---
libavcodec/dirac_dwt.c | 530 ++-------------------------------------
libavcodec/dirac_dwt.h | 20 +-
libavcodec/diracdec.c | 8 +-
libavcodec/diracdsp.c | 25 +-
libavcodec/diracdsp.h | 4 +-
libavcodec/x86/dirac_dwt.c | 65 +++--
libavcodec/x86/diracdsp_mmx.c | 4 +-
libavcodec/x86/diracdsp_yasm.asm | 4 +-
8 files changed, 111 insertions(+), 549 deletions(-)
diff --git a/libavcodec/dirac_dwt.c b/libavcodec/dirac_dwt.c
index ee3665e..ae847f7 100644
--- a/libavcodec/dirac_dwt.c
+++ b/libavcodec/dirac_dwt.c
@@ -25,523 +25,40 @@
#include "dirac_dwt.h"
#include "libavcodec/x86/dirac_dwt.h"
+#define TEMPLATE_8bit
+#include "dirac_dwt_template.c"
+#undef TEMPLATE_8bit
-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 av_always_inline
-void interleave(IDWTELEM *dst, IDWTELEM *src0, IDWTELEM *src1, int w2, int
add, int shift)
-{
- int i;
- for (i = 0; i < w2; i++) {
- dst[2*i ] = (src0[i] + add) >> shift;
- dst[2*i+1] = (src1[i] + add) >> shift;
- }
-}
-
-static void horizontal_compose_dirac53i(IDWTELEM *b, IDWTELEM *temp, int w)
-{
- const int w2 = w >> 1;
- int x;
-
- temp[0] = COMPOSE_53iL0(b[w2], b[0], b[w2]);
- for (x = 1; x < w2; x++) {
- temp[x ] = COMPOSE_53iL0 (b[x+w2-1], b[x ], b[x+w2]);
- temp[x+w2-1] = COMPOSE_DIRAC53iH0(temp[x-1], b[x+w2-1], temp[x]);
- }
- temp[w-1] = COMPOSE_DIRAC53iH0(temp[w2-1], b[w-1], temp[w2-1]);
-
- interleave(b, temp, temp+w2, w2, 1, 1);
-}
-
-static void horizontal_compose_dd97i(IDWTELEM *b, IDWTELEM *tmp, int w)
-{
- const int w2 = w >> 1;
- int x;
-
- tmp[0] = COMPOSE_53iL0(b[w2], b[0], b[w2]);
- for (x = 1; x < w2; x++)
- tmp[x] = COMPOSE_53iL0(b[x+w2-1], b[x], b[x+w2]);
-
- // extend the edges
- tmp[-1] = tmp[0];
- tmp[w2+1] = tmp[w2] = tmp[w2-1];
-
- for (x = 0; x < w2; x++) {
- b[2*x ] = (tmp[x] + 1)>>1;
- b[2*x+1] = (COMPOSE_DD97iH0(tmp[x-1], tmp[x], b[x+w2], tmp[x+1],
tmp[x+2]) + 1)>>1;
- }
-}
-
-static void horizontal_compose_dd137i(IDWTELEM *b, IDWTELEM *tmp, int w)
-{
- const int w2 = w >> 1;
- int x;
-
- tmp[0] = COMPOSE_DD137iL0(b[w2], b[w2], b[0], b[w2 ], b[w2+1]);
- tmp[1] = COMPOSE_DD137iL0(b[w2], b[w2], b[1], b[w2+1], b[w2+2]);
- for (x = 2; x < w2-1; x++)
- tmp[x] = COMPOSE_DD137iL0(b[x+w2-2], b[x+w2-1], b[x], b[x+w2],
b[x+w2+1]);
- tmp[w2-1] = COMPOSE_DD137iL0(b[w-3], b[w-2], b[w2-1], b[w-1], b[w-1]);
-
- // extend the edges
- tmp[-1] = tmp[0];
- tmp[w2+1] = tmp[w2] = tmp[w2-1];
-
- for (x = 0; x < w2; x++) {
- b[2*x ] = (tmp[x] + 1)>>1;
- b[2*x+1] = (COMPOSE_DD97iH0(tmp[x-1], tmp[x], b[x+w2], tmp[x+1],
tmp[x+2]) + 1)>>1;
- }
-}
-
-static av_always_inline
-void horizontal_compose_haari(IDWTELEM *b, IDWTELEM *temp, int w, int shift)
-{
- const int w2 = w >> 1;
- int x;
-
- for (x = 0; x < w2; x++) {
- temp[x ] = COMPOSE_HAARiL0(b[x ], b[x+w2]);
- temp[x+w2] = COMPOSE_HAARiH0(b[x+w2], temp[x]);
- }
-
- interleave(b, temp, temp+w2, w2, shift, shift);
-}
-
-static void horizontal_compose_haar0i(IDWTELEM *b, IDWTELEM *temp, int w)
-{
- horizontal_compose_haari(b, temp, w, 0);
-}
-
-static void horizontal_compose_haar1i(IDWTELEM *b, IDWTELEM *temp, int w)
-{
- horizontal_compose_haari(b, temp, w, 1);
-}
-
-static void horizontal_compose_fidelityi(IDWTELEM *b, IDWTELEM *tmp, int w)
-{
- const int w2 = w >> 1;
- int i, x;
- IDWTELEM v[8];
-
- for (x = 0; x < w2; x++) {
- for (i = 0; i < 8; i++)
- v[i] = b[av_clip(x-3+i, 0, w2-1)];
- tmp[x] = COMPOSE_FIDELITYiH0(v[0], v[1], v[2], v[3], b[x+w2], v[4],
v[5], v[6], v[7]);
- }
-
- for (x = 0; x < w2; x++) {
- for (i = 0; i < 8; i++)
- v[i] = tmp[av_clip(x-4+i, 0, w2-1)];
- tmp[x+w2] = COMPOSE_FIDELITYiL0(v[0], v[1], v[2], v[3], b[x], v[4],
v[5], v[6], v[7]);
- }
-
- interleave(b, tmp+w2, tmp, w2, 0, 0);
-}
-
-static void horizontal_compose_daub97i(IDWTELEM *b, IDWTELEM *temp, int w)
-{
- const int w2 = w >> 1;
- int x, b0, b1, b2;
-
- temp[0] = COMPOSE_DAUB97iL1(b[w2], b[0], b[w2]);
- for (x = 1; x < w2; x++) {
- temp[x ] = COMPOSE_DAUB97iL1(b[x+w2-1], b[x ], b[x+w2]);
