# HG changeset patch
# User cigaes
# Date 1285261690 0
# Node ID 0dd2b27940b57be06c399c4206223c9acd886d31
# Parent  56dadca4f36dfc959f967ccb115918fa484caff8
vf_ass: avoid divisions in alpha blending.
On a test case, it speeds up the blending by about 20%.

diff -r 56dadca4f36d -r 0dd2b27940b5 libmpcodecs/vf_ass.c
--- a/libmpcodecs/vf_ass.c	Wed Sep 22 21:11:31 2010 +0000
+++ b/libmpcodecs/vf_ass.c	Thu Sep 23 17:08:10 2010 +0000
@@ -306,11 +306,13 @@
     unsigned char y = rgba2y(color);
     unsigned char u = rgba2u(color);
     unsigned char v = rgba2v(color);
-    unsigned char opacity = 255 - _a(color);
+    unsigned opacity = 255 - _a(color);
     unsigned char *src, *dsty, *dstu, *dstv;
     int i, j;
     mp_image_t *dmpi = vf->dmpi;
 
+    opacity = (0x10203 * opacity + 0x80) >> 8; /* 0x10203 = (1<<32)/(255*255) */
+    /* 0 <= opacity <= 0x10101 */
     src = bitmap;
     dsty = dmpi->planes[0]     + dst_x + dst_y * dmpi->stride[0];
     dstu = vf->priv->planes[1] + dst_x + dst_y * vf->priv->outw;
@@ -320,10 +322,10 @@
             unsigned k = src[j];
             if (!k)
                 continue;
-            k = k * opacity / 255;
-            dsty[j] = (k * y + (255 - k) * dsty[j]) / 255;
-            dstu[j] = (k * u + (255 - k) * dstu[j]) / 255;
-            dstv[j] = (k * v + (255 - k) * dstv[j]) / 255;
+            k *= opacity; /* 0 <= k <= 0xFFFFFF */
+            dsty[j] = (k * y + (0xFFFFFF - k) * dsty[j] + 0x800000) >> 24;
+            dstu[j] = (k * u + (0xFFFFFF - k) * dstu[j] + 0x800000) >> 24;
+            dstv[j] = (k * v + (0xFFFFFF - k) * dstv[j] + 0x800000) >> 24;
         }
         src  += stride;
         dsty += dmpi->stride[0];