diff --git a/.gitignore b/.gitignore
index eac1758ed9..3e8ed0046e 100644
--- a/.gitignore
+++ b/.gitignore
@@ -16,6 +16,9 @@ ffmpeg
 ffplay
 ffprobe
 ffserver
+libavcodec/*_tablegen
+libavcodec/*_tables.c
+libavcodec/*_tables.h
 libavcodec/libavcodec*
 libavcore/libavcore*
 libavdevice/libavdevice*
diff --git a/configure b/configure
index 41a66d7ab4..6445636c6d 100755
--- a/configure
+++ b/configure
@@ -956,6 +956,7 @@ CONFIG_LIST="
     mdct
     memalign_hack
     mlib
+    mpegaudiodsp
     network
     nonfree
     pic
@@ -1244,6 +1245,7 @@ symver_if_any="symver_asm_label symver_gnu_asm"
 dct_select="rdft"
 mdct_select="fft"
 rdft_select="fft"
+mpegaudiodsp_select="dct"
 
 # decoders / encoders / hardware accelerators
 aac_decoder_select="mdct sinewin"
@@ -1297,11 +1299,16 @@ ljpeg_encoder_select="aandct"
 loco_decoder_select="golomb"
 mjpeg_encoder_select="aandct"
 mlp_decoder_select="mlp_parser"
-mp1float_decoder_select="dct"
-mp2float_decoder_select="dct"
-mp3adufloat_decoder_select="dct"
-mp3float_decoder_select="dct"
-mp3on4float_decoder_select="dct"
+mp1_decoder_select="mpegaudiodsp"
+mp2_decoder_select="mpegaudiodsp"
+mp3adu_decoder_select="mpegaudiodsp"
+mp3_decoder_select="mpegaudiodsp"
+mp3on4_decoder_select="mpegaudiodsp"
+mp1float_decoder_select="mpegaudiodsp"
+mp2float_decoder_select="mpegaudiodsp"
+mp3adufloat_decoder_select="mpegaudiodsp"
+mp3float_decoder_select="mpegaudiodsp"
+mp3on4float_decoder_select="mpegaudiodsp"
 mpeg1video_encoder_select="aandct"
 mpeg2video_encoder_select="aandct"
 mpeg4_decoder_select="h263_decoder mpeg4video_parser"
@@ -1331,7 +1338,7 @@ nellymoser_encoder_select="mdct sinewin"
 png_decoder_select="zlib"
 png_encoder_select="zlib"
 qcelp_decoder_select="lsp"
-qdm2_decoder_select="mdct rdft"
+qdm2_decoder_select="mdct rdft mpegaudiodsp"
 ra_144_encoder_select="lpc"
 rv10_decoder_select="h263_decoder"
 rv10_encoder_select="h263_encoder"
@@ -2753,7 +2760,7 @@ EOF
 
         check_yasm "pextrd [eax], xmm0, 1" && enable yasm ||
             die "yasm not found, use --disable-yasm for a crippled build"
-        check_yasm "vpaddw xmm0, xmm0, xmm0" || disable avx
+        check_yasm "vextractf128 xmm0, ymm0, 0" || disable avx
     fi
 
     case "$cpu" in
diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 6723118693..db184bcdce 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -41,6 +41,9 @@ OBJS-$(CONFIG_HUFFMAN)                 += huffman.o
 OBJS-$(CONFIG_LPC)                     += lpc.o
 OBJS-$(CONFIG_LSP)                     += lsp.o
 OBJS-$(CONFIG_MDCT)                    += mdct_fixed.o mdct_float.o
+OBJS-$(CONFIG_MPEGAUDIODSP)            += mpegaudiodsp.o                \
+                                          mpegaudiodsp_fixed.o          \
+                                          mpegaudiodsp_float.o
 RDFT-OBJS-$(CONFIG_HARDCODED_TABLES)   += sin_tables.o
 OBJS-$(CONFIG_RDFT)                    += rdft.o $(RDFT-OBJS-yes)
 OBJS-$(CONFIG_SINEWIN)                 += sinewin.o
@@ -676,24 +679,23 @@ TESTPROGS = cabac dct eval fft fft-fixed h264 iirfilter rangecoder snow
 TESTPROGS-$(HAVE_MMX) += motion
 TESTOBJS = dctref.o
 
-HOSTPROGS = costablegen
+HOSTPROGS = aac_tablegen aacps_tablegen cbrt_tablegen cos_tablegen      \
+            dv_tablegen motionpixels_tablegen mpegaudio_tablegen        \
+            pcm_tablegen qdm2_tablegen sinewin_tablegen
 
 DIRS = alpha arm bfin mlib ppc ps2 sh4 sparc x86
 
-CLEANFILES = sin_tables.c cos_tables.c *_tables.h *_tablegen$(HOSTEXESUF)
+CLEANFILES = *_tables.c *_tables.h *_tablegen$(HOSTEXESUF)
 
 include $(SUBDIR)../subdir.mak
 
 $(SUBDIR)dct-test$(EXESUF): $(SUBDIR)dctref.o
 
-$(SUBDIR)cos_tables.c: $(SUBDIR)costablegen$(HOSTEXESUF)
-	$(M)./$< > $@
+TRIG_TABLES  = cos cos_fixed sin
+TRIG_TABLES := $(TRIG_TABLES:%=$(SUBDIR)%_tables.c)
 
-$(SUBDIR)cos_fixed_tables.c: $(SUBDIR)costablegen$(HOSTEXESUF)
-	$(M)./$< cos fixed > $@
-
-$(SUBDIR)sin_tables.c: $(SUBDIR)costablegen$(HOSTEXESUF)
-	$(M)./$< sin > $@
+$(TRIG_TABLES): $(SUBDIR)%_tables.c: $(SUBDIR)cos_tablegen$(HOSTEXESUF)
+	$(M)./$< $* > $@
 
 ifdef CONFIG_SMALL
 $(SUBDIR)%_tablegen$(HOSTEXESUF): HOSTCFLAGS += -DCONFIG_SMALL=1
@@ -701,9 +703,6 @@ else
 $(SUBDIR)%_tablegen$(HOSTEXESUF): HOSTCFLAGS += -DCONFIG_SMALL=0
 endif
 
-$(SUBDIR)%_tablegen$(HOSTEXESUF): $(SUBDIR)%_tablegen.c $(SUBDIR)%_tablegen.h $(SUBDIR)tableprint.c
-	$(HOSTCC) $(HOSTCFLAGS) $(HOSTLDFLAGS) -o $@ $(filter %.c,$^) $(HOSTLIBS)
-
 GEN_HEADERS = cbrt_tables.h aacps_tables.h aac_tables.h dv_tables.h     \
               sinewin_tables.h mpegaudio_tables.h motionpixels_tables.h \
               pcm_tables.h qdm2_tables.h
diff --git a/libavcodec/costablegen.c b/libavcodec/cos_tablegen.c
similarity index 94%
rename from libavcodec/costablegen.c
rename to libavcodec/cos_tablegen.c
index 6e818252f7..1577166a46 100644
--- a/libavcodec/costablegen.c
+++ b/libavcodec/cos_tablegen.c
@@ -51,10 +51,10 @@ int main(int argc, char *argv[])
 {
     int i, j;
     int do_sin = argc > 1 && !strcmp(argv[1], "sin");
-    int fixed  = argc > 2 && !strcmp(argv[2], "fixed");
+    int fixed  = argc > 1 &&  strstr(argv[1], "fixed");
     double (*func)(double) = do_sin ? sin : cos;
 