- temp[x+w2-1] = COMPOSE_DAUB97iH1(temp[x-1], b[x+w2-1], temp[x]);
- }
- temp[w-1] = COMPOSE_DAUB97iH1(temp[w2-1], b[w-1], temp[w2-1]);
-
- // second stage combined with interleave and shift
- b0 = b2 = COMPOSE_DAUB97iL0(temp[w2], temp[0], temp[w2]);
- b[0] = (b0 + 1) >> 1;
- for (x = 1; x < w2; x++) {
- b2 = COMPOSE_DAUB97iL0(temp[x+w2-1], temp[x ], temp[x+w2]);
- b1 = COMPOSE_DAUB97iH0( b0, temp[x+w2-1], b2 );
- b[2*x-1] = (b1 + 1) >> 1;
- b[2*x ] = (b2 + 1) >> 1;
- b0 = b2;
- }
- b[w-1] = (COMPOSE_DAUB97iH0(b2, temp[w-1], b2) + 1) >> 1;
-}
-
-static void vertical_compose_dirac53iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, int width)
-{
- int i;
-
- for(i=0; i<width; i++){
- b1[i] = COMPOSE_DIRAC53iH0(b0[i], b1[i], b2[i]);
- }
-}
-
-static void vertical_compose_dd97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
- IDWTELEM *b3, IDWTELEM *b4, int width)
-{
- int i;
-
- for(i=0; i<width; i++){
- b2[i] = COMPOSE_DD97iH0(b0[i], b1[i], b2[i], b3[i], b4[i]);
- }
-}
-
-static void vertical_compose_dd137iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
- IDWTELEM *b3, IDWTELEM *b4, int width)
-{
- int i;
-
- for(i=0; i<width; i++){
- b2[i] = COMPOSE_DD137iL0(b0[i], b1[i], b2[i], b3[i], b4[i]);
- }
-}
-
-static void vertical_compose_haar(IDWTELEM *b0, IDWTELEM *b1, int width)
-{
- int i;
-
- for (i = 0; i < width; i++) {
- b0[i] = COMPOSE_HAARiL0(b0[i], b1[i]);
- b1[i] = COMPOSE_HAARiH0(b1[i], b0[i]);
- }
-}
-
-static void vertical_compose_fidelityiH0(IDWTELEM *dst, IDWTELEM *b[8], int
width)
-{
- int i;
-
- for(i=0; i<width; i++){
- dst[i] = COMPOSE_FIDELITYiH0(b[0][i], b[1][i], b[2][i], b[3][i],
dst[i], b[4][i], b[5][i], b[6][i], b[7][i]);
- }
-}
-
-static void vertical_compose_fidelityiL0(IDWTELEM *dst, IDWTELEM *b[8], int
width)
-{
- int i;
-
- for(i=0; i<width; i++){
- dst[i] = COMPOSE_FIDELITYiL0(b[0][i], b[1][i], b[2][i], b[3][i],
dst[i], b[4][i], b[5][i], b[6][i], b[7][i]);
- }
-}
-
-static void vertical_compose_daub97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, int width)
-{
- int i;
-
- for(i=0; i<width; i++){
- b1[i] = COMPOSE_DAUB97iH0(b0[i], b1[i], b2[i]);
- }
-}
-
-static void vertical_compose_daub97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, int width)
-{
- int i;
-
- for(i=0; i<width; i++){
- b1[i] = COMPOSE_DAUB97iH1(b0[i], b1[i], b2[i]);
- }
-}
-
-static void vertical_compose_daub97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, int width)
-{
- int i;
-
- for(i=0; i<width; i++){
- b1[i] = COMPOSE_DAUB97iL0(b0[i], b1[i], b2[i]);
- }
-}
-
-static void vertical_compose_daub97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, int width)
-{
- int i;
-
- for(i=0; i<width; i++){
- b1[i] = COMPOSE_DAUB97iL1(b0[i], b1[i], b2[i]);
- }
-}
-
-
-static void spatial_compose_dd97i_dy(DWTContext *d, int level, int width, int
height, int stride)
-{
- vertical_compose_3tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
- vertical_compose_5tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
- DWTCompose *cs = d->cs + level;
-
- int i, y = cs->y;
- IDWTELEM *b[8];
- for (i = 0; i < 6; i++)
- b[i] = cs->b[i];
- b[6] = d->buffer + av_clip(y+5, 0, height-2)*stride;
- b[7] = d->buffer + av_clip(y+6, 1, height-1)*stride;
-
- if(y+5<(unsigned)height) vertical_compose_l0( b[5], b[6], b[7],
width);
- if(y+1<(unsigned)height) vertical_compose_h0(b[0], b[2], b[3], b[4],
b[6], width);
-
- if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
- if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
-
- for (i = 0; i < 6; i++)
- cs->b[i] = b[i+2];
- cs->y += 2;
-}
-
-static void spatial_compose_dirac53i_dy(DWTContext *d, int level, int width,
int height, int stride)
-{
- vertical_compose_3tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
- vertical_compose_3tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
- DWTCompose *cs = d->cs + level;
-
- int y= cs->y;
- IDWTELEM *b[4] = { cs->b[0], cs->b[1] };
- b[2] = d->buffer + avpriv_mirror(y+1, height-1)*stride;
