idx] : NULL);
 
     if (p->num_channels == 2)
         if (p->joint_stereo)
             mlt_compensate_output(q, q->decode_buffer_2, &p->gains1,
-                                  p->mono_previous_buffer2, outbuffer, p->ch_idx + 1);
+                                  p->mono_previous_buffer2,
+                                  outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
         else
             mlt_compensate_output(q, q->decode_buffer_2, &p->gains2,
-                                  p->mono_previous_buffer2, outbuffer, p->ch_idx + 1);
+                                  p->mono_previous_buffer2,
+                                  outbuffer ? outbuffer[p->ch_idx + 1] : NULL);
 
     return 0;
 }
@@ -978,7 +978,7 @@ static int cook_decode_frame(AVCodecContext *avctx, void *data,
     const uint8_t *buf = avpkt->data;
     int buf_size = avpkt->size;
     COOKContext *q = avctx->priv_data;
-    float *samples = NULL;
+    float **samples = NULL;
     int i, ret;
     int offset = 0;
     int chidx = 0;
@@ -993,7 +993,7 @@ static int cook_decode_frame(AVCodecContext *avctx, void *data,
             av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
             return ret;
         }
-        samples = (float *) q->frame.data[0];
+        samples = (float **)q->frame.extended_data;
     }
 
     /* estimate subpacket sizes */
@@ -1110,6 +1110,8 @@ static av_cold int cook_decode_init(AVCodecContext *avctx)
     /* Initialize RNG. */
     av_lfg_init(&q->random_state, 0);
 
+    ff_dsputil_init(&q->dsp, avctx);
+
     while (edata_ptr < edata_ptr_end) {
         /* 8 for mono, 16 for stereo, ? for multichannel
            Swap to right endianness so we don't need to care later on. */
@@ -1290,7 +1292,7 @@ static av_cold int cook_decode_init(AVCodecContext *avctx)
         return AVERROR_PATCHWELCOME;
     }
 
-    avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
+    avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
     if (channel_mask)
         avctx->channel_layout = channel_mask;
     else
@@ -1315,4 +1317,6 @@ AVCodec ff_cook_decoder = {
     .decode         = cook_decode_frame,
     .capabilities   = CODEC_CAP_DR1,
     .long_name      = NULL_IF_CONFIG_SMALL("Cook / Cooker / Gecko (RealAudio G2)"),
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
+                                                      AV_SAMPLE_FMT_NONE },
 };
diff --git a/libavcodec/dcadec.c b/libavcodec/dcadec.c
index 4e9f957660..4d5e1152f2 100644
--- a/libavcodec/dcadec.c
+++ b/libavcodec/dcadec.c
@@ -417,11 +417,9 @@ typedef struct {
     DECLARE_ALIGNED(32, float, raXin)[32];
 
     int output;                 ///< type of output
-    float scale_bias;           ///< output scale
 
     DECLARE_ALIGNED(32, float, subband_samples)[DCA_BLOCKS_MAX][DCA_PRIM_CHANNELS_MAX][DCA_SUBBANDS][8];
-    DECLARE_ALIGNED(32, float, samples)[(DCA_PRIM_CHANNELS_MAX + 1) * 256];
-    const float *samples_chanptr[DCA_PRIM_CHANNELS_MAX + 1];
+    float *samples_chanptr[DCA_PRIM_CHANNELS_MAX + 1];
 
     uint8_t dca_buffer[DCA_MAX_FRAME_SIZE + DCA_MAX_EXSS_HEADER_SIZE + DCA_BUFFER_PADDING_SIZE];
     int dca_buffer_size;        ///< how much data is in the dca_buffer
@@ -1169,20 +1167,20 @@ static void lfe_interpolation_fir(DCAContext *s, int decimation_select,
 }
 
 /* downmixing routines */
-#define MIX_REAR1(samples, si1, rs, coef)           \
-    samples[i]     += samples[si1] * coef[rs][0];   \
-    samples[i+256] += samples[si1] * coef[rs][1];
+#define MIX_REAR1(samples, s1, rs, coef)            \
+    samples[0][i] += samples[s1][i] * coef[rs][0];  \
+    samples[1][i] += samples[s1][i] * coef[rs][1];
 
