From cd4602ecfa1622a8261cfb180af82af3d8ae7709 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Sebastian=20Dr=C3=B6ge?= <sebastian@centricular.com>
Date: Mon, 23 Dec 2013 14:54:02 +0100
Subject: [PATCH] videoconvert: Fix I420 to BGRA fast-path alpha setting

This fast-path was adding 128 to every component including
alpha while it should only be done for all components except
alpha. This caused wrong alpha values to be generated.

Also remove the high-quality I420 to BGRA fast-path as it needs
the same fix, which causes an additional instruction, which causes
orc to emit more than 96 variables, which then just crashes.
This can only be fixed in orc by breaking ABI and allowing more
variables.
---
 gst/videoconvert/gstvideoconvertorc.orc | 78 +------------------------
 gst/videoconvert/videoconvert.c         | 28 ++-------
 2 files changed, 7 insertions(+), 99 deletions(-)

diff --git a/gst/videoconvert/gstvideoconvertorc.orc b/gst/videoconvert/gstvideoconvertorc.orc
index c040395071..c055ce2571 100644
--- a/gst/videoconvert/gstvideoconvertorc.orc
+++ b/gst/videoconvert/gstvideoconvertorc.orc
@@ -725,7 +725,6 @@ x4 addb argb, x, 128
 .temp 1 r
 .temp 1 g
 .temp 1 b
-.temp 4 x
 .const 1 c8 8
 .const 1 c128 128
 
@@ -767,81 +766,10 @@ convssswb g, wg
 convssswb b, wb
 
 mergebw t1, b, g
+x2 subb t1, t1, c128
+subb    r, r, c128
 mergebw t2, r, 255
-mergewl x, t1, t2
-x4 addb argb, x, c128
-
-
-
-.function video_convert_orc_convert_I420_BGRA_avg
-.dest 4 argb guint8
-.source 1 y guint8
-.source 1 u1 guint8
-.source 1 u2 guint8
-.source 1 v1 guint8
-.source 1 v2 guint8
-.temp 2 t1
-.temp 2 t2
-.temp 1 t3
-.temp 1 t4
-.temp 2 wy
-.temp 2 wu
-.temp 2 wv
-.temp 2 wr
-.temp 2 wg
-.temp 2 wb
-.temp 1 r
-.temp 1 g
-.temp 1 b
-.temp 4 x
-.const 1 c8 8
-.const 1 c128 128
-
-subb t3, y, c128
-convsbw wy, t3
-loadupib t3, u1
-loadupib t4, u2
-avgub t3, t3, t4
-subb t3, t3, c128
-convsbw wu, t3
-loadupib t3, v1
-loadupib t4, v2
-avgub t3, t3, t4
-subb t3, t3, c128
-convsbw wv, t3
-
-mullw t1, wy, 42
-shrsw t1, t1, c8
-addssw wy, wy, t1
-
-addssw wr, wy, wv
-mullw t1, wv, 103
-shrsw t1, t1, c8
-subssw wr, wr, t1
-addssw wr, wr, wv
-
-addssw wb, wy, wu
-addssw wb, wb, wu
-mullw t1, wu, 4
-shrsw t1, t1, c8
-addssw wb, wb, t1
-
-mullw t1, wu, 100
-shrsw t1, t1, c8
-subssw wg, wy, t1
-mullw t1, wv, 104
-shrsw t1, t1, c8
-subssw wg, wg, t1
-subssw wg, wg, t1
-
-convssswb r, wr
-convssswb g, wg
-convssswb b, wb
-
-mergebw t1, b, g
-mergebw t2, r, 255
-mergewl x, t1, t2
-x4 addb argb, x, c128
+mergewl argb, t1, t2
 
 
 
diff --git a/gst/videoconvert/videoconvert.c b/gst/videoconvert/videoconvert.c
index 28e3ae60b0..b1a36834e7 100644
--- a/gst/videoconvert/videoconvert.c
+++ b/gst/videoconvert/videoconvert.c
@@ -1246,33 +1246,13 @@ convert_I420_BGRA (VideoConvert * convert, GstVideoFrame * dest,
     const GstVideoFrame * src)
 {
   int i;
-  int quality = 0;
   gint width = convert->width;
   gint height = convert->height;
 
-  if (quality > 3) {
-    for (i = 0; i < height; i++) {
-      if (i & 1) {
-        video_convert_orc_convert_I420_BGRA_avg (FRAME_GET_LINE (dest, i),
-            FRAME_GET_Y_LINE (src, i),
-            FRAME_GET_U_LINE (src, i >> 1),
-            FRAME_GET_U_LINE (src, (i >> 1) + 1),
-            FRAME_GET_V_LINE (src, i >> 1),
-            FRAME_GET_V_LINE (src, (i >> 1) + 1), width);
-      } else {
-        video_convert_orc_convert_I420_BGRA (FRAME_GET_LINE (dest, i),
-            FRAME_GET_Y_LINE (src, i),
-            FRAME_GET_U_LINE (src, i >> 1),
-            FRAME_GET_V_LINE (src, i >> 1), width);
-      }
-    }
-  } else {
-    for (i = 0; i < height; i++) {
-      video_convert_orc_convert_I420_BGRA (FRAME_GET_LINE (dest, i),
-          FRAME_GET_Y_LINE (src, i),
-          FRAME_GET_U_LINE (src, i >> 1),
-          FRAME_GET_V_LINE (src, i >> 1), width);
-    }
+  for (i = 0; i < height; i++) {
+    video_convert_orc_convert_I420_BGRA (FRAME_GET_LINE (dest, i),
+        FRAME_GET_Y_LINE (src, i),
+        FRAME_GET_U_LINE (src, i >> 1), FRAME_GET_V_LINE (src, i >> 1), width);
   }
 }
 #endif