-    printf("/* This file was generated by libavcodec/costablegen */\n");
+    printf("/* This file was automatically generated. */\n");
     printf("#define CONFIG_FFT_FLOAT %d\n", !fixed);
     printf("#include \"libavcodec/%s\"\n", do_sin ? "rdft.h" : "fft.h");
     for (i = 4; i <= BITS; i++) {
diff --git a/libavcodec/h264.c b/libavcodec/h264.c
index ae3d263535..3068db8d85 100644
--- a/libavcodec/h264.c
+++ b/libavcodec/h264.c
@@ -1002,7 +1002,7 @@ static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
                                   int linesize, int uvlinesize,
                                   int xchg, int simple, int pixel_shift){
     MpegEncContext * const s = &h->s;
-    int deblock_left;
+    int deblock_topleft;
     int deblock_top;
     int top_idx = 1;
     uint8_t *top_border_m1;
@@ -1018,11 +1018,11 @@ static inline void xchg_mb_border(H264Context *h, uint8_t *src_y,
     }
 
     if(h->deblocking_filter == 2) {
-        deblock_left = h->left_type[0];
-        deblock_top  = h->top_type;
+        deblock_topleft = h->slice_table[h->mb_xy - 1 - s->mb_stride] == h->slice_num;
+        deblock_top     = h->top_type;
     } else {
-        deblock_left = (s->mb_x > 0);
-        deblock_top =  (s->mb_y > !!MB_FIELD);
+        deblock_topleft = (s->mb_x > 0);
+        deblock_top     = (s->mb_y > !!MB_FIELD);
     }
 
     src_y  -=   linesize + 1 + pixel_shift;
@@ -1045,7 +1045,7 @@ if (xchg) AV_SWAP64(b,a);\
 else      AV_COPY64(b,a);
 
     if(deblock_top){
-        if(deblock_left){
+        if(deblock_topleft){
             XCHG(top_border_m1 + (8 << pixel_shift), src_y - (7 << pixel_shift), 1);
         }
         XCHG(top_border + (0 << pixel_shift), src_y + (1 << pixel_shift), xchg);
@@ -1056,7 +1056,7 @@ else      AV_COPY64(b,a);
     }
     if(simple || !CONFIG_GRAY || !(s->flags&CODEC_FLAG_GRAY)){
         if(deblock_top){
-            if(deblock_left){
+            if(deblock_topleft){
                 XCHG(top_border_m1 + (16 << pixel_shift), src_cb - (7 << pixel_shift), 1);
                 XCHG(top_border_m1 + (24 << pixel_shift), src_cr - (7 << pixel_shift), 1);
             }
@@ -2561,18 +2561,16 @@ static int fill_filter_caches(H264Context *h, int mb_type){
     return 0;
 }
 
-static void loop_filter(H264Context *h){
+static void loop_filter(H264Context *h, int start_x, int end_x){
     MpegEncContext * const s = &h->s;
     uint8_t  *dest_y, *dest_cb, *dest_cr;
     int linesize, uvlinesize, mb_x, mb_y;
     const int end_mb_y= s->mb_y + FRAME_MBAFF;
     const int old_slice_type= h->slice_type;
-    const int end_mb_x  = s->mb_x;
     const int pixel_shift = h->pixel_shift;
 
     if(h->deblocking_filter) {
-        int start_x= s->resync_mb_y == s->mb_y ? s->resync_mb_x : 0;
-        for(mb_x= start_x; mb_x<end_mb_x; mb_x++){
+        for(mb_x= start_x; mb_x<end_x; mb_x++){
             for(mb_y=end_mb_y - FRAME_MBAFF; mb_y<= end_mb_y; mb_y++){
                 int mb_xy, mb_type;
                 mb_xy = h->mb_xy = mb_x + mb_y*s->mb_stride;
@@ -2617,7 +2615,7 @@ static void loop_filter(H264Context *h){
         }
     }
     h->slice_type= old_slice_type;
-    s->mb_x= end_mb_x;
+    s->mb_x= end_x;
     s->mb_y= end_mb_y - FRAME_MBAFF;
     h->chroma_qp[0] = get_chroma_qp(h, 0, s->qscale);
     h->chroma_qp[1] = get_chroma_qp(h, 1, s->qscale);
@@ -2672,6 +2670,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
     H264Context *h = *(void**)arg;
     MpegEncContext * const s = &h->s;
     const int part_mask= s->partitioned_frame ? (AC_END|AC_ERROR) : 0x7F;
+    int lf_x_start = s->mb_x;
 
     s->mb_skip_run= -1;
 
@@ -2710,6 +2709,7 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
 
             if((s->workaround_bugs & FF_BUG_TRUNCATED) && h->cabac.bytestream > h->cabac.bytestream_end + 2){
                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                if (s->mb_x >= lf_x_start) loop_filter(h, lf_x_start, s->mb_x + 1);
                 return 0;
             }
             if( ret < 0 || h->cabac.bytestream > h->cabac.bytestream_end + 2) {
@@ -2719,8 +2719,8 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             }
 
             if( ++s->mb_x >= s->mb_width ) {
-                loop_filter(h);
-                s->mb_x = 0;
+                loop_filter(h, lf_x_start, s->mb_x);
+                s->mb_x = lf_x_start = 0;
                 decode_finish_row(h);
                 ++s->mb_y;
                 if(FIELD_OR_MBAFF_PICTURE) {
@@ -2731,10 +2731,9 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             }
 
             if( eos || s->mb_y >= s->mb_height ) {
-                if(s->mb_x)
-                    loop_filter(h);
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
                 return 0;
             }
         }
@@ -2756,13 +2755,12 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             if(ret<0){
                 av_log(h->s.avctx, AV_LOG_ERROR, "error while decoding MB %d %d\n", s->mb_x, s->mb_y);
                 ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x, s->mb_y, (AC_ERROR|DC_ERROR|MV_ERROR)&part_mask);
-
                 return -1;
             }
 
             if(++s->mb_x >= s->mb_width){
-                loop_filter(h);
-                s->mb_x=0;
+                loop_filter(h, lf_x_start, s->mb_x);
+                s->mb_x = lf_x_start = 0;
                 decode_finish_row(h);
                 ++s->mb_y;
                 if(FIELD_OR_MBAFF_PICTURE) {
@@ -2789,9 +2787,8 @@ static int decode_slice(struct AVCodecContext *avctx, void *arg){
             if(get_bits_count(&s->gb) >= s->gb.size_in_bits && s->mb_skip_run<=0){
                 tprintf(s->avctx, "slice end %d %d\n", get_bits_count(&s->gb), s->gb.size_in_bits);
                 if(get_bits_count(&s->gb) == s->gb.size_in_bits ){
-                    if(s->mb_x)
-                        loop_filter(h);
                     ff_er_add_slice(s, s->resync_mb_x, s->resync_mb_y, s->mb_x-1, s->mb_y, (AC_END|DC_END|MV_END)&part_mask);
+                    if (s->mb_x > lf_x_start) loop_filter(h, lf_x_start, s->mb_x);
 
                     return 0;
                 }else{
diff --git a/libavcodec/mpc.c b/libavcodec/mpc.c
index ca4c3d0dcb..c2975ec732 100644
--- a/libavcodec/mpc.c
+++ b/libavcodec/mpc.c
@@ -29,6 +29,7 @@
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
+#include "mpegaudiodsp.h"
 #include "mpegaudio.h"
 
 #include "mpc.h"
@@ -51,7 +52,8 @@ static void mpc_synth(MPCContext *c, int16_t *out, int channels)
     for(ch = 0;  ch < channels; ch++){
         samples_ptr = samples + ch;
         for(i = 0; i < SAMPLES_PER_BAND; i++) {
-            ff_mpa_synth_filter_fixed(c->synth_buf[ch], &(c->synth_buf_offset[ch]),
+            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,
                                 c->sb_samples[ch][i]);
diff --git a/libavcodec/mpc.h b/libavcodec/mpc.h
index 2d9755a593..365580ebd0 100644
--- a/libavcodec/mpc.h
+++ b/libavcodec/mpc.h
@@ -52,6 +52,7 @@ typedef struct {
 
 typedef struct {
     DSPContext dsp;
+    MPADSPContext mpadsp;
     GetBitContext gb;
     int IS, MSS, gapless;
     int lastframelen;
diff --git a/libavcodec/mpc7.c b/libavcodec/mpc7.c
index 04657e5ff2..bb21469356 100644
--- a/libavcodec/mpc7.c
+++ b/libavcodec/mpc7.c
@@ -29,7 +29,7 @@
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
-#include "mpegaudio.h"
+#include "mpegaudiodsp.h"
 #include "libavutil/audioconvert.h"
 