-#define MIX_REAR2(samples, si1, si2, rs, coef)                                     \
-    samples[i]     += samples[si1] * coef[rs][0] + samples[si2] * coef[rs + 1][0]; \
-    samples[i+256] += samples[si1] * coef[rs][1] + samples[si2] * coef[rs + 1][1];
+#define MIX_REAR2(samples, s1, s2, rs, coef)                                          \
+    samples[0][i] += samples[s1][i] * coef[rs][0] + samples[s2][i] * coef[rs + 1][0]; \
+    samples[1][i] += samples[s1][i] * coef[rs][1] + samples[s2][i] * coef[rs + 1][1];
 
 #define MIX_FRONT3(samples, coef)                                      \
-    t = samples[i + c];                                                \
-    u = samples[i + l];                                                \
-    v = samples[i + r];                                                \
-    samples[i]     = t * coef[0][0] + u * coef[1][0] + v * coef[2][0]; \
-    samples[i+256] = t * coef[0][1] + u * coef[1][1] + v * coef[2][1];
+    t = samples[c][i];                                                 \
+    u = samples[l][i];                                                 \
+    v = samples[r][i];                                                 \
+    samples[0][i] = t * coef[0][0] + u * coef[1][0] + v * coef[2][0];  \
+    samples[1][i] = t * coef[0][1] + u * coef[1][1] + v * coef[2][1];
 
 #define DOWNMIX_TO_STEREO(op1, op2)             \
     for (i = 0; i < 256; i++) {                 \
@@ -1190,7 +1188,7 @@ static void lfe_interpolation_fir(DCAContext *s, int decimation_select,
         op2                                     \
     }
 
-static void dca_downmix(float *samples, int srcfmt,
+static void dca_downmix(float **samples, int srcfmt,
                         int downmix_coef[DCA_PRIM_CHANNELS_MAX][2],
                         const int8_t *channel_mapping)
 {
@@ -1215,36 +1213,36 @@ static void dca_downmix(float *samples, int srcfmt,
     case DCA_STEREO:
         break;
     case DCA_3F:
-        c = channel_mapping[0] * 256;
-        l = channel_mapping[1] * 256;
-        r = channel_mapping[2] * 256;
+        c = channel_mapping[0];
+        l = channel_mapping[1];
+        r = channel_mapping[2];
         DOWNMIX_TO_STEREO(MIX_FRONT3(samples, coef), );
         break;
     case DCA_2F1R:
-        s = channel_mapping[2] * 256;
-        DOWNMIX_TO_STEREO(MIX_REAR1(samples, i + s, 2, coef), );
+        s = channel_mapping[2];
+        DOWNMIX_TO_STEREO(MIX_REAR1(samples, s, 2, coef), );
         break;
     case DCA_3F1R:
-        c = channel_mapping[0] * 256;
-        l = channel_mapping[1] * 256;
-        r = channel_mapping[2] * 256;
-        s = channel_mapping[3] * 256;
+        c = channel_mapping[0];
+        l = channel_mapping[1];
+        r = channel_mapping[2];
+        s = channel_mapping[3];
         DOWNMIX_TO_STEREO(MIX_FRONT3(samples, coef),
-                          MIX_REAR1(samples, i + s, 3, coef));
+                          MIX_REAR1(samples, s, 3, coef));
         break;
     case DCA_2F2R:
-        sl = channel_mapping[2] * 256;
-        sr = channel_mapping[3] * 256;
-        DOWNMIX_TO_STEREO(MIX_REAR2(samples, i + sl, i + sr, 2, coef), );
+        sl = channel_mapping[2];
+        sr = channel_mapping[3];
+        DOWNMIX_TO_STEREO(MIX_REAR2(samples, sl, sr, 2, coef), );
         break;
     case DCA_3F2R:
-        c  = channel_mapping[0] * 256;
-        l  = channel_mapping[1] * 256;
-        r  = channel_mapping[2] * 256;
-        sl = channel_mapping[3] * 256;
-        sr = channel_mapping[4] * 256;
+        c  = channel_mapping[0];
+        l  = channel_mapping[1];
+        r  = channel_mapping[2];
+        sl = channel_mapping[3];
+        sr = channel_mapping[4];
         DOWNMIX_TO_STEREO(MIX_FRONT3(samples, coef),
-                          MIX_REAR2(samples, i + sl, i + sr, 3, coef));
+                          MIX_REAR2(samples, sl, sr, 3, coef));
         break;
     }
 }
@@ -1441,21 +1439,21 @@ static int dca_filter_channels(DCAContext *s, int block_index)
 /*        static float pcm_to_double[8] = { 32768.0, 32768.0, 524288.0, 524288.0,
                                             0, 8388608.0, 8388608.0 };*/
         qmf_32_subbands(s, k, subband_samples[k],
-                        &s->samples[256 * s->channel_order_tab[k]],
-                        M_SQRT1_2 * s->scale_bias /* pcm_to_double[s->source_pcm_res] */);
+                        s->samples_chanptr[s->channel_order_tab[k]],
+                        M_SQRT1_2 / 32768.0 /* pcm_to_double[s->source_pcm_res] */);
     }
 
