bayer2rgb: Convert to Orc

Seriously faster.  Algorithm is nearly the same as bilinear, which
given the speed of this code, should be considered the baseline of
quality.  Speed appears to be limited by memory bandwidth, so I
didn't bother trying to make it any faster.

5 files changed, 3638 insertions, 258 deletions

diff --git a/gst/bayer/Makefile.am b/gst/bayer/Makefile.am
index c62ab99a2..69666ea2f 100644
--- a/gst/bayer/Makefile.am
+++ b/gst/bayer/Makefile.am
@@ -1,16 +1,23 @@
 plugin_LTLIBRARIES = libgstbayer.la
 
+ORC_SOURCE=gstbayerorc
+include $(top_srcdir)/common/orc.mak
+
 libgstbayer_la_SOURCES = \
 	gstbayer.c \
 	gstbayer2rgb.c \
 	gstrgb2bayer.c \
 	gstrgb2bayer.h
 libgstbayer_la_CFLAGS = $(GST_PLUGINS_BASE_CFLAGS) $(GST_BASE_CFLAGS) \
+    $(ORC_CFLAGS) \
     $(GST_CFLAGS)
 libgstbayer_la_LIBADD = $(GST_PLUGINS_BASE_LIBS) -lgstvideo-$(GST_MAJORMINOR) \
+    $(ORC_LIBS) \
     $(GST_BASE_LIBS)
 libgstbayer_la_LDFLAGS = $(GST_PLUGIN_LDFLAGS)
 libgstbayer_la_LIBTOOLFLAGS = --tag=disable-static
+nodist_libgstbayer_la_SOURCES = $(ORC_NODIST_SOURCES)
+
 
 Android.mk: Makefile.am $(BUILT_SOURCES)
 	androgenizer \
diff --git a/gst/bayer/gstbayer2rgb.c b/gst/bayer/gstbayer2rgb.c
index 5e4dc9aa7..8f9ed124d 100644
--- a/gst/bayer/gstbayer2rgb.c
+++ b/gst/bayer/gstbayer2rgb.c
@@ -80,7 +80,8 @@
 #include <gst/video/video.h>
 #include <string.h>
 #include <stdlib.h>
-#include "_stdint.h"
+#include <_stdint.h>
+#include "gstbayerorc.h"
 
 #define GST_CAT_DEFAULT gst_bayer2rgb_debug
 GST_DEBUG_CATEGORY_STATIC (GST_CAT_DEFAULT);
@@ -125,7 +126,6 @@ struct _GstBayer2RGBClass
   GstBaseTransformClass parent;
 };
 
-//#define SRC_CAPS GST_VIDEO_CAPS_RGBx
 #define	SRC_CAPS                                 \
   GST_VIDEO_CAPS_RGBx ";"                        \
   GST_VIDEO_CAPS_xRGB ";"                        \
@@ -134,9 +134,7 @@ struct _GstBayer2RGBClass
   GST_VIDEO_CAPS_RGBA ";"                        \
   GST_VIDEO_CAPS_ARGB ";"                        \
   GST_VIDEO_CAPS_BGRA ";"                        \
-  GST_VIDEO_CAPS_ABGR ";"                        \
-  GST_VIDEO_CAPS_RGB ";"                         \
-  GST_VIDEO_CAPS_BGR
+  GST_VIDEO_CAPS_ABGR
 
 #define SINK_CAPS "video/x-raw-bayer,format=(string){bggr,grbg,gbrg,rggb}," \
   "width=(int)[1,MAX],height=(int)[1,MAX],framerate=(fraction)[0/1,MAX]"
@@ -365,7 +363,6 @@ gst_bayer2rgb_get_unit_size (GstBaseTransform * base, GstCaps * caps,
     name = gst_structure_get_name (structure);
     /* Our name must be either video/x-raw-bayer video/x-raw-rgb */
     if (strcmp (name, "video/x-raw-rgb")) {
-      /* For bayer, we handle only BA81 (BGGR), which is BPP=24 */
       *size = GST_ROUND_UP_4 (width) * height;
       return TRUE;
     } else {
@@ -382,282 +379,108 @@ gst_bayer2rgb_get_unit_size (GstBaseTransform * base, GstCaps * caps,
   return FALSE;
 }
 
-#define RECONSTRUCT_SQUARE(x) \
-  do { \
-    int _h1 = next[i-1]; \
-    int _h2 = prev[i+1]; \
-    int _v1 = next[i+1]; \
-    int _v2 = prev[i-1]; \
-    (x) = (_h1+_h2+_v1+_v2+2)>>2; \
-  } while (0)
-
-#define RECONSTRUCT_DIAMOND(x) \
-  do { \
-    int _h1 = src[i-1]; \
-    int _h2 = src[i+1]; \
-    int _v1 = next[i]; \
-    int _v2 = prev[i]; \
-    (x) = (_h1+_h2+_v1+_v2+2)>>2; \
-  } while (0)
-
-#define RECONSTRUCT_HORIZ(x) \
-  do { \
-    (x) = (src[i-1] + src[i+1] + 1) >> 1; \
-  } while (0)
-
-#define RECONSTRUCT_VERT(x) \
-  do { \
-    (x) = (next[i] + prev[i] + 1) >> 1; \
-  } while (0)
-
-
-static void
-reconstruct_blue_green (GstBayer2RGB * bayer2rgb, uint8_t * dest,
-    uint8_t * src, int src_stride, int blue_loc)
-{
-  int i;
-  int r, g, b;
-  uint8_t *prev;
-  uint8_t *next;
-  int width = bayer2rgb->width;
-
-  prev = src - src_stride;
-  next = src + src_stride;
-
-  i = 0;
-  if ((i & 1) == blue_loc) {
-    b = src[i];
-    r = (next[i + 1] + prev[i + 1] + 1) >> 1;
-    g = (next[i] + prev[i] + 1) >> 1;
-  } else {
-    b = src[i + 1];
-    r = (next[i] + prev[i] + 1) >> 1;
-    g = src[i];
-  }
-  dest[i * 4 + bayer2rgb->r_off] = r;
-  dest[i * 4 + bayer2rgb->g_off] = g;
-  dest[i * 4 + bayer2rgb->b_off] = b;
-  for (i = 1; i < width - 1; i++) {
-    if ((i & 1) == blue_loc) {
-      b = src[i];
-      RECONSTRUCT_SQUARE (r);
-      RECONSTRUCT_DIAMOND (g);
-    } else {
-      RECONSTRUCT_HORIZ (b);
-      RECONSTRUCT_VERT (r);
-      g = src[i];
-    }
-    dest[i * 4 + bayer2rgb->r_off] = r;
-    dest[i * 4 + bayer2rgb->g_off] = g;
-    dest[i * 4 + bayer2rgb->b_off] = b;
-  }
-  if ((i & 1) == blue_loc) {
-    b = src[i];
-    r = (next[i - 1] + prev[i - 1] + 1) >> 1;
-    g = (next[i] + prev[i] + 1) >> 1;
-  } else {
-    b = src[i - 1];
-    r = (next[i] + prev[i] + 1) >> 1;
-    g = src[i];
-  }
-  dest[i * 4 + bayer2rgb->r_off] = r;
-  dest[i * 4 + bayer2rgb->g_off] = g;
-  dest[i * 4 + bayer2rgb->b_off] = b;
-}
-
 static void
-reconstruct_green_red (GstBayer2RGB * bayer2rgb, uint8_t * dest,
-    uint8_t * src, int src_stride, int red_loc)
+gst_bayer2rgb_split_and_upsample_horiz (guint8 * dest0, guint8 * dest1,
+    const guint8 * src, int n)
 {
   int i;
-  int r, g, b;
-  uint8_t *prev;
-  uint8_t *next;
-  int width = bayer2rgb->width;
-
-  prev = src - src_stride;
-  next = src + src_stride;
-
-  i = 0;
-  if ((i & 1) == red_loc) {
-    r = src[i];
-    b = (next[i + 1] + prev[i + 1] + 1) >> 1;
-    g = (next[i] + prev[i] + 1) >> 1;
-  } else {
-    r = src[i + 1];
-    b = (next[i] + prev[i] + 1) >> 1;
-    g = src[i];
-  }
-  dest[i * 4 + bayer2rgb->r_off] = r;
-  dest[i * 4 + bayer2rgb->g_off] = g;
-  dest[i * 4 + bayer2rgb->b_off] = b;
-  for (i = 1; i < width - 1; i++) {
-    if ((i & 1) == red_loc) {
-      r = src[i];
-      RECONSTRUCT_SQUARE (b);
-      RECONSTRUCT_DIAMOND (g);
-    } else {
-      RECONSTRUCT_HORIZ (r);
-      RECONSTRUCT_VERT (b);
-      g = src[i];
-    }
-    dest[i * 4 + bayer2rgb->r_off] = r;
-    dest[i * 4 + bayer2rgb->g_off] = g;
-    dest[i * 4 + bayer2rgb->b_off] = b;
-  }
-  if ((i & 1) == red_loc) {
-    r = src[i];
-    b = (next[i - 1] + prev[i - 1] + 1) >> 1;
-    g = (next[i] + prev[i] + 1) >> 1;
-  } else {
-    r = src[i - 1];
-    b = (next[i] + prev[i] + 1) >> 1;
-    g = src[i];
-  }
-  dest[i * 4 + bayer2rgb->r_off] = r;
-  dest[i * 4 + bayer2rgb->g_off] = g;
-  dest[i * 4 + bayer2rgb->b_off] = b;
-}
 
-static void
-reconstruct_blue_green_edge (GstBayer2RGB * bayer2rgb, uint8_t * dest,
-    uint8_t * src, int src_stride, int blue_loc, int offset)
-{
-  int i;
-  int r, g, b;
-  uint8_t *next;
-  int width = bayer2rgb->width;
+  dest0[0] = src[0];
+  dest1[0] = src[1];
+  dest0[1] = (src[0] + src[2] + 1) >> 1;
+  dest1[1] = src[1];
 
-  next = src + offset * src_stride;
+#if defined(__i386__) || defined(__amd64__)
+  gst_bayer_horiz_upsample_unaligned (dest0 + 2, dest1 + 2, src + 1,
+      (n - 4) >> 1);
+#else
+  gst_bayer_horiz_upsample (dest0 + 2, dest1 + 2, src + 2, (n - 4) >> 1);
+#endif
 
