videoscale: Add support for 16 bit grayscale in native endianness

7 files changed, 354 insertions, 1 deletions

diff --git a/gst/videoscale/gstvideoscale.c b/gst/videoscale/gstvideoscale.c
index bceae2125..252d8ae1d 100644
--- a/gst/videoscale/gstvideoscale.c
+++ b/gst/videoscale/gstvideoscale.c
@@ -106,6 +106,12 @@ static GstStaticCaps gst_video_scale_format_caps[] = {
       "bpp = 8, "
       "depth = 8, "
       "width = " GST_VIDEO_SIZE_RANGE ", "
+      "height = " GST_VIDEO_SIZE_RANGE ", " "framerate = " GST_VIDEO_FPS_RANGE),
+  GST_STATIC_CAPS ("video/x-raw-gray, "
+      "bpp = 16, "
+      "depth = 16, "
+      "endianness = BYTE_ORDER, "
+      "width = " GST_VIDEO_SIZE_RANGE ", "
       "height = " GST_VIDEO_SIZE_RANGE ", " "framerate = " GST_VIDEO_FPS_RANGE)
 };
 
@@ -131,7 +137,8 @@ enum
   GST_VIDEO_SCALE_YV12,
   GST_VIDEO_SCALE_RGB565,
   GST_VIDEO_SCALE_RGB555,
-  GST_VIDEO_SCALE_GRAY8
+  GST_VIDEO_SCALE_GRAY8,
+  GST_VIDEO_SCALE_GRAY16
 };
 
 #define GST_TYPE_VIDEO_SCALE_METHOD (gst_video_scale_method_get_type())
@@ -445,6 +452,10 @@ gst_video_scale_prepare_size (GstVideoScale * videoscale, gint format,
       img->stride = GST_ROUND_UP_4 (img->width);
       *size = img->stride * img->height;
       break;
+    case GST_VIDEO_SCALE_GRAY16:
+      img->stride = GST_ROUND_UP_4 (img->width * 2);
+      *size = img->stride * img->height;
+      break;
     case GST_VIDEO_SCALE_I420:
     case GST_VIDEO_SCALE_YV12:
     {
@@ -768,6 +779,9 @@ gst_video_scale_transform (GstBaseTransform * trans, GstBuffer * in,
         case GST_VIDEO_SCALE_GRAY8:
           vs_image_scale_nearest_Y (dest, src, videoscale->tmp_buf);
           break;
+        case GST_VIDEO_SCALE_GRAY16:
+          vs_image_scale_nearest_Y16 (dest, src, videoscale->tmp_buf);
+          break;
         case GST_VIDEO_SCALE_I420:
         case GST_VIDEO_SCALE_YV12:
           vs_image_scale_nearest_Y (dest, src, videoscale->tmp_buf);
@@ -814,6 +828,9 @@ gst_video_scale_transform (GstBaseTransform * trans, GstBuffer * in,
         case GST_VIDEO_SCALE_GRAY8:
           vs_image_scale_linear_Y (dest, src, videoscale->tmp_buf);
           break;
+        case GST_VIDEO_SCALE_GRAY16:
+          vs_image_scale_linear_Y16 (dest, src, videoscale->tmp_buf);
+          break;
         case GST_VIDEO_SCALE_I420:
         case GST_VIDEO_SCALE_YV12:
           vs_image_scale_linear_Y (dest, src, videoscale->tmp_buf);
@@ -860,6 +877,9 @@ gst_video_scale_transform (GstBaseTransform * trans, GstBuffer * in,
         case GST_VIDEO_SCALE_GRAY8:
           vs_image_scale_4tap_Y (dest, src, videoscale->tmp_buf);
           break;
+        case GST_VIDEO_SCALE_GRAY16:
+          vs_image_scale_4tap_Y16 (dest, src, videoscale->tmp_buf);
+          break;
         case GST_VIDEO_SCALE_I420:
         case GST_VIDEO_SCALE_YV12:
           vs_image_scale_4tap_Y (dest, src, videoscale->tmp_buf);
diff --git a/gst/videoscale/vs_4tap.c b/gst/videoscale/vs_4tap.c
index e5ffef569..c41419a5b 100644
--- a/gst/videoscale/vs_4tap.c
+++ b/gst/videoscale/vs_4tap.c
@@ -210,6 +210,123 @@ vs_image_scale_4tap_Y (const VSImage * dest, const VSImage * src,
 }
 