     /* Down mixing */
     if (s->avctx->request_channels == 2 && s->prim_channels > 2) {
-        dca_downmix(s->samples, s->amode, s->downmix_coef, s->channel_order_tab);
+        dca_downmix(s->samples_chanptr, s->amode, s->downmix_coef, s->channel_order_tab);
     }
 
     /* Generate LFE samples for this subsubframe FIXME!!! */
     if (s->output & DCA_LFE) {
         lfe_interpolation_fir(s, s->lfe, 2 * s->lfe,
                               s->lfe_data + 2 * s->lfe * (block_index + 4),
-                              &s->samples[256 * s->lfe_index],
-                              (1.0 / 256.0) * s->scale_bias);
+                              s->samples_chanptr[s->lfe_index],
+                              1.0 / (256.0 * 32768.0));
         /* Outputs 20bits pcm samples */
     }
 
@@ -2067,10 +2065,9 @@ static int dca_decode_frame(AVCodecContext *avctx, void *data,
     int lfe_samples;
     int num_core_channels = 0;
     int i, ret;
-    float *samples_flt;
+    float **samples_flt;
     float *src_chan;
     float *dst_chan;
-    int16_t *samples_s16;
     DCAContext *s = avctx->priv_data;
     int core_ss_end;
     int channels;
@@ -2081,7 +2078,6 @@ static int dca_decode_frame(AVCodecContext *avctx, void *data,
     int lavc;
     int posn;
     int j, k;
-    int ch;
     int endch;
 
     s->xch_present = 0;
@@ -2342,19 +2338,23 @@ static int dca_decode_frame(AVCodecContext *avctx, void *data,
         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
         return ret;
     }
-    samples_flt = (float *)   s->frame.data[0];
-    samples_s16 = (int16_t *) s->frame.data[0];
+    samples_flt = (float  **) s->frame.extended_data;
 
     /* filter to get final output */
     for (i = 0; i < (s->sample_blocks / 8); i++) {
+        int ch;
+
+        for (ch = 0; ch < channels; ch++)
+            s->samples_chanptr[ch] = samples_flt[ch] + i * 256;
+
         dca_filter_channels(s, i);
 
         /* If this was marked as a DTS-ES stream we need to subtract back- */
         /* channel from SL & SR to remove matrixed back-channel signal */
         if ((s->source_pcm_res & 1) && s->xch_present) {
-            float *back_chan = s->samples + s->channel_order_tab[s->xch_base_channel]     * 256;
-            float *lt_chan   = s->samples + s->channel_order_tab[s->xch_base_channel - 2] * 256;
-            float *rt_chan   = s->samples + s->channel_order_tab[s->xch_base_channel - 1] * 256;
+            float *back_chan = s->samples_chanptr[s->channel_order_tab[s->xch_base_channel]];
+            float *lt_chan   = s->samples_chanptr[s->channel_order_tab[s->xch_base_channel - 2]];
+            float *rt_chan   = s->samples_chanptr[s->channel_order_tab[s->xch_base_channel - 1]];
             s->fdsp.vector_fmac_scalar(lt_chan, back_chan, -M_SQRT1_2, 256);
             s->fdsp.vector_fmac_scalar(rt_chan, back_chan, -M_SQRT1_2, 256);
         }
@@ -2370,12 +2370,12 @@ static int dca_decode_frame(AVCodecContext *avctx, void *data,
                 /* undo downmix */
                 for (j = ch; j < endch; j++) {
                     if (mask & (1 << j)) { /* this channel has been mixed-out */
-                        src_chan = s->samples + s->channel_order_tab[j] * 256;
+                        src_chan = s->samples_chanptr[s->channel_order_tab[j]];
                         for (k = 0; k < endch; k++) {
                             achan = s->channel_order_tab[k];
                             scale = s->xxch_dmix_coeff[j][k];
                             if (scale != 0.0) {
-                                dst_chan = s->samples + achan * 256;
+                                dst_chan = s->samples_chanptr[achan];
                                 s->fdsp.vector_fmac_scalar(dst_chan, src_chan,
                                                            -scale, 256);
                             }
@@ -2388,14 +2388,14 @@ static int dca_decode_frame(AVCodecContext *avctx, void *data,
                     scale = s->xxch_dmix_sf[chset];
 
