Separate DWT from snow and dsputil

This moves the DWT functions from snow.c and dsputil.c to a file of
their own.  A new struct, DWTContext, holds the function pointers
previously part of DSPContext.

Originally committed as revision 22522 to svn://svn.ffmpeg.org/ffmpeg/trunk

11 files changed, 1043 insertions, 974 deletions

diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 2253e29cbb..47ebcb282b 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -28,6 +28,7 @@ OBJS = allcodecs.o                                                      \
 OBJS-$(CONFIG_AANDCT)                  += aandcttab.o
 OBJS-$(CONFIG_ENCODERS)                += faandct.o jfdctfst.o jfdctint.o
 OBJS-$(CONFIG_DCT)                     += dct.o
+OBJS-$(CONFIG_DWT)                     += dwt.o
 OBJS-$(CONFIG_DXVA2)                   += dxva2.o
 FFT-OBJS-$(CONFIG_HARDCODED_TABLES)    += cos_tables.o
 OBJS-$(CONFIG_FFT)                     += avfft.o fft.o $(FFT-OBJS-yes)
@@ -598,7 +599,7 @@ MMX-OBJS-$(CONFIG_CAVS_DECODER)        += x86/cavsdsp_mmx.o
 MMX-OBJS-$(CONFIG_ENCODERS)            += x86/dsputilenc_mmx.o
 MMX-OBJS-$(CONFIG_GPL)                 += x86/idct_mmx.o
 MMX-OBJS-$(CONFIG_LPC)                 += x86/lpc_mmx.o
-MMX-OBJS-$(CONFIG_SNOW_DECODER)        += x86/snowdsp_mmx.o
+MMX-OBJS-$(CONFIG_DWT)                 += x86/snowdsp_mmx.o
 MMX-OBJS-$(CONFIG_VC1_DECODER)         += x86/vc1dsp_mmx.o
 MMX-OBJS-$(CONFIG_VP3_DECODER)         += x86/vp3dsp_mmx.o              \
                                           x86/vp3dsp_sse2.o
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index cb9d38100c..8535c6d665 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -33,7 +33,6 @@
 #include "faandct.h"
 #include "faanidct.h"
 #include "mathops.h"
-#include "snow.h"
 #include "mpegvideo.h"
 #include "config.h"
 #include "lpc.h"
@@ -329,102 +328,6 @@ static int sse16_c(void *v, uint8_t *pix1, uint8_t *pix2, int line_size, int h)
     return s;
 }
 
-
-#if CONFIG_SNOW_ENCODER //dwt is in snow.c
-static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){
-    int s, i, j;
-    const int dec_count= w==8 ? 3 : 4;
-    int tmp[32*32];
-    int level, ori;
-    static const int scale[2][2][4][4]={
-      {
-        {
-            // 9/7 8x8 dec=3
-            {268, 239, 239, 213},
-            {  0, 224, 224, 152},
-            {  0, 135, 135, 110},
-        },{
-            // 9/7 16x16 or 32x32 dec=4
-            {344, 310, 310, 280},
-            {  0, 320, 320, 228},
-            {  0, 175, 175, 136},
-            {  0, 129, 129, 102},
-        }
-      },{
-        {
-            // 5/3 8x8 dec=3
-            {275, 245, 245, 218},
-            {  0, 230, 230, 156},
-            {  0, 138, 138, 113},
-        },{
-            // 5/3 16x16 or 32x32 dec=4
-            {352, 317, 317, 286},
-            {  0, 328, 328, 233},
-            {  0, 180, 180, 140},
-            {  0, 132, 132, 105},
-        }
-      }
-    };
-
-    for (i = 0; i < h; i++) {
-        for (j = 0; j < w; j+=4) {
-            tmp[32*i+j+0] = (pix1[j+0] - pix2[j+0])<<4;
-            tmp[32*i+j+1] = (pix1[j+1] - pix2[j+1])<<4;
-            tmp[32*i+j+2] = (pix1[j+2] - pix2[j+2])<<4;
-            tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4;
-        }
-        pix1 += line_size;
-        pix2 += line_size;
-    }
-
-    ff_spatial_dwt(tmp, w, h, 32, type, dec_count);
-
-    s=0;
-    assert(w==h);
-    for(level=0; level<dec_count; level++){
-        for(ori= level ? 1 : 0; ori<4; ori++){
-            int size= w>>(dec_count-level);
-            int sx= (ori&1) ? size : 0;
-            int stride= 32<<(dec_count-level);
-            int sy= (ori&2) ? stride>>1 : 0;
-
-            for(i=0; i<size; i++){
-                for(j=0; j<size; j++){
-                    int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori];
-                    s += FFABS(v);
-                }
-            }
-        }
-    }
-    assert(s>=0);
-    return s>>9;
-}
-
-static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
-    return w_c(v, pix1, pix2, line_size,  8, h, 1);
-}
-
-static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
-    return w_c(v, pix1, pix2, line_size,  8, h, 0);
-}
-
-static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
-    return w_c(v, pix1, pix2, line_size, 16, h, 1);
-}
-
-static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
-    return w_c(v, pix1, pix2, line_size, 16, h, 0);
-}
-
-int w53_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
-    return w_c(v, pix1, pix2, line_size, 32, h, 1);
-}
-
-int w97_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
-    return w_c(v, pix1, pix2, line_size, 32, h, 0);
-}
-#endif
-
 /* draw the edges of width 'w' of an image of size width, height */
 //FIXME check that this is ok for mpeg4 interlaced
 static void draw_edges_c(uint8_t *buf, int wrap, int width, int height, int w)
@@ -3531,7 +3434,7 @@ void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type){
         case FF_CMP_NSSE:
             cmp[i]= c->nsse[i];
             break;
-#if CONFIG_SNOW_ENCODER
+#if CONFIG_DWT
         case FF_CMP_W53:
             cmp[i]= c->w53[i];
             break;
@@ -4816,11 +4719,8 @@ av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
     c->vsse[5]= vsse_intra8_c;
     c->nsse[0]= nsse16_c;
     c->nsse[1]= nsse8_c;
-#if CONFIG_SNOW_ENCODER
-    c->w53[0]= w53_16_c;
-    c->w53[1]= w53_8_c;
-    c->w97[0]= w97_16_c;
-    c->w97[1]= w97_8_c;
+#if CONFIG_DWT
+    ff_dsputil_init_dwt(c);
 #endif
 
     c->ssd_int8_vs_int16 = ssd_int8_vs_int16_c;
@@ -4865,12 +4765,6 @@ av_cold void dsputil_init(DSPContext* c, AVCodecContext *avctx)
     c->try_8x8basis= try_8x8basis_c;
     c->add_8x8basis= add_8x8basis_c;
 
-#if CONFIG_SNOW_DECODER
-    c->vertical_compose97i = ff_snow_vertical_compose97i;
-    c->horizontal_compose97i = ff_snow_horizontal_compose97i;
-    c->inner_add_yblock = ff_snow_inner_add_yblock;
-#endif
-
 #if CONFIG_VORBIS_DECODER
     c->vorbis_inverse_coupling = vorbis_inverse_coupling;
 #endif
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index 97c08fe5d2..5a2246cbee 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -37,8 +37,6 @@
 //#define DEBUG
 /* dct code */
 typedef short DCTELEM;
-typedef int DWTELEM;
-typedef short IDWTELEM;
 
 void fdct_ifast (DCTELEM *data);
 void fdct_ifast248 (DCTELEM *data);
@@ -185,10 +183,6 @@ static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/;
 
-
-// for snow slices
-typedef struct slice_buffer_s slice_buffer;
-
 /**
  * Scantable.
  */
@@ -538,11 +532,6 @@ typedef struct DSPContext {
     void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
     void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]);
 
-    /* snow wavelet */
-    void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
-    void (*horizontal_compose97i)(IDWTELEM *b, int width);
-    void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
-
     void (*prefetch)(void *mem, int stride, int h);
 
     void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
@@ -681,6 +670,7 @@ void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
 