 #include "mpc.h"
@@ -68,6 +68,7 @@ static av_cold int mpc7_decode_init(AVCodecContext * avctx)
     memset(c->oldDSCF, 0, sizeof(c->oldDSCF));
     av_lfg_init(&c->rnd, 0xDEADBEEF);
     dsputil_init(&c->dsp, avctx);
+    ff_mpadsp_init(&c->mpadsp);
     c->dsp.bswap_buf((uint32_t*)buf, (const uint32_t*)avctx->extradata, 4);
     ff_mpc_init();
     init_get_bits(&gb, buf, 128);
diff --git a/libavcodec/mpc8.c b/libavcodec/mpc8.c
index cf65ffe904..51c5509425 100644
--- a/libavcodec/mpc8.c
+++ b/libavcodec/mpc8.c
@@ -29,7 +29,7 @@
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
-#include "mpegaudio.h"
+#include "mpegaudiodsp.h"
 #include "libavutil/audioconvert.h"
 
 #include "mpc.h"
@@ -120,6 +120,7 @@ static av_cold int mpc8_decode_init(AVCodecContext * avctx)
     memset(c->oldDSCF, 0, sizeof(c->oldDSCF));
     av_lfg_init(&c->rnd, 0xDEADBEEF);
     dsputil_init(&c->dsp, avctx);
+    ff_mpadsp_init(&c->mpadsp);
 
     ff_mpc_init();
 
diff --git a/libavcodec/mpegaudio.h b/libavcodec/mpegaudio.h
index 005598797d..7e2ca845d1 100644
--- a/libavcodec/mpegaudio.h
+++ b/libavcodec/mpegaudio.h
@@ -31,9 +31,6 @@
 #endif
 
 #include "avcodec.h"
-#include "get_bits.h"
-#include "dsputil.h"
-#include "dct.h"
 
 /* max frame size, in samples */
 #define MPA_FRAME_SIZE 1152
@@ -50,10 +47,6 @@
 #define MPA_DUAL    2
 #define MPA_MONO    3
 
-/* header + layer + bitrate + freq + lsf/mpeg25 */
-#define SAME_HEADER_MASK \
-   (0xffe00000 | (3 << 17) | (0xf << 12) | (3 << 10) | (3 << 19))
-
 #define MP3_MASK 0xFFFE0CCF
 
 #ifndef FRAC_BITS
@@ -65,47 +58,20 @@
 
 #define FIX(a)   ((int)((a) * FRAC_ONE))
 
-#if CONFIG_FLOAT
-typedef float OUT_INT;
-#else
-typedef int16_t OUT_INT;
-#define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
-#endif
-
 #if CONFIG_FLOAT
 #   define INTFLOAT float
 typedef float MPA_INT;
+typedef float OUT_INT;
 #elif FRAC_BITS <= 15
 #   define INTFLOAT int
 typedef int16_t MPA_INT;
+typedef int16_t OUT_INT;
 #else
 #   define INTFLOAT int
 typedef int32_t MPA_INT;
+typedef int16_t OUT_INT;
 #endif
 
-#define BACKSTEP_SIZE 512
-#define EXTRABYTES 24
-
-/* layer 3 "granule" */
-typedef struct GranuleDef {
-    uint8_t scfsi;
-    int part2_3_length;
-    int big_values;
-    int global_gain;
-    int scalefac_compress;
-    uint8_t block_type;
-    uint8_t switch_point;
-    int table_select[3];
-    int subblock_gain[3];
-    uint8_t scalefac_scale;
-    uint8_t count1table_select;
-    int region_size[3]; /* number of huffman codes in each region */
-    int preflag;
-    int short_start, long_end; /* long/short band indexes */
-    uint8_t scale_factors[40];
-    INTFLOAT sb_hybrid[SBLIMIT * 18]; /* 576 samples */
-} GranuleDef;
-
 #define MPA_DECODE_HEADER \
     int frame_size; \
     int error_protection; \
@@ -122,58 +88,8 @@ typedef struct MPADecodeHeader {
   MPA_DECODE_HEADER
 } MPADecodeHeader;
 
-typedef struct MPADecodeContext {
-    MPA_DECODE_HEADER
-    uint8_t last_buf[2*BACKSTEP_SIZE + EXTRABYTES];
-    int last_buf_size;
-    /* next header (used in free format parsing) */
-    uint32_t free_format_next_header;
-    GetBitContext gb;
-    GetBitContext in_gb;
-    DECLARE_ALIGNED(16, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
-    int synth_buf_offset[MPA_MAX_CHANNELS];
-    DECLARE_ALIGNED(16, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
-    INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
-    GranuleDef granules[2][2]; /* Used in Layer 3 */
-#ifdef DEBUG
-    int frame_count;
-#endif
-    int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
-    int dither_state;
-    int error_recognition;
-    AVCodecContext* avctx;
-#if CONFIG_FLOAT
-    DCTContext dct;
-#endif
-    void (*apply_window_mp3)(MPA_INT *synth_buf, MPA_INT *window,
-                             int *dither_state, OUT_INT *samples, int incr);
-} MPADecodeContext;
-
-/* layer 3 huffman tables */
-typedef struct HuffTable {
-    int xsize;
-    const uint8_t *bits;
-    const uint16_t *codes;
-} HuffTable;
-
 int ff_mpa_l2_select_table(int bitrate, int nb_channels, int freq, int lsf);
 int ff_mpa_decode_header(AVCodecContext *avctx, uint32_t head, int *sample_rate, int *channels, int *frame_size, int *bitrate);
-extern MPA_INT ff_mpa_synth_window_fixed[];
-void ff_mpa_synth_init_fixed(MPA_INT *window);
-void ff_mpa_synth_filter_fixed(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
-                         MPA_INT *window, int *dither_state,
-                         OUT_INT *samples, int incr,
-                         INTFLOAT sb_samples[SBLIMIT]);
-
-void ff_mpa_synth_init_float(MPA_INT *window);
-void ff_mpa_synth_filter_float(MPADecodeContext *s,
-                         MPA_INT *synth_buf_ptr, int *synth_buf_offset,
-                         MPA_INT *window, int *dither_state,
-                         OUT_INT *samples, int incr,
-                         INTFLOAT sb_samples[SBLIMIT]);
-
-void ff_mpegaudiodec_init_mmx(MPADecodeContext *s);
-void ff_mpegaudiodec_init_altivec(MPADecodeContext *s);
 
 /* fast header check for resync */
 static inline int ff_mpa_check_header(uint32_t header){
diff --git a/libavcodec/mpegaudio_parser.c b/libavcodec/mpegaudio_parser.c
index cfd92d42eb..ee54def579 100644
--- a/libavcodec/mpegaudio_parser.c
+++ b/libavcodec/mpegaudio_parser.c
@@ -35,7 +35,6 @@ typedef struct MpegAudioParseContext {
 #define MPA_HEADER_SIZE 4
 
 /* header + layer + bitrate + freq + lsf/mpeg25 */
-#undef SAME_HEADER_MASK /* mpegaudio.h defines different version */
 #define SAME_HEADER_MASK \
    (0xffe00000 | (3 << 17) | (3 << 10) | (3 << 19))
 
diff --git a/libavcodec/mpegaudiodec.c b/libavcodec/mpegaudiodec.c
index c7d830fe21..decb23e665 100644
--- a/libavcodec/mpegaudiodec.c
+++ b/libavcodec/mpegaudiodec.c
@@ -29,7 +29,7 @@
 #include "get_bits.h"
 #include "dsputil.h"
 #include "mathops.h"
-#include "dct32.h"
+#include "mpegaudiodsp.h"
 
 /*
  * TODO:
@@ -39,6 +39,52 @@
 #include "mpegaudio.h"
 #include "mpegaudiodecheader.h"
 