                     for (j = 0; j < ch; j++) {
-                        src_chan = s->samples + s->channel_order_tab[j] * 256;
+                        src_chan = s->samples_chanptr[s->channel_order_tab[j]];
                         for (k = 0; k < 256; k++)
                             src_chan[k] *= scale;
                     }
 
                     /* LFE channel is always part of core, scale if it exists */
                     if (s->lfe) {
-                        src_chan = s->samples + s->lfe_index * 256;
+                        src_chan = s->samples_chanptr[s->lfe_index];
                         for (k = 0; k < 256; k++)
                             src_chan[k] *= scale;
                     }
@@ -2405,17 +2405,6 @@ static int dca_decode_frame(AVCodecContext *avctx, void *data,
             }
 
         }
-
-        if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT) {
-            s->fmt_conv.float_interleave(samples_flt, s->samples_chanptr, 256,
-                                         channels);
-            samples_flt += 256 * channels;
-        } else {
-            s->fmt_conv.float_to_int16_interleave(samples_s16,
-                                                  s->samples_chanptr, 256,
-                                                  channels);
-            samples_s16 += 256 * channels;
-        }
     }
 
     /* update lfe history */
@@ -2440,7 +2429,6 @@ static int dca_decode_frame(AVCodecContext *avctx, void *data,
 static av_cold int dca_decode_init(AVCodecContext *avctx)
 {
     DCAContext *s = avctx->priv_data;
-    int i;
 
     s->avctx = avctx;
     dca_init_vlcs();
@@ -2451,16 +2439,7 @@ static av_cold int dca_decode_init(AVCodecContext *avctx)
     ff_dcadsp_init(&s->dcadsp);
     ff_fmt_convert_init(&s->fmt_conv, avctx);
 
-    for (i = 0; i < DCA_PRIM_CHANNELS_MAX + 1; i++)
-        s->samples_chanptr[i] = s->samples + i * 256;
-
-    if (avctx->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
-        avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
-        s->scale_bias     = 1.0 / 32768.0;
-    } else {
-        avctx->sample_fmt = AV_SAMPLE_FMT_S16;
-        s->scale_bias     = 1.0;
-    }
+    avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
 
     /* allow downmixing to stereo */
     if (avctx->channels > 0 && avctx->request_channels < avctx->channels &&
@@ -2500,8 +2479,7 @@ AVCodec ff_dca_decoder = {
     .close           = dca_decode_end,
     .long_name       = NULL_IF_CONFIG_SMALL("DCA (DTS Coherent Acoustics)"),
     .capabilities    = CODEC_CAP_CHANNEL_CONF | CODEC_CAP_DR1,
-    .sample_fmts     = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLT,
-                                                       AV_SAMPLE_FMT_S16,
+    .sample_fmts     = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
                                                        AV_SAMPLE_FMT_NONE },
     .profiles        = NULL_IF_CONFIG_SMALL(profiles),
 };
diff --git a/libavcodec/imc.c b/libavcodec/imc.c
index d12c7551e9..9da769ba38 100644
--- a/libavcodec/imc.c
+++ b/libavcodec/imc.c
@@ -242,7 +242,7 @@ static av_cold int imc_decode_init(AVCodecContext *avctx)
         return ret;
     }
     ff_dsputil_init(&q->dsp, avctx);
-    avctx->sample_fmt     = AV_SAMPLE_FMT_FLT;
+    avctx->sample_fmt     = AV_SAMPLE_FMT_FLTP;
     avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
                                                  : AV_CH_LAYOUT_STEREO;
 