 void
+vs_scanline_resample_4tap_Y16 (uint8_t * dest, uint8_t * src,
+    int n, int src_width, int *xacc, int increment)
+{
+  int i;
+  int j;
+  int acc;
+  int x;
+  int y;
+  uint16_t *d = (uint16_t *) dest, *s = (uint16_t *) src;
+
+  acc = *xacc;
+  for (i = 0; i < n; i++) {
+    j = acc >> 16;
+    x = (acc & 0xff00) >> 8;
+    if (j - 1 >= 0 && j + 2 < src_width) {
+      y = vs_4tap_taps[x][0] * s[MAX (j - 1, 0)];
+      y += vs_4tap_taps[x][1] * s[j];
+      y += vs_4tap_taps[x][2] * s[j + 1];
+      y += vs_4tap_taps[x][3] * s[j + 2];
+    } else {
+      y = vs_4tap_taps[x][0] * s[CLAMP (j - 1, 0, src_width - 1)];
+      y += vs_4tap_taps[x][1] * s[CLAMP (j, 0, src_width - 1)];
+      y += vs_4tap_taps[x][2] * s[CLAMP (j + 1, 0, src_width - 1)];
+      y += vs_4tap_taps[x][3] * s[CLAMP (j + 2, 0, src_width - 1)];
+    }
+    y += (1 << (SHIFT - 1));
+    d[i] = CLAMP (y >> SHIFT, 0, 65535);
+    acc += increment;
+  }
+  *xacc = acc;
+}
+
+void
+vs_scanline_merge_4tap_Y16 (uint8_t * dest, uint8_t * src1, uint8_t * src2,
+    uint8_t * src3, uint8_t * src4, int n, int acc)
+{
+  int i;
+  int y;
+  int a, b, c, d;
+  uint16_t *de = (uint16_t *) dest, *s1 = (uint16_t *) src1;
+  uint16_t *s2 = (uint16_t *) src2, *s3 = (uint16_t *) src3;
+  uint16_t *s4 = (uint16_t *) src4;
+
+  acc = (acc >> 8) & 0xff;
+  a = vs_4tap_taps[acc][0];
+  b = vs_4tap_taps[acc][1];
+  c = vs_4tap_taps[acc][2];
+  d = vs_4tap_taps[acc][3];
+  for (i = 0; i < n; i++) {
+    y = a * s1[i];
+    y += b * s2[i];
+    y += c * s3[i];
+    y += d * s4[i];
+    y += (1 << (SHIFT - 1));
+    de[i] = CLAMP (y >> SHIFT, 0, 65535);
+  }
+}
+
+
+void
+vs_image_scale_4tap_Y16 (const VSImage * dest, const VSImage * src,
+    uint8_t * tmpbuf)
+{
+  int yacc;
+  int y_increment;
+  int x_increment;
+  int i;
+  int j;
+  int xacc;
+  int k;
+
+  if (dest->height == 1)
+    y_increment = 0;
+  else
+    y_increment = ((src->height - 1) << 16) / (dest->height - 1);
+
+  if (dest->width == 1)
+    x_increment = 0;
+  else
+    x_increment = ((src->width - 1) << 16) / (dest->width - 1);
+
+  k = 0;
+  for (i = 0; i < 4; i++) {
+    xacc = 0;
+    vs_scanline_resample_4tap_Y16 (tmpbuf + i * dest->stride,
+        src->pixels + i * src->stride, dest->width, src->width,
+        &xacc, x_increment);
+  }
+
+  yacc = 0;
+  for (i = 0; i < dest->height; i++) {
+    uint8_t *t0, *t1, *t2, *t3;
+
+    j = yacc >> 16;
+
+    while (j > k) {
+      k++;
+      if (k + 3 < src->height) {
+        xacc = 0;
+        vs_scanline_resample_4tap_Y16 (tmpbuf + ((k + 3) & 3) * dest->stride,
+            src->pixels + (k + 3) * src->stride,
+            dest->width, src->width, &xacc, x_increment);
+      }
+    }
+
+    t0 = tmpbuf + (CLAMP (j - 1, 0, src->height - 1) & 3) * dest->stride;
+    t1 = tmpbuf + (CLAMP (j, 0, src->height - 1) & 3) * dest->stride;
+    t2 = tmpbuf + (CLAMP (j + 1, 0, src->height - 1) & 3) * dest->stride;
+    t3 = tmpbuf + (CLAMP (j + 2, 0, src->height - 1) & 3) * dest->stride;
+    vs_scanline_merge_4tap_Y16 (dest->pixels + i * dest->stride,
+        t0, t1, t2, t3, dest->width, yacc & 0xffff);
+
+    yacc += y_increment;
+  }
+}
+
+void
 vs_scanline_resample_4tap_RGBA (uint8_t * dest, uint8_t * src,
     int n, int src_width, int *xacc, int increment)
 {
diff --git a/gst/videoscale/vs_4tap.h b/gst/videoscale/vs_4tap.h
index fd370f191..3b13662d6 100644
--- a/gst/videoscale/vs_4tap.h
+++ b/gst/videoscale/vs_4tap.h
@@ -82,5 +82,12 @@ void vs_scanline_merge_4tap_RGB555 (uint8_t *dest, uint8_t *src1, uint8_t *src2,
 void vs_image_scale_4tap_RGB555 (const VSImage * dest, const VSImage * src,
     uint8_t * tmpbuf);
 
