cog: Improvements in colorspace and scaler

Add fast paths for YUV->YUV conversions and Orc code for all.
Use Orc for horizontal resampling.

diff --git a/ext/cog/cogvirtframe.c b/ext/cog/cogvirtframe.c
index f1faa8e..74336ca 100644 (file)
--- a/ext/cog/cogvirtframe.c
+++ b/ext/cog/cogvirtframe.c
@@ -630,19 +630,14 @@ cog_virt_frame_render_resample_horiz_1tap (CogFrame * frame, void *_dest,
 {
   uint8_t *dest = _dest;
   uint8_t *src;
-  int j;
   int n_src;
   int scale = frame->param1;
-  int acc;
 
   n_src = frame->virt_frame1->components[component].width;
   src = cog_virt_frame_get_line (frame->virt_frame1, component, i);
 
-  acc = 0;
-  for (j = 0; j < frame->components[component].width; j++) {
-    dest[j] = src[(acc >> 8)];
-    acc += scale;
-  }
+  cogorc_resample_horiz_1tap (dest, src, 0, scale,
+      frame->components[component].width);
 }
 
 static void
@@ -651,45 +646,14 @@ cog_virt_frame_render_resample_horiz_2tap (CogFrame * frame, void *_dest,
 {
   uint8_t *dest = _dest;
   uint8_t *src;
-  int j;
   int n_src;
   int scale = frame->param1;
-  int acc;
 
   n_src = frame->virt_frame1->components[component].width;
   src = cog_virt_frame_get_line (frame->virt_frame1, component, i);
 
-  acc = 0;
-  for (j = 0; j < frame->components[component].width - 2; j++) {
-    int src_i;
-    int y;
-    int z;
-
-    src_i = acc >> 8;
-    y = acc & 255;
-
-    z = 128;
-    z += (256 - y) * src[src_i + 0];
-    z += y * src[src_i + 1];
-    z >>= 8;
-    dest[j] = CLAMP (z, 0, 255);
-    acc += scale;
-  }
-  for (; j < frame->components[component].width; j++) {
-    int src_i;
-    int y;
-    int z;
-
-    src_i = acc >> 8;
-    y = acc & 255;
-
-    z = 128;
-    z += (256 - y) * src[CLAMP (src_i + 0, 0, n_src - 1)];
-    z += y * src[CLAMP (src_i + 1, 0, n_src - 1)];
-    z >>= 8;
-    dest[j] = CLAMP (z, 0, 255);
-    acc += scale;
-  }
+  cogorc_resample_horiz_2tap (dest, src, 0, scale,
+      frame->components[component].width);
 }
 
 static void
@@ -712,8 +676,8 @@ cog_virt_frame_render_resample_horiz_4tap (CogFrame * frame, void *_dest,
     int y;
     int z;
 
-    src_i = acc >> 8;
-    y = acc & 255;
+    src_i = acc >> 16;
+    y = (acc >> 8) & 255;
 
     z = 32;
     z += cog_resample_table_4tap[y][0] * src[CLAMP (src_i - 1, 0, n_src - 1)];
@@ -729,8 +693,8 @@ cog_virt_frame_render_resample_horiz_4tap (CogFrame * frame, void *_dest,
     int y;
     int z;
 
-    src_i = acc >> 8;
-    y = acc & 255;
+    src_i = acc >> 16;
+    y = (acc >> 8) & 255;
 
     z = 32;
     z += cog_resample_table_4tap[y][0] * src[src_i - 1];
@@ -746,8 +710,8 @@ cog_virt_frame_render_resample_horiz_4tap (CogFrame * frame, void *_dest,
     int y;
     int z;
 
-    src_i = acc >> 8;
-    y = acc & 255;
+    src_i = acc >> 16;
+    y = (acc >> 8) & 255;
 
     z = 32;
     z += cog_resample_table_4tap[y][0] * src[CLAMP (src_i - 1, 0, n_src - 1)];
@@ -775,7 +739,7 @@ cog_virt_frame_new_horiz_resample (CogFrame * vf, int width, int n_taps)
     virt_frame->render_line = cog_virt_frame_render_resample_horiz_4tap;
   }
 