@@ -662,7 +662,7 @@ static void imc_imdct256(IMCContext *q, IMCChannel *chctx, int channels)
     int i;
     float re, im;
     float *dst1 = q->out_samples;
-    float *dst2 = q->out_samples + (COEFFS - 1) * channels;
+    float *dst2 = q->out_samples + (COEFFS - 1);
 
     /* prerotation */
     for (i = 0; i < COEFFS / 2; i++) {
@@ -684,8 +684,8 @@ static void imc_imdct256(IMCContext *q, IMCChannel *chctx, int channels)
                + (q->mdct_sine_window[i * 2] * re);
         *dst2 =  (q->mdct_sine_window[i * 2] * chctx->last_fft_im[i])
                - (q->mdct_sine_window[COEFFS - 1 - i * 2] * re);
-        dst1 += channels * 2;
-        dst2 -= channels * 2;
+        dst1 += 2;
+        dst2 -= 2;
         chctx->last_fft_im[i] = im;
     }
 }
@@ -786,7 +786,6 @@ static int imc_decode_block(AVCodecContext *avctx, IMCContext *q, int ch)
         chctx->decoder_reset = 1;
 
     if (chctx->decoder_reset) {
-        memset(q->out_samples, 0, COEFFS * sizeof(*q->out_samples));
         for (i = 0; i < BANDS; i++)
             chctx->old_floor[i] = 1.0;
         for (i = 0; i < COEFFS; i++)
@@ -945,7 +944,7 @@ static int imc_decode_frame(AVCodecContext *avctx, void *data,
     }
 
     for (i = 0; i < avctx->channels; i++) {
-        q->out_samples = (float*)q->frame.data[0] + i;
+        q->out_samples = (float *)q->frame.extended_data[i];
 
         q->dsp.bswap16_buf(buf16, (const uint16_t*)buf, IMC_BLOCK_SIZE / 2);
 
@@ -958,15 +957,8 @@ static int imc_decode_frame(AVCodecContext *avctx, void *data,
     }
 
     if (avctx->channels == 2) {
-        float *src = (float*)q->frame.data[0], t1, t2;
-
-        for (i = 0; i < COEFFS; i++) {
-            t1     = src[0];
-            t2     = src[1];
-            src[0] = t1 + t2;
-            src[1] = t1 - t2;
-            src   += 2;
-        }
+        q->dsp.butterflies_float((float *)q->frame.extended_data[0],
+                                 (float *)q->frame.extended_data[1], COEFFS);
     }
 
     *got_frame_ptr   = 1;
@@ -996,6 +988,8 @@ AVCodec ff_imc_decoder = {
     .decode         = imc_decode_frame,
     .capabilities   = CODEC_CAP_DR1,
     .long_name      = NULL_IF_CONFIG_SMALL("IMC (Intel Music Coder)"),
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
+                                                      AV_SAMPLE_FMT_NONE },
 };
 #endif
 #if CONFIG_IAC_DECODER
@@ -1009,5 +1003,7 @@ AVCodec ff_iac_decoder = {
     .decode         = imc_decode_frame,
     .capabilities   = CODEC_CAP_DR1,
     .long_name      = NULL_IF_CONFIG_SMALL("IAC (Indeo Audio Coder)"),
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
+                                                      AV_SAMPLE_FMT_NONE },
 };
 #endif
diff --git a/libavcodec/mpc.c b/libavcodec/mpc.c
index 7ef437c4dc..d064924b91 100644
--- a/libavcodec/mpc.c
+++ b/libavcodec/mpc.c
@@ -43,28 +43,24 @@ void ff_mpc_init(void)
 /**
  * Process decoded Musepack data and produce PCM
  */
-static void mpc_synth(MPCContext *c, int16_t *out, int channels)
+static void mpc_synth(MPCContext *c, int16_t **out, int channels)
 {
     int dither_state = 0;
     int i, ch;
-    OUT_INT samples[MPA_MAX_CHANNELS * MPA_FRAME_SIZE], *samples_ptr;
 
     for(ch = 0;  ch < channels; ch++){
-        samples_ptr = samples + ch;
         for(i = 0; i < SAMPLES_PER_BAND; i++) {
             ff_mpa_synth_filter_fixed(&c->mpadsp,
                                 c->synth_buf[ch], &(c->synth_buf_offset[ch]),
                                 ff_mpa_synth_window_fixed, &dither_state,
-                                samples_ptr, channels,
+                                out[ch] + 32 * i, 1,
                                 c->sb_samples[ch][i]);
-            samples_ptr += 32 * channels;
         }
     }
-    for(i = 0; i < MPC_FRAME_SIZE*channels; i++)
-        *out++=samples[i];
 }
 