+void ff_dsputil_init_dwt(DSPContext *c);
 void ff_cavsdsp_init(DSPContext* c, AVCodecContext *avctx);
 void ff_rv30dsp_init(DSPContext* c, AVCodecContext *avctx);
 void ff_rv40dsp_init(DSPContext* c, AVCodecContext *avctx);
diff --git a/libavcodec/dwt.c b/libavcodec/dwt.c
new file mode 100644
index 0000000000..28b3036337
--- /dev/null
+++ b/libavcodec/dwt.c
@@ -0,0 +1,843 @@
+/*
+ * Copyright (C) 2004-2010 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/attributes.h"
+#include "dsputil.h"
+#include "dwt.h"
+
+void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer)
+{
+    int i;
+
+    buf->base_buffer = base_buffer;
+    buf->line_count = line_count;
+    buf->line_width = line_width;
+    buf->data_count = max_allocated_lines;
+    buf->line = av_mallocz (sizeof(IDWTELEM *) * line_count);
+    buf->data_stack = av_malloc (sizeof(IDWTELEM *) * max_allocated_lines);
+
+    for(i = 0; i < max_allocated_lines; i++){
+        buf->data_stack[i] = av_malloc (sizeof(IDWTELEM) * line_width);
+    }
+
+    buf->data_stack_top = max_allocated_lines - 1;
+}
+
+IDWTELEM * slice_buffer_load_line(slice_buffer * buf, int line)
+{
+    IDWTELEM * buffer;
+
+    assert(buf->data_stack_top >= 0);
+//  assert(!buf->line[line]);
+    if (buf->line[line])
+        return buf->line[line];
+
+    buffer = buf->data_stack[buf->data_stack_top];
+    buf->data_stack_top--;
+    buf->line[line] = buffer;
+
+    return buffer;
+}
+
+void slice_buffer_release(slice_buffer * buf, int line)
+{
+    IDWTELEM * buffer;
+
+    assert(line >= 0 && line < buf->line_count);
+    assert(buf->line[line]);
+
+    buffer = buf->line[line];
+    buf->data_stack_top++;
+    buf->data_stack[buf->data_stack_top] = buffer;
+    buf->line[line] = NULL;
+}
+
+void slice_buffer_flush(slice_buffer * buf)
+{
+    int i;
+    for(i = 0; i < buf->line_count; i++){
+        if (buf->line[i])
+            slice_buffer_release(buf, i);
+    }
+}
+
+void slice_buffer_destroy(slice_buffer * buf)
+{
+    int i;
+    slice_buffer_flush(buf);
+
+    for(i = buf->data_count - 1; i >= 0; i--){
+        av_freep(&buf->data_stack[i]);
+    }
+    av_freep(&buf->data_stack);
+    av_freep(&buf->line);
+}
+
+static inline int mirror(int v, int m){
+    while((unsigned)v > (unsigned)m){
+        v=-v;
+        if(v<0) v+= 2*m;
+    }
+    return v;
+}
+
+static av_always_inline void
+lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
+     int dst_step, int src_step, int ref_step,
+     int width, int mul, int add, int shift,
+     int highpass, int inverse){
+    const int mirror_left= !highpass;
+    const int mirror_right= (width&1) ^ highpass;
+    const int w= (width>>1) - 1 + (highpass & width);
+    int i;
+
+#define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref)))
+    if(mirror_left){
+        dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
+        dst += dst_step;
+        src += src_step;
+    }
+
+    for(i=0; i<w; i++){
+        dst[i*dst_step] =
+            LIFT(src[i*src_step],
+                 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift),
+                 inverse);
+    }
+
+    if(mirror_right){
+        dst[w*dst_step] =
+            LIFT(src[w*src_step],
+                 ((mul*2*ref[w*ref_step]+add)>>shift),
+                 inverse);
+    }
+}
+
+static av_always_inline void
+inv_lift(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref,
+         int dst_step, int src_step, int ref_step,
+         int width, int mul, int add, int shift,
+         int highpass, int inverse){
+    const int mirror_left= !highpass;
+    const int mirror_right= (width&1) ^ highpass;
+    const int w= (width>>1) - 1 + (highpass & width);
+    int i;
+
+#define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref)))
+    if(mirror_left){
+        dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
+        dst += dst_step;
+        src += src_step;
+    }
+
+    for(i=0; i<w; i++){
+        dst[i*dst_step] =
+            LIFT(src[i*src_step],
+                 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift),
+                 inverse);
+    }
+
+    if(mirror_right){
+        dst[w*dst_step] =
+            LIFT(src[w*src_step],
+                 ((mul*2*ref[w*ref_step]+add)>>shift),
+                 inverse);
+    }
+}
+
+#ifndef liftS
+static av_always_inline void
+liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
+      int dst_step, int src_step, int ref_step,
+      int width, int mul, int add, int shift,
+      int highpass, int inverse){
+    const int mirror_left= !highpass;
+    const int mirror_right= (width&1) ^ highpass;
+    const int w= (width>>1) - 1 + (highpass & width);
+    int i;
+
+    assert(shift == 4);
+#define LIFTS(src, ref, inv) \
+        ((inv) ? \
+            (src) + (((ref) + 4*(src))>>shift): \
+            -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
+    if(mirror_left){
+        dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
+        dst += dst_step;
+        src += src_step;
+    }
+
+    for(i=0; i<w; i++){
+        dst[i*dst_step] =
+            LIFTS(src[i*src_step],
+                  mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add,
+                  inverse);
+    }
+
+    if(mirror_right){
+        dst[w*dst_step] =
+            LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
+    }
+}
+static av_always_inline void
+inv_liftS(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref,
+          int dst_step, int src_step, int ref_step,
+          int width, int mul, int add, int shift,
+          int highpass, int inverse){
+    const int mirror_left= !highpass;
+    const int mirror_right= (width&1) ^ highpass;
+    const int w= (width>>1) - 1 + (highpass & width);
+    int i;
+
+    assert(shift == 4);
+#define LIFTS(src, ref, inv) \
+    ((inv) ? \
+        (src) + (((ref) + 4*(src))>>shift): \
+        -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
+    if(mirror_left){
+        dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
+        dst += dst_step;
+        src += src_step;
+    }
+
+    for(i=0; i<w; i++){
+        dst[i*dst_step] =
+            LIFTS(src[i*src_step],
+                  mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add,
+                  inverse);
+    }
+
+    if(mirror_right){
+        dst[w*dst_step] =
+            LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
+    }
+}
+#endif /* ! liftS */
+
+static void horizontal_decompose53i(DWTELEM *b, int width){
+    DWTELEM temp[width];
+    const int width2= width>>1;
+    int x;
+    const int w2= (width+1)>>1;
+
+    for(x=0; x<width2; x++){
+        temp[x   ]= b[2*x    ];
+        temp[x+w2]= b[2*x + 1];
+    }
+    if(width&1)
+        temp[x   ]= b[2*x    ];
+#if 0
+    {
+    int A1,A2,A3,A4;
+    A2= temp[1       ];
+    A4= temp[0       ];
+    A1= temp[0+width2];
+    A1 -= (A2 + A4)>>1;
+    A4 += (A1 + 1)>>1;
+    b[0+width2] = A1;
+    b[0       ] = A4;
+    for(x=1; x+1<width2; x+=2){
+        A3= temp[x+width2];
+        A4= temp[x+1     ];
+        A3 -= (A2 + A4)>>1;
+        A2 += (A1 + A3 + 2)>>2;
+        b[x+width2] = A3;
+        b[x       ] = A2;
+
+        A1= temp[x+1+width2];
+        A2= temp[x+2       ];
+        A1 -= (A2 + A4)>>1;
+        A4 += (A1 + A3 + 2)>>2;
+        b[x+1+width2] = A1;
+        b[x+1       ] = A4;
+    }
+    A3= temp[width-1];
+    A3 -= A2;
+    A2 += (A1 + A3 + 2)>>2;
+    b[width -1] = A3;
+    b[width2-1] = A2;
+    }
+#else
+    lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0);
+    lift(b   , temp   , b+w2, 1, 1, 1, width,  1, 2, 2, 0, 0);
+#endif /* 0 */
+}
+
+static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] -= (b0[i] + b2[i])>>1;
+    }
+}
+
+static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] += (b0[i] + b2[i] + 2)>>2;
+    }
+}
+
+static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){
+    int y;
+    DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride;
+    DWTELEM *b1= buffer + mirror(-2  , height-1)*stride;
+
+    for(y=-2; y<height; y+=2){
+        DWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
+        DWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
+
+        if(y+1<(unsigned)height) horizontal_decompose53i(b2, width);
+        if(y+2<(unsigned)height) horizontal_decompose53i(b3, width);
+
+        if(y+1<(unsigned)height) vertical_decompose53iH0(b1, b2, b3, width);
+        if(y+0<(unsigned)height) vertical_decompose53iL0(b0, b1, b2, width);
+
+        b0=b2;
+        b1=b3;
+    }
+}
+
+static void horizontal_decompose97i(DWTELEM *b, int width){
+    DWTELEM temp[width];
+    const int w2= (width+1)>>1;
+
+    lift (temp+w2, b    +1, b      , 1, 2, 2, width,  W_AM, W_AO, W_AS, 1, 1);
+    liftS(temp   , b      , temp+w2, 1, 2, 1, width,  W_BM, W_BO, W_BS, 0, 0);
+    lift (b   +w2, temp+w2, temp   , 1, 1, 1, width,  W_CM, W_CO, W_CS, 1, 0);
+    lift (b      , temp   , b   +w2, 1, 1, 1, width,  W_DM, W_DO, W_DS, 0, 0);
+}
+
+
+static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
+    }
+}
+
+static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
+    }
+}
+
+static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+#ifdef liftS
+        b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
+#else
+        b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + W_BO*5 + (5<<27)) / (5*16) - (1<<23);
+#endif
+    }
+}
+
+static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
+    }
+}
+