-  virt_frame->param1 = 256 * vf->width / width;
+  virt_frame->param1 = 65536 * vf->width / width;
 
   return virt_frame;
 }
@@ -1429,7 +1393,6 @@ cog_virt_frame_new_color_matrix_YCbCr_to_RGB (CogFrame * vf,
     CogColorMatrix color_matrix, int bits)
 {
   CogFrame *virt_frame;
-  //int *matrix = frame->virt_priv2;
 
   virt_frame = cog_frame_new_virtual (NULL, COG_FRAME_FORMAT_U8_444,
       vf->width, vf->height);

diff --git a/ext/cog/gstcogcolorspace.c b/ext/cog/gstcogcolorspace.c
index bbd0109..96081f7 100644 (file)
--- a/ext/cog/gstcogcolorspace.c
+++ b/ext/cog/gstcogcolorspace.c
@@ -37,6 +37,7 @@
 #include <math.h>
 #include <cog/cogvirtframe.h>
 #include "gstcogutils.h"
+#include "gstcogorc.h"
 
 #define GST_TYPE_COGCOLORSPACE \
   (gst_cogcolorspace_get_type())
@@ -383,6 +384,425 @@ gst_cogcolorspace_caps_get_chroma_site (GstCaps * caps)
   return COG_CHROMA_SITE_MPEG2;
 }
 