+void vs_scanline_resample_4tap_Y16 (uint8_t *dest, uint8_t *src,
+    int n, int src_width, int *xacc, int increment);
+void vs_scanline_merge_4tap_Y16 (uint8_t *dest, uint8_t *src1, uint8_t *src2,
+    uint8_t *src3, uint8_t *src4, int n, int acc);
+void vs_image_scale_4tap_Y16 (const VSImage * dest, const VSImage * src,
+    uint8_t * tmpbuf);
+
 #endif
 
diff --git a/gst/videoscale/vs_image.c b/gst/videoscale/vs_image.c
index 58401835b..4991d7392 100644
--- a/gst/videoscale/vs_image.c
+++ b/gst/videoscale/vs_image.c
@@ -692,6 +692,136 @@ vs_image_scale_linear_Y (const VSImage * dest, const VSImage * src,
   }
 }
 
+void
+vs_image_scale_nearest_Y16 (const VSImage * dest, const VSImage * src,
+    uint8_t * tmpbuf)
+{
+  int acc;
+  int y_increment;
+  int x_increment;
+  int i;
+  int j;
+  int x;
+  int xacc;
+
+  if (dest->height == 1)
+    y_increment = 0;
+  else
+    y_increment = ((src->height - 1) << 16) / (dest->height - 1);
+
+  if (dest->width == 1)
+    x_increment = 0;
+  else
+    x_increment = ((src->width - 1) << 16) / (dest->width - 1);
+
+  acc = 0;
+  for (i = 0; i < dest->height; i++) {
+    j = acc >> 16;
+    x = acc & 0xffff;
+
+    xacc = 0;
+    vs_scanline_resample_nearest_Y16 (dest->pixels + i * dest->stride,
+        src->pixels + j * src->stride, src->width, dest->width, &xacc,
+        x_increment);
+
+    acc += y_increment;
+  }
+}
+
+void
+vs_image_scale_linear_Y16 (const VSImage * dest, const VSImage * src,
+    uint8_t * tmpbuf)
+{
+  int acc;
+  int y_increment;
+  int x_increment;
+  uint8_t *tmp1;
+  uint8_t *tmp2;
+  int y1;
+  int y2;
+  int i;
+  int j;
+  int x;
+  int dest_size;
+  int xacc;
+
+  if (dest->height == 1)
+    y_increment = 0;
+  else
+    y_increment = ((src->height - 1) << 16) / (dest->height - 1);
+
+  if (dest->width == 1)
+    x_increment = 0;
+  else
+    x_increment = ((src->width - 1) << 16) / (dest->width - 1);
+
+  dest_size = 2 * dest->width;
+
+  tmp1 = tmpbuf;
+  tmp2 = tmpbuf + dest_size;
+
+  acc = 0;
+  xacc = 0;
+  y2 = -1;
+  vs_scanline_resample_linear_Y16 (tmp1, src->pixels, src->width, dest->width,
+      &xacc, x_increment);
+  y1 = 0;
+  for (i = 0; i < dest->height; i++) {
+    j = acc >> 16;
+    x = acc & 0xffff;
+
+    if (x == 0) {
+      if (j == y1) {
+        memcpy (dest->pixels + i * dest->stride, tmp1, dest_size);