+#define BACKSTEP_SIZE 512
+#define EXTRABYTES 24
+
+/* layer 3 "granule" */
+typedef struct GranuleDef {
+    uint8_t scfsi;
+    int part2_3_length;
+    int big_values;
+    int global_gain;
+    int scalefac_compress;
+    uint8_t block_type;
+    uint8_t switch_point;
+    int table_select[3];
+    int subblock_gain[3];
+    uint8_t scalefac_scale;
+    uint8_t count1table_select;
+    int region_size[3]; /* number of huffman codes in each region */
+    int preflag;
+    int short_start, long_end; /* long/short band indexes */
+    uint8_t scale_factors[40];
+    INTFLOAT sb_hybrid[SBLIMIT * 18]; /* 576 samples */
+} GranuleDef;
+
+typedef struct MPADecodeContext {
+    MPA_DECODE_HEADER
+    uint8_t last_buf[2*BACKSTEP_SIZE + EXTRABYTES];
+    int last_buf_size;
+    /* next header (used in free format parsing) */
+    uint32_t free_format_next_header;
+    GetBitContext gb;
+    GetBitContext in_gb;
+    DECLARE_ALIGNED(16, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512 * 2];
+    int synth_buf_offset[MPA_MAX_CHANNELS];
+    DECLARE_ALIGNED(16, INTFLOAT, sb_samples)[MPA_MAX_CHANNELS][36][SBLIMIT];
+    INTFLOAT mdct_buf[MPA_MAX_CHANNELS][SBLIMIT * 18]; /* previous samples, for layer 3 MDCT */
+    GranuleDef granules[2][2]; /* Used in Layer 3 */
+#ifdef DEBUG
+    int frame_count;
+#endif
+    int adu_mode; ///< 0 for standard mp3, 1 for adu formatted mp3
+    int dither_state;
+    int error_recognition;
+    AVCodecContext* avctx;
+    MPADSPContext mpadsp;
+} MPADecodeContext;
+
 #if CONFIG_FLOAT
 #   define SHR(a,b)       ((a)*(1.0f/(1<<(b))))
 #   define FIXR_OLD(a)    ((int)((a) * FRAC_ONE + 0.5))
@@ -68,8 +114,6 @@
 #include "mpegaudiodectab.h"
 
 static void RENAME(compute_antialias)(MPADecodeContext *s, GranuleDef *g);
-static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,
-                               int *dither_state, OUT_INT *samples, int incr);
 
 /* vlc structure for decoding layer 3 huffman tables */
 static VLC huff_vlc[16];
@@ -119,8 +163,6 @@ static const int32_t scale_factor_mult2[3][3] = {
     SCALE_GEN(4.0 / 9.0), /* 9 steps */
 };
 
-DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256];
-
 /**
  * Convert region offsets to region sizes and truncate
  * size to big_values.
@@ -259,14 +301,8 @@ static av_cold int decode_init(AVCodecContext * avctx)
     int i, j, k;
 
     s->avctx = avctx;
-    s->apply_window_mp3 = apply_window_mp3_c;
-#if HAVE_MMX && CONFIG_FLOAT
-    ff_mpegaudiodec_init_mmx(s);
-#endif
-#if CONFIG_FLOAT
-    ff_dct_init(&s->dct, 5, DCT_II);
-#endif
-    if (HAVE_ALTIVEC && CONFIG_FLOAT) ff_mpegaudiodec_init_altivec(s);
+
+    ff_mpadsp_init(&s->mpadsp);
 
     avctx->sample_fmt= OUT_FMT;
     s->error_recognition= avctx->error_recognition;
@@ -461,183 +497,6 @@ static av_cold int decode_init(AVCodecContext * avctx)
     return 0;
 }
 
-
-#if CONFIG_FLOAT
-static inline float round_sample(float *sum)
-{
-    float sum1=*sum;
-    *sum = 0;
-    return sum1;
-}
-
-/* signed 16x16 -> 32 multiply add accumulate */
-#define MACS(rt, ra, rb) rt+=(ra)*(rb)
-
-/* signed 16x16 -> 32 multiply */
-#define MULS(ra, rb) ((ra)*(rb))
-
-#define MLSS(rt, ra, rb) rt-=(ra)*(rb)
-
-#else
-
-static inline int round_sample(int64_t *sum)
-{
-    int sum1;
-    sum1 = (int)((*sum) >> OUT_SHIFT);
-    *sum &= (1<<OUT_SHIFT)-1;
-    return av_clip_int16(sum1);
-}
-
-#   define MULS(ra, rb) MUL64(ra, rb)
-#   define MACS(rt, ra, rb) MAC64(rt, ra, rb)
-#   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
-#endif
-
-#define SUM8(op, sum, w, p)               \
-{                                         \
-    op(sum, (w)[0 * 64], (p)[0 * 64]);    \
-    op(sum, (w)[1 * 64], (p)[1 * 64]);    \
-    op(sum, (w)[2 * 64], (p)[2 * 64]);    \
-    op(sum, (w)[3 * 64], (p)[3 * 64]);    \
-    op(sum, (w)[4 * 64], (p)[4 * 64]);    \
-    op(sum, (w)[5 * 64], (p)[5 * 64]);    \
-    op(sum, (w)[6 * 64], (p)[6 * 64]);    \
-    op(sum, (w)[7 * 64], (p)[7 * 64]);    \
-}
-
-#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
-{                                               \
-    INTFLOAT tmp;\
-    tmp = p[0 * 64];\
-    op1(sum1, (w1)[0 * 64], tmp);\
-    op2(sum2, (w2)[0 * 64], tmp);\
-    tmp = p[1 * 64];\
-    op1(sum1, (w1)[1 * 64], tmp);\
-    op2(sum2, (w2)[1 * 64], tmp);\
-    tmp = p[2 * 64];\
-    op1(sum1, (w1)[2 * 64], tmp);\
-    op2(sum2, (w2)[2 * 64], tmp);\
-    tmp = p[3 * 64];\
-    op1(sum1, (w1)[3 * 64], tmp);\
-    op2(sum2, (w2)[3 * 64], tmp);\
-    tmp = p[4 * 64];\
-    op1(sum1, (w1)[4 * 64], tmp);\
-    op2(sum2, (w2)[4 * 64], tmp);\
-    tmp = p[5 * 64];\
-    op1(sum1, (w1)[5 * 64], tmp);\
-    op2(sum2, (w2)[5 * 64], tmp);\
-    tmp = p[6 * 64];\
-    op1(sum1, (w1)[6 * 64], tmp);\
-    op2(sum2, (w2)[6 * 64], tmp);\
-    tmp = p[7 * 64];\
-    op1(sum1, (w1)[7 * 64], tmp);\
-    op2(sum2, (w2)[7 * 64], tmp);\
-}
-
-void av_cold RENAME(ff_mpa_synth_init)(MPA_INT *window)
-{
-    int i, j;
-
-    /* max = 18760, max sum over all 16 coefs : 44736 */
-    for(i=0;i<257;i++) {
-        INTFLOAT v;
-        v = ff_mpa_enwindow[i];
-#if CONFIG_FLOAT
-        v *= 1.0 / (1LL<<(16 + FRAC_BITS));
-#endif
-        window[i] = v;
-        if ((i & 63) != 0)
-            v = -v;
-        if (i != 0)
-            window[512 - i] = v;
-    }
-
-    // Needed for avoiding shuffles in ASM implementations
-    for(i=0; i < 8; i++)
-        for(j=0; j < 16; j++)
-            window[512+16*i+j] = window[64*i+32-j];
-
-    for(i=0; i < 8; i++)
-        for(j=0; j < 16; j++)
-            window[512+128+16*i+j] = window[64*i+48-j];
-}
-
-static void apply_window_mp3_c(MPA_INT *synth_buf, MPA_INT *window,
-                               int *dither_state, OUT_INT *samples, int incr)
-{
-    register const MPA_INT *w, *w2, *p;
-    int j;
-    OUT_INT *samples2;
-#if CONFIG_FLOAT
-    float sum, sum2;
-#else
-    int64_t sum, sum2;
-#endif
-
-    /* copy to avoid wrap */
-    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));
-
-    samples2 = samples + 31 * incr;
-    w = window;
-    w2 = window + 31;
-
-    sum = *dither_state;
-    p = synth_buf + 16;
-    SUM8(MACS, sum, w, p);
-    p = synth_buf + 48;
-    SUM8(MLSS, sum, w + 32, p);
-    *samples = round_sample(&sum);
-    samples += incr;
-    w++;
-
-    /* we calculate two samples at the same time to avoid one memory
-       access per two sample */
-    for(j=1;j<16;j++) {
-        sum2 = 0;
-        p = synth_buf + 16 + j;
-        SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);
-        p = synth_buf + 48 - j;
-        SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);
-
-        *samples = round_sample(&sum);
-        samples += incr;
-        sum += sum2;
-        *samples2 = round_sample(&sum);
-        samples2 -= incr;
-        w++;
-        w2--;
-    }
-
-    p = synth_buf + 32;
-    SUM8(MLSS, sum, w + 32, p);
-    *samples = round_sample(&sum);
-    *dither_state= sum;
-}
-
-
-/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
-   32 samples. */
-/* XXX: optimize by avoiding ring buffer usage */
-#if !CONFIG_FLOAT
-void ff_mpa_synth_filter_fixed(MPA_INT *synth_buf_ptr, int *synth_buf_offset,
-                         MPA_INT *window, int *dither_state,
-                         OUT_INT *samples, int incr,
-                         INTFLOAT sb_samples[SBLIMIT])
-{
-    register MPA_INT *synth_buf;
-    int offset;
-
-    offset = *synth_buf_offset;
-    synth_buf = synth_buf_ptr + offset;
-
-    ff_dct32_fixed(synth_buf, sb_samples);
-    apply_window_mp3_c(synth_buf, window, dither_state, samples, incr);
-
-    offset = (offset - 32) & 511;
-    *synth_buf_offset = offset;
-}
-#endif
-
 #define C3 FIXHR(0.86602540378443864676/2)
 