+static void
+convert_I420_YUY2 (CogFrame * dest, CogFrame * src)
+{
+  int i;
+
+  for (i = 0; i < dest->height; i += 2) {
+    cogorc_convert_I420_YUY2 (COG_FRAME_DATA_GET_LINE (dest->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (dest->components + 0, i + 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i + 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 1, i >> 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 2, i >> 1),
+        (dest->width + 1) / 2);
+  }
+}
+
+static void
+convert_I420_UYVY (CogFrame * dest, CogFrame * src)
+{
+  int i;
+
+  for (i = 0; i < dest->height; i += 2) {
+    cogorc_convert_I420_UYVY (COG_FRAME_DATA_GET_LINE (dest->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (dest->components + 0, i + 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i + 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 1, i >> 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 2, i >> 1),
+        (dest->width + 1) / 2);
+  }
+}
+
+static void
+convert_I420_AYUV (CogFrame * dest, CogFrame * src)
+{
+  int i;
+
+  for (i = 0; i < dest->height; i += 2) {
+    cogorc_convert_I420_AYUV (COG_FRAME_DATA_GET_LINE (dest->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (dest->components + 0, i + 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i + 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 1, i >> 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 2, i >> 1), dest->width);
+  }
+}
+
+static void
+convert_I420_Y42B (CogFrame * dest, CogFrame * src)
+{
+  cogorc_memcpy_2d (dest->components[0].data, dest->components[0].stride,
+      src->components[0].data, src->components[0].stride,
+      dest->width, dest->height);
+
+  cogorc_planar_chroma_420_422 (dest->components[1].data,
+      2 * dest->components[1].stride,
+      COG_FRAME_DATA_GET_LINE (dest->components + 2, 1),
+      2 * dest->components[1].stride, src->components[1].data,
+      src->components[1].stride, (dest->width + 1) / 2, dest->height / 2);
+
+  cogorc_planar_chroma_420_422 (dest->components[2].data,
+      2 * dest->components[2].stride,
+      COG_FRAME_DATA_GET_LINE (dest->components + 2, 1),
+      2 * dest->components[2].stride, src->components[2].data,
+      src->components[2].stride, (dest->width + 1) / 2, dest->height / 2);
+}
+
+static void
+convert_I420_Y444 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_memcpy_2d (dest->components[0].data, dest->components[0].stride,
+      src->components[0].data, src->components[0].stride,
+      dest->width, dest->height);
+
+  cogorc_planar_chroma_420_444 (dest->components[1].data,
+      2 * dest->components[1].stride,
+      COG_FRAME_DATA_GET_LINE (dest->components + 1, 1),
+      2 * dest->components[1].stride, src->components[1].data,
+      src->components[1].stride, (dest->width + 1) / 2, (dest->height + 1) / 2);
+
+  cogorc_planar_chroma_420_444 (dest->components[2].data,
+      2 * dest->components[2].stride,
+      COG_FRAME_DATA_GET_LINE (dest->components + 2, 1),
+      2 * dest->components[2].stride, src->components[2].data,
+      src->components[2].stride, (dest->width + 1) / 2, (dest->height + 1) / 2);
+}
+
+static void
+convert_YUY2_I420 (CogFrame * dest, CogFrame * src)
+{
+  int i;
+
+  for (i = 0; i < dest->height; i += 2) {
+    cogorc_convert_YUY2_I420 (COG_FRAME_DATA_GET_LINE (dest->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (dest->components + 0, i + 1),
+        COG_FRAME_DATA_GET_LINE (dest->components + 1, i >> 1),
+        COG_FRAME_DATA_GET_LINE (dest->components + 2, i >> 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i + 1),
+        (dest->width + 1) / 2);
+  }
+}
+
+static void
+convert_YUY2_AYUV (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_YUY2_AYUV (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_YUY2_Y42B (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_YUY2_Y42B (dest->components[0].data,
+      dest->components[0].stride, dest->components[1].data,
+      dest->components[1].stride, dest->components[2].data,
+      dest->components[2].stride, src->components[0].data,
+      src->components[0].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_YUY2_Y444 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_YUY2_Y444 (dest->components[0].data,
+      dest->components[0].stride, dest->components[1].data,
+      dest->components[1].stride, dest->components[2].data,
+      dest->components[2].stride, src->components[0].data,
+      src->components[0].stride, (dest->width + 1) / 2, dest->height);
+}
+
+
+static void
+convert_UYVY_I420 (CogFrame * dest, CogFrame * src)
+{
+  int i;
+
+  for (i = 0; i < dest->height; i += 2) {
+    cogorc_convert_UYVY_I420 (COG_FRAME_DATA_GET_LINE (dest->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (dest->components + 0, i + 1),
+        COG_FRAME_DATA_GET_LINE (dest->components + 1, i >> 1),
+        COG_FRAME_DATA_GET_LINE (dest->components + 2, i >> 1),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i),
+        COG_FRAME_DATA_GET_LINE (src->components + 0, i + 1),
+        (dest->width + 1) / 2);
+  }
+}
+
+static void