-  i = 0;
-  if ((i & 1) == blue_loc) {
-    b = src[i];
-    r = next[i + 1];
-    g = next[i];
-  } else {
-    b = src[i + 1];
-    r = next[i];
-    g = src[i];
-  }
-  dest[i * 4 + bayer2rgb->r_off] = r;
-  dest[i * 4 + bayer2rgb->g_off] = g;
-  dest[i * 4 + bayer2rgb->b_off] = b;
-  for (i = 1; i < width - 1; i++) {
-    if ((i & 1) == blue_loc) {
-      b = src[i];
-      r = (next[i - 1] + next[i + 1] + 1) >> 1;
-      g = (src[i - 1] + src[i + 1] + 1) >> 1;
+  for (i = n - 2; i < n; i++) {
+    if ((i & 1) == 0) {
+      dest0[i] = src[i];
+      dest1[i] = src[i - 1];
     } else {
-      b = (src[i - 1] + src[i + 1] + 1) >> 1;
-      r = next[i];
-      g = src[i];
+      dest0[i] = src[i - 1];
+      dest1[i] = src[i];
     }
-    dest[i * 4 + bayer2rgb->r_off] = r;
-    dest[i * 4 + bayer2rgb->g_off] = g;
-    dest[i * 4 + bayer2rgb->b_off] = b;
   }
-  if ((i & 1) == blue_loc) {
-    b = src[i];
-    r = next[i - 1];
-    g = next[i];
-  } else {
-    b = src[i - 1];
-    r = next[i];
-    g = src[i];
-  }
-  dest[i * 4 + bayer2rgb->r_off] = r;
-  dest[i * 4 + bayer2rgb->g_off] = g;
-  dest[i * 4 + bayer2rgb->b_off] = b;
 }
 
+typedef void (*process_func) (guint8 * d0, const guint8 * s0, const guint8 * s1,
+    const guint8 * s2, const guint8 * s3, const guint8 * s4, const guint8 * s5,
+    int n);
+
 static void
-reconstruct_green_red_edge (GstBayer2RGB * bayer2rgb, uint8_t * dest,
-    uint8_t * src, int src_stride, int red_loc, int offset)
+gst_bayer2rgb_process (GstBayer2RGB * bayer2rgb, uint8_t * dest,
+    int dest_stride, uint8_t * src, int src_stride)
 {
-  int i;
-  int r, g, b;
-  uint8_t *next;
-  int width = bayer2rgb->width;
-
-  next = src + offset * src_stride;
-
-  i = 0;
-  if ((i & 1) == red_loc) {
-    r = src[i];
-    b = next[i + 1];
-    g = next[i];
-  } else {
-    r = src[i + 1];
-    b = next[i];
-    g = src[i];
+  int j;
+  guint8 *tmp;
+  process_func merge[2] = { NULL, NULL };
+  int r_off, g_off, b_off;
+
+  /* We exploit some symmetry in the functions here.  The base functions
+   * are all named for the BGGR arrangement.  For RGGB, we swap the
+   * red offset and blue offset in the output.  For GRBG, we swap the
+   * order of the merge functions.  For GBRG, do both. */
+  r_off = bayer2rgb->r_off;
+  g_off = bayer2rgb->g_off;
+  b_off = bayer2rgb->b_off;
+  if (bayer2rgb->format == GST_BAYER_2_RGB_FORMAT_RGGB ||
+      bayer2rgb->format == GST_BAYER_2_RGB_FORMAT_GBRG) {
+    r_off = bayer2rgb->b_off;
+    b_off = bayer2rgb->r_off;
   }
-  dest[i * 4 + bayer2rgb->r_off] = r;
-  dest[i * 4 + bayer2rgb->g_off] = g;
-  dest[i * 4 + bayer2rgb->b_off] = b;
-  for (i = 1; i < width - 1; i++) {
-    if ((i & 1) == red_loc) {
-      r = src[i];
-      b = (next[i - 1] + next[i + 1] + 1) >> 1;
-      g = (src[i - 1] + src[i + 1] + 1) >> 1;
-    } else {
-      r = (src[i - 1] + src[i + 1] + 1) >> 1;
-      b = next[i];
-      g = src[i];
-    }
-    dest[i * 4 + bayer2rgb->r_off] = r;
-    dest[i * 4 + bayer2rgb->g_off] = g;
-    dest[i * 4 + bayer2rgb->b_off] = b;
+
+  if (r_off == 2 && g_off == 1 && b_off == 0) {
+    merge[0] = gst_bayer_merge_bg_bgra;
+    merge[1] = gst_bayer_merge_gr_bgra;
+  } else if (r_off == 3 && g_off == 2 && b_off == 1) {
+    merge[0] = gst_bayer_merge_bg_abgr;
+    merge[1] = gst_bayer_merge_gr_abgr;
+  } else if (r_off == 1 && g_off == 2 && b_off == 3) {
+    merge[0] = gst_bayer_merge_bg_argb;
+    merge[1] = gst_bayer_merge_gr_argb;
+  } else if (r_off == 0 && g_off == 1 && b_off == 2) {
+    merge[0] = gst_bayer_merge_bg_rgba;
+    merge[1] = gst_bayer_merge_gr_rgba;
   }
-  if ((i & 1) == red_loc) {
-    r = src[i];
-    b = next[i - 1];
-    g = next[i];
-  } else {
-    r = src[i - 1];
-    b = next[i];
-    g = src[i];
+  if (bayer2rgb->format == GST_BAYER_2_RGB_FORMAT_GRBG ||
+      bayer2rgb->format == GST_BAYER_2_RGB_FORMAT_GBRG) {
+    process_func tmp = merge[0];
+    merge[0] = merge[1];
+    merge[1] = tmp;
   }
-  dest[i * 4 + bayer2rgb->r_off] = r;
-  dest[i * 4 + bayer2rgb->g_off] = g;
-  dest[i * 4 + bayer2rgb->b_off] = b;
-}
 
-static void
-gst_bayer2rgb_process_ref (GstBayer2RGB * bayer2rgb, uint8_t * dest,
-    int dest_stride, uint8_t * src, int src_stride)
-{
-  int j;
-  int format = bayer2rgb->format;
+  tmp = g_malloc (2 * 4 * bayer2rgb->width);
+#define LINE(x) (tmp + ((x)&7) * bayer2rgb->width)
 
+  gst_bayer2rgb_split_and_upsample_horiz (LINE (3 * 2 + 0), LINE (3 * 2 + 1),
+      src + 1 * src_stride, bayer2rgb->width);
   j = 0;
-  if ((j & 1) == (format & 2) >> 1) {
-    reconstruct_blue_green_edge (bayer2rgb, dest + j * dest_stride,
-        src + j * src_stride, src_stride, format & 1, 1);
-  } else {
-    reconstruct_green_red_edge (bayer2rgb, dest + j * dest_stride,
-        src + j * src_stride, src_stride, (format & 1) ^ 1, 1);
-  }
-  for (j = 1; j < bayer2rgb->height - 1; j++) {
-    if ((j & 1) == (format & 2) >> 1) {
-      reconstruct_blue_green (bayer2rgb, dest + j * dest_stride,
-          src + j * src_stride, src_stride, format & 1);
-    } else {
-      reconstruct_green_red (bayer2rgb, dest + j * dest_stride,
-          src + j * src_stride, src_stride, (format & 1) ^ 1);
+  gst_bayer2rgb_split_and_upsample_horiz (LINE (j * 2 + 0), LINE (j * 2 + 1),
+      src + j * src_stride, bayer2rgb->width);
+
+  for (j = 0; j < bayer2rgb->height; j++) {
+    if (j < bayer2rgb->height - 1) {
+      gst_bayer2rgb_split_and_upsample_horiz (LINE ((j + 1) * 2 + 0),
+          LINE ((j + 1) * 2 + 1), src + (j + 1) * src_stride, bayer2rgb->width);
     }
-  }
-  if ((j & 1) == (format & 2) >> 1) {
-    reconstruct_blue_green_edge (bayer2rgb, dest + j * dest_stride,
-        src + j * src_stride, src_stride, format & 1, -1);
-  } else {
-    reconstruct_green_red_edge (bayer2rgb, dest + j * dest_stride,
-        src + j * src_stride, src_stride, (format & 1) ^ 1, -1);
+
+    merge[j & 1] (dest + j * dest_stride,
+        LINE (j * 2 - 2), LINE (j * 2 - 1),
+        LINE (j * 2 + 0), LINE (j * 2 + 1),
+        LINE (j * 2 + 2), LINE (j * 2 + 3), bayer2rgb->width >> 1);
   }
 
+  g_free (tmp);
 }
 
 
+
+
 static GstFlowReturn
 gst_bayer2rgb_transform (GstBaseTransform * base, GstBuffer * inbuf,
     GstBuffer * outbuf)
@@ -675,7 +498,7 @@ gst_bayer2rgb_transform (GstBaseTransform * base, GstBuffer * inbuf,
   GST_DEBUG ("transforming buffer");
   input = (uint8_t *) GST_BUFFER_DATA (inbuf);
   output = (uint8_t *) GST_BUFFER_DATA (outbuf);
-  gst_bayer2rgb_process_ref (filter, output, filter->width * 4,
+  gst_bayer2rgb_process (filter, output, filter->width * 4,
       input, filter->width);
 