- b[3] = d->buffer + avpriv_mirror(y+2, height-1)*stride;
-
- if(y+1<(unsigned)height) vertical_compose_l0(b[1], b[2], b[3], width);
- if(y+0<(unsigned)height) vertical_compose_h0(b[0], b[1], b[2], width);
-
- if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
- if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
-
- cs->b[0] = b[2];
- cs->b[1] = b[3];
- cs->y += 2;
-}
-
-
-static void spatial_compose_dd137i_dy(DWTContext *d, int level, int width, int
height, int stride)
-{
- vertical_compose_5tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
- vertical_compose_5tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
- DWTCompose *cs = d->cs + level;
-
- int i, y = cs->y;
- IDWTELEM *b[10];
- for (i = 0; i < 8; i++)
- b[i] = cs->b[i];
- b[8] = d->buffer + av_clip(y+7, 0, height-2)*stride;
- b[9] = d->buffer + av_clip(y+8, 1, height-1)*stride;
-
- if(y+5<(unsigned)height) vertical_compose_l0(b[3], b[5], b[6], b[7],
b[9], width);
- if(y+1<(unsigned)height) vertical_compose_h0(b[0], b[2], b[3], b[4],
b[6], width);
+#define TEMPLATE_10bit
+#include "dirac_dwt_template.c"
+#undef TEMPLATE_10bit
- if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
- if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
+#define TEMPLATE_12bit
+#include "dirac_dwt_template.c"
+#undef TEMPLATE_12bit
- for (i = 0; i < 8; i++)
- cs->b[i] = b[i+2];
- cs->y += 2;
-}
-
-// haar makes the assumption that height is even (always true for dirac)
-static void spatial_compose_haari_dy(DWTContext *d, int level, int width, int
height, int stride)
-{
- vertical_compose_2tap vertical_compose = (void*)d->vertical_compose;
- int y = d->cs[level].y;
- IDWTELEM *b0 = d->buffer + (y-1)*stride;
- IDWTELEM *b1 = d->buffer + (y )*stride;
-
- vertical_compose(b0, b1, width);
- d->horizontal_compose(b0, d->temp, width);
- d->horizontal_compose(b1, d->temp, width);
-
- d->cs[level].y += 2;
-}
-
-// Don't do sliced idwt for fidelity; the 9 tap filter makes it a bit annoying
-// Fortunately, this filter isn't used in practice.
-static void spatial_compose_fidelity(DWTContext *d, int level, int width, int
height, int stride)
-{
- vertical_compose_9tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
- vertical_compose_9tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
- int i, y;
- IDWTELEM *b[8];
-
- for (y = 1; y < height; y += 2) {
- for (i = 0; i < 8; i++)
- b[i] = d->buffer + av_clip((y-7 + 2*i), 0, height-2)*stride;
- vertical_compose_h0(d->buffer + y*stride, b, width);
- }
-
- for (y = 0; y < height; y += 2) {
- for (i = 0; i < 8; i++)
- b[i] = d->buffer + av_clip((y-7 + 2*i), 1, height-1)*stride;
- vertical_compose_l0(d->buffer + y*stride, b, width);
- }
-
- for (y = 0; y < height; y++)
- d->horizontal_compose(d->buffer + y*stride, d->temp, width);
-
- d->cs[level].y = height+1;
-}
-
-static void spatial_compose_daub97i_dy(DWTContext *d, int level, int width,
int height, int stride)
-{
- vertical_compose_3tap vertical_compose_l0 = (void*)d->vertical_compose_l0;
- vertical_compose_3tap vertical_compose_h0 = (void*)d->vertical_compose_h0;
- vertical_compose_3tap vertical_compose_l1 = (void*)d->vertical_compose_l1;
- vertical_compose_3tap vertical_compose_h1 = (void*)d->vertical_compose_h1;
- DWTCompose *cs = d->cs + level;
-
- int i, y = cs->y;
- IDWTELEM *b[6];
- for (i = 0; i < 4; i++)
- b[i] = cs->b[i];
- b[4] = d->buffer + avpriv_mirror(y+3, height-1)*stride;
- b[5] = d->buffer + avpriv_mirror(y+4, height-1)*stride;
-
- if(y+3<(unsigned)height) vertical_compose_l1(b[3], b[4], b[5], width);
- if(y+2<(unsigned)height) vertical_compose_h1(b[2], b[3], b[4], width);
- if(y+1<(unsigned)height) vertical_compose_l0(b[1], b[2], b[3], width);
- if(y+0<(unsigned)height) vertical_compose_h0(b[0], b[1], b[2], width);
-
- if(y-1<(unsigned)height) d->horizontal_compose(b[0], d->temp, width);
- if(y+0<(unsigned)height) d->horizontal_compose(b[1], d->temp, width);
-
- for (i = 0; i < 4; i++)
- cs->b[i] = b[i+2];
- cs->y += 2;
