Replace ASMALIGN() with .p2align
[ffmpeg.git] / libavcodec / snow.c
index b7d205a..18039d2 100644 (file)
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+#include "libavutil/intmath.h"
 #include "avcodec.h"
 #include "dsputil.h"
+#include "dwt.h"
 #include "snow.h"
 
 #include "rangecoder.h"
 #include "mathops.h"
 
 #include "mpegvideo.h"
+#include "h263.h"
 
 #undef NDEBUG
 #include <assert.h>
@@ -438,6 +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;
@@ -492,85 +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
@@ -578,14 +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;
 
@@ -707,688 +624,102 @@ 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;
-
-#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
-    inv_lift(temp   , b   , b+w2, 1, 1, 1, width,  1, 2, 2, 0, 1);
-    inv_lift(temp+w2, b+w2, temp, 1, 1, 1, width, -1, 0, 1, 1, 1);
-#endif /* 0 */
-    for(x=0; x<width2; x++){
-        b[2*x    ]= temp[x   ];
-        b[2*x + 1]= temp[x+w2];
-    }
-    if(width&1)
-        b[2*x    ]= temp[x   ];
-}
-
-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) 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;
-
-    inv_lift (temp   , b      , b   +w2, 1, 1, 1, width,  W_DM, W_DO, W_DS, 0, 1);
-    inv_lift (temp+w2, b   +w2, temp   , 1, 1, 1, width,  W_CM, W_CO, W_CS, 1, 1);
-    inv_liftS(b      , temp   , temp+w2, 2, 1, 1, width,  W_BM, W_BO, W_BS, 0, 1);
-    inv_lift (b+1    , temp+w2, b      , 2, 1, 2, width,  W_AM, W_AO, W_AS, 1, 0);
-}
-
-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;
     int x,y;
 
-        int run, runs;
-        x_and_coeff *xc= b->x_coeff;
-        x_and_coeff *prev_xc= NULL;
-        x_and_coeff *prev2_xc= xc;
-        x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
-        x_and_coeff *prev_parent_xc= parent_xc;
+    int run, runs;
+    x_and_coeff *xc= b->x_coeff;
+    x_and_coeff *prev_xc= NULL;
+    x_and_coeff *prev2_xc= xc;
+    x_and_coeff *parent_xc= parent ? parent->x_coeff : NULL;
+    x_and_coeff *prev_parent_xc= parent_xc;
 
-        runs= get_symbol2(&s->c, b->state[30], 0);
-        if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
-        else           run= INT_MAX;
+    runs= get_symbol2(&s->c, b->state[30], 0);
+    if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
+    else           run= INT_MAX;
 
-        for(y=0; y<h; y++){
-            int v=0;
-            int lt=0, t=0, rt=0;
+    for(y=0; y<h; y++){
+        int v=0;
+        int lt=0, t=0, rt=0;
 
-            if(y && prev_xc->x == 0){
-                rt= prev_xc->coeff;
-            }
-            for(x=0; x<w; x++){
-                int p=0;
-                const int l= v;
+        if(y && prev_xc->x == 0){
+            rt= prev_xc->coeff;
+        }
+        for(x=0; x<w; x++){
+            int p=0;
+            const int l= v;
 
-                lt= t; t= rt;
+            lt= t; t= rt;
 
-                if(y){
-                    if(prev_xc->x <= x)
-                        prev_xc++;
-                    if(prev_xc->x == x + 1)
-                        rt= prev_xc->coeff;
-                    else
-                        rt=0;
+            if(y){
+                if(prev_xc->x <= x)
+                    prev_xc++;
+                if(prev_xc->x == x + 1)
+                    rt= prev_xc->coeff;
+                else
+                    rt=0;
+            }
+            if(parent_xc){
+                if(x>>1 > parent_xc->x){
+                    parent_xc++;
                 }
-                if(parent_xc){
-                    if(x>>1 > parent_xc->x){
-                        parent_xc++;
-                    }
-                    if(x>>1 == parent_xc->x){
-                        p= parent_xc->coeff;
-                    }
+                if(x>>1 == parent_xc->x){
+                    p= parent_xc->coeff;
                 }
-                if(/*ll|*/l|lt|t|rt|p){
-                    int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
+            }
+            if(/*ll|*/l|lt|t|rt|p){
+                int context= av_log2(/*FFABS(ll) + */3*(l>>1) + (lt>>1) + (t&~1) + (rt>>1) + (p>>1));
 