   GST_OBJECT_UNLOCK (filter);
diff --git a/gst/bayer/gstbayerorc-dist.c b/gst/bayer/gstbayerorc-dist.c
new file mode 100644
index 000000000..a8bc11716
--- /dev/null
+++ b/gst/bayer/gstbayerorc-dist.c
@@ -0,0 +1,3212 @@
+
+/* autogenerated from gstbayerorc.orc */
+
+#ifdef HAVE_CONFIG_H
+#include "config.h"
+#endif
+#include <glib.h>
+
+#ifndef _ORC_INTEGER_TYPEDEFS_
+#define _ORC_INTEGER_TYPEDEFS_
+#if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
+#include <stdint.h>
+typedef int8_t orc_int8;
+typedef int16_t orc_int16;
+typedef int32_t orc_int32;
+typedef int64_t orc_int64;
+typedef uint8_t orc_uint8;
+typedef uint16_t orc_uint16;
+typedef uint32_t orc_uint32;
+typedef uint64_t orc_uint64;
+#define ORC_UINT64_C(x) UINT64_C(x)
+#elif defined(_MSC_VER)
+typedef signed __int8 orc_int8;
+typedef signed __int16 orc_int16;
+typedef signed __int32 orc_int32;
+typedef signed __int64 orc_int64;
+typedef unsigned __int8 orc_uint8;
+typedef unsigned __int16 orc_uint16;
+typedef unsigned __int32 orc_uint32;
+typedef unsigned __int64 orc_uint64;
+#define ORC_UINT64_C(x) (x##Ui64)
+#define inline __inline
+#else
+#include <limits.h>
+typedef signed char orc_int8;
+typedef short orc_int16;
+typedef int orc_int32;
+typedef unsigned char orc_uint8;
+typedef unsigned short orc_uint16;
+typedef unsigned int orc_uint32;
+#if INT_MAX == LONG_MAX
+typedef long long orc_int64;
+typedef unsigned long long orc_uint64;
+#define ORC_UINT64_C(x) (x##ULL)
+#else
+typedef long orc_int64;
+typedef unsigned long orc_uint64;
+#define ORC_UINT64_C(x) (x##UL)
+#endif
+#endif
+typedef union
+{
+  orc_int16 i;
+  orc_int8 x2[2];
+} orc_union16;
+typedef union
+{
+  orc_int32 i;
+  float f;
+  orc_int16 x2[2];
+  orc_int8 x4[4];
+} orc_union32;
+typedef union
+{
+  orc_int64 i;
+  double f;
+  orc_int32 x2[2];
+  float x2f[2];
+  orc_int16 x4[4];
+} orc_union64;
+#endif
+#ifndef ORC_RESTRICT
+#if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
+#define ORC_RESTRICT restrict
+#elif defined(__GNUC__) && __GNUC__ >= 4
+#define ORC_RESTRICT __restrict__
+#else
+#define ORC_RESTRICT
+#endif
+#endif
+
+#ifndef DISABLE_ORC
+#include <orc/orc.h>
+#endif
+void gst_bayer_horiz_upsample_unaligned (guint8 * ORC_RESTRICT d1,
+    guint8 * ORC_RESTRICT d2, const guint8 * ORC_RESTRICT s1, int n);
+void gst_bayer_horiz_upsample (guint8 * ORC_RESTRICT d1,
+    guint8 * ORC_RESTRICT d2, const guint8 * ORC_RESTRICT s1, int n);
+void gst_bayer_merge_bg_bgra (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_gr_bgra (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_bg_abgr (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_gr_abgr (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_bg_rgba (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_gr_rgba (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_bg_argb (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_gr_argb (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+
+
+/* begin Orc C target preamble */
+#define ORC_CLAMP(x,a,b) ((x)<(a) ? (a) : ((x)>(b) ? (b) : (x)))
+#define ORC_ABS(a) ((a)<0 ? -(a) : (a))
+#define ORC_MIN(a,b) ((a)<(b) ? (a) : (b))
+#define ORC_MAX(a,b) ((a)>(b) ? (a) : (b))
+#define ORC_SB_MAX 127
+#define ORC_SB_MIN (-1-ORC_SB_MAX)
+#define ORC_UB_MAX 255
+#define ORC_UB_MIN 0
+#define ORC_SW_MAX 32767
+#define ORC_SW_MIN (-1-ORC_SW_MAX)
+#define ORC_UW_MAX 65535
+#define ORC_UW_MIN 0
+#define ORC_SL_MAX 2147483647
+#define ORC_SL_MIN (-1-ORC_SL_MAX)
+#define ORC_UL_MAX 4294967295U
+#define ORC_UL_MIN 0
+#define ORC_CLAMP_SB(x) ORC_CLAMP(x,ORC_SB_MIN,ORC_SB_MAX)
+#define ORC_CLAMP_UB(x) ORC_CLAMP(x,ORC_UB_MIN,ORC_UB_MAX)
+#define ORC_CLAMP_SW(x) ORC_CLAMP(x,ORC_SW_MIN,ORC_SW_MAX)
+#define ORC_CLAMP_UW(x) ORC_CLAMP(x,ORC_UW_MIN,ORC_UW_MAX)
+#define ORC_CLAMP_SL(x) ORC_CLAMP(x,ORC_SL_MIN,ORC_SL_MAX)
+#define ORC_CLAMP_UL(x) ORC_CLAMP(x,ORC_UL_MIN,ORC_UL_MAX)
+#define ORC_SWAP_W(x) ((((x)&0xff)<<8) | (((x)&0xff00)>>8))
+#define ORC_SWAP_L(x) ((((x)&0xff)<<24) | (((x)&0xff00)<<8) | (((x)&0xff0000)>>8) | (((x)&0xff000000)>>24))
+#define ORC_SWAP_Q(x) ((((x)&ORC_UINT64_C(0xff))<<56) | (((x)&ORC_UINT64_C(0xff00))<<40) | (((x)&ORC_UINT64_C(0xff0000))<<24) | (((x)&ORC_UINT64_C(0xff000000))<<8) | (((x)&ORC_UINT64_C(0xff00000000))>>8) | (((x)&ORC_UINT64_C(0xff0000000000))>>24) | (((x)&ORC_UINT64_C(0xff000000000000))>>40) | (((x)&ORC_UINT64_C(0xff00000000000000))>>56))
+#define ORC_PTR_OFFSET(ptr,offset) ((void *)(((unsigned char *)(ptr)) + (offset)))
+#define ORC_DENORMAL(x) ((x) & ((((x)&0x7f800000) == 0) ? 0xff800000 : 0xffffffff))
+#define ORC_ISNAN(x) ((((x)&0x7f800000) == 0x7f800000) && (((x)&0x007fffff) != 0))
+#define ORC_DENORMAL_DOUBLE(x) ((x) & ((((x)&ORC_UINT64_C(0x7ff0000000000000)) == 0) ? ORC_UINT64_C(0xfff0000000000000) : ORC_UINT64_C(0xffffffffffffffff)))
+#define ORC_ISNAN_DOUBLE(x) ((((x)&ORC_UINT64_C(0x7ff0000000000000)) == ORC_UINT64_C(0x7ff0000000000000)) && (((x)&ORC_UINT64_C(0x000fffffffffffff)) != 0))
+#ifndef ORC_RESTRICT
+#if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
+#define ORC_RESTRICT restrict
+#elif defined(__GNUC__) && __GNUC__ >= 4
+#define ORC_RESTRICT __restrict__
+#else
+#define ORC_RESTRICT
+#endif
+#endif
+/* end Orc C target preamble */
+
+
+
+/* gst_bayer_horiz_upsample_unaligned */
+#ifdef DISABLE_ORC
+void
+gst_bayer_horiz_upsample_unaligned (guint8 * ORC_RESTRICT d1,
+    guint8 * ORC_RESTRICT d2, const guint8 * ORC_RESTRICT s1, int n)
+{
+  int i;
+  orc_union16 *ORC_RESTRICT ptr0;
+  orc_union16 *ORC_RESTRICT ptr1;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_int8 var40;
+  orc_int8 var41;
+  orc_union16 var42;
+  orc_int8 var43;
+  orc_int8 var44;
+  orc_int8 var45;
+  orc_int8 var46;
+
+  ptr0 = (orc_union16 *) d1;
+  ptr1 = (orc_union16 *) d2;
+  ptr4 = (orc_union16 *) s1;
+
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var37.i;
+      var40 = _src.x2[1];
+      var41 = _src.x2[0];
+    }
+    /* 2: loadoffw */
+    var42 = ptr4[i + 1];
+    /* 3: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var42.i;
+      var43 = _src.x2[1];
+      var44 = _src.x2[0];
+    }
+    /* 4: avgub */
+    var45 = ((orc_uint8) var40 + (orc_uint8) var43 + 1) >> 1;
+    /* 5: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var40;
+      _dest.x2[1] = var45;
+      var38.i = _dest.i;
+    }
+    /* 6: storew */
+    ptr0[i] = var38;
+    /* 7: avgub */
+    var46 = ((orc_uint8) var41 + (orc_uint8) var44 + 1) >> 1;
+    /* 8: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var46;
+      _dest.x2[1] = var44;
+      var39.i = _dest.i;
+    }
+    /* 9: storew */
+    ptr1[i] = var39;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_horiz_upsample_unaligned (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union16 *ORC_RESTRICT ptr0;
+  orc_union16 *ORC_RESTRICT ptr1;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_int8 var40;
+  orc_int8 var41;
+  orc_union16 var42;
+  orc_int8 var43;
+  orc_int8 var44;
+  orc_int8 var45;
+  orc_int8 var46;
+
+  ptr0 = (orc_union16 *) ex->arrays[0];
+  ptr1 = (orc_union16 *) ex->arrays[1];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var37.i;
+      var40 = _src.x2[1];
+      var41 = _src.x2[0];
+    }
+    /* 2: loadoffw */
+    var42 = ptr4[i + 1];
+    /* 3: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var42.i;
+      var43 = _src.x2[1];
+      var44 = _src.x2[0];
+    }
+    /* 4: avgub */
+    var45 = ((orc_uint8) var40 + (orc_uint8) var43 + 1) >> 1;
+    /* 5: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var40;
+      _dest.x2[1] = var45;
+      var38.i = _dest.i;
+    }
+    /* 6: storew */
+    ptr0[i] = var38;
+    /* 7: avgub */
+    var46 = ((orc_uint8) var41 + (orc_uint8) var44 + 1) >> 1;
+    /* 8: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var46;
+      _dest.x2[1] = var44;
+      var39.i = _dest.i;
+    }
+    /* 9: storew */
+    ptr1[i] = var39;
+  }
+
+}
+
+void
+gst_bayer_horiz_upsample_unaligned (guint8 * ORC_RESTRICT d1,
+    guint8 * ORC_RESTRICT d2, const guint8 * ORC_RESTRICT s1, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_horiz_upsample_unaligned");
+      orc_program_set_backup_function (p,
+          _backup_gst_bayer_horiz_upsample_unaligned);
+      orc_program_add_destination (p, 2, "d1");
+      orc_program_add_destination (p, 2, "d2");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_constant (p, 4, 0x00000001, "c1");
+      orc_program_add_temporary (p, 2, "t1");
+      orc_program_add_temporary (p, 1, "t2");
+      orc_program_add_temporary (p, 1, "t3");
+      orc_program_add_temporary (p, 1, "t4");
+      orc_program_add_temporary (p, 1, "t5");
+
+      orc_program_append_2 (p, "splitwb", 0, ORC_VAR_T3, ORC_VAR_T2, ORC_VAR_S1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "loadoffw", 0, ORC_VAR_T1, ORC_VAR_S1,
+          ORC_VAR_C1, ORC_VAR_D1);
+      orc_program_append_2 (p, "splitwb", 0, ORC_VAR_T5, ORC_VAR_T4, ORC_VAR_T1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 0, ORC_VAR_T5, ORC_VAR_T3, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 0, ORC_VAR_D1, ORC_VAR_T3, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 0, ORC_VAR_T2, ORC_VAR_T2, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 0, ORC_VAR_D2, ORC_VAR_T2, ORC_VAR_T4,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_D2] = d2;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
+
+
+/* gst_bayer_horiz_upsample */
+#ifdef DISABLE_ORC
+void
+gst_bayer_horiz_upsample (guint8 * ORC_RESTRICT d1, guint8 * ORC_RESTRICT d2,
+    const guint8 * ORC_RESTRICT s1, int n)
+{
+  int i;
+  orc_union16 *ORC_RESTRICT ptr0;
+  orc_union16 *ORC_RESTRICT ptr1;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_int8 var43;
+  orc_int8 var44;
+  orc_int8 var45;
+  orc_int8 var46;
+  orc_union16 var47;
+  orc_int8 var48;
+  orc_int8 var49;
+  orc_int8 var50;