+convert_UYVY_AYUV (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_UYVY_AYUV (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_UYVY_YUY2 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_UYVY_YUY2 (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_UYVY_Y42B (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_UYVY_Y42B (dest->components[0].data,
+      dest->components[0].stride, dest->components[1].data,
+      dest->components[1].stride, dest->components[2].data,
+      dest->components[2].stride, src->components[0].data,
+      src->components[0].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_UYVY_Y444 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_UYVY_Y444 (dest->components[0].data,
+      dest->components[0].stride, dest->components[1].data,
+      dest->components[1].stride, dest->components[2].data,
+      dest->components[2].stride, src->components[0].data,
+      src->components[0].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_AYUV_I420 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_AYUV_I420 (COG_FRAME_DATA_GET_LINE (dest->components + 0, 0),
+      2 * dest->components[0].stride,
+      COG_FRAME_DATA_GET_LINE (dest->components + 0, 1),
+      2 * dest->components[0].stride,
+      dest->components[1].data, dest->components[1].stride,
+      dest->components[2].data, dest->components[2].stride,
+      COG_FRAME_DATA_GET_LINE (src->components + 0, 0),
+      /* FIXME why not 2* ? */
+      src->components[0].stride,
+      COG_FRAME_DATA_GET_LINE (src->components + 0, 1),
+      src->components[0].stride, dest->width / 2, dest->height / 2);
+}
+
+static void
+convert_AYUV_YUY2 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_AYUV_YUY2 (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, dest->width / 2, dest->height);
+}
+
+static void
+convert_AYUV_UYVY (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_AYUV_UYVY (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, dest->width / 2, dest->height);
+}
+
+static void
+convert_AYUV_Y42B (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_AYUV_Y42B (dest->components[0].data,
+      dest->components[0].stride, dest->components[1].data,
+      dest->components[1].stride, dest->components[2].data,
+      dest->components[2].stride, src->components[0].data,
+      src->components[0].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_AYUV_Y444 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_AYUV_Y444 (dest->components[0].data,
+      dest->components[0].stride, dest->components[1].data,
+      dest->components[1].stride, dest->components[2].data,
+      dest->components[2].stride, src->components[0].data,
+      src->components[0].stride, dest->width, dest->height);
+}
+
+static void
+convert_Y42B_I420 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_memcpy_2d (dest->components[0].data, dest->components[0].stride,
+      src->components[0].data, src->components[0].stride,
+      dest->width, dest->height);
+
+  cogorc_planar_chroma_422_420 (dest->components[1].data,
+      dest->components[1].stride, src->components[1].data,
+      2 * src->components[1].stride,
+      COG_FRAME_DATA_GET_LINE (src->components + 1, 1),
+      2 * src->components[1].stride, (dest->width + 1) / 2,
+      (dest->height + 1) / 2);
+
+  cogorc_planar_chroma_422_420 (dest->components[2].data,
+      dest->components[2].stride, src->components[2].data,
+      2 * src->components[2].stride,
+      COG_FRAME_DATA_GET_LINE (src->components + 2, 1),
+      2 * src->components[2].stride, (dest->width + 1) / 2,
+      (dest->height + 1) / 2);
+}
+
+static void
+convert_Y42B_Y444 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_memcpy_2d (dest->components[0].data, dest->components[0].stride,
+      src->components[0].data, src->components[0].stride,
+      dest->width, dest->height);
+
+  cogorc_planar_chroma_422_444 (dest->components[1].data,
+      dest->components[1].stride, src->components[1].data,
+      src->components[1].stride, (dest->width + 1) / 2, dest->height);
+
+  cogorc_planar_chroma_422_444 (dest->components[2].data,
+      dest->components[2].stride, src->components[2].data,
+      src->components[2].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_Y42B_YUY2 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_Y42B_YUY2 (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, src->components[1].data,
+      src->components[1].stride, src->components[2].data,
+      src->components[2].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_Y42B_UYVY (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_Y42B_UYVY (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, src->components[1].data,
+      src->components[1].stride, src->components[2].data,
+      src->components[2].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_Y42B_AYUV (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_Y42B_AYUV (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, src->components[1].data,
+      src->components[1].stride, src->components[2].data,
+      src->components[2].stride, (dest->width) / 2, dest->height);
+}
+
+static void
