Changed to use a new variant of WHT

The commit changed to use a new variant of Walsh-Hadamard Transform
by Tim Terriberry. This new variant has the best compression among a
number of variants that developed by Tim.

Change-Id: Icb3a88515463cfc644b17ca046fcd139db2557e9

diff --git a/vp9/common/vp9_idct.c b/vp9/common/vp9_idct.c
index 026ba913df6bba82bf7c55e767fd7b85e5efec99..dcc7f03307eaa2d27bfdf248b6bce1342e8ddcb4 100644 (file)
--- a/vp9/common/vp9_idct.c
+++ b/vp9/common/vp9_idct.c
@@ -32,20 +32,17 @@ void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
     c1 = ip[1] >> WHT_UPSCALE_FACTOR;
     d1 = ip[2] >> WHT_UPSCALE_FACTOR;
     b1 = ip[3] >> WHT_UPSCALE_FACTOR;
-
-    c1 = a1 - c1;
-    b1 += d1;
-    e1 = (c1 - b1) >> 1;
-    a1 -= e1;
-    d1 += e1;
-    b1 = a1 - b1;
-    c1 -= d1;
-
+    a1 += c1;
+    d1 -= b1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= b1;
+    d1 += c1;
     op[0] = a1;
     op[1] = b1;
     op[2] = c1;
     op[3] = d1;
-
     ip += 4;
     op += 4;
   }
@@ -56,15 +53,13 @@ void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
     c1 = ip[4 * 1];
     d1 = ip[4 * 2];
     b1 = ip[4 * 3];
-
-    c1 = a1 - c1;
-    b1 += d1;
-    e1 = (c1 - b1) >> 1;
-    a1 -= e1;
-    d1 += e1;
-    b1 = a1 - b1;
-    c1 -= d1;
-
+    a1 += c1;
+    d1 -= b1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= b1;
+    d1 += c1;
     dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] + a1);
     dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + b1);
     dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + c1);
@@ -84,16 +79,17 @@ void vp9_short_iwalsh4x4_1_add_c(int16_t *in, uint8_t *dest, int dest_stride) {
 
   a1 = ip[0] >> WHT_UPSCALE_FACTOR;
   e1 = a1 >> 1;
-  op[0] = op[1] = op[2] = a1 - e1;
-  op[3] = e1;
+  a1 -= e1;
+  op[0] = a1;
+  op[1] = op[2] = op[3] = e1;
 
   ip = tmp;
   for (i = 0; i < 4; i++) {
     e1 = ip[0] >> 1;
     a1 = ip[0] - e1;
     dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] + a1);
-    dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + a1);
-    dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + a1);
+    dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] + e1);
+    dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] + e1);
     dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] + e1);
     ip++;
     dest++;

diff --git a/vp9/common/vp9_idct.h b/vp9/common/vp9_idct.h
index 176bf5da4311ad2fbf30921f852cc1772ff17a5a..af35432c44cc8998ccea332801a5fc124ad2aa69 100644 (file)
--- a/vp9/common/vp9_idct.h
+++ b/vp9/common/vp9_idct.h
@@ -17,6 +17,7 @@
 #include "vpx/vpx_integer.h"
 #include "vp9/common/vp9_common.h"
 
+
 // Constants and Macros used by all idct/dct functions
 #define DCT_CONST_BITS 14
 #define DCT_CONST_ROUNDING  (1 << (DCT_CONST_BITS - 1))

diff --git a/vp9/encoder/vp9_dct.c b/vp9/encoder/vp9_dct.c
index d22644424e17c0f7c28a287a8bcb61c9025da516..8d4eec1394bf36e8bcce7ab24e3bf189696a02fa 100644 (file)
--- a/vp9/encoder/vp9_dct.c
+++ b/vp9/encoder/vp9_dct.c
@@ -606,14 +606,13 @@ void vp9_short_walsh4x4_c(short *input, short *output, int pitch) {
     c1 = ip[2 * pitch_short];
     d1 = ip[3 * pitch_short];
 
-    b1 = a1 - b1;
-    c1 += d1;
-    e1 = (c1 - b1) >> 1;
-    a1 += e1;
-    d1 -= e1;
-    c1 = a1 - c1;
-    b1 -= d1;
-
+    a1 += b1;
+    d1 = d1 - c1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= c1;
+    d1 += b1;
     op[0] = a1;
     op[4] = c1;
     op[8] = d1;
@@ -631,14 +630,13 @@ void vp9_short_walsh4x4_c(short *input, short *output, int pitch) {
     c1 = ip[2];
     d1 = ip[3];
 
-    b1 = a1 - b1;
-    c1 += d1;
-    e1 = (c1 - b1) >> 1;
-    a1 += e1;
-    d1 -= e1;
-    c1 = a1 - c1;
-    b1 -= d1;
-
+    a1 += b1;
+    d1 -= c1;
+    e1 = (a1 - d1) >> 1;
+    b1 = e1 - b1;
+    c1 = e1 - c1;
+    a1 -= c1;
+    d1 += b1;
     op[0] = a1 << WHT_UPSCALE_FACTOR;
     op[1] = c1 << WHT_UPSCALE_FACTOR;
     op[2] = d1 << WHT_UPSCALE_FACTOR;