+  orc_int8 var51;
+
+  ptr0 = (orc_union16 *) d1;
+  ptr1 = (orc_union16 *) d2;
+  ptr4 = (orc_union16 *) s1;
+
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadoffw */
+    var42 = ptr4[i + -1];
+    /* 1: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var42.i;
+      var43 = _src.x2[1];
+      var44 = _src.x2[0];
+    }
+    /* 2: loadw */
+    var39 = ptr4[i];
+    /* 3: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var39.i;
+      var45 = _src.x2[1];
+      var46 = _src.x2[0];
+    }
+    /* 4: loadoffw */
+    var47 = ptr4[i + 1];
+    /* 5: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var47.i;
+      var48 = _src.x2[1];
+      var49 = _src.x2[0];
+    }
+    /* 6: avgub */
+    var50 = ((orc_uint8) var46 + (orc_uint8) var49 + 1) >> 1;
+    /* 7: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var46;
+      _dest.x2[1] = var50;
+      var40.i = _dest.i;
+    }
+    /* 8: storew */
+    ptr0[i] = var40;
+    /* 9: avgub */
+    var51 = ((orc_uint8) var43 + (orc_uint8) var45 + 1) >> 1;
+    /* 10: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var51;
+      _dest.x2[1] = var45;
+      var41.i = _dest.i;
+    }
+    /* 11: storew */
+    ptr1[i] = var41;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_horiz_upsample (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union16 *ORC_RESTRICT ptr0;
+  orc_union16 *ORC_RESTRICT ptr1;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_int8 var43;
+  orc_int8 var44;
+  orc_int8 var45;
+  orc_int8 var46;
+  orc_union16 var47;
+  orc_int8 var48;
+  orc_int8 var49;
+  orc_int8 var50;
+  orc_int8 var51;
+
+  ptr0 = (orc_union16 *) ex->arrays[0];
+  ptr1 = (orc_union16 *) ex->arrays[1];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadoffw */
+    var42 = ptr4[i + -1];
+    /* 1: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var42.i;
+      var43 = _src.x2[1];
+      var44 = _src.x2[0];
+    }
+    /* 2: loadw */
+    var39 = ptr4[i];
+    /* 3: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var39.i;
+      var45 = _src.x2[1];
+      var46 = _src.x2[0];
+    }
+    /* 4: loadoffw */
+    var47 = ptr4[i + 1];
+    /* 5: splitwb */
+    {
+      orc_union16 _src;
+      _src.i = var47.i;
+      var48 = _src.x2[1];
+      var49 = _src.x2[0];
+    }
+    /* 6: avgub */
+    var50 = ((orc_uint8) var46 + (orc_uint8) var49 + 1) >> 1;
+    /* 7: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var46;
+      _dest.x2[1] = var50;
+      var40.i = _dest.i;
+    }
+    /* 8: storew */
+    ptr0[i] = var40;
+    /* 9: avgub */
+    var51 = ((orc_uint8) var43 + (orc_uint8) var45 + 1) >> 1;
+    /* 10: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var51;
+      _dest.x2[1] = var45;
+      var41.i = _dest.i;
+    }
+    /* 11: storew */
+    ptr1[i] = var41;
+  }
+
+}
+
+void
+gst_bayer_horiz_upsample (guint8 * ORC_RESTRICT d1, guint8 * ORC_RESTRICT d2,
+    const guint8 * ORC_RESTRICT s1, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_horiz_upsample");
+      orc_program_set_backup_function (p, _backup_gst_bayer_horiz_upsample);
+      orc_program_add_destination (p, 2, "d1");
+      orc_program_add_destination (p, 2, "d2");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_constant (p, 4, 0xffffffff, "c1");
+      orc_program_add_constant (p, 4, 0x00000001, "c2");
+      orc_program_add_temporary (p, 2, "t1");
+      orc_program_add_temporary (p, 1, "t2");
+      orc_program_add_temporary (p, 1, "t3");
+      orc_program_add_temporary (p, 1, "t4");
+      orc_program_add_temporary (p, 1, "t5");
+      orc_program_add_temporary (p, 1, "t6");
+      orc_program_add_temporary (p, 1, "t7");
+
+      orc_program_append_2 (p, "loadoffw", 0, ORC_VAR_T1, ORC_VAR_S1,
+          ORC_VAR_C1, ORC_VAR_D1);
+      orc_program_append_2 (p, "splitwb", 0, ORC_VAR_T3, ORC_VAR_T2, ORC_VAR_T1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "splitwb", 0, ORC_VAR_T5, ORC_VAR_T4, ORC_VAR_S1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "loadoffw", 0, ORC_VAR_T1, ORC_VAR_S1,
+          ORC_VAR_C2, ORC_VAR_D1);
+      orc_program_append_2 (p, "splitwb", 0, ORC_VAR_T7, ORC_VAR_T6, ORC_VAR_T1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 0, ORC_VAR_T6, ORC_VAR_T4, ORC_VAR_T6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 0, ORC_VAR_D1, ORC_VAR_T4, ORC_VAR_T6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 0, ORC_VAR_T3, ORC_VAR_T3, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 0, ORC_VAR_D2, ORC_VAR_T3, ORC_VAR_T5,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_D2] = d2;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
+
+
+/* gst_bayer_merge_bg_bgra */
+#ifdef DISABLE_ORC
+void
+gst_bayer_merge_bg_bgra (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  int i;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) d1;
+  ptr4 = (orc_union16 *) s1;
+  ptr5 = (orc_union16 *) s2;
+  ptr6 = (orc_union16 *) s3;
+  ptr7 = (orc_union16 *) s4;
+  ptr8 = (orc_union16 *) s5;
+  ptr9 = (orc_union16 *) s6;
+
+  /* 9: loadpw */
+  var42.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 11: loadpw */
+  var43.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 16: loadpb */
+  var45.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var45.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr5[i];
+    /* 1: loadw */
+    var38 = ptr9[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr4[i];
+    /* 4: loadw */
+    var40 = ptr8[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr7[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 14: loadw */
+    var44 = ptr6[i];
+    /* 15: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var53.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var53.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_merge_bg_bgra (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) ex->arrays[0];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+  ptr5 = (orc_union16 *) ex->arrays[5];
+  ptr6 = (orc_union16 *) ex->arrays[6];
+  ptr7 = (orc_union16 *) ex->arrays[7];
+  ptr8 = (orc_union16 *) ex->arrays[8];
+  ptr9 = (orc_union16 *) ex->arrays[9];
+
+  /* 9: loadpw */
+  var42.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 11: loadpw */
+  var43.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 16: loadpb */
+  var45.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var45.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr5[i];
+    /* 1: loadw */
+    var38 = ptr9[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr4[i];
+    /* 4: loadw */
+    var40 = ptr8[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr7[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 14: loadw */
+    var44 = ptr6[i];
+    /* 15: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var53.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var53.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+void
+gst_bayer_merge_bg_bgra (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_merge_bg_bgra");
+      orc_program_set_backup_function (p, _backup_gst_bayer_merge_bg_bgra);
+      orc_program_add_destination (p, 8, "d1");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_source (p, 2, "s2");
+      orc_program_add_source (p, 2, "s3");
+      orc_program_add_source (p, 2, "s4");
+      orc_program_add_source (p, 2, "s5");
+      orc_program_add_source (p, 2, "s6");
+      orc_program_add_constant (p, 4, 0x000000ff, "c1");
+      orc_program_add_constant (p, 4, 0x0000ff00, "c2");
+      orc_program_add_temporary (p, 4, "t1");
+      orc_program_add_temporary (p, 4, "t2");
+      orc_program_add_temporary (p, 2, "t3");
+      orc_program_add_temporary (p, 2, "t4");
+      orc_program_add_temporary (p, 2, "t5");
+
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T3, ORC_VAR_S2, ORC_VAR_S6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_S1, ORC_VAR_S5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "copyw", 0, ORC_VAR_T5, ORC_VAR_S4, ORC_VAR_D1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T5, ORC_VAR_T5, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "orw", 0, ORC_VAR_T4, ORC_VAR_T5, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T2, ORC_VAR_S3, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T1, ORC_VAR_T3, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergewl", 1, ORC_VAR_D1, ORC_VAR_T2, ORC_VAR_T1,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+  ex->arrays[ORC_VAR_S2] = (void *) s2;
+  ex->arrays[ORC_VAR_S3] = (void *) s3;
+  ex->arrays[ORC_VAR_S4] = (void *) s4;
+  ex->arrays[ORC_VAR_S5] = (void *) s5;
+  ex->arrays[ORC_VAR_S6] = (void *) s6;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
+
+
+/* gst_bayer_merge_gr_bgra */
+#ifdef DISABLE_ORC
+void
+gst_bayer_merge_gr_bgra (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  int i;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) d1;
+  ptr4 = (orc_union16 *) s1;
+  ptr5 = (orc_union16 *) s2;
+  ptr6 = (orc_union16 *) s3;
+  ptr7 = (orc_union16 *) s4;
+  ptr8 = (orc_union16 *) s5;
+  ptr9 = (orc_union16 *) s6;
+
+  /* 9: loadpw */
+  var42.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 11: loadpw */
+  var43.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 16: loadpb */
+  var45.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var45.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: loadw */
+    var38 = ptr8[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr5[i];
+    /* 4: loadw */
+    var40 = ptr9[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr6[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 14: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[0];
+      _dest.x2[1] = var53.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[1];
+      _dest.x2[1] = var53.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 15: loadw */
+    var44 = ptr7[i];