+convert_Y444_I420 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_memcpy_2d (dest->components[0].data, dest->components[0].stride,
+      src->components[0].data, src->components[0].stride,
+      dest->width, dest->height);
+
+  cogorc_planar_chroma_444_420 (dest->components[1].data,
+      dest->components[1].stride, src->components[1].data,
+      2 * src->components[1].stride,
+      COG_FRAME_DATA_GET_LINE (src->components + 1, 1),
+      2 * src->components[1].stride, (dest->width + 1) / 2,
+      (dest->height + 1) / 2);
+
+  cogorc_planar_chroma_444_420 (dest->components[2].data,
+      dest->components[2].stride, src->components[2].data,
+      2 * src->components[2].stride,
+      COG_FRAME_DATA_GET_LINE (src->components + 2, 1),
+      2 * src->components[2].stride, (dest->width + 1) / 2,
+      (dest->height + 1) / 2);
+}
+
+static void
+convert_Y444_Y42B (CogFrame * dest, CogFrame * src)
+{
+  cogorc_memcpy_2d (dest->components[0].data, dest->components[0].stride,
+      src->components[0].data, src->components[0].stride,
+      dest->width, dest->height);
+
+  cogorc_planar_chroma_444_422 (dest->components[1].data,
+      dest->components[1].stride, src->components[1].data,
+      src->components[1].stride, (dest->width + 1) / 2, dest->height);
+
+  cogorc_planar_chroma_444_422 (dest->components[2].data,
+      dest->components[2].stride, src->components[2].data,
+      src->components[2].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_Y444_YUY2 (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_Y444_YUY2 (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, src->components[1].data,
+      src->components[1].stride, src->components[2].data,
+      src->components[2].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_Y444_UYVY (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_Y444_UYVY (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, src->components[1].data,
+      src->components[1].stride, src->components[2].data,
+      src->components[2].stride, (dest->width + 1) / 2, dest->height);
+}
+
+static void
+convert_Y444_AYUV (CogFrame * dest, CogFrame * src)
+{
+  cogorc_convert_Y444_AYUV (dest->components[0].data,
+      dest->components[0].stride, src->components[0].data,
+      src->components[0].stride, src->components[1].data,
+      src->components[1].stride, src->components[2].data,
+      src->components[2].stride, dest->width, dest->height);
+}
+
+
+
+
+typedef struct
+{
+  uint32_t in_format;
+  uint32_t out_format;
+  void (*convert) (CogFrame * dest, CogFrame * src);
+} CogColorspaceTransform;
+static CogColorspaceTransform transforms[] = {
+  {GST_VIDEO_FORMAT_I420, GST_VIDEO_FORMAT_YUY2, convert_I420_YUY2},
+  {GST_VIDEO_FORMAT_I420, GST_VIDEO_FORMAT_UYVY, convert_I420_UYVY},
+  {GST_VIDEO_FORMAT_I420, GST_VIDEO_FORMAT_AYUV, convert_I420_AYUV},
+  {GST_VIDEO_FORMAT_I420, GST_VIDEO_FORMAT_Y42B, convert_I420_Y42B},
+  {GST_VIDEO_FORMAT_I420, GST_VIDEO_FORMAT_Y444, convert_I420_Y444},
+
+  {GST_VIDEO_FORMAT_YUY2, GST_VIDEO_FORMAT_I420, convert_YUY2_I420},
+  {GST_VIDEO_FORMAT_YUY2, GST_VIDEO_FORMAT_UYVY, convert_UYVY_YUY2},    /* alias */
+  {GST_VIDEO_FORMAT_YUY2, GST_VIDEO_FORMAT_AYUV, convert_YUY2_AYUV},
+  {GST_VIDEO_FORMAT_YUY2, GST_VIDEO_FORMAT_Y42B, convert_YUY2_Y42B},
+  {GST_VIDEO_FORMAT_YUY2, GST_VIDEO_FORMAT_Y444, convert_YUY2_Y444},
+
+  {GST_VIDEO_FORMAT_UYVY, GST_VIDEO_FORMAT_I420, convert_UYVY_I420},
+  {GST_VIDEO_FORMAT_UYVY, GST_VIDEO_FORMAT_YUY2, convert_UYVY_YUY2},
+  {GST_VIDEO_FORMAT_UYVY, GST_VIDEO_FORMAT_AYUV, convert_UYVY_AYUV},
+  {GST_VIDEO_FORMAT_UYVY, GST_VIDEO_FORMAT_Y42B, convert_UYVY_Y42B},
+  {GST_VIDEO_FORMAT_UYVY, GST_VIDEO_FORMAT_Y444, convert_UYVY_Y444},
+
+  {GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_I420, convert_AYUV_I420},
+  {GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_YUY2, convert_AYUV_YUY2},
+  {GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_UYVY, convert_AYUV_UYVY},
+  {GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_Y42B, convert_AYUV_Y42B},
+  {GST_VIDEO_FORMAT_AYUV, GST_VIDEO_FORMAT_Y444, convert_AYUV_Y444},
+
+  {GST_VIDEO_FORMAT_Y42B, GST_VIDEO_FORMAT_I420, convert_Y42B_I420},
+  {GST_VIDEO_FORMAT_Y42B, GST_VIDEO_FORMAT_YUY2, convert_Y42B_YUY2},
+  {GST_VIDEO_FORMAT_Y42B, GST_VIDEO_FORMAT_UYVY, convert_Y42B_UYVY},
+  {GST_VIDEO_FORMAT_Y42B, GST_VIDEO_FORMAT_AYUV, convert_Y42B_AYUV},
+  {GST_VIDEO_FORMAT_Y42B, GST_VIDEO_FORMAT_Y444, convert_Y42B_Y444},
+
+  {GST_VIDEO_FORMAT_Y444, GST_VIDEO_FORMAT_I420, convert_Y444_I420},
+  {GST_VIDEO_FORMAT_Y444, GST_VIDEO_FORMAT_YUY2, convert_Y444_YUY2},
+  {GST_VIDEO_FORMAT_Y444, GST_VIDEO_FORMAT_UYVY, convert_Y444_UYVY},
+  {GST_VIDEO_FORMAT_Y444, GST_VIDEO_FORMAT_AYUV, convert_Y444_AYUV},
+  {GST_VIDEO_FORMAT_Y444, GST_VIDEO_FORMAT_Y42B, convert_Y444_Y42B},
+};
+
 static GstFlowReturn
 gst_cogcolorspace_transform (GstBaseTransform * base_transform,
     GstBuffer * inbuf, GstBuffer * outbuf)
@@ -421,6 +841,28 @@ gst_cogcolorspace_transform (GstBaseTransform * base_transform,
   out_frame = gst_cog_buffer_wrap (gst_buffer_ref (outbuf),
       out_format, width, height);
 