-                    v=get_rac(&s->c, &b->state[0][context]);
-                    if(v){
-                        v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
-                        v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]);
+                v=get_rac(&s->c, &b->state[0][context]);
+                if(v){
+                    v= 2*(get_symbol2(&s->c, b->state[context + 2], context-4) + 1);
+                    v+=get_rac(&s->c, &b->state[0][16 + 1 + 3 + quant3bA[l&0xFF] + 3*quant3bA[t&0xFF]]);
 
-                        xc->x=x;
-                        (xc++)->coeff= v;
-                    }
+                    xc->x=x;
+                    (xc++)->coeff= v;
+                }
+            }else{
+                if(!run){
+                    if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
+                    else           run= INT_MAX;
+                    v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
+                    v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
+
+                    xc->x=x;
+                    (xc++)->coeff= v;
                 }else{
-                    if(!run){
-                        if(runs-- > 0) run= get_symbol2(&s->c, b->state[1], 3);
-                        else           run= INT_MAX;
-                        v= 2*(get_symbol2(&s->c, b->state[0 + 2], 0-4) + 1);
-                        v+=get_rac(&s->c, &b->state[0][16 + 1 + 3]);
-
-                        xc->x=x;
-                        (xc++)->coeff= v;
-                    }else{
-                        int max_run;
-                        run--;
-                        v=0;
-
-                        if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
-                        else  max_run= FFMIN(run, w-x-1);
-                        if(parent_xc)
-                            max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
-                        x+= max_run;
-                        run-= max_run;
-                    }
+                    int max_run;
+                    run--;
+                    v=0;
+
+                    if(y) max_run= FFMIN(run, prev_xc->x - x - 2);
+                    else  max_run= FFMIN(run, w-x-1);
+                    if(parent_xc)
+                        max_run= FFMIN(max_run, 2*parent_xc->x - x - 1);
+                    x+= max_run;
+                    run-= max_run;
                 }
             }
-            (xc++)->x= w+1; //end marker
-            prev_xc= prev2_xc;
-            prev2_xc= xc;
+        }
+        (xc++)->x= w+1; //end marker
+        prev_xc= prev2_xc;
+        prev2_xc= xc;
 
-            if(parent_xc){
-                if(y&1){
-                    while(parent_xc->x != parent->width+1)
-                        parent_xc++;
+        if(parent_xc){
+            if(y&1){
+                while(parent_xc->x != parent->width+1)
                     parent_xc++;
-                    prev_parent_xc= parent_xc;
-                }else{
-                    parent_xc= prev_parent_xc;
-                }
+                parent_xc++;
+                prev_parent_xc= parent_xc;
+            }else{
+                parent_xc= prev_parent_xc;
             }
         }
+    }
 
-        (xc++)->x= w+1; //end marker
+    (xc++)->x= w+1; //end marker
 }
 
 static inline void decode_subband_slice_buffered(SnowContext *s, SubBand *b, slice_buffer * sb, int start_y, int h, int save_state[1]){
@@ -1589,7 +920,7 @@ static void decode_blocks(SnowContext *s){
     }
 }
 
-static void mc_block(Plane *p, uint8_t *dst, const uint8_t *src, uint8_t *tmp, int stride, int b_w, int b_h, int dx, int dy){
+static void mc_block(Plane *p, uint8_t *dst, const uint8_t *src, int stride, int b_w, int b_h, int dx, int dy){
     static const uint8_t weight[64]={
     8,7,6,5,4,3,2,1,
     7,7,0,0,0,0,0,1,
@@ -1775,9 +1106,8 @@ static void mc_block(Plane *p, uint8_t *dst, const uint8_t *src, uint8_t *tmp, i
 