-}
-
-
-static void spatial_compose97i_init2(DWTCompose *cs, IDWTELEM *buffer, int
height, int stride)
-{
- cs->b[0] = buffer + avpriv_mirror(-3-1, height-1)*stride;
- cs->b[1] = buffer + avpriv_mirror(-3 , height-1)*stride;
- cs->b[2] = buffer + avpriv_mirror(-3+1, height-1)*stride;
- cs->b[3] = buffer + avpriv_mirror(-3+2, height-1)*stride;
- cs->y = -3;
-}
-
-static void spatial_compose53i_init2(DWTCompose *cs, IDWTELEM *buffer, int
height, int stride)
-{
- cs->b[0] = buffer + avpriv_mirror(-1-1, height-1)*stride;
- cs->b[1] = buffer + avpriv_mirror(-1 , height-1)*stride;
- cs->y = -1;
-}
-
-static void spatial_compose_dd97i_init(DWTCompose *cs, IDWTELEM *buffer, int
height, int stride)
-{
- cs->b[0] = buffer + av_clip(-5-1, 0, height-2)*stride;
- cs->b[1] = buffer + av_clip(-5 , 1, height-1)*stride;
- cs->b[2] = buffer + av_clip(-5+1, 0, height-2)*stride;
- cs->b[3] = buffer + av_clip(-5+2, 1, height-1)*stride;
- cs->b[4] = buffer + av_clip(-5+3, 0, height-2)*stride;
- cs->b[5] = buffer + av_clip(-5+4, 1, height-1)*stride;
- cs->y = -5;
-}
-
-static void spatial_compose_dd137i_init(DWTCompose *cs, IDWTELEM *buffer, int
height, int stride)
-{
- cs->b[0] = buffer + av_clip(-5-1, 0, height-2)*stride;
- cs->b[1] = buffer + av_clip(-5 , 1, height-1)*stride;
- cs->b[2] = buffer + av_clip(-5+1, 0, height-2)*stride;
- cs->b[3] = buffer + av_clip(-5+2, 1, height-1)*stride;
- cs->b[4] = buffer + av_clip(-5+3, 0, height-2)*stride;
- cs->b[5] = buffer + av_clip(-5+4, 1, height-1)*stride;
- cs->b[6] = buffer + av_clip(-5+5, 0, height-2)*stride;
- cs->b[7] = buffer + av_clip(-5+6, 1, height-1)*stride;
- cs->y = -5;
-}
-
-int ff_spatial_idwt_init2(DWTContext *d, IDWTELEM *buffer, int width, int
height,
+int ff_spatial_idwt_init2(DWTContext *d, uint8_t *buffer, int width, int
height,
int stride, enum dwt_type type, int
decomposition_count,
- IDWTELEM *temp)
+ uint8_t *temp, int bit_depth)
{
- int level;
+ int ret = 0;
- d->buffer = buffer;
- d->width = width;
- d->height = height;
- d->stride = stride;
- d->decomposition_count = decomposition_count;
- d->temp = temp + 8;
+ if (bit_depth == 8)
+ ret = ff_spatial_idwt_init2_8bit(d, buffer, width, height, stride,
type, decomposition_count, temp);
+ else if (bit_depth == 10)
+ ret = ff_spatial_idwt_init2_10bit(d, buffer, width, height, stride,
type, decomposition_count, temp);
+ else if (bit_depth == 12)
+ ret = ff_spatial_idwt_init2_12bit(d, buffer, width, height, stride,
type, decomposition_count, temp);
+ else
+ av_log(NULL, AV_LOG_WARNING, "Unsupported bit depth = %i\n",
bit_depth);
- for(level=decomposition_count-1; level>=0; level--){
- int hl = height >> level;
- int stride_l = stride << level;
-
- switch(type){
- case DWT_DIRAC_DD9_7:
- spatial_compose_dd97i_init(d->cs+level, buffer, hl, stride_l);
- break;
- case DWT_DIRAC_LEGALL5_3:
- spatial_compose53i_init2(d->cs+level, buffer, hl, stride_l);
- break;
- case DWT_DIRAC_DD13_7:
- spatial_compose_dd137i_init(d->cs+level, buffer, hl, stride_l);
- break;
- case DWT_DIRAC_HAAR0:
- case DWT_DIRAC_HAAR1:
- d->cs[level].y = 1;
- break;
- case DWT_DIRAC_DAUB9_7:
- spatial_compose97i_init2(d->cs+level, buffer, hl, stride_l);
- break;
- default:
- d->cs[level].y = 0;
- break;
- }
- }
-
- switch (type) {
- case DWT_DIRAC_DD9_7:
- d->spatial_compose = spatial_compose_dd97i_dy;
- d->vertical_compose_l0 = (void*)vertical_compose53iL0;
- d->vertical_compose_h0 = (void*)vertical_compose_dd97iH0;
- d->horizontal_compose = horizontal_compose_dd97i;
- d->support = 7;
- break;
- case DWT_DIRAC_LEGALL5_3:
- d->spatial_compose = spatial_compose_dirac53i_dy;
- d->vertical_compose_l0 = (void*)vertical_compose53iL0;
- d->vertical_compose_h0 = (void*)vertical_compose_dirac53iH0;
- d->horizontal_compose = horizontal_compose_dirac53i;
- d->support = 3;
- break;
- case DWT_DIRAC_DD13_7:
- d->spatial_compose = spatial_compose_dd137i_dy;
- d->vertical_compose_l0 = (void*)vertical_compose_dd137iL0;