+  if (in_format == out_format) {
+    memcpy (GST_BUFFER_DATA (outbuf), GST_BUFFER_DATA (inbuf),
+        GST_BUFFER_SIZE (outbuf));
+  }
+
+  {
+    int i;
+
+    for (i = 0; i < sizeof (transforms) / sizeof (transforms[0]); i++) {
+      if (transforms[i].in_format == in_format &&
+          transforms[i].out_format == out_format) {
+        transforms[i].convert (out_frame, frame);
+        cog_frame_unref (frame);
+        cog_frame_unref (out_frame);
+
+        return GST_FLOW_OK;
+      }
+    }
+
+    GST_ERROR ("no match");
+  }
+
   switch (out_format) {
     case GST_VIDEO_FORMAT_YUY2:
     case GST_VIDEO_FORMAT_UYVY:

diff --git a/ext/cog/gstcogorc.orc b/ext/cog/gstcogorc.orc
index 445a798..e125e43 100644 (file)
--- a/ext/cog/gstcogorc.orc
+++ b/ext/cog/gstcogorc.orc
@@ -1,4 +1,11 @@
 
+.function cogorc_memcpy_2d
+.flags 2d
+.dest 1 d1
+.source 1 s1
+
+copyb d1, s1
+
 
 .function cogorc_downsample_horiz_cosite_1tap
 .dest 1 d1
@@ -492,7 +499,7 @@ convubw t2, s2
 mullw t2, t2, p2
 addw t1, t1, t2
 shruw t1, t1, 8
-convuuswb d1, t1
+convsuswb d1, t1
 
 
 .function cogorc_combine4_u8
@@ -560,3 +567,489 @@ select1lw t1, s1
 select1wb d1, t1
 