-void ff_mpc_dequantize_and_synth(MPCContext * c, int maxband, void *data, int channels)
+void ff_mpc_dequantize_and_synth(MPCContext * c, int maxband, int16_t **out,
+                                 int channels)
 {
     int i, j, ch;
     Band *bands = c->bands;
@@ -100,5 +96,5 @@ void ff_mpc_dequantize_and_synth(MPCContext * c, int maxband, void *data, int ch
         }
     }
 
-    mpc_synth(c, data, channels);
+    mpc_synth(c, out, channels);
 }
diff --git a/libavcodec/mpc.h b/libavcodec/mpc.h
index 8b4deef689..34ef263b40 100644
--- a/libavcodec/mpc.h
+++ b/libavcodec/mpc.h
@@ -73,6 +73,6 @@ typedef struct {
 } MPCContext;
 
 void ff_mpc_init(void);
-void ff_mpc_dequantize_and_synth(MPCContext *c, int maxband, void *dst, int channels);
+void ff_mpc_dequantize_and_synth(MPCContext *c, int maxband, int16_t **out, int channels);
 
 #endif /* AVCODEC_MPC_H */
diff --git a/libavcodec/mpc7.c b/libavcodec/mpc7.c
index cab4eee03d..9f3cf6c828 100644
--- a/libavcodec/mpc7.c
+++ b/libavcodec/mpc7.c
@@ -90,7 +90,7 @@ static av_cold int mpc7_decode_init(AVCodecContext * avctx)
             c->IS, c->MSS, c->gapless, c->lastframelen, c->maxbands);
     c->frames_to_skip = 0;
 
-    avctx->sample_fmt = AV_SAMPLE_FMT_S16;
+    avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
     avctx->channel_layout = AV_CH_LAYOUT_STEREO;
 
     avcodec_get_frame_defaults(&c->frame);
@@ -293,7 +293,7 @@ static int mpc7_decode_frame(AVCodecContext * avctx, void *data,
         for(ch = 0; ch < 2; ch++)
             idx_to_quant(c, &gb, bands[i].res[ch], c->Q[ch] + off);
 
-    ff_mpc_dequantize_and_synth(c, mb, c->frame.data[0], 2);
+    ff_mpc_dequantize_and_synth(c, mb, (int16_t **)c->frame.extended_data, 2);
     if(last_frame)
         c->frame.nb_samples = c->lastframelen;
 
@@ -342,4 +342,6 @@ AVCodec ff_mpc7_decoder = {
     .flush          = mpc7_decode_flush,
     .capabilities   = CODEC_CAP_DR1,
     .long_name      = NULL_IF_CONFIG_SMALL("Musepack SV7"),
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
+                                                      AV_SAMPLE_FMT_NONE },
 };
diff --git a/libavcodec/mpc8.c b/libavcodec/mpc8.c
index cb30b2c28b..64d19193b0 100644
--- a/libavcodec/mpc8.c
+++ b/libavcodec/mpc8.c
@@ -139,7 +139,7 @@ static av_cold int mpc8_decode_init(AVCodecContext * avctx)
     c->MSS = get_bits1(&gb);
     c->frames = 1 << (get_bits(&gb, 3) * 2);
 
-    avctx->sample_fmt = AV_SAMPLE_FMT_S16;
+    avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
     avctx->channel_layout = (channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
     avctx->channels = channels;
 
@@ -413,7 +413,8 @@ static int mpc8_decode_frame(AVCodecContext * avctx, void *data,
         }
     }
 