+static void spatial_decompose97i(DWTELEM *buffer, int width, int height, int stride){
+    int y;
+    DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride;
+    DWTELEM *b1= buffer + mirror(-4  , height-1)*stride;
+    DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride;
+    DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride;
+
+    for(y=-4; y<height; y+=2){
+        DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
+        DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
+
+        if(y+3<(unsigned)height) horizontal_decompose97i(b4, width);
+        if(y+4<(unsigned)height) horizontal_decompose97i(b5, width);
+
+        if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width);
+        if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width);
+        if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width);
+        if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width);
+
+        b0=b2;
+        b1=b3;
+        b2=b4;
+        b3=b5;
+    }
+}
+
+void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
+    int level;
+
+    for(level=0; level<decomposition_count; level++){
+        switch(type){
+        case DWT_97: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break;
+        case DWT_53: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break;
+        }
+    }
+}
+
+static void horizontal_compose53i(IDWTELEM *b, int width){
+    IDWTELEM temp[width];
+    const int width2= width>>1;
+    const int w2= (width+1)>>1;
+    int x;
+
+    for(x=0; x<width2; x++){
+        temp[2*x    ]= b[x   ];
+        temp[2*x + 1]= b[x+w2];
+    }
+    if(width&1)
+        temp[2*x    ]= b[x   ];
+
+    b[0] = temp[0] - ((temp[1]+1)>>1);
+    for(x=2; x<width-1; x+=2){
+        b[x  ] = temp[x  ] - ((temp[x-1] + temp[x+1]+2)>>2);
+        b[x-1] = temp[x-1] + ((b   [x-2] + b   [x  ]+1)>>1);
+    }
+    if(width&1){
+        b[x  ] = temp[x  ] - ((temp[x-1]+1)>>1);
+        b[x-1] = temp[x-1] + ((b   [x-2] + b  [x  ]+1)>>1);
+    }else
+        b[x-1] = temp[x-1] + b[x-2];
+}
+
+static void vertical_compose53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] += (b0[i] + b2[i])>>1;
+    }
+}
+
+static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] -= (b0[i] + b2[i] + 2)>>2;
+    }
+}
+
+static void spatial_compose53i_buffered_init(DWTCompose *cs, slice_buffer * sb, int height, int stride_line){
+    cs->b0 = slice_buffer_get_line(sb, mirror(-1-1, height-1) * stride_line);
+    cs->b1 = slice_buffer_get_line(sb, mirror(-1  , height-1) * stride_line);
+    cs->y = -1;
+}
+
+static void spatial_compose53i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride){
+    cs->b0 = buffer + mirror(-1-1, height-1)*stride;
+    cs->b1 = buffer + mirror(-1  , height-1)*stride;
+    cs->y = -1;
+}
+
+static void spatial_compose53i_dy_buffered(DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line){
+    int y= cs->y;
+
+    IDWTELEM *b0= cs->b0;
+    IDWTELEM *b1= cs->b1;
+    IDWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line);
+    IDWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line);
+
+    if(y+1<(unsigned)height && y<(unsigned)height){
+        int x;
+
+        for(x=0; x<width; x++){
+            b2[x] -= (b1[x] + b3[x] + 2)>>2;
+            b1[x] += (b0[x] + b2[x])>>1;
+        }
+    }else{
+        if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
+        if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
+    }
+
+        if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
+        if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
+
+    cs->b0 = b2;
+    cs->b1 = b3;
+    cs->y += 2;
+}
+
+static void spatial_compose53i_dy(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride){
+    int y= cs->y;
+    IDWTELEM *b0= cs->b0;
+    IDWTELEM *b1= cs->b1;
+    IDWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
+    IDWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
+
+        if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
+        if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
+
+        if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
+        if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
+
+    cs->b0 = b2;
+    cs->b1 = b3;
+    cs->y += 2;
+}
+
+static void av_unused spatial_compose53i(IDWTELEM *buffer, int width, int height, int stride){
+    DWTCompose cs;
+    spatial_compose53i_init(&cs, buffer, height, stride);
+    while(cs.y <= height)
+        spatial_compose53i_dy(&cs, buffer, width, height, stride);
+}
+
+
+void ff_snow_horizontal_compose97i(IDWTELEM *b, int width){
+    IDWTELEM temp[width];
+    const int w2= (width+1)>>1;
+
+#if 0 //maybe more understadable but slower
+    inv_lift (temp   , b      , b   +w2, 2, 1, 1, width,  W_DM, W_DO, W_DS, 0, 1);
+    inv_lift (temp+1 , b   +w2, temp   , 2, 1, 2, width,  W_CM, W_CO, W_CS, 1, 1);
+
+    inv_liftS(b      , temp   , temp+1 , 2, 2, 2, width,  W_BM, W_BO, W_BS, 0, 1);
+    inv_lift (b+1    , temp+1 , b      , 2, 2, 2, width,  W_AM, W_AO, W_AS, 1, 0);
+#else
+    int x;
+    temp[0] = b[0] - ((3*b[w2]+2)>>2);
+    for(x=1; x<(width>>1); x++){
+        temp[2*x  ] = b[x     ] - ((3*(b   [x+w2-1] + b[x+w2])+4)>>3);
+        temp[2*x-1] = b[x+w2-1] - temp[2*x-2] - temp[2*x];
+    }
+    if(width&1){
+        temp[2*x  ] = b[x     ] - ((3*b   [x+w2-1]+2)>>2);
+        temp[2*x-1] = b[x+w2-1] - temp[2*x-2] - temp[2*x];
+    }else
+        temp[2*x-1] = b[x+w2-1] - 2*temp[2*x-2];
+
+    b[0] = temp[0] + ((2*temp[0] + temp[1]+4)>>3);
+    for(x=2; x<width-1; x+=2){
+        b[x  ] = temp[x  ] + ((4*temp[x  ] + temp[x-1] + temp[x+1]+8)>>4);
+        b[x-1] = temp[x-1] + ((3*(b  [x-2] + b   [x  ] ))>>1);
+    }
+    if(width&1){
+        b[x  ] = temp[x  ] + ((2*temp[x  ] + temp[x-1]+4)>>3);
+        b[x-1] = temp[x-1] + ((3*(b  [x-2] + b   [x  ] ))>>1);
+    }else
+        b[x-1] = temp[x-1] + 3*b [x-2];
+#endif
+}
+
+static void vertical_compose97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
+    }
+}
+
+static void vertical_compose97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
+    }
+}
+
+static void vertical_compose97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+#ifdef liftS
+        b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
+#else
+        b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS;
+#endif
+    }
+}
+
+static void vertical_compose97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
+    }
+}
+
+void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
+    int i;
+
+    for(i=0; i<width; i++){
+        b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
+        b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
+#ifdef liftS
+        b2[i] += (W_BM*(b1[i] + b3[i])+W_BO)>>W_BS;
+#else
+        b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
+#endif
+        b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
+    }
+}
+
+static void spatial_compose97i_buffered_init(DWTCompose *cs, slice_buffer * sb, int height, int stride_line){
+    cs->b0 = slice_buffer_get_line(sb, mirror(-3-1, height-1) * stride_line);
+    cs->b1 = slice_buffer_get_line(sb, mirror(-3  , height-1) * stride_line);
+    cs->b2 = slice_buffer_get_line(sb, mirror(-3+1, height-1) * stride_line);
+    cs->b3 = slice_buffer_get_line(sb, mirror(-3+2, height-1) * stride_line);
+    cs->y = -3;
+}
+
+static void spatial_compose97i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride){
+    cs->b0 = buffer + mirror(-3-1, height-1)*stride;
+    cs->b1 = buffer + mirror(-3  , height-1)*stride;
+    cs->b2 = buffer + mirror(-3+1, height-1)*stride;
+    cs->b3 = buffer + mirror(-3+2, height-1)*stride;
+    cs->y = -3;
+}
+
+static void spatial_compose97i_dy_buffered(DWTContext *dsp, DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line){
+    int y = cs->y;
+
+    IDWTELEM *b0= cs->b0;
+    IDWTELEM *b1= cs->b1;
+    IDWTELEM *b2= cs->b2;
+    IDWTELEM *b3= cs->b3;
+    IDWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line);
+    IDWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line);
+
+    if(y>0 && y+4<height){
+        dsp->vertical_compose97i(b0, b1, b2, b3, b4, b5, width);
+    }else{
+        if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
+        if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
+        if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
+        if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
+    }
+
+    if(y-1<(unsigned)height) dsp->horizontal_compose97i(b0, width);
+    if(y+0<(unsigned)height) dsp->horizontal_compose97i(b1, width);
+
+    cs->b0=b2;
+    cs->b1=b3;
+    cs->b2=b4;
+    cs->b3=b5;
+    cs->y += 2;
+}
+
+static void spatial_compose97i_dy(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride){
+    int y = cs->y;
+    IDWTELEM *b0= cs->b0;
+    IDWTELEM *b1= cs->b1;
+    IDWTELEM *b2= cs->b2;
+    IDWTELEM *b3= cs->b3;
+    IDWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
+    IDWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
+
+    if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
+    if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
+    if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
+    if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
+
+    if(y-1<(unsigned)height) ff_snow_horizontal_compose97i(b0, width);
+    if(y+0<(unsigned)height) ff_snow_horizontal_compose97i(b1, width);
+
+    cs->b0=b2;
+    cs->b1=b3;
+    cs->b2=b4;
+    cs->b3=b5;
+    cs->y += 2;