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_merge_gr_bgra (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) ex->arrays[0];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+  ptr5 = (orc_union16 *) ex->arrays[5];
+  ptr6 = (orc_union16 *) ex->arrays[6];
+  ptr7 = (orc_union16 *) ex->arrays[7];
+  ptr8 = (orc_union16 *) ex->arrays[8];
+  ptr9 = (orc_union16 *) ex->arrays[9];
+
+  /* 9: loadpw */
+  var42.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 11: loadpw */
+  var43.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 16: loadpb */
+  var45.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var45.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: loadw */
+    var38 = ptr8[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr5[i];
+    /* 4: loadw */
+    var40 = ptr9[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr6[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 14: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[0];
+      _dest.x2[1] = var53.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[1];
+      _dest.x2[1] = var53.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 15: loadw */
+    var44 = ptr7[i];
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+void
+gst_bayer_merge_gr_bgra (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_merge_gr_bgra");
+      orc_program_set_backup_function (p, _backup_gst_bayer_merge_gr_bgra);
+      orc_program_add_destination (p, 8, "d1");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_source (p, 2, "s2");
+      orc_program_add_source (p, 2, "s3");
+      orc_program_add_source (p, 2, "s4");
+      orc_program_add_source (p, 2, "s5");
+      orc_program_add_source (p, 2, "s6");
+      orc_program_add_constant (p, 4, 0x0000ff00, "c1");
+      orc_program_add_constant (p, 4, 0x000000ff, "c2");
+      orc_program_add_temporary (p, 4, "t1");
+      orc_program_add_temporary (p, 4, "t2");
+      orc_program_add_temporary (p, 2, "t3");
+      orc_program_add_temporary (p, 2, "t4");
+      orc_program_add_temporary (p, 2, "t5");
+
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T3, ORC_VAR_S1, ORC_VAR_S5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_S2, ORC_VAR_S6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "copyw", 0, ORC_VAR_T5, ORC_VAR_S3, ORC_VAR_D1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T5, ORC_VAR_T5, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "orw", 0, ORC_VAR_T4, ORC_VAR_T5, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T2, ORC_VAR_T3, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T1, ORC_VAR_S4, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergewl", 1, ORC_VAR_D1, ORC_VAR_T2, ORC_VAR_T1,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+  ex->arrays[ORC_VAR_S2] = (void *) s2;
+  ex->arrays[ORC_VAR_S3] = (void *) s3;
+  ex->arrays[ORC_VAR_S4] = (void *) s4;
+  ex->arrays[ORC_VAR_S5] = (void *) s5;
+  ex->arrays[ORC_VAR_S6] = (void *) s6;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
+
+
+/* gst_bayer_merge_bg_abgr */
+#ifdef DISABLE_ORC
+void
+gst_bayer_merge_bg_abgr (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  int i;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) d1;
+  ptr4 = (orc_union16 *) s1;
+  ptr5 = (orc_union16 *) s2;
+  ptr6 = (orc_union16 *) s3;
+  ptr7 = (orc_union16 *) s4;
+  ptr8 = (orc_union16 *) s5;
+  ptr9 = (orc_union16 *) s6;
+
+  /* 9: loadpw */
+  var42.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 11: loadpw */
+  var43.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 14: loadpb */
+  var44.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var44.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr5[i];
+    /* 1: loadw */
+    var38 = ptr9[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr4[i];
+    /* 4: loadw */
+    var40 = ptr8[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr7[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 15: loadw */
+    var45 = ptr6[i];
+    /* 16: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[0];
+      _dest.x2[1] = var47.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[1];
+      _dest.x2[1] = var47.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_merge_bg_abgr (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) ex->arrays[0];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+  ptr5 = (orc_union16 *) ex->arrays[5];
+  ptr6 = (orc_union16 *) ex->arrays[6];
+  ptr7 = (orc_union16 *) ex->arrays[7];
+  ptr8 = (orc_union16 *) ex->arrays[8];
+  ptr9 = (orc_union16 *) ex->arrays[9];
+
+  /* 9: loadpw */
+  var42.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 11: loadpw */
+  var43.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 14: loadpb */
+  var44.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var44.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr5[i];
+    /* 1: loadw */
+    var38 = ptr9[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr4[i];
+    /* 4: loadw */
+    var40 = ptr8[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr7[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 15: loadw */
+    var45 = ptr6[i];
+    /* 16: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[0];
+      _dest.x2[1] = var47.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[1];
+      _dest.x2[1] = var47.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+void
+gst_bayer_merge_bg_abgr (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_merge_bg_abgr");
+      orc_program_set_backup_function (p, _backup_gst_bayer_merge_bg_abgr);
+      orc_program_add_destination (p, 8, "d1");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_source (p, 2, "s2");
+      orc_program_add_source (p, 2, "s3");
+      orc_program_add_source (p, 2, "s4");
+      orc_program_add_source (p, 2, "s5");
+      orc_program_add_source (p, 2, "s6");
+      orc_program_add_constant (p, 4, 0x000000ff, "c1");
+      orc_program_add_constant (p, 4, 0x0000ff00, "c2");
+      orc_program_add_temporary (p, 4, "t1");
+      orc_program_add_temporary (p, 4, "t2");
+      orc_program_add_temporary (p, 2, "t3");
+      orc_program_add_temporary (p, 2, "t4");
+      orc_program_add_temporary (p, 2, "t5");
+
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T3, ORC_VAR_S2, ORC_VAR_S6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_S1, ORC_VAR_S5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "copyw", 0, ORC_VAR_T5, ORC_VAR_S4, ORC_VAR_D1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T5, ORC_VAR_T5, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "orw", 0, ORC_VAR_T4, ORC_VAR_T5, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T1, ORC_VAR_C1, ORC_VAR_S3,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T2, ORC_VAR_T4, ORC_VAR_T3,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergewl", 1, ORC_VAR_D1, ORC_VAR_T1, ORC_VAR_T2,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+  ex->arrays[ORC_VAR_S2] = (void *) s2;
+  ex->arrays[ORC_VAR_S3] = (void *) s3;
+  ex->arrays[ORC_VAR_S4] = (void *) s4;
+  ex->arrays[ORC_VAR_S5] = (void *) s5;
+  ex->arrays[ORC_VAR_S6] = (void *) s6;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
+
+
+/* gst_bayer_merge_gr_abgr */
+#ifdef DISABLE_ORC
+void
+gst_bayer_merge_gr_abgr (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  int i;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) d1;
+  ptr4 = (orc_union16 *) s1;
+  ptr5 = (orc_union16 *) s2;
+  ptr6 = (orc_union16 *) s3;
+  ptr7 = (orc_union16 *) s4;
+  ptr8 = (orc_union16 *) s5;
+  ptr9 = (orc_union16 *) s6;
+
+  /* 9: loadpw */
+  var42.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 11: loadpw */
+  var43.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 14: loadpb */
+  var44.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var44.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: loadw */
+    var38 = ptr8[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr5[i];
+    /* 4: loadw */
+    var40 = ptr9[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr6[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 15: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var47.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var47.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 16: loadw */
+    var45 = ptr7[i];
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_merge_gr_abgr (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) ex->arrays[0];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+  ptr5 = (orc_union16 *) ex->arrays[5];
+  ptr6 = (orc_union16 *) ex->arrays[6];
+  ptr7 = (orc_union16 *) ex->arrays[7];
+  ptr8 = (orc_union16 *) ex->arrays[8];
+  ptr9 = (orc_union16 *) ex->arrays[9];
+
+  /* 9: loadpw */
+  var42.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 11: loadpw */
+  var43.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 14: loadpb */
+  var44.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var44.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: loadw */
+    var38 = ptr8[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr5[i];
+    /* 4: loadw */
+    var40 = ptr9[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr6[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 15: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var47.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var47.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 16: loadw */
+    var45 = ptr7[i];
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+void
+gst_bayer_merge_gr_abgr (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_merge_gr_abgr");
+      orc_program_set_backup_function (p, _backup_gst_bayer_merge_gr_abgr);
+      orc_program_add_destination (p, 8, "d1");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_source (p, 2, "s2");
+      orc_program_add_source (p, 2, "s3");