-    ff_mpc_dequantize_and_synth(c, maxband - 1, c->frame.data[0],
+    ff_mpc_dequantize_and_synth(c, maxband - 1,
+                                (int16_t **)c->frame.extended_data,
                                 avctx->channels);
 
     c->cur_frame++;
@@ -446,4 +447,6 @@ AVCodec ff_mpc8_decoder = {
     .flush          = mpc8_decode_flush,
     .capabilities   = CODEC_CAP_DR1,
     .long_name      = NULL_IF_CONFIG_SMALL("Musepack SV8"),
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
+                                                      AV_SAMPLE_FMT_NONE },
 };
diff --git a/libavcodec/ralf.c b/libavcodec/ralf.c
index 6866f79609..3244424f74 100644
--- a/libavcodec/ralf.c
+++ b/libavcodec/ralf.c
@@ -149,7 +149,7 @@ static av_cold int decode_init(AVCodecContext *avctx)
                avctx->sample_rate, avctx->channels);
         return AVERROR_INVALIDDATA;
     }
-    avctx->sample_fmt     = AV_SAMPLE_FMT_S16;
+    avctx->sample_fmt     = AV_SAMPLE_FMT_S16P;
     avctx->channel_layout = (avctx->channels == 2) ? AV_CH_LAYOUT_STEREO
                                                    : AV_CH_LAYOUT_MONO;
 
@@ -338,7 +338,8 @@ static void apply_lpc(RALFContext *ctx, int ch, int length, int bits)
     }
 }
 
-static int decode_block(AVCodecContext *avctx, GetBitContext *gb, int16_t *dst)
+static int decode_block(AVCodecContext *avctx, GetBitContext *gb,
+                        int16_t *dst0, int16_t *dst1)
 {
     RALFContext *ctx = avctx->priv_data;
     int len, ch, ret;
@@ -382,35 +383,35 @@ static int decode_block(AVCodecContext *avctx, GetBitContext *gb, int16_t *dst)
     switch (dmode) {
     case 0:
         for (i = 0; i < len; i++)
-            *dst++ = ch0[i] + ctx->bias[0];
+            dst0[i] = ch0[i] + ctx->bias[0];
         break;
     case 1:
         for (i = 0; i < len; i++) {
-            *dst++ = ch0[i] + ctx->bias[0];
-            *dst++ = ch1[i] + ctx->bias[1];
+            dst0[i] = ch0[i] + ctx->bias[0];
+            dst1[i] = ch1[i] + ctx->bias[1];
         }
         break;
     case 2:
         for (i = 0; i < len; i++) {
             ch0[i] += ctx->bias[0];
-            *dst++ = ch0[i];
-            *dst++ = ch0[i] - (ch1[i] + ctx->bias[1]);
+            dst0[i] = ch0[i];
+            dst1[i] = ch0[i] - (ch1[i] + ctx->bias[1]);
         }
         break;
     case 3:
         for (i = 0; i < len; i++) {
             t  = ch0[i] + ctx->bias[0];
             t2 = ch1[i] + ctx->bias[1];
-            *dst++ = t + t2;
-            *dst++ = t;
+            dst0[i] = t + t2;
+            dst1[i] = t;
         }
         break;
     case 4:
         for (i = 0; i < len; i++) {
             t  =   ch1[i] + ctx->bias[1];
             t2 = ((ch0[i] + ctx->bias[0]) << 1) | (t & 1);
-            *dst++ = (t2 + t) / 2;
-            *dst++ = (t2 - t) / 2;
+            dst0[i] = (t2 + t) / 2;
+            dst1[i] = (t2 - t) / 2;
         }
         break;
     }
@@ -424,7 +425,8 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
                         AVPacket *avpkt)
 {
     RALFContext *ctx = avctx->priv_data;
-    int16_t *samples;
+    int16_t *samples0;
+    int16_t *samples1;
     int ret;
     GetBitContext gb;
     int table_size, table_bytes, i;
@@ -465,7 +467,8 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
         av_log(avctx, AV_LOG_ERROR, "Me fail get_buffer()? That's unpossible!\n");
         return ret;
     }
-    samples = (int16_t*)ctx->frame.data[0];
+    samples0 = (int16_t *)ctx->frame.data[0];
+    samples1 = (int16_t *)ctx->frame.data[1];
 