+}
+
+static void av_unused spatial_compose97i(IDWTELEM *buffer, int width, int height, int stride){
+    DWTCompose cs;
+    spatial_compose97i_init(&cs, buffer, height, stride);
+    while(cs.y <= height)
+        spatial_compose97i_dy(&cs, buffer, width, height, stride);
+}
+
+void ff_spatial_idwt_buffered_init(DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line, int type, int decomposition_count){
+    int level;
+    for(level=decomposition_count-1; level>=0; level--){
+        switch(type){
+        case DWT_97: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
+        case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
+        }
+    }
+}
+
+void ff_spatial_idwt_buffered_slice(DWTContext *dsp, DWTCompose *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){
+    const int support = type==1 ? 3 : 5;
+    int level;
+    if(type==2) return;
+
+    for(level=decomposition_count-1; level>=0; level--){
+        while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
+            switch(type){
+            case DWT_97: spatial_compose97i_dy_buffered(dsp, cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
+                break;
+            case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
+                break;
+            }
+        }
+    }
+}
+
+void ff_spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
+    int level;
+    for(level=decomposition_count-1; level>=0; level--){
+        switch(type){
+        case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break;
+        case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break;
+        }
+    }
+}
+
+void ff_spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){
+    const int support = type==1 ? 3 : 5;
+    int level;
+    if(type==2) return;
+
+    for(level=decomposition_count-1; level>=0; level--){
+        while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
+            switch(type){
+            case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
+                break;
+            case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
+                break;
+            }
+        }
+    }
+}
+
+void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
+        DWTCompose cs[MAX_DECOMPOSITIONS];
+        int y;
+        ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count);
+        for(y=0; y<height; y+=4)
+            ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y);
+}
+
+static inline int w_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int w, int h, int type){
+    int s, i, j;
+    const int dec_count= w==8 ? 3 : 4;
+    int tmp[32*32];
+    int level, ori;
+    static const int scale[2][2][4][4]={
+      {
+        {
+            // 9/7 8x8 dec=3
+            {268, 239, 239, 213},
+            {  0, 224, 224, 152},
+            {  0, 135, 135, 110},
+        },{
+            // 9/7 16x16 or 32x32 dec=4
+            {344, 310, 310, 280},
+            {  0, 320, 320, 228},
+            {  0, 175, 175, 136},
+            {  0, 129, 129, 102},
+        }
+      },{
+        {
+            // 5/3 8x8 dec=3
+            {275, 245, 245, 218},
+            {  0, 230, 230, 156},
+            {  0, 138, 138, 113},
+        },{
+            // 5/3 16x16 or 32x32 dec=4
+            {352, 317, 317, 286},
+            {  0, 328, 328, 233},
+            {  0, 180, 180, 140},
+            {  0, 132, 132, 105},
+        }
+      }
+    };
+
+    for (i = 0; i < h; i++) {
+        for (j = 0; j < w; j+=4) {
+            tmp[32*i+j+0] = (pix1[j+0] - pix2[j+0])<<4;
+            tmp[32*i+j+1] = (pix1[j+1] - pix2[j+1])<<4;
+            tmp[32*i+j+2] = (pix1[j+2] - pix2[j+2])<<4;
+            tmp[32*i+j+3] = (pix1[j+3] - pix2[j+3])<<4;
+        }
+        pix1 += line_size;
+        pix2 += line_size;
+    }
+
+    ff_spatial_dwt(tmp, w, h, 32, type, dec_count);
+
+    s=0;
+    assert(w==h);
+    for(level=0; level<dec_count; level++){
+        for(ori= level ? 1 : 0; ori<4; ori++){
+            int size= w>>(dec_count-level);
+            int sx= (ori&1) ? size : 0;
+            int stride= 32<<(dec_count-level);
+            int sy= (ori&2) ? stride>>1 : 0;
+
+            for(i=0; i<size; i++){
+                for(j=0; j<size; j++){
+                    int v= tmp[sx + sy + i*stride + j] * scale[type][dec_count-3][level][ori];
+                    s += FFABS(v);
+                }
+            }
+        }
+    }
+    assert(s>=0);
+    return s>>9;
+}
+
+static int w53_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
+    return w_c(v, pix1, pix2, line_size,  8, h, 1);
+}
+
+static int w97_8_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
+    return w_c(v, pix1, pix2, line_size,  8, h, 0);
+}
+
+static int w53_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
+    return w_c(v, pix1, pix2, line_size, 16, h, 1);
+}
+
+static int w97_16_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
+    return w_c(v, pix1, pix2, line_size, 16, h, 0);
+}
+
+int w53_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
+    return w_c(v, pix1, pix2, line_size, 32, h, 1);
+}
+
+int w97_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h){
+    return w_c(v, pix1, pix2, line_size, 32, h, 0);
+}
+
+void ff_dsputil_init_dwt(DSPContext *c)
+{
+    c->w53[0]= w53_16_c;
+    c->w53[1]= w53_8_c;
+    c->w97[0]= w97_16_c;
+    c->w97[1]= w97_8_c;
+}
+
+void ff_dwt_init(DWTContext *c)
+{
+    c->vertical_compose97i = ff_snow_vertical_compose97i;
+    c->horizontal_compose97i = ff_snow_horizontal_compose97i;
+    c->inner_add_yblock = ff_snow_inner_add_yblock;
+
+    if (ARCH_X86) ff_dwt_init_x86(c);
+}
diff --git a/libavcodec/dwt.h b/libavcodec/dwt.h
new file mode 100644
index 0000000000..d9a8b16135
--- /dev/null
+++ b/libavcodec/dwt.h
@@ -0,0 +1,156 @@
+/*
+ * Copyright (C) 2004-2010 Michael Niedermayer <michaelni@gmx.at>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVCODEC_DWT_H
+#define AVCODEC_DWT_H
+
+#include <stdint.h>
+
+typedef int DWTELEM;
+typedef short IDWTELEM;
+
+typedef struct {
+    IDWTELEM *b0;
+    IDWTELEM *b1;
+    IDWTELEM *b2;
+    IDWTELEM *b3;
+    int y;
+} DWTCompose;
+
+/** Used to minimize the amount of memory used in order to optimize cache performance. **/
+typedef struct slice_buffer_s {
+    IDWTELEM * * line; ///< For use by idwt and predict_slices.
+    IDWTELEM * * data_stack; ///< Used for internal purposes.
+    int data_stack_top;
+    int line_count;
+    int line_width;
+    int data_count;
+    IDWTELEM * base_buffer; ///< Buffer that this structure is caching.
+} slice_buffer;
+
+typedef struct DWTContext {
+    void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
+    void (*horizontal_compose97i)(IDWTELEM *b, int width);
+    void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
+} DWTContext;
+
+#define MAX_DECOMPOSITIONS 8
+
+#define DWT_97 0
+#define DWT_53 1
+
+#define liftS lift
+#if 1
+#define W_AM 3
+#define W_AO 0
+#define W_AS 1
+
+#undef liftS
+#define W_BM 1
+#define W_BO 8
+#define W_BS 4
+
+#define W_CM 1
+#define W_CO 0
+#define W_CS 0
+
+#define W_DM 3
+#define W_DO 4
+#define W_DS 3
+#elif 0
+#define W_AM 55
+#define W_AO 16
+#define W_AS 5
+
+#define W_BM 3
+#define W_BO 32
+#define W_BS 6
+
+#define W_CM 127
+#define W_CO 64
+#define W_CS 7
+
+#define W_DM 7
+#define W_DO 8
+#define W_DS 4
+#elif 0
+#define W_AM 97
+#define W_AO 32
+#define W_AS 6
+
+#define W_BM 63
+#define W_BO 512
+#define W_BS 10
+
+#define W_CM 13
+#define W_CO 8
+#define W_CS 4
+
+#define W_DM 15
+#define W_DO 16
+#define W_DS 5
+
+#else
+
+#define W_AM 203
+#define W_AO 64
+#define W_AS 7
+
+#define W_BM 217
+#define W_BO 2048
+#define W_BS 12
+
+#define W_CM 113
+#define W_CO 64
+#define W_CS 7
+
+#define W_DM 227
+#define W_DO 128
+#define W_DS 9
+#endif
+
+#define slice_buffer_get_line(slice_buf, line_num) ((slice_buf)->line[line_num] ? (slice_buf)->line[line_num] : slice_buffer_load_line((slice_buf), (line_num)))
+//#define slice_buffer_get_line(slice_buf, line_num) (slice_buffer_load_line((slice_buf), (line_num)))
+
+void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer);
+void slice_buffer_release(slice_buffer * buf, int line);
+void slice_buffer_flush(slice_buffer * buf);
+void slice_buffer_destroy(slice_buffer * buf);
+IDWTELEM * slice_buffer_load_line(slice_buffer * buf, int line);
+
+void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
+void ff_snow_horizontal_compose97i(IDWTELEM *b, int width);
+void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
+
+int w53_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h);
+int w97_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h);
+
+void ff_spatial_dwt(int *buffer, int width, int height, int stride, int type, int decomposition_count);
+
+void ff_spatial_idwt_buffered_init(DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line, int type, int decomposition_count);
+void ff_spatial_idwt_buffered_slice(DWTContext *dsp, DWTCompose *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y);
+void ff_spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count);
+void ff_spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y);
+void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count);
+
+void ff_dwt_init(DWTContext *c);
+void ff_dwt_init_x86(DWTContext *c);
+
+#endif /* AVCODEC_DWT_H */
diff --git a/libavcodec/ivi_dsp.c b/libavcodec/ivi_dsp.c
index f18f378aef..5aa8f94c99 100644
--- a/libavcodec/ivi_dsp.c
+++ b/libavcodec/ivi_dsp.c
@@ -28,6 +28,7 @@
 
