From: Timothy B. Terriberry <tterribe@xiph.org>
Date: Fri, 17 May 2013 17:11:30 +0000 (-0700)
Subject: Reduce WHT complexity.
X-Git-Tag: v1.3.0~1104^2~132
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=95339d68258b8f752e114e6c81e6e19f99cde5a4;p=libvpx

Reduce WHT complexity.

Saves 1 add, 3 shifts (and a shift bias) per 1-D transform.

Change-Id: I1104bb1679fe342b2f9677df8a9cdc0cb9699e7d
---

diff --git a/vp9/common/vp9_idct.c b/vp9/common/vp9_idct.c
index f61230c80..026ba913d 100644
--- a/vp9/common/vp9_idct.c
+++ b/vp9/common/vp9_idct.c
@@ -19,22 +19,32 @@
 #include "vp9/common/vp9_idct.h"
 
 void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
+/* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
+   0.5 shifts per pixel. */
   int i;
   int16_t output[16];
-  int a1, b1, c1, d1;
+  int a1, b1, c1, d1, e1;
   int16_t *ip = input;
   int16_t *op = output;
 
   for (i = 0; i < 4; i++) {
-    a1 = (ip[0] + ip[3]) >> WHT_UPSCALE_FACTOR;
-    b1 = (ip[1] + ip[2]) >> WHT_UPSCALE_FACTOR;
-    c1 = (ip[1] - ip[2]) >> WHT_UPSCALE_FACTOR;
-    d1 = (ip[0] - ip[3]) >> WHT_UPSCALE_FACTOR;
-
-    op[0] = (a1 + b1 + 1) >> 1;
-    op[1] = (c1 + d1) >> 1;
-    op[2] = (a1 - b1) >> 1;
-    op[3] = (d1 - c1) >> 1;
+    a1 = ip[0] >> WHT_UPSCALE_FACTOR;
+    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;
+
+    op[0] = a1;
+    op[1] = b1;
+    op[2] = c1;
+    op[3] = d1;
 
     ip += 4;
     op += 4;
@@ -42,20 +52,23 @@ void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
 
   ip = output;
   for (i = 0; i < 4; i++) {
-    a1 = ip[4 * 0] + ip[4 * 3];
-    b1 = ip[4 * 1] + ip[4 * 2];
-    c1 = ip[4 * 1] - ip[4 * 2];
-    d1 = ip[4 * 0] - ip[4 * 3];
-
-
-    dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] +
-                                       ((a1 + b1 + 1) >> 1));
-    dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] +
-                                       ((c1 + d1) >> 1));
-    dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] +
-                                       ((a1 - b1) >> 1));
-    dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] +
-                                       ((d1 - c1) >> 1));
+    a1 = ip[4 * 0];
+    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;
+
+    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);
+    dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] + d1);
 
     ip++;
     dest++;
@@ -64,23 +77,24 @@ void vp9_short_iwalsh4x4_add_c(int16_t *input, uint8_t *dest, int dest_stride) {
 
 void vp9_short_iwalsh4x4_1_add_c(int16_t *in, uint8_t *dest, int dest_stride) {
   int i;
+  int a1, e1;
   int16_t tmp[4];
   int16_t *ip = in;
   int16_t *op = tmp;
 
-  op[0] = ((ip[0] >> WHT_UPSCALE_FACTOR) + 1) >> 1;
-  op[1] = op[2] = op[3] = (ip[0] >> WHT_UPSCALE_FACTOR) >> 1;
+  a1 = ip[0] >> WHT_UPSCALE_FACTOR;
+  e1 = a1 >> 1;
+  op[0] = op[1] = op[2] = a1 - e1;
+  op[3] = e1;
 
   ip = tmp;
   for (i = 0; i < 4; i++) {
-    dest[dest_stride * 0] = clip_pixel(dest[dest_stride * 0] +
-                                       ((ip[0] + 1) >> 1));
-    dest[dest_stride * 1] = clip_pixel(dest[dest_stride * 1] +
-                                       (ip[0] >> 1));
-    dest[dest_stride * 2] = clip_pixel(dest[dest_stride * 2] +
-                                       (ip[0] >> 1));
-    dest[dest_stride * 3] = clip_pixel(dest[dest_stride * 3] +
-                                       (ip[0] >> 1));
+    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 * 3] = clip_pixel(dest[dest_stride * 3] + e1);
     ip++;
     dest++;
   }
diff --git a/vp9/encoder/vp9_dct.c b/vp9/encoder/vp9_dct.c
index ebf40e4e6..d22644424 100644
--- a/vp9/encoder/vp9_dct.c
+++ b/vp9/encoder/vp9_dct.c
@@ -591,23 +591,33 @@ void vp9_short_fht8x8_c(int16_t *input, int16_t *output,
   }
 }
 
+/* 4-point reversible, orthonormal Walsh-Hadamard in 3.5 adds, 0.5 shifts per
+   pixel. */
 void vp9_short_walsh4x4_c(short *input, short *output, int pitch) {
   int i;
-  int a1, b1, c1, d1;
+  int a1, b1, c1, d1, e1;
   short *ip = input;
   short *op = output;
   int pitch_short = pitch >> 1;
 
   for (i = 0; i < 4; i++) {
-    a1 = ip[0 * pitch_short] + ip[3 * pitch_short];
-    b1 = ip[1 * pitch_short] + ip[2 * pitch_short];
-    c1 = ip[1 * pitch_short] - ip[2 * pitch_short];
-    d1 = ip[0 * pitch_short] - ip[3 * pitch_short];
-
-    op[0] = (a1 + b1 + 1) >> 1;
-    op[4] = (c1 + d1) >> 1;
-    op[8] = (a1 - b1) >> 1;
-    op[12] = (d1 - c1) >> 1;
+    a1 = ip[0 * pitch_short];
+    b1 = ip[1 * pitch_short];
+    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;
+
+    op[0] = a1;
+    op[4] = c1;
+    op[8] = d1;
+    op[12] = b1;
 
     ip++;
     op++;
@@ -616,15 +626,23 @@ void vp9_short_walsh4x4_c(short *input, short *output, int pitch) {
   op = output;
 
   for (i = 0; i < 4; i++) {
-    a1 = ip[0] + ip[3];
-    b1 = ip[1] + ip[2];
-    c1 = ip[1] - ip[2];
-    d1 = ip[0] - ip[3];
-
-    op[0] = ((a1 + b1 + 1) >> 1) << WHT_UPSCALE_FACTOR;
-    op[1] = ((c1 + d1) >> 1) << WHT_UPSCALE_FACTOR;
-    op[2] = ((a1 - b1) >> 1) << WHT_UPSCALE_FACTOR;
-    op[3] = ((d1 - c1) >> 1) << WHT_UPSCALE_FACTOR;
+    a1 = ip[0];
+    b1 = ip[1];
+    c1 = ip[2];
+    d1 = ip[3];
+
+    b1 = a1 - b1;
+    c1 += d1;
+    e1 = (c1 - b1) >> 1;
+    a1 += e1;
+    d1 -= e1;
+    c1 = a1 - c1;
+    b1 -= d1;
+
+    op[0] = a1 << WHT_UPSCALE_FACTOR;
+    op[1] = c1 << WHT_UPSCALE_FACTOR;
+    op[2] = d1 << WHT_UPSCALE_FACTOR;
+    op[3] = b1 << WHT_UPSCALE_FACTOR;
 
     ip += 4;
     op += 4;