+      orc_program_add_source (p, 2, "s4");
+      orc_program_add_source (p, 2, "s5");
+      orc_program_add_source (p, 2, "s6");
+      orc_program_add_constant (p, 4, 0x0000ff00, "c1");
+      orc_program_add_constant (p, 4, 0x000000ff, "c2");
+      orc_program_add_temporary (p, 4, "t1");
+      orc_program_add_temporary (p, 4, "t2");
+      orc_program_add_temporary (p, 2, "t3");
+      orc_program_add_temporary (p, 2, "t4");
+      orc_program_add_temporary (p, 2, "t5");
+
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T3, ORC_VAR_S1, ORC_VAR_S5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_S2, ORC_VAR_S6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "copyw", 0, ORC_VAR_T5, ORC_VAR_S3, ORC_VAR_D1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T5, ORC_VAR_T5, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "orw", 0, ORC_VAR_T4, ORC_VAR_T5, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T1, ORC_VAR_C2, ORC_VAR_T3,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T2, ORC_VAR_T4, ORC_VAR_S4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergewl", 1, ORC_VAR_D1, ORC_VAR_T1, ORC_VAR_T2,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+  ex->arrays[ORC_VAR_S2] = (void *) s2;
+  ex->arrays[ORC_VAR_S3] = (void *) s3;
+  ex->arrays[ORC_VAR_S4] = (void *) s4;
+  ex->arrays[ORC_VAR_S5] = (void *) s5;
+  ex->arrays[ORC_VAR_S6] = (void *) s6;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
+
+
+/* gst_bayer_merge_bg_rgba */
+#ifdef DISABLE_ORC
+void
+gst_bayer_merge_bg_rgba (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  int i;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) d1;
+  ptr4 = (orc_union16 *) s1;
+  ptr5 = (orc_union16 *) s2;
+  ptr6 = (orc_union16 *) s3;
+  ptr7 = (orc_union16 *) s4;
+  ptr8 = (orc_union16 *) s5;
+  ptr9 = (orc_union16 *) s6;
+
+  /* 9: loadpw */
+  var42.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 11: loadpw */
+  var43.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 16: loadpb */
+  var45.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var45.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr5[i];
+    /* 1: loadw */
+    var38 = ptr9[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr4[i];
+    /* 4: loadw */
+    var40 = ptr8[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr7[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 14: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[0];
+      _dest.x2[1] = var53.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[1];
+      _dest.x2[1] = var53.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 15: loadw */
+    var44 = ptr6[i];
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_merge_bg_rgba (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) ex->arrays[0];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+  ptr5 = (orc_union16 *) ex->arrays[5];
+  ptr6 = (orc_union16 *) ex->arrays[6];
+  ptr7 = (orc_union16 *) ex->arrays[7];
+  ptr8 = (orc_union16 *) ex->arrays[8];
+  ptr9 = (orc_union16 *) ex->arrays[9];
+
+  /* 9: loadpw */
+  var42.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 11: loadpw */
+  var43.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 16: loadpb */
+  var45.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var45.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr5[i];
+    /* 1: loadw */
+    var38 = ptr9[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr4[i];
+    /* 4: loadw */
+    var40 = ptr8[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr7[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 14: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[0];
+      _dest.x2[1] = var53.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[1];
+      _dest.x2[1] = var53.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 15: loadw */
+    var44 = ptr6[i];
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+void
+gst_bayer_merge_bg_rgba (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_merge_bg_rgba");
+      orc_program_set_backup_function (p, _backup_gst_bayer_merge_bg_rgba);
+      orc_program_add_destination (p, 8, "d1");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_source (p, 2, "s2");
+      orc_program_add_source (p, 2, "s3");
+      orc_program_add_source (p, 2, "s4");
+      orc_program_add_source (p, 2, "s5");
+      orc_program_add_source (p, 2, "s6");
+      orc_program_add_constant (p, 4, 0x000000ff, "c1");
+      orc_program_add_constant (p, 4, 0x0000ff00, "c2");
+      orc_program_add_temporary (p, 4, "t1");
+      orc_program_add_temporary (p, 4, "t2");
+      orc_program_add_temporary (p, 2, "t3");
+      orc_program_add_temporary (p, 2, "t4");
+      orc_program_add_temporary (p, 2, "t5");
+
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T3, ORC_VAR_S2, ORC_VAR_S6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_S1, ORC_VAR_S5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "copyw", 0, ORC_VAR_T5, ORC_VAR_S4, ORC_VAR_D1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T5, ORC_VAR_T5, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "orw", 0, ORC_VAR_T4, ORC_VAR_T5, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T1, ORC_VAR_T3, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T2, ORC_VAR_S3, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergewl", 1, ORC_VAR_D1, ORC_VAR_T1, ORC_VAR_T2,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+  ex->arrays[ORC_VAR_S2] = (void *) s2;
+  ex->arrays[ORC_VAR_S3] = (void *) s3;
+  ex->arrays[ORC_VAR_S4] = (void *) s4;
+  ex->arrays[ORC_VAR_S5] = (void *) s5;
+  ex->arrays[ORC_VAR_S6] = (void *) s6;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
+
+
+/* gst_bayer_merge_gr_rgba */
+#ifdef DISABLE_ORC
+void
+gst_bayer_merge_gr_rgba (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  int i;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) d1;
+  ptr4 = (orc_union16 *) s1;
+  ptr5 = (orc_union16 *) s2;
+  ptr6 = (orc_union16 *) s3;
+  ptr7 = (orc_union16 *) s4;
+  ptr8 = (orc_union16 *) s5;
+  ptr9 = (orc_union16 *) s6;
+
+  /* 9: loadpw */
+  var42.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 11: loadpw */
+  var43.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 16: loadpb */
+  var45.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var45.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: loadw */
+    var38 = ptr8[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr5[i];
+    /* 4: loadw */
+    var40 = ptr9[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr6[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 14: loadw */
+    var44 = ptr7[i];
+    /* 15: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var53.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var53.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_merge_gr_rgba (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) ex->arrays[0];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+  ptr5 = (orc_union16 *) ex->arrays[5];
+  ptr6 = (orc_union16 *) ex->arrays[6];
+  ptr7 = (orc_union16 *) ex->arrays[7];
+  ptr8 = (orc_union16 *) ex->arrays[8];
+  ptr9 = (orc_union16 *) ex->arrays[9];
+
+  /* 9: loadpw */
+  var42.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 11: loadpw */
+  var43.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 16: loadpb */
+  var45.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var45.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: loadw */
+    var38 = ptr8[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr5[i];
+    /* 4: loadw */
+    var40 = ptr9[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr6[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 14: loadw */
+    var44 = ptr7[i];
+    /* 15: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var53.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var53.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var47.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+void
+gst_bayer_merge_gr_rgba (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_merge_gr_rgba");
+      orc_program_set_backup_function (p, _backup_gst_bayer_merge_gr_rgba);
+      orc_program_add_destination (p, 8, "d1");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_source (p, 2, "s2");
+      orc_program_add_source (p, 2, "s3");
+      orc_program_add_source (p, 2, "s4");
+      orc_program_add_source (p, 2, "s5");
+      orc_program_add_source (p, 2, "s6");
+      orc_program_add_constant (p, 4, 0x0000ff00, "c1");
+      orc_program_add_constant (p, 4, 0x000000ff, "c2");
+      orc_program_add_temporary (p, 4, "t1");
+      orc_program_add_temporary (p, 4, "t2");
+      orc_program_add_temporary (p, 2, "t3");
+      orc_program_add_temporary (p, 2, "t4");
+      orc_program_add_temporary (p, 2, "t5");
+
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T3, ORC_VAR_S1, ORC_VAR_S5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_S2, ORC_VAR_S6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "copyw", 0, ORC_VAR_T5, ORC_VAR_S3, ORC_VAR_D1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T5, ORC_VAR_T5, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "orw", 0, ORC_VAR_T4, ORC_VAR_T5, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T1, ORC_VAR_S4, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T2, ORC_VAR_T3, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergewl", 1, ORC_VAR_D1, ORC_VAR_T1, ORC_VAR_T2,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+  ex->arrays[ORC_VAR_S2] = (void *) s2;
+  ex->arrays[ORC_VAR_S3] = (void *) s3;
+  ex->arrays[ORC_VAR_S4] = (void *) s4;
+  ex->arrays[ORC_VAR_S5] = (void *) s5;
+  ex->arrays[ORC_VAR_S6] = (void *) s6;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
+
+
+/* gst_bayer_merge_bg_argb */
+#ifdef DISABLE_ORC
+void