 #include "avcodec.h"
 #include "dsputil.h"
+#include "dwt.h"
 #include "ivi_common.h"
 #include "ivi_dsp.h"
 
diff --git a/libavcodec/snow.c b/libavcodec/snow.c
index d51c998097..a4758b69eb 100644
--- a/libavcodec/snow.c
+++ b/libavcodec/snow.c
@@ -21,6 +21,7 @@
 #include "libavutil/intmath.h"
 #include "avcodec.h"
 #include "dsputil.h"
+#include "dwt.h"
 #include "snow.h"
 
 #include "rangecoder.h"
@@ -440,6 +441,7 @@ typedef struct SnowContext{
     AVCodecContext *avctx;
     RangeCoder c;
     DSPContext dsp;
+    DWTContext dwt;
     AVFrame new_picture;
     AVFrame input_picture;              ///< new_picture with the internal linesizes
     AVFrame current_picture;
@@ -494,85 +496,6 @@ typedef struct SnowContext{
     uint8_t *scratchbuf;
 }SnowContext;
 
-typedef struct {
-    IDWTELEM *b0;
-    IDWTELEM *b1;
-    IDWTELEM *b2;
-    IDWTELEM *b3;
-    int y;
-} DWTCompose;
-
-#define slice_buffer_get_line(slice_buf, line_num) ((slice_buf)->line[line_num] ? (slice_buf)->line[line_num] : slice_buffer_load_line((slice_buf), (line_num)))
-//#define slice_buffer_get_line(slice_buf, line_num) (slice_buffer_load_line((slice_buf), (line_num)))
-
-static void slice_buffer_init(slice_buffer * buf, int line_count, int max_allocated_lines, int line_width, IDWTELEM * base_buffer)
-{
-    int i;
-
-    buf->base_buffer = base_buffer;
-    buf->line_count = line_count;
-    buf->line_width = line_width;
-    buf->data_count = max_allocated_lines;
-    buf->line = av_mallocz (sizeof(IDWTELEM *) * line_count);
-    buf->data_stack = av_malloc (sizeof(IDWTELEM *) * max_allocated_lines);
-
-    for(i = 0; i < max_allocated_lines; i++){
-        buf->data_stack[i] = av_malloc (sizeof(IDWTELEM) * line_width);
-    }
-
-    buf->data_stack_top = max_allocated_lines - 1;
-}
-
-static IDWTELEM * slice_buffer_load_line(slice_buffer * buf, int line)
-{
-    IDWTELEM * buffer;
-
-    assert(buf->data_stack_top >= 0);
-//  assert(!buf->line[line]);
-    if (buf->line[line])
-        return buf->line[line];
-
-    buffer = buf->data_stack[buf->data_stack_top];
-    buf->data_stack_top--;
-    buf->line[line] = buffer;
-
-    return buffer;
-}
-
-static void slice_buffer_release(slice_buffer * buf, int line)
-{
-    IDWTELEM * buffer;
-
-    assert(line >= 0 && line < buf->line_count);
-    assert(buf->line[line]);
-
-    buffer = buf->line[line];
-    buf->data_stack_top++;
-    buf->data_stack[buf->data_stack_top] = buffer;
-    buf->line[line] = NULL;
-}
-
-static void slice_buffer_flush(slice_buffer * buf)
-{
-    int i;
-    for(i = 0; i < buf->line_count; i++){
-        if (buf->line[i])
-            slice_buffer_release(buf, i);
-    }
-}
-
-static void slice_buffer_destroy(slice_buffer * buf)
-{
-    int i;
-    slice_buffer_flush(buf);
-
-    for(i = buf->data_count - 1; i >= 0; i--){
-        av_freep(&buf->data_stack[i]);
-    }
-    av_freep(&buf->data_stack);
-    av_freep(&buf->line);
-}
-
 #ifdef __sgi
 // Avoid a name clash on SGI IRIX
 #undef qexp
@@ -580,14 +503,6 @@ static void slice_buffer_destroy(slice_buffer * buf)
 #define QEXPSHIFT (7-FRAC_BITS+8) //FIXME try to change this to 0
 static uint8_t qexp[QROOT];
 
-static inline int mirror(int v, int m){
-    while((unsigned)v > (unsigned)m){
-        v=-v;
-        if(v<0) v+= 2*m;
-    }
-    return v;
-}
-
 static inline void put_symbol(RangeCoder *c, uint8_t *state, int v, int is_signed){
     int i;
 
@@ -709,605 +624,6 @@ static inline int get_symbol2(RangeCoder *c, uint8_t *state, int log2){
     return v;
 }
 