- d->vertical_compose_h0 = (void*)vertical_compose_dd97iH0;
- d->horizontal_compose = horizontal_compose_dd137i;
- d->support = 7;
- break;
- case DWT_DIRAC_HAAR0:
- case DWT_DIRAC_HAAR1:
- d->spatial_compose = spatial_compose_haari_dy;
- d->vertical_compose = (void*)vertical_compose_haar;
- if (type == DWT_DIRAC_HAAR0)
- d->horizontal_compose = horizontal_compose_haar0i;
- else
- d->horizontal_compose = horizontal_compose_haar1i;
- d->support = 1;
- break;
- case DWT_DIRAC_FIDELITY:
- d->spatial_compose = spatial_compose_fidelity;
- d->vertical_compose_l0 = (void*)vertical_compose_fidelityiL0;
- d->vertical_compose_h0 = (void*)vertical_compose_fidelityiH0;
- d->horizontal_compose = horizontal_compose_fidelityi;
- d->support = 0; // not really used
- break;
- case DWT_DIRAC_DAUB9_7:
- d->spatial_compose = spatial_compose_daub97i_dy;
- d->vertical_compose_l0 = (void*)vertical_compose_daub97iL0;
- d->vertical_compose_h0 = (void*)vertical_compose_daub97iH0;
- d->vertical_compose_l1 = (void*)vertical_compose_daub97iL1;
- d->vertical_compose_h1 = (void*)vertical_compose_daub97iH1;
- d->horizontal_compose = horizontal_compose_daub97i;
- d->support = 5;
- break;
- default:
+ if (ret) {
av_log(NULL, AV_LOG_ERROR, "Unknown wavelet type %d\n", type);
return AVERROR_INVALIDDATA;
}
- if (HAVE_MMX) ff_spatial_idwt_init_mmx(d, type);
-
+ if (bit_depth == 8 && HAVE_MMX)
+ ff_spatial_idwt_init_mmx(d, type);
return 0;
}
@@ -558,4 +75,3 @@ void ff_spatial_idwt_slice2(DWTContext *d, int y)
d->spatial_compose(d, level, wl, hl, stride_l);
}
}
-
diff --git a/libavcodec/dirac_dwt.h b/libavcodec/dirac_dwt.h
index e5e447b..41842b5 100644
--- a/libavcodec/dirac_dwt.h
+++ b/libavcodec/dirac_dwt.h
@@ -30,21 +30,21 @@ typedef short IDWTELEM;
#define MAX_DECOMPOSITIONS 8
typedef struct DWTCompose {
- IDWTELEM *b[MAX_DWT_SUPPORT];
+ uint8_t *b[MAX_DWT_SUPPORT];
int y;
} DWTCompose;
struct DWTContext;
// Possible prototypes for vertical_compose functions
-typedef void (*vertical_compose_2tap)(IDWTELEM *b0, IDWTELEM *b1, int width);
-typedef void (*vertical_compose_3tap)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, int width);
-typedef void (*vertical_compose_5tap)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, IDWTELEM *b3, IDWTELEM *b4, int width);
-typedef void (*vertical_compose_9tap)(IDWTELEM *dst, IDWTELEM *b[8], int
width);
+typedef void (*vertical_compose_2tap)(uint8_t *b0, uint8_t *b1, int width);
+typedef void (*vertical_compose_3tap)(uint8_t *b0, uint8_t *b1, uint8_t *b2,
int width);
+typedef void (*vertical_compose_5tap)(uint8_t *b0, uint8_t *b1, uint8_t *b2,
uint8_t *b3, uint8_t *b4, int width);
+typedef void (*vertical_compose_9tap)(uint8_t *dst, uint8_t *b[8], int width);
typedef struct DWTContext {
- IDWTELEM *buffer;
- IDWTELEM *temp;
+ uint8_t *buffer;
+ uint8_t *temp;
int width;
int height;
int stride;
@@ -57,7 +57,7 @@ typedef struct DWTContext {
void (*vertical_compose_l1)(void);
void (*vertical_compose_h1)(void);
void (*vertical_compose)(void); ///< one set of lowpass and highpass
combined
- void (*horizontal_compose)(IDWTELEM *b, IDWTELEM *tmp, int width);
+ void (*horizontal_compose)(uint8_t *b, uint8_t *tmp, int width);
DWTCompose cs[MAX_DECOMPOSITIONS];
} DWTContext;
@@ -76,9 +76,9 @@ enum dwt_type {
};
// -1 if an error occurred, e.g. the dwt_type isn't recognized
-int ff_spatial_idwt_init2(DWTContext *d, IDWTELEM *buffer, int width, int
height,
+int ff_spatial_idwt_init2(DWTContext *d, uint8_t *buffer, int width, int
height,
int stride, enum dwt_type type, int
decomposition_count,
- IDWTELEM *temp);
+ uint8_t *temp, int bit_depth);
void ff_spatial_idwt_slice2(DWTContext *d, int y);
diff --git a/libavcodec/diracdec.c b/libavcodec/diracdec.c
index 62f375f..4c02259 100644
--- a/libavcodec/diracdec.c
+++ b/libavcodec/diracdec.c
@@ -1687,16 +1687,16 @@ static int dirac_decode_frame_internal(DiracContext *s)
memset(p->idwt_buf, 0, p->idwt_stride * p->idwt_height);
decode_component(s, comp); /* [DIRAC_STD] 13.4.1