+gst_bayer_merge_bg_argb (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  int i;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) d1;
+  ptr4 = (orc_union16 *) s1;
+  ptr5 = (orc_union16 *) s2;
+  ptr6 = (orc_union16 *) s3;
+  ptr7 = (orc_union16 *) s4;
+  ptr8 = (orc_union16 *) s5;
+  ptr9 = (orc_union16 *) s6;
+
+  /* 9: loadpw */
+  var42.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 11: loadpw */
+  var43.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 14: loadpb */
+  var44.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var44.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr5[i];
+    /* 1: loadw */
+    var38 = ptr9[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr4[i];
+    /* 4: loadw */
+    var40 = ptr8[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr7[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 15: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var47.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var47.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 16: loadw */
+    var45 = ptr6[i];
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_merge_bg_argb (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) ex->arrays[0];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+  ptr5 = (orc_union16 *) ex->arrays[5];
+  ptr6 = (orc_union16 *) ex->arrays[6];
+  ptr7 = (orc_union16 *) ex->arrays[7];
+  ptr8 = (orc_union16 *) ex->arrays[8];
+  ptr9 = (orc_union16 *) ex->arrays[9];
+
+  /* 9: loadpw */
+  var42.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 11: loadpw */
+  var43.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 14: loadpb */
+  var44.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var44.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr5[i];
+    /* 1: loadw */
+    var38 = ptr9[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr4[i];
+    /* 4: loadw */
+    var40 = ptr8[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr7[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 15: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var47.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var47.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 16: loadw */
+    var45 = ptr6[i];
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+void
+gst_bayer_merge_bg_argb (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_merge_bg_argb");
+      orc_program_set_backup_function (p, _backup_gst_bayer_merge_bg_argb);
+      orc_program_add_destination (p, 8, "d1");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_source (p, 2, "s2");
+      orc_program_add_source (p, 2, "s3");
+      orc_program_add_source (p, 2, "s4");
+      orc_program_add_source (p, 2, "s5");
+      orc_program_add_source (p, 2, "s6");
+      orc_program_add_constant (p, 4, 0x000000ff, "c1");
+      orc_program_add_constant (p, 4, 0x0000ff00, "c2");
+      orc_program_add_temporary (p, 4, "t1");
+      orc_program_add_temporary (p, 4, "t2");
+      orc_program_add_temporary (p, 2, "t3");
+      orc_program_add_temporary (p, 2, "t4");
+      orc_program_add_temporary (p, 2, "t5");
+
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T3, ORC_VAR_S2, ORC_VAR_S6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_S1, ORC_VAR_S5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "copyw", 0, ORC_VAR_T5, ORC_VAR_S4, ORC_VAR_D1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T5, ORC_VAR_T5, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "orw", 0, ORC_VAR_T4, ORC_VAR_T5, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T1, ORC_VAR_C1, ORC_VAR_T3,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T2, ORC_VAR_T4, ORC_VAR_S3,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergewl", 1, ORC_VAR_D1, ORC_VAR_T1, ORC_VAR_T2,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+  ex->arrays[ORC_VAR_S2] = (void *) s2;
+  ex->arrays[ORC_VAR_S3] = (void *) s3;
+  ex->arrays[ORC_VAR_S4] = (void *) s4;
+  ex->arrays[ORC_VAR_S5] = (void *) s5;
+  ex->arrays[ORC_VAR_S6] = (void *) s6;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
+
+
+/* gst_bayer_merge_gr_argb */
+#ifdef DISABLE_ORC
+void
+gst_bayer_merge_gr_argb (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  int i;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) d1;
+  ptr4 = (orc_union16 *) s1;
+  ptr5 = (orc_union16 *) s2;
+  ptr6 = (orc_union16 *) s3;
+  ptr7 = (orc_union16 *) s4;
+  ptr8 = (orc_union16 *) s5;
+  ptr9 = (orc_union16 *) s6;
+
+  /* 9: loadpw */
+  var42.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 11: loadpw */
+  var43.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 14: loadpb */
+  var44.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var44.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: loadw */
+    var38 = ptr8[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr5[i];
+    /* 4: loadw */
+    var40 = ptr9[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr6[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 15: loadw */
+    var45 = ptr7[i];
+    /* 16: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[0];
+      _dest.x2[1] = var47.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[1];
+      _dest.x2[1] = var47.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+#else
+static void
+_backup_gst_bayer_merge_gr_argb (OrcExecutor * ORC_RESTRICT ex)
+{
+  int i;
+  int n = ex->n;
+  orc_union64 *ORC_RESTRICT ptr0;
+  const orc_union16 *ORC_RESTRICT ptr4;
+  const orc_union16 *ORC_RESTRICT ptr5;
+  const orc_union16 *ORC_RESTRICT ptr6;
+  const orc_union16 *ORC_RESTRICT ptr7;
+  const orc_union16 *ORC_RESTRICT ptr8;
+  const orc_union16 *ORC_RESTRICT ptr9;
+  orc_union16 var37;
+  orc_union16 var38;
+  orc_union16 var39;
+  orc_union16 var40;
+  orc_union16 var41;
+  orc_union16 var42;
+  orc_union16 var43;
+  orc_union16 var44;
+  orc_union16 var45;
+  orc_union64 var46;
+  orc_union16 var47;
+  orc_union16 var48;
+  orc_union16 var49;
+  orc_union16 var50;
+  orc_union16 var51;
+  orc_union16 var52;
+  orc_union16 var53;
+  orc_union32 var54;
+  orc_union32 var55;
+
+  ptr0 = (orc_union64 *) ex->arrays[0];
+  ptr4 = (orc_union16 *) ex->arrays[4];
+  ptr5 = (orc_union16 *) ex->arrays[5];
+  ptr6 = (orc_union16 *) ex->arrays[6];
+  ptr7 = (orc_union16 *) ex->arrays[7];
+  ptr8 = (orc_union16 *) ex->arrays[8];
+  ptr9 = (orc_union16 *) ex->arrays[9];
+
+  /* 9: loadpw */
+  var42.i = (int) 0x0000ff00;   /* 65280 or 3.22526e-319f */
+  /* 11: loadpw */
+  var43.i = (int) 0x000000ff;   /* 255 or 1.25987e-321f */
+  /* 14: loadpb */
+  var44.x2[0] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+  var44.x2[1] = (int) 0x000000ff;       /* 255 or 1.25987e-321f */
+
+  for (i = 0; i < n; i++) {
+    /* 0: loadw */
+    var37 = ptr4[i];
+    /* 1: loadw */
+    var38 = ptr8[i];
+    /* 2: avgub */
+    var47.x2[0] = ((orc_uint8) var37.x2[0] + (orc_uint8) var38.x2[0] + 1) >> 1;
+    var47.x2[1] = ((orc_uint8) var37.x2[1] + (orc_uint8) var38.x2[1] + 1) >> 1;
+    /* 3: loadw */
+    var39 = ptr5[i];
+    /* 4: loadw */
+    var40 = ptr9[i];
+    /* 5: avgub */
+    var48.x2[0] = ((orc_uint8) var39.x2[0] + (orc_uint8) var40.x2[0] + 1) >> 1;
+    var48.x2[1] = ((orc_uint8) var39.x2[1] + (orc_uint8) var40.x2[1] + 1) >> 1;
+    /* 6: loadw */
+    var41 = ptr6[i];
+    /* 7: copyw */
+    var49.i = var41.i;
+    /* 8: avgub */
+    var50.x2[0] = ((orc_uint8) var48.x2[0] + (orc_uint8) var49.x2[0] + 1) >> 1;
+    var50.x2[1] = ((orc_uint8) var48.x2[1] + (orc_uint8) var49.x2[1] + 1) >> 1;
+    /* 10: andw */
+    var51.i = var50.i & var42.i;
+    /* 12: andw */
+    var52.i = var49.i & var43.i;
+    /* 13: orw */
+    var53.i = var52.i | var51.i;
+    /* 15: loadw */
+    var45 = ptr7[i];
+    /* 16: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[0];
+      _dest.x2[1] = var45.x2[0];
+      var54.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var44.x2[1];
+      _dest.x2[1] = var45.x2[1];
+      var54.x2[1] = _dest.i;
+    }
+    /* 17: mergebw */
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[0];
+      _dest.x2[1] = var47.x2[0];
+      var55.x2[0] = _dest.i;
+    }
+    {
+      orc_union16 _dest;
+      _dest.x2[0] = var53.x2[1];
+      _dest.x2[1] = var47.x2[1];
+      var55.x2[1] = _dest.i;
+    }
+    /* 18: mergewl */
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[0];
+      _dest.x2[1] = var55.x2[0];
+      var46.x2[0] = _dest.i;
+    }
+    {
+      orc_union32 _dest;
+      _dest.x2[0] = var54.x2[1];
+      _dest.x2[1] = var55.x2[1];
+      var46.x2[1] = _dest.i;
+    }
+    /* 19: storeq */
+    ptr0[i] = var46;
+  }
+
+}
+
+void
+gst_bayer_merge_gr_argb (guint8 * ORC_RESTRICT d1,
+    const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2,
+    const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4,
+    const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n)
+{
+  OrcExecutor _ex, *ex = &_ex;
+  static int p_inited = 0;
+  static OrcProgram *p = 0;
+  void (*func) (OrcExecutor *);
+
+  if (!p_inited) {
+    orc_once_mutex_lock ();
+    if (!p_inited) {
+
+      p = orc_program_new ();
+      orc_program_set_name (p, "gst_bayer_merge_gr_argb");
+      orc_program_set_backup_function (p, _backup_gst_bayer_merge_gr_argb);
+      orc_program_add_destination (p, 8, "d1");
+      orc_program_add_source (p, 2, "s1");
+      orc_program_add_source (p, 2, "s2");
+      orc_program_add_source (p, 2, "s3");
+      orc_program_add_source (p, 2, "s4");
+      orc_program_add_source (p, 2, "s5");
+      orc_program_add_source (p, 2, "s6");
+      orc_program_add_constant (p, 4, 0x0000ff00, "c1");
+      orc_program_add_constant (p, 4, 0x000000ff, "c2");
+      orc_program_add_temporary (p, 4, "t1");
+      orc_program_add_temporary (p, 4, "t2");
+      orc_program_add_temporary (p, 2, "t3");
+      orc_program_add_temporary (p, 2, "t4");