-static av_always_inline void
-lift(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
-     int dst_step, int src_step, int ref_step,
-     int width, int mul, int add, int shift,
-     int highpass, int inverse){
-    const int mirror_left= !highpass;
-    const int mirror_right= (width&1) ^ highpass;
-    const int w= (width>>1) - 1 + (highpass & width);
-    int i;
-
-#define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref)))
-    if(mirror_left){
-        dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
-        dst += dst_step;
-        src += src_step;
-    }
-
-    for(i=0; i<w; i++){
-        dst[i*dst_step] =
-            LIFT(src[i*src_step],
-                 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift),
-                 inverse);
-    }
-
-    if(mirror_right){
-        dst[w*dst_step] =
-            LIFT(src[w*src_step],
-                 ((mul*2*ref[w*ref_step]+add)>>shift),
-                 inverse);
-    }
-}
-
-static av_always_inline void
-inv_lift(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref,
-         int dst_step, int src_step, int ref_step,
-         int width, int mul, int add, int shift,
-         int highpass, int inverse){
-    const int mirror_left= !highpass;
-    const int mirror_right= (width&1) ^ highpass;
-    const int w= (width>>1) - 1 + (highpass & width);
-    int i;
-
-#define LIFT(src, ref, inv) ((src) + ((inv) ? - (ref) : + (ref)))
-    if(mirror_left){
-        dst[0] = LIFT(src[0], ((mul*2*ref[0]+add)>>shift), inverse);
-        dst += dst_step;
-        src += src_step;
-    }
-
-    for(i=0; i<w; i++){
-        dst[i*dst_step] =
-            LIFT(src[i*src_step],
-                 ((mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add)>>shift),
-                 inverse);
-    }
-
-    if(mirror_right){
-        dst[w*dst_step] =
-            LIFT(src[w*src_step],
-                 ((mul*2*ref[w*ref_step]+add)>>shift),
-                 inverse);
-    }
-}
-
-#ifndef liftS
-static av_always_inline void
-liftS(DWTELEM *dst, DWTELEM *src, DWTELEM *ref,
-      int dst_step, int src_step, int ref_step,
-      int width, int mul, int add, int shift,
-      int highpass, int inverse){
-    const int mirror_left= !highpass;
-    const int mirror_right= (width&1) ^ highpass;
-    const int w= (width>>1) - 1 + (highpass & width);
-    int i;
-
-    assert(shift == 4);
-#define LIFTS(src, ref, inv) \
-        ((inv) ? \
-            (src) + (((ref) + 4*(src))>>shift): \
-            -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
-    if(mirror_left){
-        dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
-        dst += dst_step;
-        src += src_step;
-    }
-
-    for(i=0; i<w; i++){
-        dst[i*dst_step] =
-            LIFTS(src[i*src_step],
-                  mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add,
-                  inverse);
-    }
-
-    if(mirror_right){
-        dst[w*dst_step] =
-            LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
-    }
-}
-static av_always_inline void
-inv_liftS(IDWTELEM *dst, IDWTELEM *src, IDWTELEM *ref,
-          int dst_step, int src_step, int ref_step,
-          int width, int mul, int add, int shift,
-          int highpass, int inverse){
-    const int mirror_left= !highpass;
-    const int mirror_right= (width&1) ^ highpass;
-    const int w= (width>>1) - 1 + (highpass & width);
-    int i;
-
-    assert(shift == 4);
-#define LIFTS(src, ref, inv) \
-    ((inv) ? \
-        (src) + (((ref) + 4*(src))>>shift): \
-        -((-16*(src) + (ref) + add/4 + 1 + (5<<25))/(5*4) - (1<<23)))
-    if(mirror_left){
-        dst[0] = LIFTS(src[0], mul*2*ref[0]+add, inverse);
-        dst += dst_step;
-        src += src_step;
-    }
-
-    for(i=0; i<w; i++){
-        dst[i*dst_step] =
-            LIFTS(src[i*src_step],
-                  mul*(ref[i*ref_step] + ref[(i+1)*ref_step])+add,
-                  inverse);
-    }
-
-    if(mirror_right){
-        dst[w*dst_step] =
-            LIFTS(src[w*src_step], mul*2*ref[w*ref_step]+add, inverse);
-    }
-}
-#endif /* ! liftS */
-
-static void horizontal_decompose53i(DWTELEM *b, int width){
-    DWTELEM temp[width];
-    const int width2= width>>1;
-    int x;
-    const int w2= (width+1)>>1;
-
-    for(x=0; x<width2; x++){
-        temp[x   ]= b[2*x    ];
-        temp[x+w2]= b[2*x + 1];
-    }
-    if(width&1)
-        temp[x   ]= b[2*x    ];
-#if 0
-    {
-    int A1,A2,A3,A4;
-    A2= temp[1       ];
-    A4= temp[0       ];
-    A1= temp[0+width2];
-    A1 -= (A2 + A4)>>1;
-    A4 += (A1 + 1)>>1;
-    b[0+width2] = A1;
-    b[0       ] = A4;
-    for(x=1; x+1<width2; x+=2){
-        A3= temp[x+width2];
-        A4= temp[x+1     ];
-        A3 -= (A2 + A4)>>1;
-        A2 += (A1 + A3 + 2)>>2;
-        b[x+width2] = A3;
-        b[x       ] = A2;
-
-        A1= temp[x+1+width2];
-        A2= temp[x+2       ];
-        A1 -= (A2 + A4)>>1;
-        A4 += (A1 + A3 + 2)>>2;
-        b[x+1+width2] = A1;
-        b[x+1       ] = A4;
-    }
-    A3= temp[width-1];
-    A3 -= A2;
-    A2 += (A1 + A3 + 2)>>2;
-    b[width -1] = A3;
-    b[width2-1] = A2;
-    }
-#else
-    lift(b+w2, temp+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 0);
-    lift(b   , temp   , b+w2, 1, 1, 1, width,  1, 2, 2, 0, 0);
-#endif /* 0 */
-}
-
-static void vertical_decompose53iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] -= (b0[i] + b2[i])>>1;
-    }
-}
-
-static void vertical_decompose53iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] += (b0[i] + b2[i] + 2)>>2;
-    }
-}
-
-static void spatial_decompose53i(DWTELEM *buffer, int width, int height, int stride){
-    int y;
-    DWTELEM *b0= buffer + mirror(-2-1, height-1)*stride;
-    DWTELEM *b1= buffer + mirror(-2  , height-1)*stride;
-
-    for(y=-2; y<height; y+=2){
-        DWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
-        DWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
-
-        if(y+1<(unsigned)height) horizontal_decompose53i(b2, width);
-        if(y+2<(unsigned)height) horizontal_decompose53i(b3, width);
-
-        if(y+1<(unsigned)height) vertical_decompose53iH0(b1, b2, b3, width);
-        if(y+0<(unsigned)height) vertical_decompose53iL0(b0, b1, b2, width);
-
-        b0=b2;
-        b1=b3;
-    }
-}
-
-static void horizontal_decompose97i(DWTELEM *b, int width){
-    DWTELEM temp[width];
-    const int w2= (width+1)>>1;
-
-    lift (temp+w2, b    +1, b      , 1, 2, 2, width,  W_AM, W_AO, W_AS, 1, 1);
-    liftS(temp   , b      , temp+w2, 1, 2, 1, width,  W_BM, W_BO, W_BS, 0, 0);
-    lift (b   +w2, temp+w2, temp   , 1, 1, 1, width,  W_CM, W_CO, W_CS, 1, 0);
-    lift (b      , temp   , b   +w2, 1, 1, 1, width,  W_DM, W_DO, W_DS, 0, 0);
-}
-
-
-static void vertical_decompose97iH0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] -= (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
-    }
-}
-
-static void vertical_decompose97iH1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] += (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
-    }
-}
-
-static void vertical_decompose97iL0(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-#ifdef liftS
-        b1[i] -= (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
-#else
-        b1[i] = (16*4*b1[i] - 4*(b0[i] + b2[i]) + W_BO*5 + (5<<27)) / (5*16) - (1<<23);
-#endif
-    }
-}
-
-static void vertical_decompose97iL1(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] += (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
-    }
-}
-
-static void spatial_decompose97i(DWTELEM *buffer, int width, int height, int stride){
-    int y;
-    DWTELEM *b0= buffer + mirror(-4-1, height-1)*stride;
-    DWTELEM *b1= buffer + mirror(-4  , height-1)*stride;
-    DWTELEM *b2= buffer + mirror(-4+1, height-1)*stride;
-    DWTELEM *b3= buffer + mirror(-4+2, height-1)*stride;
-
-    for(y=-4; y<height; y+=2){
-        DWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
-        DWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
-
-        if(y+3<(unsigned)height) horizontal_decompose97i(b4, width);
-        if(y+4<(unsigned)height) horizontal_decompose97i(b5, width);
-
-        if(y+3<(unsigned)height) vertical_decompose97iH0(b3, b4, b5, width);
-        if(y+2<(unsigned)height) vertical_decompose97iL0(b2, b3, b4, width);
-        if(y+1<(unsigned)height) vertical_decompose97iH1(b1, b2, b3, width);
-        if(y+0<(unsigned)height) vertical_decompose97iL1(b0, b1, b2, width);
-
-        b0=b2;
-        b1=b3;
-        b2=b4;
-        b3=b5;
-    }
-}
-
-void ff_spatial_dwt(DWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
-    int level;
-
-    for(level=0; level<decomposition_count; level++){
-        switch(type){
-        case DWT_97: spatial_decompose97i(buffer, width>>level, height>>level, stride<<level); break;
-        case DWT_53: spatial_decompose53i(buffer, width>>level, height>>level, stride<<level); break;
-        }
-    }
-}
-
-static void horizontal_compose53i(IDWTELEM *b, int width){
-    IDWTELEM temp[width];
-    const int width2= width>>1;
-    const int w2= (width+1)>>1;
-    int x;
-
-    for(x=0; x<width2; x++){
-        temp[2*x    ]= b[x   ];
-        temp[2*x + 1]= b[x+w2];
-    }
-    if(width&1)
-        temp[2*x    ]= b[x   ];
-
-    b[0] = temp[0] - ((temp[1]+1)>>1);
-    for(x=2; x<width-1; x+=2){
-        b[x  ] = temp[x  ] - ((temp[x-1] + temp[x+1]+2)>>2);
-        b[x-1] = temp[x-1] + ((b   [x-2] + b   [x  ]+1)>>1);
-    }
-    if(width&1){
-        b[x  ] = temp[x  ] - ((temp[x-1]+1)>>1);
-        b[x-1] = temp[x-1] + ((b   [x-2] + b  [x  ]+1)>>1);
-    }else
-        b[x-1] = temp[x-1] + b[x-2];
-}
-
-static void vertical_compose53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] += (b0[i] + b2[i])>>1;
