[FFmpeg-devel] [PATCH 1/2] avcodec/snow_dwt: Support correct rounding of dwt decomposition of non even sizes

[Michael Niedermayer]

Signed-off-by: Michael Niedermayer <michael at niedermayer.cc>
---
 libavcodec/snow_dwt.c | 61 ++++++++++++++++++++++++-------------------
 libavcodec/snow_dwt.h |  3 +++
 2 files changed, 37 insertions(+), 27 deletions(-)

diff --git a/libavcodec/snow_dwt.c b/libavcodec/snow_dwt.c
index 965f409002..b27b07f58a 100644
--- a/libavcodec/snow_dwt.c
+++ b/libavcodec/snow_dwt.c
@@ -322,15 +322,17 @@ void ff_spatial_dwt(DWTELEM *buffer, DWTELEM *temp, int width, int height,
     int level;
 
     for (level = 0; level < decomposition_count; level++) {
-        switch (type) {
+        int w = DWT_RSHIFT(width , level);
+        int h = DWT_RSHIFT(height, level);
+        switch (type & 3) {
         case DWT_97:
             spatial_decompose97i(buffer, temp,
-                                 width >> level, height >> level,
+                                 w, h,
                                  stride << level);
             break;
         case DWT_53:
             spatial_decompose53i(buffer, temp,
-                                 width >> level, height >> level,
+                                 w, h,
                                  stride << level);
             break;
         }
@@ -642,13 +644,14 @@ void ff_spatial_idwt_buffered_init(DWTCompose *cs, slice_buffer *sb, int width,
 {
     int level;
     for (level = decomposition_count - 1; level >= 0; level--) {
-        switch (type) {
+        int h = DWT_RSHIFT(height, level);
+        switch (type & 3) {
         case DWT_97:
-            spatial_compose97i_buffered_init(cs + level, sb, height >> level,
+            spatial_compose97i_buffered_init(cs + level, sb, h,
                                              stride_line << level);
             break;
         case DWT_53:
-            spatial_compose53i_buffered_init(cs + level, sb, height >> level,
+            spatial_compose53i_buffered_init(cs + level, sb, h,
                                              stride_line << level);
             break;
         }
@@ -660,28 +663,29 @@ void ff_spatial_idwt_buffered_slice(SnowDWTContext *dsp, DWTCompose *cs,
                                     int width, int height, int stride_line,
                                     int type, int decomposition_count, int y)
 {
-    const int support = type == 1 ? 3 : 5;
+    const int support = (type & 3) == 1 ? 3 : 5;
     int level;
-    if (type == 2)
+    if ((type & 3) == 2)
         return;
 
-    for (level = decomposition_count - 1; level >= 0; level--)
-        while (cs[level].y <= FFMIN((y >> level) + support, height >> level)) {
-            switch (type) {
+    for (level = decomposition_count - 1; level >= 0; level--) {
+        int w = DWT_RSHIFT(width , level);
+        int h = DWT_RSHIFT(height, level);
+        while (cs[level].y <= FFMIN((y >> level) + support, h)) {
+            switch (type & 3) {
             case DWT_97:
                 spatial_compose97i_dy_buffered(dsp, cs + level, slice_buf, temp,
-                                               width >> level,
-                                               height >> level,
+                                               w, h,
                                                stride_line << level);
                 break;
             case DWT_53:
                 spatial_compose53i_dy_buffered(cs + level, slice_buf, temp,
-                                               width >> level,
-                                               height >> level,
+                                               w, h,
                                                stride_line << level);
                 break;
             }
         }
+    }
 }
 
 static void spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width,
@@ -690,13 +694,14 @@ static void spatial_idwt_init(DWTCompose *cs, IDWTELEM *buffer, int width,
 {
     int level;
     for (level = decomposition_count - 1; level >= 0; level--) {
-        switch (type) {
+        int h = DWT_RSHIFT(height, level);
+        switch (type & 3) {
         case DWT_97:
-            spatial_compose97i_init(cs + level, buffer, height >> level,
+            spatial_compose97i_init(cs + level, buffer, h,
                                     stride << level);
             break;
         case DWT_53:
-            spatial_compose53i_init(cs + level, buffer, height >> level,
+            spatial_compose53i_init(cs + level, buffer, h,
                                     stride << level);
             break;
         }
@@ -708,24 +713,25 @@ static void spatial_idwt_slice(DWTCompose *cs, IDWTELEM *buffer,
                                   int stride, int type,
                                   int decomposition_count, int y)
 {
-    const int support = type == 1 ? 3 : 5;
+    const int support = (type & 3) == 1 ? 3 : 5;
     int level;
-    if (type == 2)
+    if ((type & 3) == 2)
         return;
 
-    for (level = decomposition_count - 1; level >= 0; level--)
-        while (cs[level].y <= FFMIN((y >> level) + support, height >> level)) {
-            switch (type) {
+    for (level = decomposition_count - 1; level >= 0; level--) {
+        int w = DWT_RSHIFT(width , level);
+        int h = DWT_RSHIFT(height, level);
+        while (cs[level].y <= FFMIN((y >> level) + support, h)) {
+            switch (type & 3) {
             case DWT_97:
-                spatial_compose97i_dy(cs + level, buffer, temp, width >> level,
-                                      height >> level, stride << level);
+                spatial_compose97i_dy(cs + level, buffer, temp, w, h, stride << level);
                 break;
             case DWT_53:
-                spatial_compose53i_dy(cs + level, buffer, temp, width >> level,
-                                      height >> level, stride << level);
+                spatial_compose53i_dy(cs + level, buffer, temp, w, h, stride << level);
                 break;
             }
         }
+    }
 }
 
 void ff_spatial_idwt(IDWTELEM *buffer, IDWTELEM *temp, int width, int height,
@@ -788,6 +794,7 @@ static inline int w_c(struct MpegEncContext *v, const uint8_t *pix1, const uint8
     }
 
     ff_spatial_dwt(tmp, tmp2, w, h, 32, type, dec_count);
+    type &= 3;
 
     s = 0;
     av_assert1(w == h);
diff --git a/libavcodec/snow_dwt.h b/libavcodec/snow_dwt.h
index 15b8a3007b..48acfd46c5 100644
--- a/libavcodec/snow_dwt.h
+++ b/libavcodec/snow_dwt.h
@@ -67,6 +67,9 @@ typedef struct SnowDWTContext {
 
 #define DWT_97 0
 #define DWT_53 1
+#define DWT_NEW_ROUNDING_FLAG 4
+
+#define DWT_RSHIFT(a, b) (((type) & DWT_NEW_ROUNDING_FLAG) ? -(-(a) >> (b)) : (a) >> (b))
 
 #define liftS lift
 #define W_AM 3
-- 
2.17.1