From 4d4d0e817607bce2a3a0806540f19d5d2b3b9376 Mon Sep 17 00:00:00 2001
From: Kieran Kunhya <kierank@ob-encoder.com>
Date: Thu, 22 Sep 2011 18:21:05 -0500
Subject: Fix unnecessary shift with 9/10bit vertical scaling

Signed-off-by: Diego Biurrun <diego@biurrun.de>
---
 libswscale/swscale.c               | 18 +++++++++---------
 tests/ref/lavfi/pixfmts_scale      | 12 ++++++------
 tests/ref/vsynth1/dnxhd_720p_10bit |  4 ++--
 tests/ref/vsynth2/dnxhd_720p_10bit |  4 ++--
 4 files changed, 19 insertions(+), 19 deletions(-)

diff --git a/libswscale/swscale.c b/libswscale/swscale.c
index 733f57b049..5d90250acf 100644
--- a/libswscale/swscale.c
+++ b/libswscale/swscale.c
@@ -267,7 +267,7 @@ yuv2yuvX10_c_template(const int16_t *lumFilter, const int16_t **lumSrc,
     int i;
     uint16_t *yDest = dest[0], *uDest = dest[1], *vDest = dest[2],
              *aDest = CONFIG_SWSCALE_ALPHA ? dest[3] : NULL;
-    int shift = 11 + 16 - output_bits - 1;
+    int shift = 11 + 16 - output_bits;
 
 #define output_pixel(pos, val) \
     if (big_endian) { \
@@ -276,24 +276,24 @@ yuv2yuvX10_c_template(const int16_t *lumFilter, const int16_t **lumSrc,
         AV_WL16(pos, av_clip_uintp2(val >> shift, output_bits)); \
     }
     for (i = 0; i < dstW; i++) {
-        int val = 1 << (26-output_bits - 1);
+        int val = 1 << (26-output_bits);
         int j;
 
         for (j = 0; j < lumFilterSize; j++)
-            val += (lumSrc[j][i] * lumFilter[j]) >> 1;
+            val += lumSrc[j][i] * lumFilter[j];
 
         output_pixel(&yDest[i], val);
     }
 
     if (uDest) {
         for (i = 0; i < chrDstW; i++) {
-            int u = 1 << (26-output_bits - 1);
-            int v = 1 << (26-output_bits - 1);
+            int u = 1 << (26-output_bits);
+            int v = 1 << (26-output_bits);
             int j;
 
             for (j = 0; j < chrFilterSize; j++) {
-                u += (chrUSrc[j][i] * chrFilter[j]) >> 1;
-                v += (chrVSrc[j][i] * chrFilter[j]) >> 1;
+                u += chrUSrc[j][i] * chrFilter[j];
+                v += chrVSrc[j][i] * chrFilter[j];
             }
 
             output_pixel(&uDest[i], u);
@@ -303,11 +303,11 @@ yuv2yuvX10_c_template(const int16_t *lumFilter, const int16_t **lumSrc,
 
     if (CONFIG_SWSCALE_ALPHA && aDest) {
         for (i = 0; i < dstW; i++) {
-            int val = 1 << (26-output_bits - 1);
+            int val = 1 << (26-output_bits);
             int j;
 
             for (j = 0; j < lumFilterSize; j++)
-                val += (alpSrc[j][i] * lumFilter[j]) >> 1;
+                val += alpSrc[j][i] * lumFilter[j];
 