core_transform_data() */
}
- ret = ff_spatial_idwt_init2(&d, (int16_t*)p->idwt_buf, p->idwt_width,
p->idwt_height, p->idwt_stride >> 1,
- s->wavelet_idx+2, s->wavelet_depth,
(int16_t*)p->idwt_tmp);
+ ret = ff_spatial_idwt_init2(&d, p->idwt_buf, p->idwt_width,
p->idwt_height, p->idwt_stride,
+ s->wavelet_idx+2, s->wavelet_depth,
p->idwt_tmp, s->bit_depth);
if (ret < 0)
return ret;
if (!s->num_refs) { /* intra */
for (y = 0; y < p->height; y += 16) {
ff_spatial_idwt_slice2(&d, y+16); /* decode */
- s->diracdsp.put_signed_rect_clamped(frame + y*p->stride,
p->stride,
- (int16_t*)(p->idwt_buf) +
y*(p->idwt_stride >> 1), (p->idwt_stride >> 1), p->width, 16);
+ s->diracdsp.put_signed_rect_clamped[s->pshift](frame +
y*p->stride, p->stride,
+ p->idwt_buf +
y*p->idwt_stride, p->idwt_stride, p->width, 16);
}
} else { /* inter */
int rowheight = p->ybsep*p->stride;
diff --git a/libavcodec/diracdsp.c b/libavcodec/diracdsp.c
index 6b02779..e82b587 100644
--- a/libavcodec/diracdsp.c
+++ b/libavcodec/diracdsp.c
@@ -135,9 +135,10 @@ ADD_OBMC(8)
ADD_OBMC(16)
ADD_OBMC(32)
-static void put_signed_rect_clamped_c(uint8_t *dst, int dst_stride, const
int16_t *src, int src_stride, int width, int height)
+static void put_signed_rect_clamped_8bit_c(uint8_t *dst, int dst_stride, const
uint8_t *_src, int src_stride, int width, int height)
{
int x, y;
+ int16_t *src = (int16_t *)_src;
for (y = 0; y < height; y++) {
for (x = 0; x < width; x+=4) {
dst[x ] = av_clip_uint8(src[x ] + 128);
@@ -146,7 +147,24 @@ static void put_signed_rect_clamped_c(uint8_t *dst, int
dst_stride, const int16_
dst[x+3] = av_clip_uint8(src[x+3] + 128);
}
dst += dst_stride;
- src += src_stride;
+ src += src_stride >> 1;
+ }
+}
+
+static void put_signed_rect_clamped_10bit_c(uint8_t *_dst, int dst_stride,
const uint8_t *_src, int src_stride, int width, int height)
+{
+ int x, y;
+ uint16_t *dst = (uint16_t *)_dst;
+ int32_t *src = (int32_t *)_src;
+ for (y = 0; y < height; y++) {
+ for (x = 0; x < width; x+=4) {
+ dst[x ] = av_clip(src[x ] + 512, 0, (1 << 10) - 1);
+ dst[x+1] = av_clip(src[x+1] + 512, 0, (1 << 10) - 1);
+ dst[x+2] = av_clip(src[x+2] + 512, 0, (1 << 10) - 1);
+ dst[x+3] = av_clip(src[x+3] + 512, 0, (1 << 10) - 1);
+ }
+ dst += dst_stride >> 1;
+ src += src_stride >> 2;
}
}
@@ -177,7 +195,8 @@ av_cold void ff_diracdsp_init(DiracDSPContext *c)
{
c->dirac_hpel_filter = dirac_hpel_filter;
c->add_rect_clamped = add_rect_clamped_c;
- c->put_signed_rect_clamped = put_signed_rect_clamped_c;
+ c->put_signed_rect_clamped[0] = put_signed_rect_clamped_8bit_c;
+ c->put_signed_rect_clamped[1] = put_signed_rect_clamped_10bit_c;
c->add_dirac_obmc[0] = add_obmc8_c;
c->add_dirac_obmc[1] = add_obmc16_c;
diff --git a/libavcodec/diracdsp.h b/libavcodec/diracdsp.h
index 613ca5b..439bda2 100644
--- a/libavcodec/diracdsp.h
+++ b/libavcodec/diracdsp.h
@@ -41,8 +41,8 @@ typedef struct {
void (*put_dirac_pixels_tab[3][4])(uint8_t *dst, const uint8_t *src[5],
int stride, int h);
void (*avg_dirac_pixels_tab[3][4])(uint8_t *dst, const uint8_t *src[5],
int stride, int h);
- void (*put_signed_rect_clamped)(uint8_t *dst/*align 16*/, int dst_stride,
const int16_t *src/*align 16*/, int src_stride, int width, int height/*mod 2*/);
- void (*put_rect_clamped)(uint8_t *dst/*align 16*/, int dst_stride, const
int16_t *src/*align 16*/, int src_stride, int width, int height/*mod 2*/);
+ void (*put_signed_rect_clamped[3])(uint8_t *dst/*align 16*/, int
dst_stride, const uint8_t *src/*align 16*/, int src_stride, int width, int
height/*mod 2*/);
+ void (*put_rect_clamped)(uint8_t *dst/*align 16*/, int dst_stride, const
uint8_t *src/*align 16*/, int src_stride, int width, int height/*mod 2*/);
void (*add_rect_clamped)(uint8_t *dst/*align 16*/, const uint16_t
*src/*align 16*/, int stride, const int16_t *idwt/*align 16*/, int idwt_stride,
int width, int height/*mod 2*/);
void (*add_dirac_obmc[3])(uint16_t *dst, const uint8_t *src, int stride,