 
+.function cogorc_resample_horiz_1tap
+.dest 1 d1
+.source 1 s1
+.param 4 p1
+.param 4 p2
+
+ldresnearb d1, s1, p1, p2
+
+
+.function cogorc_resample_horiz_2tap
+.dest 1 d1
+.source 1 s1
+.param 4 p1
+.param 4 p2
+
+ldreslinb d1, s1, p1, p2
+
+
+.function cogorc_convert_I420_UYVY
+.dest 4 d1
+.dest 4 d2
+.source 2 y1
+.source 2 y2
+.source 1 u
+.source 1 v
+.temp 2 uv
+
+mergebw uv, u, v
+x2 mergebw d1, uv, y1
+x2 mergebw d2, uv, y2
+
+
+.function cogorc_convert_I420_YUY2
+.dest 4 d1
+.dest 4 d2
+.source 2 y1
+.source 2 y2
+.source 1 u
+.source 1 v
+.temp 2 uv
+
+mergebw uv, u, v
+x2 mergebw d1, y1, uv
+x2 mergebw d2, y2, uv
+
+
+
+.function cogorc_convert_I420_AYUV
+.dest 4 d1
+.dest 4 d2
+.source 1 y1
+.source 1 y2
+.source 1 u
+.source 1 v
+.const 1 c255 255
+.temp 2 uv
+.temp 2 ay
+.temp 1 tu
+.temp 1 tv
+
+loadupdb tu, u
+loadupdb tv, v
+mergebw uv, tu, tv
+mergebw ay, c255, y1
+mergewl d1, ay, uv
+mergebw ay, c255, y2
+mergewl d2, ay, uv
+
+
+.function cogorc_convert_YUY2_I420
+.dest 2 y1
+.dest 2 y2
+.dest 1 u
+.dest 1 v
+.source 4 yuv1
+.source 4 yuv2
+.temp 2 t1
+.temp 2 t2
+.temp 2 ty
+
+x2 splitwb t1, ty, yuv1
+storew y1, ty
+x2 splitwb t2, ty, yuv2
+storew y2, ty
+x2 avgub t1, t1, t2
+splitwb v, u, t1
+
+
+.function cogorc_convert_UYVY_YUY2
+.flags 2d
+.dest 4 yuy2
+.source 4 uyvy
+
+x2 swapw yuy2, uyvy
+
+
+.function cogorc_planar_chroma_420_422
+.flags 2d
+.dest 1 d1
+.dest 1 d2
+.source 1 s
+
+copyb d1, s
+copyb d2, s
+
+
+.function cogorc_planar_chroma_420_444
+.flags 2d
+.dest 2 d1
+.dest 2 d2
+.source 1 s
+.temp 2 t
+
+splatbw t, s
+storew d1, t
+storew d2, t
+
+
+.function cogorc_planar_chroma_422_444
+.flags 2d
+.dest 2 d1
+.source 1 s
+.temp 2 t
+
+splatbw t, s
+storew d1, t
+
+
+.function cogorc_planar_chroma_444_422
+.flags 2d
+.dest 1 d
+.source 2 s
+.temp 1 t1
+.temp 1 t2
+
+splitwb t1, t2, s
+avgub d, t1, t2
+
+
+.function cogorc_planar_chroma_444_420
+.flags 2d
+.dest 1 d
+.source 2 s1
+.source 2 s2
+.temp 2 t
+.temp 1 t1
+.temp 1 t2
+
+x2 avgub t, s1, s2
+splitwb t1, t2, t
+avgub d, t1, t2
+
+
+.function cogorc_planar_chroma_422_420
+.flags 2d
+.dest 1 d
+.source 1 s1
+.source 1 s2
+
+avgub d, s1, s2
+
+
+.function cogorc_convert_YUY2_AYUV
+.flags 2d
+.dest 8 ayuv
+.source 4 yuy2
+.const 2 c255 0xff
+.temp 2 yy
+.temp 2 uv
+.temp 4 ayay
+.temp 4 uvuv
+
+x2 splitwb uv, yy, yuy2
+x2 mergebw ayay, c255, yy
+mergewl uvuv, uv, uv
+x2 mergewl ayuv, ayay, uvuv
+
+
+.function cogorc_convert_UYVY_AYUV
+.flags 2d
+.dest 8 ayuv
+.source 4 uyvy
+.const 2 c255 0xff
+.temp 2 yy
+.temp 2 uv
+.temp 4 ayay
+.temp 4 uvuv
+
+x2 splitwb yy, uv, uyvy
+x2 mergebw ayay, c255, yy
+mergewl uvuv, uv, uv
+x2 mergewl ayuv, ayay, uvuv
+
+
+.function cogorc_convert_YUY2_Y42B
+.flags 2d
+.dest 2 y
+.dest 1 u
+.dest 1 v
+.source 4 yuy2
+.temp 2 uv
+
+x2 splitwb uv, y, yuy2
+splitwb v, u, uv
+
+
+.function cogorc_convert_UYVY_Y42B
+.flags 2d
+.dest 2 y
+.dest 1 u
+.dest 1 v
+.source 4 uyvy
+.temp 2 uv
+
+x2 splitwb y, uv, uyvy
+splitwb v, u, uv
+
+
+.function cogorc_convert_YUY2_Y444
+.flags 2d
+.dest 2 y
+.dest 2 uu
+.dest 2 vv
+.source 4 yuy2
+.temp 2 uv
+.temp 1 u
+.temp 1 v
+
+x2 splitwb uv, y, yuy2
+splitwb v, u, uv
+splatbw uu, u
+splatbw vv, v
+
+
+.function cogorc_convert_UYVY_Y444
+.flags 2d