     if (src_size < 5) {
         av_log(avctx, AV_LOG_ERROR, "too short packets are too short!\n");
@@ -498,8 +501,8 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr,
             break;
         }
         init_get_bits(&gb, block_pointer, ctx->block_size[i] * 8);
-        if (decode_block(avctx, &gb, samples + ctx->sample_offset
-                                               * avctx->channels) < 0) {
+        if (decode_block(avctx, &gb, samples0 + ctx->sample_offset,
+                                     samples1 + ctx->sample_offset) < 0) {
             av_log(avctx, AV_LOG_ERROR, "Sir, I got carsick in your office. Not decoding the rest of packet.\n");
             break;
         }
@@ -533,4 +536,6 @@ AVCodec ff_ralf_decoder = {
     .flush          = decode_flush,
     .capabilities   = CODEC_CAP_DR1,
     .long_name      = NULL_IF_CONFIG_SMALL("RealAudio Lossless"),
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S16P,
+                                                      AV_SAMPLE_FMT_NONE },
 };
diff --git a/libavcodec/twinvq.c b/libavcodec/twinvq.c
index 070f498cae..f47f4c30f1 100644
--- a/libavcodec/twinvq.c
+++ b/libavcodec/twinvq.c
@@ -666,7 +666,7 @@ static void imdct_and_window(TwinContext *tctx, enum FrameType ftype, int wtype,
 }
 
 static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
-                         float *out)
+                         float **out)
 {
     const ModeTab *mtab = tctx->mtab;
     int size1, size2;
@@ -685,24 +685,15 @@ static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
 
     size2 = tctx->last_block_pos[0];
     size1 = mtab->size - size2;
+
+    memcpy(&out[0][0    ], prev_buf,         size1 * sizeof(out[0][0]));
+    memcpy(&out[0][size1], tctx->curr_frame, size2 * sizeof(out[0][0]));
+
     if (tctx->avctx->channels == 2) {
-        tctx->dsp.butterflies_float_interleave(out, prev_buf,
-                                               &prev_buf[2*mtab->size],
-                                               size1);
-
-        out += 2 * size1;
-
-        tctx->dsp.butterflies_float_interleave(out, tctx->curr_frame,
-                                               &tctx->curr_frame[2*mtab->size],
-                                               size2);
-    } else {
-        memcpy(out, prev_buf, size1 * sizeof(*out));
-
-        out += size1;
-
-        memcpy(out, tctx->curr_frame, size2 * sizeof(*out));
+        memcpy(&out[1][0],     &prev_buf[2*mtab->size],         size1 * sizeof(out[1][0]));
+        memcpy(&out[1][size1], &tctx->curr_frame[2*mtab->size], size2 * sizeof(out[1][0]));
+        tctx->dsp.butterflies_float(out[0], out[1], mtab->size);
     }
-
 }
 
 static void dec_bark_env(TwinContext *tctx, const uint8_t *in, int use_hist,
@@ -825,7 +816,7 @@ static int twin_decode_frame(AVCodecContext * avctx, void *data,
     TwinContext *tctx = avctx->priv_data;
     GetBitContext gb;
     const ModeTab *mtab = tctx->mtab;
-    float *out = NULL;
+    float **out = NULL;
     enum FrameType ftype;
     int window_type, ret;
     static const enum FrameType wtype_to_ftype_table[] = {
@@ -846,7 +837,7 @@ static int twin_decode_frame(AVCodecContext * avctx, void *data,
             av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
             return ret;
         }
-        out = (float *)tctx->frame.data[0];
+        out = (float **)tctx->frame.extended_data;
     }
 
     init_get_bits(&gb, buf, buf_size * 8);
@@ -1119,7 +1110,7 @@ static av_cold int twin_decode_init(AVCodecContext *avctx)
     int isampf, ibps;
 
     tctx->avctx       = avctx;
-    avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
+    avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
 
     if (!avctx->extradata || avctx->extradata_size < 12) {
         av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
@@ -1184,4 +1175,6 @@ AVCodec ff_twinvq_decoder = {
     .decode         = twin_decode_frame,
     .capabilities   = CODEC_CAP_DR1,
     .long_name      = NULL_IF_CONFIG_SMALL("VQF TwinVQ"),
+    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
+                                                      AV_SAMPLE_FMT_NONE },
 };