 #define mca(dx,dy,b_w)\
 static void mc_block_hpel ## dx ## dy ## b_w(uint8_t *dst, const uint8_t *src, int stride, int h){\
-    uint8_t tmp[stride*(b_w+HTAPS_MAX-1)];\
     assert(h==b_w);\
-    mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, tmp, stride, b_w, b_w, dx, dy);\
+    mc_block(NULL, dst, src-(HTAPS_MAX/2-1)-(HTAPS_MAX/2-1)*stride, stride, b_w, b_w, dx, dy);\
 }
 
 mca( 0, 0,16)
@@ -1849,7 +1179,7 @@ static void pred_block(SnowContext *s, uint8_t *dst, uint8_t *tmp, int stride, i
         assert(b_w>1 && b_h>1);
         assert((tab_index>=0 && tab_index<4) || b_w==32);
         if((dx&3) || (dy&3) || !(b_w == b_h || 2*b_w == b_h || b_w == 2*b_h) || (b_w&(b_w-1)) || !s->plane[plane_index].fast_mc )
-            mc_block(&s->plane[plane_index], dst, src, tmp, stride, b_w, b_h, dx, dy);
+            mc_block(&s->plane[plane_index], dst, src, stride, b_w, b_h, dx, dy);
         else if(b_w==32){
             int y;
             for(y=0; y<b_h; y+=16){
@@ -2028,7 +1358,7 @@ 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
@@ -2351,6 +1681,7 @@ 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]=\
@@ -2749,8 +2080,8 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPac
     common_init_after_header(avctx);
 
     // realloc slice buffer for the case that spatial_decomposition_count changed
-    slice_buffer_destroy(&s->sb);
-    slice_buffer_init(&s->sb, s->plane[0].height, (MB_SIZE >> s->block_max_depth) + s->spatial_decomposition_count * 8 + 1, s->plane[0].width, s->spatial_idwt_buffer);
+    ff_slice_buffer_destroy(&s->sb);
+    ff_slice_buffer_init(&s->sb, s->plane[0].height, (MB_SIZE >> s->block_max_depth) + s->spatial_decomposition_count * 8 + 1, s->plane[0].width, s->spatial_idwt_buffer);
 
     for(plane_index=0; plane_index<3; plane_index++){
         Plane *p= &s->plane[plane_index];
@@ -2850,7 +2181,7 @@ 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){
@@ -2867,10 +2198,10 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, AVPac
             y = FFMIN(p->height, slice_starty);
             end_y = FFMIN(p->height, slice_h);
             while(y < end_y)
-                slice_buffer_release(&s->sb, y++);
+                ff_slice_buffer_release(&s->sb, y++);
         }
 
-        slice_buffer_flush(&s->sb);
+        ff_slice_buffer_flush(&s->sb);
         }
 
     }
@@ -2896,7 +2227,7 @@ static av_cold int decode_end(AVCodecContext *avctx)
 {
     SnowContext *s = avctx->priv_data;
 
-    slice_buffer_destroy(&s->sb);
+    ff_slice_buffer_destroy(&s->sb);
 
     common_end(s);
 
@@ -2905,7 +2236,7 @@ static av_cold int decode_end(AVCodecContext *avctx)
 
 AVCodec snow_decoder = {
     "snow",
-    CODEC_TYPE_VIDEO,
+    AVMEDIA_TYPE_VIDEO,
     CODEC_ID_SNOW,
     sizeof(SnowContext),
     decode_init,
@@ -3497,9 +2828,9 @@ static int get_block_rd(SnowContext *s, int mb_x, int mb_y, int plane_index, con
          * to improve the score of the whole frame, thus iterative motion
          * estimation does not always converge. */
         if(s->avctx->me_cmp == FF_CMP_W97)
-            distortion = w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
+            distortion = ff_w97_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
         else if(s->avctx->me_cmp == FF_CMP_W53)
-            distortion = w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
+            distortion = ff_w53_32_c(&s->m, src + sx + sy*ref_stride, dst + sx + sy*ref_stride, ref_stride, 32);
         else{
             distortion = 0;
             for(i=0; i<4; i++){
@@ -3592,41 +2923,6 @@ 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;
@@ -4691,7 +3987,7 @@ static av_cold int encode_end(AVCodecContext *avctx)
 
 AVCodec snow_encoder = {
     "snow",
-    CODEC_TYPE_VIDEO,
+    AVMEDIA_TYPE_VIDEO,
     CODEC_ID_SNOW,
     sizeof(SnowContext),
     encode_init,