-    }
-}
-
-static void vertical_compose53iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] -= (b0[i] + b2[i] + 2)>>2;
-    }
-}
-
-static void spatial_compose53i_buffered_init(DWTCompose *cs, slice_buffer * sb, int height, int stride_line){
-    cs->b0 = slice_buffer_get_line(sb, mirror(-1-1, height-1) * stride_line);
-    cs->b1 = slice_buffer_get_line(sb, mirror(-1  , height-1) * stride_line);
-    cs->y = -1;
-}
-
-static void spatial_compose53i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride){
-    cs->b0 = buffer + mirror(-1-1, height-1)*stride;
-    cs->b1 = buffer + mirror(-1  , height-1)*stride;
-    cs->y = -1;
-}
-
-static void spatial_compose53i_dy_buffered(DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line){
-    int y= cs->y;
-
-    IDWTELEM *b0= cs->b0;
-    IDWTELEM *b1= cs->b1;
-    IDWTELEM *b2= slice_buffer_get_line(sb, mirror(y+1, height-1) * stride_line);
-    IDWTELEM *b3= slice_buffer_get_line(sb, mirror(y+2, height-1) * stride_line);
-
-    if(y+1<(unsigned)height && y<(unsigned)height){
-        int x;
-
-        for(x=0; x<width; x++){
-            b2[x] -= (b1[x] + b3[x] + 2)>>2;
-            b1[x] += (b0[x] + b2[x])>>1;
-        }
-    }else{
-        if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
-        if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
-    }
-
-        if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
-        if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
-
-    cs->b0 = b2;
-    cs->b1 = b3;
-    cs->y += 2;
-}
-
-static void spatial_compose53i_dy(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride){
-    int y= cs->y;
-    IDWTELEM *b0= cs->b0;
-    IDWTELEM *b1= cs->b1;
-    IDWTELEM *b2= buffer + mirror(y+1, height-1)*stride;
-    IDWTELEM *b3= buffer + mirror(y+2, height-1)*stride;
-
-        if(y+1<(unsigned)height) vertical_compose53iL0(b1, b2, b3, width);
-        if(y+0<(unsigned)height) vertical_compose53iH0(b0, b1, b2, width);
-
-        if(y-1<(unsigned)height) horizontal_compose53i(b0, width);
-        if(y+0<(unsigned)height) horizontal_compose53i(b1, width);
-
-    cs->b0 = b2;
-    cs->b1 = b3;
-    cs->y += 2;
-}
-
-static void av_unused spatial_compose53i(IDWTELEM *buffer, int width, int height, int stride){
-    DWTCompose cs;
-    spatial_compose53i_init(&cs, buffer, height, stride);
-    while(cs.y <= height)
-        spatial_compose53i_dy(&cs, buffer, width, height, stride);
-}
-
-
-void ff_snow_horizontal_compose97i(IDWTELEM *b, int width){
-    IDWTELEM temp[width];
-    const int w2= (width+1)>>1;
-
-#if 0 //maybe more understadable but slower
-    inv_lift (temp   , b      , b   +w2, 2, 1, 1, width,  W_DM, W_DO, W_DS, 0, 1);
-    inv_lift (temp+1 , b   +w2, temp   , 2, 1, 2, width,  W_CM, W_CO, W_CS, 1, 1);
-
-    inv_liftS(b      , temp   , temp+1 , 2, 2, 2, width,  W_BM, W_BO, W_BS, 0, 1);
-    inv_lift (b+1    , temp+1 , b      , 2, 2, 2, width,  W_AM, W_AO, W_AS, 1, 0);
-#else
-    int x;
-    temp[0] = b[0] - ((3*b[w2]+2)>>2);
-    for(x=1; x<(width>>1); x++){
-        temp[2*x  ] = b[x     ] - ((3*(b   [x+w2-1] + b[x+w2])+4)>>3);
-        temp[2*x-1] = b[x+w2-1] - temp[2*x-2] - temp[2*x];
-    }
-    if(width&1){
-        temp[2*x  ] = b[x     ] - ((3*b   [x+w2-1]+2)>>2);
-        temp[2*x-1] = b[x+w2-1] - temp[2*x-2] - temp[2*x];
-    }else
-        temp[2*x-1] = b[x+w2-1] - 2*temp[2*x-2];
-
-    b[0] = temp[0] + ((2*temp[0] + temp[1]+4)>>3);
-    for(x=2; x<width-1; x+=2){
-        b[x  ] = temp[x  ] + ((4*temp[x  ] + temp[x-1] + temp[x+1]+8)>>4);
-        b[x-1] = temp[x-1] + ((3*(b  [x-2] + b   [x  ] ))>>1);
-    }
-    if(width&1){
-        b[x  ] = temp[x  ] + ((2*temp[x  ] + temp[x-1]+4)>>3);
-        b[x-1] = temp[x-1] + ((3*(b  [x-2] + b   [x  ] ))>>1);
-    }else
-        b[x-1] = temp[x-1] + 3*b [x-2];
-#endif
-}
-
-static void vertical_compose97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
-    }
-}
-
-static void vertical_compose97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] -= (W_CM*(b0[i] + b2[i])+W_CO)>>W_CS;
-    }
-}
-
-static void vertical_compose97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-#ifdef liftS
-        b1[i] += (W_BM*(b0[i] + b2[i])+W_BO)>>W_BS;
-#else
-        b1[i] += (W_BM*(b0[i] + b2[i])+4*b1[i]+W_BO)>>W_BS;
-#endif
-    }
-}
-
-static void vertical_compose97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b1[i] -= (W_DM*(b0[i] + b2[i])+W_DO)>>W_DS;
-    }
-}
-
-void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
-    int i;
-
-    for(i=0; i<width; i++){
-        b4[i] -= (W_DM*(b3[i] + b5[i])+W_DO)>>W_DS;
-        b3[i] -= (W_CM*(b2[i] + b4[i])+W_CO)>>W_CS;
-#ifdef liftS
-        b2[i] += (W_BM*(b1[i] + b3[i])+W_BO)>>W_BS;
-#else
-        b2[i] += (W_BM*(b1[i] + b3[i])+4*b2[i]+W_BO)>>W_BS;
-#endif
-        b1[i] += (W_AM*(b0[i] + b2[i])+W_AO)>>W_AS;
-    }
-}
-
-static void spatial_compose97i_buffered_init(DWTCompose *cs, slice_buffer * sb, int height, int stride_line){
-    cs->b0 = slice_buffer_get_line(sb, mirror(-3-1, height-1) * stride_line);
-    cs->b1 = slice_buffer_get_line(sb, mirror(-3  , height-1) * stride_line);
-    cs->b2 = slice_buffer_get_line(sb, mirror(-3+1, height-1) * stride_line);
-    cs->b3 = slice_buffer_get_line(sb, mirror(-3+2, height-1) * stride_line);
-    cs->y = -3;
-}
-
-static void spatial_compose97i_init(DWTCompose *cs, IDWTELEM *buffer, int height, int stride){
-    cs->b0 = buffer + mirror(-3-1, height-1)*stride;
-    cs->b1 = buffer + mirror(-3  , height-1)*stride;
-    cs->b2 = buffer + mirror(-3+1, height-1)*stride;
-    cs->b3 = buffer + mirror(-3+2, height-1)*stride;
-    cs->y = -3;
-}
-
-static void spatial_compose97i_dy_buffered(DSPContext *dsp, DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line){
-    int y = cs->y;
-
-    IDWTELEM *b0= cs->b0;
-    IDWTELEM *b1= cs->b1;
-    IDWTELEM *b2= cs->b2;
-    IDWTELEM *b3= cs->b3;
-    IDWTELEM *b4= slice_buffer_get_line(sb, mirror(y + 3, height - 1) * stride_line);
-    IDWTELEM *b5= slice_buffer_get_line(sb, mirror(y + 4, height - 1) * stride_line);
-
-    if(y>0 && y+4<height){
-        dsp->vertical_compose97i(b0, b1, b2, b3, b4, b5, width);
-    }else{
-        if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
-        if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
-        if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
-        if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
-    }
-
-    if(y-1<(unsigned)height) dsp->horizontal_compose97i(b0, width);
-    if(y+0<(unsigned)height) dsp->horizontal_compose97i(b1, width);
-
-    cs->b0=b2;
-    cs->b1=b3;
-    cs->b2=b4;
-    cs->b3=b5;
-    cs->y += 2;
-}
-
-static void spatial_compose97i_dy(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride){
-    int y = cs->y;
-    IDWTELEM *b0= cs->b0;
-    IDWTELEM *b1= cs->b1;
-    IDWTELEM *b2= cs->b2;
-    IDWTELEM *b3= cs->b3;
-    IDWTELEM *b4= buffer + mirror(y+3, height-1)*stride;
-    IDWTELEM *b5= buffer + mirror(y+4, height-1)*stride;
-
-    if(y+3<(unsigned)height) vertical_compose97iL1(b3, b4, b5, width);
-    if(y+2<(unsigned)height) vertical_compose97iH1(b2, b3, b4, width);
-    if(y+1<(unsigned)height) vertical_compose97iL0(b1, b2, b3, width);
-    if(y+0<(unsigned)height) vertical_compose97iH0(b0, b1, b2, width);
-
-    if(y-1<(unsigned)height) ff_snow_horizontal_compose97i(b0, width);
-    if(y+0<(unsigned)height) ff_snow_horizontal_compose97i(b1, width);
-
-    cs->b0=b2;
-    cs->b1=b3;
-    cs->b2=b4;
-    cs->b3=b5;
-    cs->y += 2;
-}
-
-static void av_unused spatial_compose97i(IDWTELEM *buffer, int width, int height, int stride){
-    DWTCompose cs;
-    spatial_compose97i_init(&cs, buffer, height, stride);
-    while(cs.y <= height)
-        spatial_compose97i_dy(&cs, buffer, width, height, stride);
-}
-
-static void ff_spatial_idwt_buffered_init(DWTCompose *cs, slice_buffer * sb, int width, int height, int stride_line, int type, int decomposition_count){
-    int level;
-    for(level=decomposition_count-1; level>=0; level--){
-        switch(type){
-        case DWT_97: spatial_compose97i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
-        case DWT_53: spatial_compose53i_buffered_init(cs+level, sb, height>>level, stride_line<<level); break;
-        }
-    }
-}
-
-static void ff_spatial_idwt_buffered_slice(DSPContext *dsp, DWTCompose *cs, slice_buffer * slice_buf, int width, int height, int stride_line, int type, int decomposition_count, int y){
-    const int support = type==1 ? 3 : 5;
-    int level;
-    if(type==2) return;
-
-    for(level=decomposition_count-1; level>=0; level--){
-        while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
-            switch(type){
-            case DWT_97: spatial_compose97i_dy_buffered(dsp, cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
-                break;
-            case DWT_53: spatial_compose53i_dy_buffered(cs+level, slice_buf, width>>level, height>>level, stride_line<<level);
-                break;
-            }
-        }
-    }
-}
-
 static inline void unpack_coeffs(SnowContext *s, SubBand *b, SubBand * parent, int orientation){
     const int w= b->width;
     const int h= b->height;
@@ -2043,7 +1359,7 @@ static av_always_inline void add_yblock(SnowContext *s, int sliced, slice_buffer
     }
 #else
     if(sliced){
-        s->dsp.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
+        s->dwt.inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
     }else{
         for(y=0; y<b_h; y++){
             //FIXME ugly misuse of obmc_stride
@@ -2366,6 +1682,7 @@ static av_cold int common_init(AVCodecContext *avctx){
     s->max_ref_frames=1; //just make sure its not an invalid value in case of no initial keyframe
 
