From 25442736e52662be97d93cbde71a713f6a3b8a60 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Sebastian=20Dr=C3=B6ge?= <slomo@circular-chaos.org>
Date: Fri, 20 Jun 2008 14:48:40 +0000
Subject: [PATCH] gst/deinterlace2/tvtime/vfir.c: Make it possible to use the
 vfir method on X86 CPUs without MMXEXT too but use the MM...

Original commit message from CVS:
* gst/deinterlace2/tvtime/vfir.c: (deinterlace_line_mmxext),
(deinterlace_line_c), (deinterlace_scanline_vfir):
Make it possible to use the vfir method on X86 CPUs without MMXEXT too
but use the MMXEXT optimized code whenever possible.
---
 ChangeLog                      |  7 ++++++
 gst/deinterlace2/tvtime/vfir.c | 41 +++++++++++++++++++++-------------
 2 files changed, 33 insertions(+), 15 deletions(-)

diff --git a/ChangeLog b/ChangeLog
index 3eea2871e0..59abd0de9c 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,10 @@
+2008-06-20  Sebastian Dröge  <sebastian.droege@collabora.co.uk>
+
+	* gst/deinterlace2/tvtime/vfir.c: (deinterlace_line_mmxext),
+	(deinterlace_line_c), (deinterlace_scanline_vfir):
+	Make it possible to use the vfir method on X86 CPUs without MMXEXT too
+	but use the MMXEXT optimized code whenever possible.
+
 2008-06-20  Sebastian Dröge  <sebastian.droege@collabora.co.uk>
 
 	* gst/deinterlace2/gstdeinterlace2.c:
diff --git a/gst/deinterlace2/tvtime/vfir.c b/gst/deinterlace2/tvtime/vfir.c
index e9b57b96f3..f32be65475 100644
--- a/gst/deinterlace2/tvtime/vfir.c
+++ b/gst/deinterlace2/tvtime/vfir.c
@@ -39,7 +39,6 @@
 # include "config.h"
 #endif
 
-#include "mmx.h"
 #include "speedy.h"
 #include "gstdeinterlace2.h"
 
@@ -50,12 +49,13 @@
  * filter taps here are: [-1 4 2 4 -1].
  */
 
+#ifdef HAVE_CPU_I386
+#include "mmx.h"
 static void
-deinterlace_line (uint8_t * dst, uint8_t * lum_m4,
+deinterlace_line_mmxext (uint8_t * dst, uint8_t * lum_m4,
     uint8_t * lum_m3, uint8_t * lum_m2,
     uint8_t * lum_m1, uint8_t * lum, int size)
 {
-#ifdef HAVE_CPU_I386
   mmx_t rounder;
 
   rounder.uw[0] = 4;
@@ -94,10 +94,17 @@ deinterlace_line (uint8_t * dst, uint8_t * lum_m4,
     dst += 4;
   }
   emms ();
-#else
-    /**
-     * C implementation.
-     */
+}
+#endif
+
+/**
+  * C implementation.
+  */
+static void
+deinterlace_line_c (uint8_t * dst, uint8_t * lum_m4,
+    uint8_t * lum_m3, uint8_t * lum_m2,
+    uint8_t * lum_m1, uint8_t * lum, int size)
+{
   int sum;
 
   for (; size > 0; size--) {
@@ -114,10 +121,8 @@ deinterlace_line (uint8_t * dst, uint8_t * lum_m4,
     lum++;
     dst++;
   }
-#endif
 }
 
-
 /*
  * The commented-out method below that uses the bottom_field member is more
  * like the filter as specified in the MPEG2 spec, but it doesn't seem to
@@ -128,8 +133,18 @@ static void
 deinterlace_scanline_vfir (GstDeinterlace2 * object,
     deinterlace_scanline_data_t * data, uint8_t * output)
 {
-  deinterlace_line (output, data->tt1, data->t0, data->m1, data->b0, data->bb1,
-      object->frame_width * 2);
+#ifdef HAVE_CPU_I386
+  if (object->cpu_feature_flags & OIL_IMPL_FLAG_MMXEXT) {
+    deinterlace_line_mmxext (output, data->tt1, data->t0, data->m1, data->b0,
+        data->bb1, object->frame_width * 2);
+  } else {
+    deinterlace_line_c (output, data->tt1, data->t0, data->m1, data->b0,
+        data->bb1, object->frame_width * 2);
+  }
+#else
+  deinterlace_line_c (output, data->tt1, data->t0, data->m1, data->b0,
+      data->bb1, object->frame_width * 2);
+#endif
   // blit_packed422_scanline( output, data->m1, width );
 }
 
@@ -153,11 +168,7 @@ static deinterlace_method_t vfirmethod = {
   "Blur: Vertical",
   "BlurVertical",
   2,
-#ifdef HAVE_CPU_I386
-  OIL_IMPL_FLAG_MMXEXT,
-#else
   0,
-#endif
   0,
   0,
   0,