             output_pixel(&aDest[i], val);
         }
diff --git a/tests/ref/lavfi/pixfmts_scale b/tests/ref/lavfi/pixfmts_scale
index 4787614bd2..27636284a0 100644
--- a/tests/ref/lavfi/pixfmts_scale
+++ b/tests/ref/lavfi/pixfmts_scale
@@ -31,15 +31,15 @@ uyvy422             314bd486277111a95d9369b944fa0400
 yuv410p             7df8f6d69b56a8dcb6c7ee908e5018b5
 yuv411p             1143e7c5cc28fe0922b051b17733bc4c
 yuv420p             fdad2d8df8985e3d17e73c71f713cb14
-yuv420p10be         d7695b9117d5b52819c569459e42669b
-yuv420p10le         0ac6d448db2df5f3d1346aa81f2b5f50
+yuv420p10be         c143e77e97d2f7d62c3b518857ba9f9b
+yuv420p10le         72d90eccf5c34691ff057dafb7447aa2
 yuv420p16be         9688e33e03b8c8275ab2fb1df0f06bee
 yuv420p16le         cba8b390ad5e7b8678e419b8ce79c008
-yuv420p9be          8fa6e007b1a40f34eaa3e2beb73ea8af
-yuv420p9le          a7b131a7dd06906a5aef2e36d117b972
+yuv420p9be          bb87fddca65d1742412c8d2b1caf96c6
+yuv420p9le          828eec50014a41258a5423c1fe56ac97
 yuv422p             918e37701ee7377d16a8a6c119c56a40
-yuv422p10be         35206fcd7e00ee582a8c366b37d57d1d
-yuv422p10le         396f930e2da02f149ab9dd5b781cbe8d
+yuv422p10be         cea7ca6b0e66d6f29539885896c88603
+yuv422p10le         a10c4a5837547716f13cd61918b145f9
 yuv422p16be         285993ee0c0f4f8e511ee46f93c5f38c
 yuv422p16le         61bfcee8e54465f760164f5a75d40b5e
 yuv440p             461503fdb9b90451020aa3b25ddf041c
diff --git a/tests/ref/vsynth1/dnxhd_720p_10bit b/tests/ref/vsynth1/dnxhd_720p_10bit
index cdab77c16a..ad97b66b7d 100644
--- a/tests/ref/vsynth1/dnxhd_720p_10bit
+++ b/tests/ref/vsynth1/dnxhd_720p_10bit
@@ -1,4 +1,4 @@
-cb29b6ae4e1562d95f9311991fef98df *./tests/data/vsynth1/dnxhd-720p-10bit.dnxhd
+b5e24a055af02edec8674333260214fd *./tests/data/vsynth1/dnxhd-720p-10bit.dnxhd
  2293760 ./tests/data/vsynth1/dnxhd-720p-10bit.dnxhd
-2f45bb1af7da5dd3dca870ac87237b7d *./tests/data/dnxhd_720p_10bit.vsynth1.out.yuv
+4466ff3d73d01bbe75ea25001d379b63 *./tests/data/dnxhd_720p_10bit.vsynth1.out.yuv
 stddev:    6.27 PSNR: 32.18 MAXDIFF:   64 bytes:   760320/  7603200
diff --git a/tests/ref/vsynth2/dnxhd_720p_10bit b/tests/ref/vsynth2/dnxhd_720p_10bit
index 81f53d8353..60c0d84a34 100644
--- a/tests/ref/vsynth2/dnxhd_720p_10bit
+++ b/tests/ref/vsynth2/dnxhd_720p_10bit
@@ -1,4 +1,4 @@
-8648511257afb816b5b911706ca391db *./tests/data/vsynth2/dnxhd-720p-10bit.dnxhd
+4b57da2c0c1280469ff3579f7151c227 *./tests/data/vsynth2/dnxhd-720p-10bit.dnxhd
  2293760 ./tests/data/vsynth2/dnxhd-720p-10bit.dnxhd
-391b6f5aa7c7b488b479cb43d420b860 *./tests/data/dnxhd_720p_10bit.vsynth2.out.yuv
+31a6aa8b8702e85fa3b48e73f035c4e4 *./tests/data/dnxhd_720p_10bit.vsynth2.out.yuv
 stddev:    1.35 PSNR: 45.46 MAXDIFF:   23 bytes:   760320/  7603200
-- 
cgit v1.2.3