+      orc_program_add_temporary (p, 2, "t5");
+
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T3, ORC_VAR_S1, ORC_VAR_S5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_S2, ORC_VAR_S6,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "copyw", 0, ORC_VAR_T5, ORC_VAR_S3, ORC_VAR_D1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "avgub", 1, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_T5,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T4, ORC_VAR_T4, ORC_VAR_C1,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "andw", 0, ORC_VAR_T5, ORC_VAR_T5, ORC_VAR_C2,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "orw", 0, ORC_VAR_T4, ORC_VAR_T5, ORC_VAR_T4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T1, ORC_VAR_C2, ORC_VAR_S4,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergebw", 1, ORC_VAR_T2, ORC_VAR_T4, ORC_VAR_T3,
+          ORC_VAR_D1);
+      orc_program_append_2 (p, "mergewl", 1, ORC_VAR_D1, ORC_VAR_T1, ORC_VAR_T2,
+          ORC_VAR_D1);
+
+      orc_program_compile (p);
+    }
+    p_inited = TRUE;
+    orc_once_mutex_unlock ();
+  }
+  ex->program = p;
+
+  ex->n = n;
+  ex->arrays[ORC_VAR_D1] = d1;
+  ex->arrays[ORC_VAR_S1] = (void *) s1;
+  ex->arrays[ORC_VAR_S2] = (void *) s2;
+  ex->arrays[ORC_VAR_S3] = (void *) s3;
+  ex->arrays[ORC_VAR_S4] = (void *) s4;
+  ex->arrays[ORC_VAR_S5] = (void *) s5;
+  ex->arrays[ORC_VAR_S6] = (void *) s6;
+
+  func = p->code_exec;
+  func (ex);
+}
+#endif
diff --git a/gst/bayer/gstbayerorc-dist.h b/gst/bayer/gstbayerorc-dist.h
new file mode 100644
index 000000000..5e8b86556
--- /dev/null
+++ b/gst/bayer/gstbayerorc-dist.h
@@ -0,0 +1,86 @@
+
+/* autogenerated from gstbayerorc.orc */
+
+#ifndef _GSTBAYERORC_H_
+#define _GSTBAYERORC_H_
+
+#include <glib.h>
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+
+#ifndef _ORC_INTEGER_TYPEDEFS_
+#define _ORC_INTEGER_TYPEDEFS_
+#if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
+#include <stdint.h>
+typedef int8_t orc_int8;
+typedef int16_t orc_int16;
+typedef int32_t orc_int32;
+typedef int64_t orc_int64;
+typedef uint8_t orc_uint8;
+typedef uint16_t orc_uint16;
+typedef uint32_t orc_uint32;
+typedef uint64_t orc_uint64;
+#define ORC_UINT64_C(x) UINT64_C(x)
+#elif defined(_MSC_VER)
+typedef signed __int8 orc_int8;
+typedef signed __int16 orc_int16;
+typedef signed __int32 orc_int32;
+typedef signed __int64 orc_int64;
+typedef unsigned __int8 orc_uint8;
+typedef unsigned __int16 orc_uint16;
+typedef unsigned __int32 orc_uint32;
+typedef unsigned __int64 orc_uint64;
+#define ORC_UINT64_C(x) (x##Ui64)
+#define inline __inline
+#else
+#include <limits.h>
+typedef signed char orc_int8;
+typedef short orc_int16;
+typedef int orc_int32;
+typedef unsigned char orc_uint8;
+typedef unsigned short orc_uint16;
+typedef unsigned int orc_uint32;
+#if INT_MAX == LONG_MAX
+typedef long long orc_int64;
+typedef unsigned long long orc_uint64;
+#define ORC_UINT64_C(x) (x##ULL)
+#else
+typedef long orc_int64;
+typedef unsigned long orc_uint64;
+#define ORC_UINT64_C(x) (x##UL)
+#endif
+#endif
+typedef union { orc_int16 i; orc_int8 x2[2]; } orc_union16;
+typedef union { orc_int32 i; float f; orc_int16 x2[2]; orc_int8 x4[4]; } orc_union32;
+typedef union { orc_int64 i; double f; orc_int32 x2[2]; float x2f[2]; orc_int16 x4[4]; } orc_union64;
+#endif
+#ifndef ORC_RESTRICT
+#if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
+#define ORC_RESTRICT restrict
+#elif defined(__GNUC__) && __GNUC__ >= 4
+#define ORC_RESTRICT __restrict__
+#else
+#define ORC_RESTRICT
+#endif
+#endif
+void gst_bayer_horiz_upsample_unaligned (guint8 * ORC_RESTRICT d1, guint8 * ORC_RESTRICT d2, const guint8 * ORC_RESTRICT s1, int n);
+void gst_bayer_horiz_upsample (guint8 * ORC_RESTRICT d1, guint8 * ORC_RESTRICT d2, const guint8 * ORC_RESTRICT s1, int n);
+void gst_bayer_merge_bg_bgra (guint8 * ORC_RESTRICT d1, const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2, const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4, const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_gr_bgra (guint8 * ORC_RESTRICT d1, const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2, const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4, const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_bg_abgr (guint8 * ORC_RESTRICT d1, const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2, const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4, const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_gr_abgr (guint8 * ORC_RESTRICT d1, const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2, const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4, const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_bg_rgba (guint8 * ORC_RESTRICT d1, const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2, const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4, const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_gr_rgba (guint8 * ORC_RESTRICT d1, const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2, const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4, const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_bg_argb (guint8 * ORC_RESTRICT d1, const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2, const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4, const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+void gst_bayer_merge_gr_argb (guint8 * ORC_RESTRICT d1, const guint8 * ORC_RESTRICT s1, const guint8 * ORC_RESTRICT s2, const guint8 * ORC_RESTRICT s3, const guint8 * ORC_RESTRICT s4, const guint8 * ORC_RESTRICT s5, const guint8 * ORC_RESTRICT s6, int n);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
+
diff --git a/gst/bayer/gstbayerorc.orc b/gst/bayer/gstbayerorc.orc
new file mode 100644
index 000000000..e57436b7f
--- /dev/null
+++ b/gst/bayer/gstbayerorc.orc
@@ -0,0 +1,252 @@
+
+
+.function gst_bayer_horiz_upsample_unaligned
+.dest 2 d0 guint8
+.dest 2 d1 guint8
+.source 2 s guint8
+.temp 2 t
+.temp 1 b
+.temp 1 c
+.temp 1 d
+.temp 1 e
+
+splitwb c, b, s
+loadoffw t, s, 1
+splitwb e, d, t
+avgub e, c, e
+mergebw d0, c, e
+avgub b, b, d
+mergebw d1, b, d
+
+
+.function gst_bayer_horiz_upsample
+.dest 2 d0 guint8
+.dest 2 d1 guint8
+.source 2 s guint8
+.temp 2 t
+.temp 1 a
+.temp 1 b
+.temp 1 c
+.temp 1 d
+.temp 1 e
+.temp 1 f
+
+loadoffw t, s, -1
+splitwb b, a, t
+splitwb d, c, s
+loadoffw t, s, 1
+splitwb f, e, t
+avgub e, c, e
+mergebw d0, c, e
+avgub b, b, d
+mergebw d1, b, d
+
+
+.function gst_bayer_merge_bg_bgra
+.dest 8 d guint8
+.source 2 g0 guint8
+.source 2 r0 guint8
+.source 2 b1 guint8
+.source 2 g1 guint8
+.source 2 g2 guint8
+.source 2 r2 guint8
+.temp 4 ra
+.temp 4 bg
+.temp 2 r
+.temp 2 g
+.temp 2 t
+
+x2 avgub r, r0, r2
+x2 avgub g, g0, g2
+copyw t, g1
+x2 avgub g, g, t
+andw g, g, 255
+andw t, t, 65280
+orw g, t, g
+x2 mergebw bg, b1, g
+x2 mergebw ra, r, 255
+x2 mergewl d, bg, ra
+
+
+.function gst_bayer_merge_gr_bgra
+.dest 8 d guint8
+.source 2 b0 guint8
+.source 2 g0 guint8
+.source 2 g1 guint8
+.source 2 r1 guint8
+.source 2 b2 guint8
+.source 2 g2 guint8
+.temp 4 ra
+.temp 4 bg
+.temp 2 b
+.temp 2 g
+.temp 2 t
+
+x2 avgub b, b0, b2
+x2 avgub g, g0, g2
+copyw t, g1
+x2 avgub g, g, t
+andw g, g, 65280
+andw t, t, 255
+orw g, t, g
+x2 mergebw bg, b, g
+x2 mergebw ra, r1, 255
+x2 mergewl d, bg, ra
+
+
+.function gst_bayer_merge_bg_abgr
+.dest 8 d guint8
+.source 2 g0 guint8
+.source 2 r0 guint8
+.source 2 b1 guint8
+.source 2 g1 guint8
+.source 2 g2 guint8
+.source 2 r2 guint8
+.temp 4 ab
+.temp 4 gr
+.temp 2 r
+.temp 2 g
+.temp 2 t
+
+x2 avgub r, r0, r2
+x2 avgub g, g0, g2
+copyw t, g1
+x2 avgub g, g, t
+andw g, g, 255
+andw t, t, 65280
+orw g, t, g
+x2 mergebw ab, 255, b1
+x2 mergebw gr, g, r
+x2 mergewl d, ab, gr
+
+
+.function gst_bayer_merge_gr_abgr
+.dest 8 d guint8
+.source 2 b0 guint8
+.source 2 g0 guint8
+.source 2 g1 guint8
+.source 2 r1 guint8
+.source 2 b2 guint8
+.source 2 g2 guint8
+.temp 4 ab
+.temp 4 gr
+.temp 2 b
+.temp 2 g
+.temp 2 t
+
+x2 avgub b, b0, b2
+x2 avgub g, g0, g2
+copyw t, g1
+x2 avgub g, g, t
+andw g, g, 65280
+andw t, t, 255
+orw g, t, g
+x2 mergebw ab, 255, b
+x2 mergebw gr, g, r1
+x2 mergewl d, ab, gr
+
+
+.function gst_bayer_merge_bg_rgba
+.dest 8 d guint8
+.source 2 g0 guint8
+.source 2 r0 guint8
+.source 2 b1 guint8
+.source 2 g1 guint8
+.source 2 g2 guint8
+.source 2 r2 guint8
+.temp 4 rg
+.temp 4 ba
+.temp 2 r
+.temp 2 g
+.temp 2 t
+
+x2 avgub r, r0, r2
+x2 avgub g, g0, g2
+copyw t, g1
+x2 avgub g, g, t
+andw g, g, 255
+andw t, t, 65280
+orw g, t, g
+x2 mergebw rg, r, g
+x2 mergebw ba, b1, 255
+x2 mergewl d, rg, ba
+
+
+.function gst_bayer_merge_gr_rgba
+.dest 8 d guint8
+.source 2 b0 guint8
+.source 2 g0 guint8
+.source 2 g1 guint8
+.source 2 r1 guint8
+.source 2 b2 guint8
+.source 2 g2 guint8
+.temp 4 rg
+.temp 4 ba
+.temp 2 b
+.temp 2 g
+.temp 2 t
+
+x2 avgub b, b0, b2
+x2 avgub g, g0, g2
+copyw t, g1
+x2 avgub g, g, t
+andw g, g, 65280
+andw t, t, 255
+orw g, t, g
+x2 mergebw rg, r1, g
+x2 mergebw ba, b, 255
+x2 mergewl d, rg, ba
+
+
+.function gst_bayer_merge_bg_argb
+.dest 8 d guint8
+.source 2 g0 guint8
+.source 2 r0 guint8
+.source 2 b1 guint8
+.source 2 g1 guint8
+.source 2 g2 guint8
+.source 2 r2 guint8
+.temp 4 ar
+.temp 4 gb
+.temp 2 r
+.temp 2 g
+.temp 2 t
+
+x2 avgub r, r0, r2
+x2 avgub g, g0, g2
+copyw t, g1
+x2 avgub g, g, t
+andw g, g, 255
+andw t, t, 65280
+orw g, t, g
+x2 mergebw ar, 255, r
+x2 mergebw gb, g, b1
+x2 mergewl d, ar, gb
+
+
+.function gst_bayer_merge_gr_argb
+.dest 8 d guint8
+.source 2 b0 guint8
+.source 2 g0 guint8
+.source 2 g1 guint8
+.source 2 r1 guint8
+.source 2 b2 guint8
+.source 2 g2 guint8
+.temp 4 ar
+.temp 4 gb
+.temp 2 b
+.temp 2 g
+.temp 2 t
+
+x2 avgub b, b0, b2
+x2 avgub g, g0, g2
+copyw t, g1
+x2 avgub g, g, t
+andw g, g, 65280
+andw t, t, 255
+orw g, t, g
+x2 mergebw ar, 255, r1
+x2 mergebw gb, g, b
+x2 mergewl d, ar, gb
+
+