const uint8_t *obmc_weight, int yblen);
diff --git a/libavcodec/x86/dirac_dwt.c b/libavcodec/x86/dirac_dwt.c
index 3c51ea6..c9c7093 100644
--- a/libavcodec/x86/dirac_dwt.c
+++ b/libavcodec/x86/dirac_dwt.c
@@ -25,17 +25,20 @@
#include "dirac_dwt.h"
#define COMPOSE_VERTICAL(ext, align) \
-void ff_vertical_compose53iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
int width); \
-void ff_vertical_compose_dirac53iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, int width); \
-void ff_vertical_compose_dd137iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, IDWTELEM *b3, IDWTELEM *b4, int width); \
-void ff_vertical_compose_dd97iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, IDWTELEM *b3, IDWTELEM *b4, int width); \
-void ff_vertical_compose_haar##ext(IDWTELEM *b0, IDWTELEM *b1, int width); \
-void ff_horizontal_compose_haar0i##ext(IDWTELEM *b, IDWTELEM *tmp, int w);\
-void ff_horizontal_compose_haar1i##ext(IDWTELEM *b, IDWTELEM *tmp, int w);\
-\
-static void vertical_compose53iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, int width) \
+void ff_vertical_compose53iL0##ext(int16_t *b0, int16_t *b1, int16_t *b2, int
width); \
+void ff_vertical_compose_dirac53iH0##ext(int16_t *b0, int16_t *b1, int16_t
*b2, int width); \
+void ff_vertical_compose_dd137iL0##ext(int16_t *b0, int16_t *b1, int16_t *b2,
int16_t *b3, int16_t *b4, int width); \
+void ff_vertical_compose_dd97iH0##ext(int16_t *b0, int16_t *b1, int16_t *b2,
int16_t *b3, int16_t *b4, int width); \
+void ff_vertical_compose_haar##ext(int16_t *b0, int16_t *b1, int width); \
+void ff_horizontal_compose_haar0i##ext(int16_t *b, int16_t *tmp, int w);\
+void ff_horizontal_compose_haar1i##ext(int16_t *b, int16_t *tmp, int w);\
+\
+static void vertical_compose53iL0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t
*_b2, int width) \
{ \
int i, width_align = width&~(align-1); \
+ int16_t *b0 = (int16_t *)_b0; \
+ int16_t *b1 = (int16_t *)_b1; \
+ int16_t *b2 = (int16_t *)_b2; \
\
for(i=width_align; i<width; i++) \
b1[i] = COMPOSE_53iL0(b0[i], b1[i], b2[i]); \
@@ -43,9 +46,12 @@ static void vertical_compose53iL0##ext(IDWTELEM *b0,
IDWTELEM *b1, IDWTELEM *b2,
ff_vertical_compose53iL0##ext(b0, b1, b2, width_align); \
} \
\
-static void vertical_compose_dirac53iH0##ext(IDWTELEM *b0, IDWTELEM *b1,
IDWTELEM *b2, int width) \
+static void vertical_compose_dirac53iH0##ext(uint8_t *_b0, uint8_t *_b1,
uint8_t *_b2, int width) \
{ \
int i, width_align = width&~(align-1); \
+ int16_t *b0 = (int16_t *)_b0; \
+ int16_t *b1 = (int16_t *)_b1; \
+ int16_t *b2 = (int16_t *)_b2; \
\
for(i=width_align; i<width; i++) \
b1[i] = COMPOSE_DIRAC53iH0(b0[i], b1[i], b2[i]); \
@@ -53,10 +59,15 @@ static void vertical_compose_dirac53iH0##ext(IDWTELEM *b0,
IDWTELEM *b1, IDWTELE
ff_vertical_compose_dirac53iH0##ext(b0, b1, b2, width_align); \
} \
\
-static void vertical_compose_dd137iL0##ext(IDWTELEM *b0, IDWTELEM *b1,
IDWTELEM *b2, \
- IDWTELEM *b3, IDWTELEM *b4, int
width) \
+static void vertical_compose_dd137iL0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t
*_b2, \
+ uint8_t *_b3, uint8_t *_b4, int
width) \
{ \
int i, width_align = width&~(align-1); \
+ int16_t *b0 = (int16_t *)_b0; \
+ int16_t *b1 = (int16_t *)_b1; \
+ int16_t *b2 = (int16_t *)_b2; \
+ int16_t *b3 = (int16_t *)_b3; \
+ int16_t *b4 = (int16_t *)_b4; \
\
for(i=width_align; i<width; i++) \
b2[i] = COMPOSE_DD137iL0(b0[i], b1[i], b2[i], b3[i], b4[i]); \
@@ -64,19 +75,26 @@ static void vertical_compose_dd137iL0##ext(IDWTELEM *b0,
IDWTELEM *b1, IDWTELEM
ff_vertical_compose_dd137iL0##ext(b0, b1, b2, b3, b4, width_align); \
} \
\
-static void vertical_compose_dd97iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2, \
- IDWTELEM *b3, IDWTELEM *b4, int
width) \
+static void vertical_compose_dd97iH0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t
*_b2, \
+ uint8_t *_b3, uint8_t *_b4, int
width) \
{ \
int i, width_align = width&~(align-1); \
+ int16_t *b0 = (int16_t *)_b0; \