 /* 0.5 / cos(pi*(2*i+1)/36) */
@@ -1914,9 +1773,7 @@ static int mp_decode_frame(MPADecodeContext *s,
         samples_ptr = samples + ch;
         for(i=0;i<nb_frames;i++) {
             RENAME(ff_mpa_synth_filter)(
-#if CONFIG_FLOAT
-                         s,
-#endif
+                         &s->mpadsp,
                          s->synth_buf[ch], &(s->synth_buf_offset[ch]),
                          RENAME(ff_mpa_synth_window), &s->dither_state,
                          samples_ptr, s->nb_channels,
diff --git a/libavcodec/mpegaudiodec_float.c b/libavcodec/mpegaudiodec_float.c
index 183e5540c2..26454190f5 100644
--- a/libavcodec/mpegaudiodec_float.c
+++ b/libavcodec/mpegaudiodec_float.c
@@ -22,25 +22,6 @@
 #define CONFIG_FLOAT 1
 #include "mpegaudiodec.c"
 
-void ff_mpa_synth_filter_float(MPADecodeContext *s, float *synth_buf_ptr,
-                               int *synth_buf_offset,
-                               float *window, int *dither_state,
-                               float *samples, int incr,
-                               float sb_samples[SBLIMIT])
-{
-    float *synth_buf;
-    int offset;
-
-    offset = *synth_buf_offset;
-    synth_buf = synth_buf_ptr + offset;
-
-    s->dct.dct32(synth_buf, sb_samples);
-    s->apply_window_mp3(synth_buf, window, dither_state, samples, incr);
-
-    offset = (offset - 32) & 511;
-    *synth_buf_offset = offset;
-}
-
 static void compute_antialias_float(MPADecodeContext *s,
                               GranuleDef *g)
 {
diff --git a/libavcodec/mpegaudiodectab.h b/libavcodec/mpegaudiodectab.h
index 234a70e474..4dd8a7cfc9 100644
--- a/libavcodec/mpegaudiodectab.h
+++ b/libavcodec/mpegaudiodectab.h
@@ -33,6 +33,13 @@
 /*******************************************************/
 /* layer 3 tables */
 
+/* layer 3 huffman tables */
+typedef struct HuffTable {
+    int xsize;
+    const uint8_t *bits;
+    const uint16_t *codes;
+} HuffTable;
+
 /* layer3 scale factor size */
 static const uint8_t slen_table[2][16] = {
     { 0, 0, 0, 0, 3, 1, 1, 1, 2, 2, 2, 3, 3, 3, 4, 4 },
diff --git a/libavcodec/tableprint.c b/libavcodec/mpegaudiodsp.c
similarity index 56%
rename from libavcodec/tableprint.c
rename to libavcodec/mpegaudiodsp.c
index 52f6ac2a7c..064acd1e74 100644
--- a/libavcodec/tableprint.c
+++ b/libavcodec/mpegaudiodsp.c
@@ -1,7 +1,5 @@
 /*
- * Generate a file for hardcoded tables
- *
- * Copyright (c) 2009 Reimar Döffinger <Reimar.Doeffinger@gmx.de>
+ * Copyright (c) 2011 Mans Rullgard
  *
  * This file is part of FFmpeg.
  *
@@ -20,22 +18,23 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-#include <stdio.h>
-#include <inttypes.h>
-#include "tableprint.h"
+#include "config.h"
+#include "mpegaudiodsp.h"
+#include "dct.h"
+#include "dct32.h"
 
-WRITE_1D_FUNC(int8_t,   "%3"PRIi8, 15)
-WRITE_1D_FUNC(uint8_t,  "0x%02"PRIx8, 15)
-WRITE_1D_FUNC(uint16_t, "0x%08"PRIx16, 7)
-WRITE_1D_FUNC(uint32_t, "0x%08"PRIx32, 7)
-WRITE_1D_FUNC(float,    "%.18e", 3)
+void ff_mpadsp_init(MPADSPContext *s)
+{
+    DCTContext dct;
 
-WRITE_2D_FUNC(int8_t)
-WRITE_2D_FUNC(uint8_t)
-WRITE_2D_FUNC(uint32_t)
-WRITE_2D_FUNC(float)
+    ff_dct_init(&dct, 5, DCT_II);
 