     dsputil_init(&s->dsp, avctx);
+    ff_dwt_init(&s->dwt);
 
 #define mcf(dx,dy)\
     s->dsp.put_qpel_pixels_tab       [0][dy+dx/4]=\
@@ -2865,7 +2182,7 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPac
             }
 
             for(; yd<slice_h; yd+=4){
-                ff_spatial_idwt_buffered_slice(&s->dsp, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd);
+                ff_spatial_idwt_buffered_slice(&s->dwt, cs, &s->sb, w, h, 1, s->spatial_decomposition_type, s->spatial_decomposition_count, yd);
             }
 
             if(s->qlog == LOSSLESS_QLOG){
@@ -3607,41 +2924,6 @@ static int get_4block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index){
     return distortion + rate*penalty_factor;
 }
 
-static void ff_spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
-    int level;
-    for(level=decomposition_count-1; level>=0; level--){
-        switch(type){
-        case DWT_97: spatial_compose97i_init(cs+level, buffer, height>>level, stride<<level); break;
-        case DWT_53: spatial_compose53i_init(cs+level, buffer, height>>level, stride<<level); break;
-        }
-    }
-}
-
-static void ff_spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count, int y){
-    const int support = type==1 ? 3 : 5;
-    int level;
-    if(type==2) return;
-
-    for(level=decomposition_count-1; level>=0; level--){
-        while(cs[level].y <= FFMIN((y>>level)+support, height>>level)){
-            switch(type){
-            case DWT_97: spatial_compose97i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
-                break;
-            case DWT_53: spatial_compose53i_dy(cs+level, buffer, width>>level, height>>level, stride<<level);
-                break;
-            }
-        }
-    }
-}
-
-static void ff_spatial_idwt(IDWTELEM *buffer, int width, int height, int stride, int type, int decomposition_count){
-        DWTCompose cs[MAX_DECOMPOSITIONS];
-        int y;
-        ff_spatial_idwt_init(cs, buffer, width, height, stride, type, decomposition_count);
-        for(y=0; y<height; y+=4)
-            ff_spatial_idwt_slice(cs, buffer, width, height, stride, type, decomposition_count, y);
-}
-
 static int encode_subband_c0run(SnowContext *s, SubBand *b, IDWTELEM *src, IDWTELEM *parent, int stride, int orientation){
     const int w= b->width;
     const int h= b->height;
diff --git a/libavcodec/snow.h b/libavcodec/snow.h
index ee0d0730d0..7d847e4b37 100644
--- a/libavcodec/snow.h
+++ b/libavcodec/snow.h
@@ -23,10 +23,10 @@
 #define AVCODEC_SNOW_H
 
 #include "dsputil.h"
+#include "dwt.h"
 
 #define MID_STATE 128
 
-#define MAX_DECOMPOSITIONS 8
 #define MAX_PLANES 4
 #define QSHIFT 5
 #define QROOT (1<<QSHIFT)
@@ -37,101 +37,6 @@
 #define LOG2_OBMC_MAX 8
 #define OBMC_MAX (1<<(LOG2_OBMC_MAX))
 
-#define DWT_97 0
-#define DWT_53 1
-
-/** Used to minimize the amount of memory used in order to optimize cache performance. **/
-struct slice_buffer_s {
-    IDWTELEM * * line; ///< For use by idwt and predict_slices.
-    IDWTELEM * * data_stack; ///< Used for internal purposes.
-    int data_stack_top;
-    int line_count;
-    int line_width;
-    int data_count;
-    IDWTELEM * base_buffer; ///< Buffer that this structure is caching.
-};
-
-#define liftS lift
-#if 1
-#define W_AM 3
-#define W_AO 0
-#define W_AS 1
-
-#undef liftS
-#define W_BM 1
-#define W_BO 8
-#define W_BS 4
-
-#define W_CM 1
-#define W_CO 0
-#define W_CS 0
-
-#define W_DM 3
-#define W_DO 4
-#define W_DS 3
-#elif 0
-#define W_AM 55
-#define W_AO 16
-#define W_AS 5
-
-#define W_BM 3
-#define W_BO 32
-#define W_BS 6
-
-#define W_CM 127
-#define W_CO 64
-#define W_CS 7
-
-#define W_DM 7
-#define W_DO 8
-#define W_DS 4
-#elif 0
-#define W_AM 97
-#define W_AO 32
-#define W_AS 6
-
-#define W_BM 63
-#define W_BO 512
-#define W_BS 10
-
-#define W_CM 13
-#define W_CO 8
-#define W_CS 4
-
-#define W_DM 15
-#define W_DO 16
-#define W_DS 5
-
-#else
-
-#define W_AM 203
-#define W_AO 64
-#define W_AS 7
-
-#define W_BM 217
-#define W_BO 2048
-#define W_BS 12
-
-#define W_CM 113
-#define W_CO 64
-#define W_CS 7
-
-#define W_DM 227
-#define W_DO 128
-#define W_DS 9
-#endif
-
-void ff_snow_vertical_compose97i(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
-void ff_snow_horizontal_compose97i(IDWTELEM *b, int width);
-void ff_snow_inner_add_yblock(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
-
-#if CONFIG_SNOW_ENCODER
-int w53_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h);
-int w97_32_c(void *v, uint8_t * pix1, uint8_t * pix2, int line_size, int h);
-#endif
-
-void ff_spatial_dwt(int *buffer, int width, int height, int stride, int type, int decomposition_count);
-
 /* C bits used by mmx/sse2/altivec */
 
 static av_always_inline void snow_interleave_line_header(int * i, int width, IDWTELEM * low, IDWTELEM * high){
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index 3b4b59319f..216cb4318d 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -2894,25 +2894,6 @@ void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx)
         }
 #endif
 
-#if CONFIG_SNOW_DECODER
-        if(mm_flags & FF_MM_SSE2 & 0){
-            c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
-#if HAVE_7REGS
-            c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
-#endif
-            c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
-        }
-        else{
-            if(mm_flags & FF_MM_MMX2){
-            c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
-#if HAVE_7REGS
-            c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
-#endif
-            }
-            c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
-        }
-#endif
-
         if(mm_flags & FF_MM_3DNOW){
             c->vorbis_inverse_coupling = vorbis_inverse_coupling_3dnow;
             c->vector_fmul = vector_fmul_3dnow;
diff --git a/libavcodec/x86/dsputil_mmx.h b/libavcodec/x86/dsputil_mmx.h
index 42d9c12c48..7d1bf7fae6 100644
--- a/libavcodec/x86/dsputil_mmx.h
+++ b/libavcodec/x86/dsputil_mmx.h
@@ -167,15 +167,6 @@ void ff_vc1dsp_init_mmx(DSPContext* dsp, AVCodecContext *avctx);
 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd);
 void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd);
 
-void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width);
-void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width);
-void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
-void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
-void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
-                                   int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
-void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
-                                  int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
-
 void ff_lpc_compute_autocorr_sse2(const int32_t *data, int len, int lag,
                                    double *autoc);
 
diff --git a/libavcodec/x86/snowdsp_mmx.c b/libavcodec/x86/snowdsp_mmx.c
index 1e1d2e22d8..263f0bbf69 100644
--- a/libavcodec/x86/snowdsp_mmx.c
+++ b/libavcodec/x86/snowdsp_mmx.c
@@ -22,9 +22,10 @@
 #include "libavutil/x86_cpu.h"
 #include "libavcodec/avcodec.h"
 #include "libavcodec/snow.h"
+#include "libavcodec/dwt.h"
 #include "dsputil_mmx.h"
 
-void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width){
+static void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width){
     const int w2= (width+1)>>1;
     DECLARE_ALIGNED(16, IDWTELEM, temp)[width>>1];
     const int w_l= (width>>1);
@@ -213,7 +214,7 @@ void ff_snow_horizontal_compose97i_sse2(IDWTELEM *b, int width){
     }
 }
 
-void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width){
+static void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width){
     const int w2= (width+1)>>1;
     IDWTELEM temp[width >> 1];
     const int w_l= (width>>1);
@@ -436,7 +437,7 @@ void ff_snow_horizontal_compose97i_mmx(IDWTELEM *b, int width){
         "movdqa %%"s2", %%"t2" \n\t"\
         "movdqa %%"s3", %%"t3" \n\t"
 
-void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
+static void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
     x86_reg i = width;
 
     while(i & 0x1F)
@@ -534,7 +535,7 @@ void ff_snow_vertical_compose97i_sse2(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
         "movq %%"s3", %%"t3" \n\t"
 
 
-void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
+static void ff_snow_vertical_compose97i_mmx(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width){
     x86_reg i = width;
     while(i & 15)
     {
@@ -847,7 +848,7 @@ snow_inner_add_yblock_mmx_mix("16", "8")
 snow_inner_add_yblock_mmx_end("32")
 }
 
-void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
+static void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
                            int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
 
     if (b_w == 16)
@@ -861,7 +862,7 @@ void ff_snow_inner_add_yblock_sse2(const uint8_t *obmc, const int obmc_stride, u
          ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
 }
 
-void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
+static void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h,
                           int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8){
     if (b_w == 16)
         inner_add_yblock_bw_16_obmc_32_mmx(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
@@ -870,3 +871,27 @@ void ff_snow_inner_add_yblock_mmx(const uint8_t *obmc, const int obmc_stride, ui
     else
         ff_snow_inner_add_yblock(obmc, obmc_stride, block, b_w, b_h, src_x,src_y, src_stride, sb, add, dst8);
 }
+
+void ff_dwt_init_x86(DWTContext *c)
+{
+    mm_flags = mm_support();
+
+    if (mm_flags & FF_MM_MMX) {
+        if(mm_flags & FF_MM_SSE2 & 0){
+            c->horizontal_compose97i = ff_snow_horizontal_compose97i_sse2;
+#if HAVE_7REGS
+            c->vertical_compose97i = ff_snow_vertical_compose97i_sse2;
+#endif
+            c->inner_add_yblock = ff_snow_inner_add_yblock_sse2;
+        }
+        else{
+            if(mm_flags & FF_MM_MMX2){
+            c->horizontal_compose97i = ff_snow_horizontal_compose97i_mmx;
+#if HAVE_7REGS
+            c->vertical_compose97i = ff_snow_vertical_compose97i_mmx;
+#endif
+            }
+            c->inner_add_yblock = ff_snow_inner_add_yblock_mmx;
+        }
+    }
+}