+.dest 2 y
+.dest 2 uu
+.dest 2 vv
+.source 4 uyvy
+.temp 2 uv
+.temp 1 u
+.temp 1 v
+
+x2 splitwb y, uv, uyvy
+splitwb v, u, uv
+splatbw uu, u
+splatbw vv, v
+
+
+.function cogorc_convert_UYVY_I420
+.dest 2 y1
+.dest 2 y2
+.dest 1 u
+.dest 1 v
+.source 4 yuv1
+.source 4 yuv2
+.temp 2 t1
+.temp 2 t2
+.temp 2 ty
+
+x2 splitwb ty, t1, yuv1
+storew y1, ty
+x2 splitwb ty, t2, yuv2
+storew y2, ty
+x2 avgub t1, t1, t2
+splitwb v, u, t1
+
+
+
+.function cogorc_convert_AYUV_I420
+.flags 2d
+.dest 2 y1
+.dest 2 y2
+.dest 1 u
+.dest 1 v
+.source 8 ayuv1
+.source 8 ayuv2
+.temp 4 ay
+.temp 4 uv1
+.temp 4 uv2
+.temp 4 uv
+.temp 2 uu
+.temp 2 vv
+.temp 1 t1
+.temp 1 t2
+
+x2 splitlw uv1, ay, ayuv1
+x2 select1wb y1, ay
+x2 splitlw uv2, ay, ayuv2
+x2 select1wb y2, ay
+x4 avgub uv, uv1, uv2
+x2 splitwb vv, uu, uv
+splitwb t1, t2, uu
+avgub u, t1, t2
+splitwb t1, t2, vv
+avgub v, t1, t2
+
+
+
+.function cogorc_convert_AYUV_YUY2
+.flags 2d
+.dest 4 yuy2
+.source 8 ayuv
+.temp 2 yy
+.temp 2 uv1
+.temp 2 uv2
+.temp 4 ayay
+.temp 4 uvuv
+
+x2 splitlw uvuv, ayay, ayuv
+splitlw uv1, uv2, uvuv
+x2 avgub uv1, uv1, uv2
+x2 select1wb yy, ayay
+x2 mergebw yuy2, yy, uv1
+
+
+.function cogorc_convert_AYUV_UYVY
+.flags 2d
+.dest 4 yuy2
+.source 8 ayuv
+.temp 2 yy
+.temp 2 uv1
+.temp 2 uv2
+.temp 4 ayay
+.temp 4 uvuv
+
+x2 splitlw uvuv, ayay, ayuv
+splitlw uv1, uv2, uvuv
+x2 avgub uv1, uv1, uv2
+x2 select1wb yy, ayay
+x2 mergebw yuy2, uv1, yy
+
+
+
+.function cogorc_convert_AYUV_Y42B
+.flags 2d
+.dest 2 y
+.dest 1 u
+.dest 1 v
+.source 8 ayuv
+.temp 4 ayay
+.temp 4 uvuv
+.temp 2 uv1
+.temp 2 uv2
+
+x2 splitlw uvuv, ayay, ayuv
+splitlw uv1, uv2, uvuv
+x2 avgub uv1, uv1, uv2
+splitwb v, u, uv1
+x2 select1wb y, ayay
+
+
+.function cogorc_convert_AYUV_Y444
+.flags 2d
+.dest 1 y
+.dest 1 u
+.dest 1 v
+.source 4 ayuv
+.temp 2 ay
+.temp 2 uv
+
+splitlw uv, ay, ayuv
+splitwb v, u, uv
+select1wb y, ay
+
+
+.function cogorc_convert_Y42B_YUY2
+.flags 2d
+.dest 4 yuy2
+.source 2 y
+.source 1 u
+.source 1 v
+.temp 2 uv
+
+mergebw uv, u, v
+x2 mergebw yuy2, y, uv
+
+
+.function cogorc_convert_Y42B_UYVY
+.flags 2d
+.dest 4 uyvy
+.source 2 y
+.source 1 u
+.source 1 v
+.temp 2 uv
+
+mergebw uv, u, v
+x2 mergebw uyvy, uv, y
+
+
+.function cogorc_convert_Y42B_AYUV
+.flags 2d
+.dest 8 ayuv
+.source 2 yy
+.source 1 u
+.source 1 v
+.const 1 c255 255
+.temp 2 uv
+.temp 2 ay
+.temp 4 uvuv
+.temp 4 ayay
+
+mergebw uv, u, v
+x2 mergebw ayay, c255, yy
+mergewl uvuv, uv, uv
+x2 mergewl ayuv, ayay, uvuv
+
+
+.function cogorc_convert_Y444_YUY2
+.flags 2d
+.dest 4 yuy2
+.source 2 y
+.source 2 u
+.source 2 v
+.temp 2 uv
+.temp 4 uvuv
+.temp 2 uv1
+.temp 2 uv2
+
+x2 mergebw uvuv, u, v
+splitlw uv1, uv2, uvuv
+x2 avgub uv, uv1, uv2
+x2 mergebw yuy2, y, uv
+
+
+.function cogorc_convert_Y444_UYVY
+.flags 2d
+.dest 4 uyvy
+.source 2 y
+.source 2 u
+.source 2 v
+.temp 2 uv
+.temp 4 uvuv
+.temp 2 uv1
+.temp 2 uv2
+
+x2 mergebw uvuv, u, v
+splitlw uv1, uv2, uvuv
+x2 avgub uv, uv1, uv2
+x2 mergebw uyvy, uv, y
+
+
+.function cogorc_convert_Y444_AYUV
+.flags 2d
+.dest 4 ayuv
+.source 1 yy
+.source 1 u
+.source 1 v
+.const 1 c255 255
+.temp 2 uv
+.temp 2 ay
+
+mergebw uv, u, v
+mergebw ay, c255, yy
+mergewl ayuv, ay, uv
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+