+ int16_t *b1 = (int16_t *)_b1; \
+ int16_t *b2 = (int16_t *)_b2; \
+ int16_t *b3 = (int16_t *)_b3; \
+ int16_t *b4 = (int16_t *)_b4; \
\
for(i=width_align; i<width; i++) \
b2[i] = COMPOSE_DD97iH0(b0[i], b1[i], b2[i], b3[i], b4[i]); \
\
ff_vertical_compose_dd97iH0##ext(b0, b1, b2, b3, b4, width_align); \
} \
-static void vertical_compose_haar##ext(IDWTELEM *b0, IDWTELEM *b1, int width) \
+static void vertical_compose_haar##ext(uint8_t *_b0, uint8_t *_b1, int width) \
{ \
int i, width_align = width&~(align-1); \
+ int16_t *b0 = (int16_t *)_b0; \
+ int16_t *b1 = (int16_t *)_b1; \
\
for(i=width_align; i<width; i++) { \
b0[i] = COMPOSE_HAARiL0(b0[i], b1[i]); \
@@ -85,10 +103,13 @@ static void vertical_compose_haar##ext(IDWTELEM *b0,
IDWTELEM *b1, int width) \
\
ff_vertical_compose_haar##ext(b0, b1, width_align); \
} \
-static void horizontal_compose_haar0i##ext(IDWTELEM *b, IDWTELEM *tmp, int w)\
+static void horizontal_compose_haar0i##ext(uint8_t *_b, uint8_t *_tmp, int w)\
{\
int w2= w>>1;\
int x= w2 - (w2&(align-1));\
+ int16_t *b = (int16_t *)_b; \
+ int16_t *tmp = (int16_t *)_tmp; \
+\
ff_horizontal_compose_haar0i##ext(b, tmp, w);\
\
for (; x < w2; x++) {\
@@ -96,10 +117,13 @@ static void horizontal_compose_haar0i##ext(IDWTELEM *b,
IDWTELEM *tmp, int w)\
b[2*x+1] = COMPOSE_HAARiH0(b[x+w2], tmp[x]);\
}\
}\
-static void horizontal_compose_haar1i##ext(IDWTELEM *b, IDWTELEM *tmp, int w)\
+static void horizontal_compose_haar1i##ext(uint8_t *_b, uint8_t *_tmp, int w)\
{\
int w2= w>>1;\
int x= w2 - (w2&(align-1));\
+ int16_t *b = (int16_t *)_b; \
+ int16_t *tmp = (int16_t *)_tmp; \
+\
ff_horizontal_compose_haar1i##ext(b, tmp, w);\
\
for (; x < w2; x++) {\
@@ -116,12 +140,15 @@ COMPOSE_VERTICAL(_mmx, 4)
COMPOSE_VERTICAL(_sse2, 8)
-void ff_horizontal_compose_dd97i_ssse3(IDWTELEM *b, IDWTELEM *tmp, int w);
+void ff_horizontal_compose_dd97i_ssse3(int16_t *_b, int16_t *_tmp, int w);
-static void horizontal_compose_dd97i_ssse3(IDWTELEM *b, IDWTELEM *tmp, int w)
+static void horizontal_compose_dd97i_ssse3(uint8_t *_b, uint8_t *_tmp, int w)
{
int w2= w>>1;
int x= w2 - (w2&7);
+ int16_t *b = (int16_t *)_b;
+ int16_t *tmp = (int16_t *)_tmp;
+
ff_horizontal_compose_dd97i_ssse3(b, tmp, w);
for (; x < w2; x++) {
diff --git a/libavcodec/x86/diracdsp_mmx.c b/libavcodec/x86/diracdsp_mmx.c
index 11df5e3..b2e949c 100644
--- a/libavcodec/x86/diracdsp_mmx.c
+++ b/libavcodec/x86/diracdsp_mmx.c
@@ -130,7 +130,7 @@ void ff_diracdsp_init_mmx(DiracDSPContext* c)
c->add_dirac_obmc[2] = ff_add_dirac_obmc32_mmx;
c->dirac_hpel_filter = dirac_hpel_filter_mmx;
c->add_rect_clamped = ff_add_rect_clamped_mmx;
- c->put_signed_rect_clamped = ff_put_signed_rect_clamped_mmx;
+ c->put_signed_rect_clamped[0] = (void *)ff_put_signed_rect_clamped_mmx;
#endif
PIXFUNC(put, 0, mmx);
PIXFUNC(avg, 0, mmx);
@@ -143,7 +143,7 @@ void ff_diracdsp_init_mmx(DiracDSPContext* c)
if (EXTERNAL_SSE2(mm_flags)) {
c->dirac_hpel_filter = dirac_hpel_filter_sse2;
c->add_rect_clamped = ff_add_rect_clamped_sse2;
- c->put_signed_rect_clamped = ff_put_signed_rect_clamped_sse2;
+ c->put_signed_rect_clamped[0] = (void
*)ff_put_signed_rect_clamped_sse2;
c->add_dirac_obmc[1] = ff_add_dirac_obmc16_sse2;
c->add_dirac_obmc[2] = ff_add_dirac_obmc32_sse2;
diff --git a/libavcodec/x86/diracdsp_yasm.asm b/libavcodec/x86/diracdsp_yasm.asm
index 40fe2c8..a042413 100644
--- a/libavcodec/x86/diracdsp_yasm.asm
+++ b/libavcodec/x86/diracdsp_yasm.asm
@@ -150,7 +150,7 @@ cglobal put_signed_rect_clamped_%1, 5,9,3, dst, dst_stride,
src, src_stride, w,
%endif
.loopy:
- lea src2q, [srcq+src_strideq*2]
+ lea src2q, [srcq+src_strideq]
lea dst2q, [dstq+dst_strideq]
.loopx:
sub wd, mmsize
@@ -164,7 +164,7 @@ cglobal put_signed_rect_clamped_%1, 5,9,3, dst, dst_stride,
src, src_stride, w,
mova [dst2q+wq], m2
jg .loopx
- lea srcq, [srcq+src_strideq*4]
+ lea srcq, [srcq+src_strideq*2]
lea dstq, [dstq+dst_strideq*2]
sub hd, 2
mov wd, wspill
--
1.9.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