-void write_fileheader(void) {
-    printf("/* This file was generated by libavcodec/tableprint */\n");
-    printf("#include <stdint.h>\n");
+    s->apply_window_float = ff_mpadsp_apply_window_float;
+    s->apply_window_fixed = ff_mpadsp_apply_window_fixed;
+
+    s->dct32_float = dct.dct32;
+    s->dct32_fixed = ff_dct32_fixed;
+
+    if (HAVE_MMX)     ff_mpadsp_init_mmx(s);
+    if (HAVE_ALTIVEC) ff_mpadsp_init_altivec(s);
 }
diff --git a/libavcodec/mpegaudiodsp.h b/libavcodec/mpegaudiodsp.h
new file mode 100644
index 0000000000..597e2533f5
--- /dev/null
+++ b/libavcodec/mpegaudiodsp.h
@@ -0,0 +1,63 @@
+/*
+ * This file is part of Libav.
+ *
+ * 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.
+ *
+ * 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_MPEGAUDIODSP_H
+#define AVCODEC_MPEGAUDIODSP_H
+
+#include <stdint.h>
+
+typedef struct MPADSPContext {
+    void (*apply_window_float)(float *synth_buf, float *window,
+                               int *dither_state, float *samples, int incr);
+    void (*apply_window_fixed)(int32_t *synth_buf, int32_t *window,
+                               int *dither_state, int16_t *samples, int incr);
+    void (*dct32_float)(float *dst, const float *src);
+    void (*dct32_fixed)(int *dst, const int *src);
+} MPADSPContext;
+
+void ff_mpadsp_init(MPADSPContext *s);
+
+extern int32_t ff_mpa_synth_window_fixed[];
+extern float   ff_mpa_synth_window_float[];
+
+void ff_mpa_synth_filter_fixed(MPADSPContext *s,
+                               int32_t *synth_buf_ptr, int *synth_buf_offset,
+                               int32_t *window, int *dither_state,
+                               int16_t *samples, int incr,
+                               int32_t *sb_samples);
+
+void ff_mpa_synth_filter_float(MPADSPContext *s,
+                               float *synth_buf_ptr, int *synth_buf_offset,
+                               float *window, int *dither_state,
+                               float *samples, int incr,
+                               float *sb_samples);
+
+void ff_mpadsp_init_mmx(MPADSPContext *s);
+void ff_mpadsp_init_altivec(MPADSPContext *s);
+
+void ff_mpa_synth_init_float(float *window);
+void ff_mpa_synth_init_fixed(int32_t *window);
+
+void ff_mpadsp_apply_window_float(float *synth_buf, float *window,
+                                  int *dither_state, float *samples,
+                                  int incr);
+void ff_mpadsp_apply_window_fixed(int32_t *synth_buf, int32_t *window,
+                                  int *dither_state, int16_t *samples,
+                                  int incr);
+
+#endif
diff --git a/libavcodec/mpegaudiodsp_fixed.c b/libavcodec/mpegaudiodsp_fixed.c
new file mode 100644
index 0000000000..3c49a568b7
--- /dev/null
+++ b/libavcodec/mpegaudiodsp_fixed.c
@@ -0,0 +1,20 @@
+/*
+ * This file is part of Libav.
+ *
+ * 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.
+ *
+ * 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#define CONFIG_FLOAT 0
+#include "mpegaudiodsp_template.c"
diff --git a/libavcodec/mpegaudiodsp_float.c b/libavcodec/mpegaudiodsp_float.c
new file mode 100644
index 0000000000..2d8d53ea26
--- /dev/null
+++ b/libavcodec/mpegaudiodsp_float.c
@@ -0,0 +1,20 @@
+/*
+ * This file is part of Libav.
+ *
+ * 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.
+ *
+ * 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#define CONFIG_FLOAT 1
+#include "mpegaudiodsp_template.c"
diff --git a/libavcodec/mpegaudiodsp_template.c b/libavcodec/mpegaudiodsp_template.c
new file mode 100644
index 0000000000..5561c46135
--- /dev/null
+++ b/libavcodec/mpegaudiodsp_template.c
@@ -0,0 +1,205 @@
+/*
+ * Copyright (c) 2001, 2002 Fabrice Bellard
+ *
+ * This file is part of Libav.
+ *
+ * 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.
+ *
+ * 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 Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "libavutil/mem.h"
+#include "dct32.h"
+#include "mathops.h"
+#include "mpegaudiodsp.h"
+#include "mpegaudio.h"
+#include "mpegaudiodata.h"
+
+#if CONFIG_FLOAT
+#define RENAME(n) n##_float
+
+static inline float round_sample(float *sum)
+{
+    float sum1=*sum;
+    *sum = 0;
+    return sum1;
+}
+
+#define MACS(rt, ra, rb) rt+=(ra)*(rb)
+#define MULS(ra, rb) ((ra)*(rb))
+#define MLSS(rt, ra, rb) rt-=(ra)*(rb)
+
+#else
+
+#define RENAME(n) n##_fixed
+#define OUT_SHIFT (WFRAC_BITS + FRAC_BITS - 15)
+
+static inline int round_sample(int64_t *sum)
+{
+    int sum1;
+    sum1 = (int)((*sum) >> OUT_SHIFT);
+    *sum &= (1<<OUT_SHIFT)-1;
+    return av_clip_int16(sum1);
+}
+
+#   define MULS(ra, rb) MUL64(ra, rb)
+#   define MACS(rt, ra, rb) MAC64(rt, ra, rb)
+#   define MLSS(rt, ra, rb) MLS64(rt, ra, rb)
+#endif
+
+DECLARE_ALIGNED(16, MPA_INT, RENAME(ff_mpa_synth_window))[512+256];
+
+#define SUM8(op, sum, w, p)               \
+{                                         \
+    op(sum, (w)[0 * 64], (p)[0 * 64]);    \
+    op(sum, (w)[1 * 64], (p)[1 * 64]);    \
+    op(sum, (w)[2 * 64], (p)[2 * 64]);    \
+    op(sum, (w)[3 * 64], (p)[3 * 64]);    \
+    op(sum, (w)[4 * 64], (p)[4 * 64]);    \
+    op(sum, (w)[5 * 64], (p)[5 * 64]);    \
+    op(sum, (w)[6 * 64], (p)[6 * 64]);    \
+    op(sum, (w)[7 * 64], (p)[7 * 64]);    \
+}
+
+#define SUM8P2(sum1, op1, sum2, op2, w1, w2, p) \
+{                                               \
+    INTFLOAT tmp;\
+    tmp = p[0 * 64];\
+    op1(sum1, (w1)[0 * 64], tmp);\
+    op2(sum2, (w2)[0 * 64], tmp);\
+    tmp = p[1 * 64];\
+    op1(sum1, (w1)[1 * 64], tmp);\
+    op2(sum2, (w2)[1 * 64], tmp);\
+    tmp = p[2 * 64];\
+    op1(sum1, (w1)[2 * 64], tmp);\
+    op2(sum2, (w2)[2 * 64], tmp);\
+    tmp = p[3 * 64];\
+    op1(sum1, (w1)[3 * 64], tmp);\
+    op2(sum2, (w2)[3 * 64], tmp);\
+    tmp = p[4 * 64];\
+    op1(sum1, (w1)[4 * 64], tmp);\
+    op2(sum2, (w2)[4 * 64], tmp);\
+    tmp = p[5 * 64];\
+    op1(sum1, (w1)[5 * 64], tmp);\
+    op2(sum2, (w2)[5 * 64], tmp);\
+    tmp = p[6 * 64];\
+    op1(sum1, (w1)[6 * 64], tmp);\
+    op2(sum2, (w2)[6 * 64], tmp);\
+    tmp = p[7 * 64];\
+    op1(sum1, (w1)[7 * 64], tmp);\
+    op2(sum2, (w2)[7 * 64], tmp);\
+}
+
+void RENAME(ff_mpadsp_apply_window)(MPA_INT *synth_buf, MPA_INT *window,
+                                  int *dither_state, OUT_INT *samples,
+                                  int incr)
+{
+    register const MPA_INT *w, *w2, *p;
+    int j;
+    OUT_INT *samples2;
+#if CONFIG_FLOAT
+    float sum, sum2;
+#else
+    int64_t sum, sum2;
+#endif
+
+    /* copy to avoid wrap */
+    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(*synth_buf));
+
+    samples2 = samples + 31 * incr;
+    w = window;
+    w2 = window + 31;
+
+    sum = *dither_state;
+    p = synth_buf + 16;
+    SUM8(MACS, sum, w, p);
+    p = synth_buf + 48;
+    SUM8(MLSS, sum, w + 32, p);
+    *samples = round_sample(&sum);
+    samples += incr;
+    w++;
+
+    /* we calculate two samples at the same time to avoid one memory
+       access per two sample */
+    for(j=1;j<16;j++) {
+        sum2 = 0;
+        p = synth_buf + 16 + j;
+        SUM8P2(sum, MACS, sum2, MLSS, w, w2, p);
+        p = synth_buf + 48 - j;
+        SUM8P2(sum, MLSS, sum2, MLSS, w + 32, w2 + 32, p);
+
+        *samples = round_sample(&sum);
+        samples += incr;
+        sum += sum2;
+        *samples2 = round_sample(&sum);
+        samples2 -= incr;
+        w++;
+        w2--;
+    }
+
+    p = synth_buf + 32;
+    SUM8(MLSS, sum, w + 32, p);
+    *samples = round_sample(&sum);
+    *dither_state= sum;
+}
+
+/* 32 sub band synthesis filter. Input: 32 sub band samples, Output:
+   32 samples. */
+void RENAME(ff_mpa_synth_filter)(MPADSPContext *s, MPA_INT *synth_buf_ptr,
+                                 int *synth_buf_offset,
+                                 MPA_INT *window, int *dither_state,
+                                 OUT_INT *samples, int incr,
+                                 MPA_INT *sb_samples)
+{
+    MPA_INT *synth_buf;
+    int offset;
+
+    offset = *synth_buf_offset;
+    synth_buf = synth_buf_ptr + offset;
+
+    s->RENAME(dct32)(synth_buf, sb_samples);
+    s->RENAME(apply_window)(synth_buf, window, dither_state, samples, incr);
+
+    offset = (offset - 32) & 511;
+    *synth_buf_offset = offset;
+}
+
+void av_cold RENAME(ff_mpa_synth_init)(MPA_INT *window)
+{
+    int i, j;
+
+    /* max = 18760, max sum over all 16 coefs : 44736 */
+    for(i=0;i<257;i++) {
+        INTFLOAT v;
+        v = ff_mpa_enwindow[i];
+#if CONFIG_FLOAT
+        v *= 1.0 / (1LL<<(16 + FRAC_BITS));
+#endif
+        window[i] = v;
+        if ((i & 63) != 0)
+            v = -v;
+        if (i != 0)
+            window[512 - i] = v;
+    }
+
+    // Needed for avoiding shuffles in ASM implementations
+    for(i=0; i < 8; i++)
+        for(j=0; j < 16; j++)
+            window[512+16*i+j] = window[64*i+32-j];
+
+    for(i=0; i < 8; i++)
+        for(j=0; j < 16; j++)
+            window[512+128+16*i+j] = window[64*i+48-j];
+}
diff --git a/libavcodec/mpegaudioenc.c b/libavcodec/mpegaudioenc.c
index 515da6f670..c3d86137fa 100644
--- a/libavcodec/mpegaudioenc.c
+++ b/libavcodec/mpegaudioenc.c
@@ -782,5 +782,3 @@ AVCodec ff_mp2_encoder = {
     .supported_samplerates= (const int[]){44100, 48000,  32000, 22050, 24000, 16000, 0},
     .long_name = NULL_IF_CONFIG_SMALL("MP2 (MPEG audio layer 2)"),
 };
-
-#undef FIX
diff --git a/libavcodec/ppc/Makefile b/libavcodec/ppc/Makefile
index 35ea0c38f8..8e37fc791d 100644
--- a/libavcodec/ppc/Makefile
+++ b/libavcodec/ppc/Makefile
@@ -7,11 +7,7 @@ ALTIVEC-OBJS-$(CONFIG_VP5_DECODER)     += ppc/vp3dsp_altivec.o
 ALTIVEC-OBJS-$(CONFIG_VP6_DECODER)     += ppc/vp3dsp_altivec.o
 ALTIVEC-OBJS-$(CONFIG_VP8_DECODER)     += ppc/vp8dsp_altivec.o
 
-ALTIVEC-OBJS-$(CONFIG_MP1FLOAT_DECODER)    += ppc/mpegaudiodec_altivec.o
-ALTIVEC-OBJS-$(CONFIG_MP2FLOAT_DECODER)    += ppc/mpegaudiodec_altivec.o
-ALTIVEC-OBJS-$(CONFIG_MP3FLOAT_DECODER)    += ppc/mpegaudiodec_altivec.o
-ALTIVEC-OBJS-$(CONFIG_MP3ON4FLOAT_DECODER) += ppc/mpegaudiodec_altivec.o
-ALTIVEC-OBJS-$(CONFIG_MP3ADUFLOAT_DECODER) += ppc/mpegaudiodec_altivec.o
+ALTIVEC-OBJS-$(CONFIG_MPEGAUDIODSP)    += ppc/mpegaudiodec_altivec.o
 
 FFT-OBJS-$(HAVE_GNU_AS)                += ppc/fft_altivec_s.o           \
 
diff --git a/libavcodec/ppc/mpegaudiodec_altivec.c b/libavcodec/ppc/mpegaudiodec_altivec.c
index e087d4add1..2de5dd133a 100644
--- a/libavcodec/ppc/mpegaudiodec_altivec.c
+++ b/libavcodec/ppc/mpegaudiodec_altivec.c
@@ -21,9 +21,8 @@
 
 #include "dsputil_altivec.h"
 #include "util_altivec.h"
-
-#define CONFIG_FLOAT 1
-#include "libavcodec/mpegaudio.h"
+#include "libavcodec/dsputil.h"
+#include "libavcodec/mpegaudiodsp.h"
 
 #define MACS(rt, ra, rb) rt+=(ra)*(rb)
 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)
@@ -124,7 +123,7 @@ static void apply_window_mp3(float *in, float *win, int *unused, float *out,
     *out = sum;
 }
 
-void ff_mpegaudiodec_init_altivec(MPADecodeContext *s)
+void ff_mpadsp_init_altivec(MPADSPContext *s)
 {
-    s->apply_window_mp3 = apply_window_mp3;
+    s->apply_window_float = apply_window_mp3;
 }
diff --git a/libavcodec/qdm2.c b/libavcodec/qdm2.c
index e1165074f7..b9252bab40 100644
--- a/libavcodec/qdm2.c
+++ b/libavcodec/qdm2.c
@@ -39,6 +39,7 @@
 #include "get_bits.h"
 #include "dsputil.h"
 #include "rdft.h"
+#include "mpegaudiodsp.h"
 #include "mpegaudio.h"
 
 #include "qdm2data.h"
@@ -170,9 +171,10 @@ typedef struct {
     float output_buffer[1024];
 
     /// Synthesis filter
-    DECLARE_ALIGNED(16, MPA_INT, synth_buf)[MPA_MAX_CHANNELS][512*2];
+    MPADSPContext mpadsp;
+    DECLARE_ALIGNED(32, float, synth_buf)[MPA_MAX_CHANNELS][512*2];
     int synth_buf_offset[MPA_MAX_CHANNELS];
-    DECLARE_ALIGNED(16, int32_t, sb_samples)[MPA_MAX_CHANNELS][128][SBLIMIT];
+    DECLARE_ALIGNED(32, float, sb_samples)[MPA_MAX_CHANNELS][128][SBLIMIT];
 
     /// Mixed temporary data used in decoding
     float tone_level[MPA_MAX_CHANNELS][30][64];
@@ -329,11 +331,6 @@ static av_cold void qdm2_init_vlc(void)
     }
 }
 
-
-/* for floating point to fixed point conversion */
-static const float f2i_scale = (float) (1 << (FRAC_BITS - 15));
-
-
 static int qdm2_get_vlc (GetBitContext *gb, VLC *vlc, int flag, int depth)
 {
     int value;
@@ -482,8 +479,8 @@ static void build_sb_samples_from_noise (QDM2Context *q, int sb)
 
     for (ch = 0; ch < q->nb_channels; ch++)
         for (j = 0; j < 64; j++) {
-            q->sb_samples[ch][j * 2][sb] = (int32_t)(f2i_scale * SB_DITHERING_NOISE(sb,q->noise_idx) * q->tone_level[ch][sb][j] + .5);
-            q->sb_samples[ch][j * 2 + 1][sb] = (int32_t)(f2i_scale * SB_DITHERING_NOISE(sb,q->noise_idx) * q->tone_level[ch][sb][j] + .5);
+            q->sb_samples[ch][j * 2][sb] = SB_DITHERING_NOISE(sb,q->noise_idx) * q->tone_level[ch][sb][j];
+            q->sb_samples[ch][j * 2 + 1][sb] = SB_DITHERING_NOISE(sb,q->noise_idx) * q->tone_level[ch][sb][j];
         }
 }
 
@@ -923,11 +920,11 @@ static void synthfilt_build_sb_samples (QDM2Context *q, GetBitContext *gb, int l
                     for (chs = 0; chs < q->nb_channels; chs++)
                         for (k = 0; k < run; k++)
                             if ((j + k) < 128)
-                                q->sb_samples[chs][j + k][sb] = (int32_t)(f2i_scale * q->tone_level[chs][sb][((j + k)/2)] * tmp[k][chs] + .5);
+                                q->sb_samples[chs][j + k][sb] = q->tone_level[chs][sb][((j + k)/2)] * tmp[k][chs];
                 } else {
                     for (k = 0; k < run; k++)
                         if ((j + k) < 128)
-                            q->sb_samples[ch][j + k][sb] = (int32_t)(f2i_scale * q->tone_level[ch][sb][(j + k)/2] * samples[k] + .5);
+                            q->sb_samples[ch][j + k][sb] = q->tone_level[ch][sb][(j + k)/2] * samples[k];
                 }
 
                 j += run;
@@ -1601,7 +1598,7 @@ static void qdm2_calculate_fft (QDM2Context *q, int channel, int sub_packet)
  */
 static void qdm2_synthesis_filter (QDM2Context *q, int index)
 {
-    OUT_INT samples[MPA_MAX_CHANNELS * MPA_FRAME_SIZE];
+    float samples[MPA_MAX_CHANNELS * MPA_FRAME_SIZE];
     int i, k, ch, sb_used, sub_sampling, dither_state = 0;
 
     /* copy sb_samples */
@@ -1613,11 +1610,12 @@ static void qdm2_synthesis_filter (QDM2Context *q, int index)
                 q->sb_samples[ch][(8 * index) + i][k] = 0;
 
     for (ch = 0; ch < q->nb_channels; ch++) {
-        OUT_INT *samples_ptr = samples + ch;
+        float *samples_ptr = samples + ch;
 
         for (i = 0; i < 8; i++) {
-            ff_mpa_synth_filter_fixed(q->synth_buf[ch], &(q->synth_buf_offset[ch]),
-                ff_mpa_synth_window_fixed, &dither_state,
+            ff_mpa_synth_filter_float(&q->mpadsp,
+                q->synth_buf[ch], &(q->synth_buf_offset[ch]),
+                ff_mpa_synth_window_float, &dither_state,
                 samples_ptr, q->nb_channels,
                 q->sb_samples[ch][(8 * index) + i]);
             samples_ptr += 32 * q->nb_channels;
@@ -1629,7 +1627,7 @@ static void qdm2_synthesis_filter (QDM2Context *q, int index)
 
     for (ch = 0; ch < q->channels; ch++)
         for (i = 0; i < q->frame_size; i++)
-            q->output_buffer[q->channels * i + ch] += (float)(samples[q->nb_channels * sub_sampling * i + ch] >> (sizeof(OUT_INT)*8-16));
+            q->output_buffer[q->channels * i + ch] += (1 << 23) * samples[q->nb_channels * sub_sampling * i + ch];
 }
 
 
@@ -1646,7 +1644,7 @@ static av_cold void qdm2_init(QDM2Context *q) {
     initialized = 1;
 
     qdm2_init_vlc();
-    ff_mpa_synth_init_fixed(ff_mpa_synth_window_fixed);
+    ff_mpa_synth_init_float(ff_mpa_synth_window_float);
     softclip_table_init();
     rnd_table_init();
     init_noise_samples();
@@ -1863,6 +1861,7 @@ static av_cold int qdm2_decode_init(AVCodecContext *avctx)
     }
 
     ff_rdft_init(&s->rdft_ctx, s->fft_order, IDFT_C2R);
+    ff_mpadsp_init(&s->mpadsp);
 
     qdm2_init(s);
 
diff --git a/libavcodec/tableprint.h b/libavcodec/tableprint.h
index 97a667db33..d81b9a387b 100644
--- a/libavcodec/tableprint.h
+++ b/libavcodec/tableprint.h
@@ -23,8 +23,9 @@
 #ifndef AVCODEC_TABLEPRINT_H
 #define AVCODEC_TABLEPRINT_H
 
-#include <stdint.h>
+#include <inttypes.h>
 #include <stdio.h>
+
 #include "libavutil/common.h"
 
 #define WRITE_1D_FUNC_ARGV(type, linebrk, fmtstr, ...)\
@@ -70,9 +71,6 @@ void write_uint32_t_2d_array(const void *, int, int);
 void write_float_2d_array   (const void *, int, int);
 /** \} */ // end of printfuncs group
 
-/** Write a standard file header */
-void write_fileheader(void);
-
 #define WRITE_ARRAY(prefix, type, name)                 \
     do {                                                \
         const size_t array_size = FF_ARRAY_ELEMS(name); \
@@ -92,4 +90,22 @@ void write_fileheader(void);
         printf("};\n");                                                 \
     } while(0)
 
+
+WRITE_1D_FUNC(int8_t,   "%3"PRIi8, 15)
+WRITE_1D_FUNC(uint8_t,  "0x%02"PRIx8, 15)
+WRITE_1D_FUNC(uint16_t, "0x%08"PRIx16, 7)
+WRITE_1D_FUNC(uint32_t, "0x%08"PRIx32, 7)
+WRITE_1D_FUNC(float,    "%.18e", 3)
+
+WRITE_2D_FUNC(int8_t)
+WRITE_2D_FUNC(uint8_t)
+WRITE_2D_FUNC(uint32_t)
+WRITE_2D_FUNC(float)
+
+static inline void write_fileheader(void)
+{
+    printf("/* This file was automatically generated. */\n");
+    printf("#include <stdint.h>\n");
+}
+
 #endif /* AVCODEC_TABLEPRINT_H */
diff --git a/libavcodec/x86/Makefile b/libavcodec/x86/Makefile
index 4775853697..578ce04d77 100644
--- a/libavcodec/x86/Makefile
+++ b/libavcodec/x86/Makefile
@@ -21,11 +21,7 @@ YASM-OBJS-$(CONFIG_VC1_DECODER)        += x86/vc1dsp_yasm.o
 MMX-OBJS-$(CONFIG_AC3DSP)              += x86/ac3dsp_mmx.o
 YASM-OBJS-$(CONFIG_AC3DSP)             += x86/ac3dsp.o
 MMX-OBJS-$(CONFIG_CAVS_DECODER)        += x86/cavsdsp_mmx.o
-MMX-OBJS-$(CONFIG_MP1FLOAT_DECODER)    += x86/mpegaudiodec_mmx.o
-MMX-OBJS-$(CONFIG_MP2FLOAT_DECODER)    += x86/mpegaudiodec_mmx.o
-MMX-OBJS-$(CONFIG_MP3FLOAT_DECODER)    += x86/mpegaudiodec_mmx.o
-MMX-OBJS-$(CONFIG_MP3ON4FLOAT_DECODER) += x86/mpegaudiodec_mmx.o
-MMX-OBJS-$(CONFIG_MP3ADUFLOAT_DECODER) += x86/mpegaudiodec_mmx.o
+MMX-OBJS-$(CONFIG_MPEGAUDIODSP)        += x86/mpegaudiodec_mmx.o
 MMX-OBJS-$(CONFIG_PNG_DECODER)         += x86/png_mmx.o
 MMX-OBJS-$(CONFIG_ENCODERS)            += x86/dsputilenc_mmx.o
 YASM-OBJS-$(CONFIG_ENCODERS)           += x86/dsputilenc_yasm.o
diff --git a/libavcodec/x86/mpegaudiodec_mmx.c b/libavcodec/x86/mpegaudiodec_mmx.c
index 2f34281510..d7f8a0a142 100644
--- a/libavcodec/x86/mpegaudiodec_mmx.c
+++ b/libavcodec/x86/mpegaudiodec_mmx.c
@@ -21,9 +21,8 @@
 
 #include "libavutil/cpu.h"
 #include "libavutil/x86_cpu.h"
-
-#define CONFIG_FLOAT 1
-#include "libavcodec/mpegaudio.h"
+#include "libavcodec/dsputil.h"
+#include "libavcodec/mpegaudiodsp.h"
 
 #define MACS(rt, ra, rb) rt+=(ra)*(rb)
 #define MLSS(rt, ra, rb) rt-=(ra)*(rb)
@@ -148,11 +147,11 @@ static void apply_window_mp3(float *in, float *win, int *unused, float *out,
     *out = sum;
 }
 
-void ff_mpegaudiodec_init_mmx(MPADecodeContext *s)
+void ff_mpadsp_init_mmx(MPADSPContext *s)
 {
     int mm_flags = av_get_cpu_flags();
 
     if (mm_flags & AV_CPU_FLAG_SSE2) {
-        s->apply_window_mp3 = apply_window_mp3;
+        s->apply_window_float = apply_window_mp3;
     }
 }
diff --git a/libavformat/asfdec.c b/libavformat/asfdec.c
index 84881592b6..3564bf465f 100644
--- a/libavformat/asfdec.c
+++ b/libavformat/asfdec.c
@@ -21,6 +21,7 @@
 
 //#define DEBUG
 
+#include "libavutil/bswap.h"
 #include "libavutil/common.h"
 #include "libavutil/avstring.h"
 #include "libavcodec/mpegaudio.h"
diff --git a/libavformat/avformat.h b/libavformat/avformat.h
index f9091f0afd..2574bbc424 100644
--- a/libavformat/avformat.h
+++ b/libavformat/avformat.h
@@ -23,8 +23,7 @@
 
 
 /**
- * I return the LIBAVFORMAT_VERSION_INT constant.  You got
- * a fucking problem with that, douchebag?
+ * Return the LIBAVFORMAT_VERSION_INT constant.
  */
 unsigned avformat_version(void);