+      } else if (j == y2) {
+        memcpy (dest->pixels + i * dest->stride, tmp2, dest_size);
+      } else {
+        xacc = 0;
+        vs_scanline_resample_linear_Y16 (tmp1, src->pixels + j * src->stride,
+            src->width, dest->width, &xacc, x_increment);
+        y1 = j;
+        memcpy (dest->pixels + i * dest->stride, tmp1, dest_size);
+      }
+    } else {
+      if (j == y1) {
+        if (j + 1 != y2) {
+          xacc = 0;
+          vs_scanline_resample_linear_Y16 (tmp2,
+              src->pixels + (j + 1) * src->stride, src->width, dest->width,
+              &xacc, x_increment);
+          y2 = j + 1;
+        }
+        vs_scanline_merge_linear_Y16 (dest->pixels + i * dest->stride,
+            tmp1, tmp2, dest->width, x);
+      } else if (j == y2) {
+        if (j + 1 != y1) {
+          xacc = 0;
+          vs_scanline_resample_linear_Y16 (tmp1,
+              src->pixels + (j + 1) * src->stride, src->width, dest->width,
+              &xacc, x_increment);
+          y1 = j + 1;
+        }
+        vs_scanline_merge_linear_Y16 (dest->pixels + i * dest->stride,
+            tmp2, tmp1, dest->width, x);
+      } else {
+        xacc = 0;
+        vs_scanline_resample_linear_Y16 (tmp1, src->pixels + j * src->stride,
+            src->width, dest->width, &xacc, x_increment);
+        y1 = j;
+        xacc = 0;
+        vs_scanline_resample_linear_Y16 (tmp2,
+            src->pixels + (j + 1) * src->stride, src->width, dest->width, &xacc,
+            x_increment);
+        y2 = (j + 1);
+        vs_scanline_merge_linear_Y16 (dest->pixels + i * dest->stride,
+            tmp1, tmp2, dest->width, x);
+      }
+    }
+
+    acc += y_increment;
+  }
+}
+
 /* RGB565 */
 
 void
diff --git a/gst/videoscale/vs_image.h b/gst/videoscale/vs_image.h
index ba912f929..9771c774a 100644
--- a/gst/videoscale/vs_image.h
+++ b/gst/videoscale/vs_image.h
@@ -74,5 +74,10 @@ void vs_image_scale_nearest_RGB555 (const VSImage *dest, const VSImage *src,
 void vs_image_scale_linear_RGB555 (const VSImage *dest, const VSImage *src,
     uint8_t *tmpbuf);
 
+void vs_image_scale_nearest_Y16 (const VSImage *dest, const VSImage *src,
+    uint8_t *tmpbuf);
+void vs_image_scale_linear_Y16 (const VSImage *dest, const VSImage *src,
+    uint8_t *tmpbuf);
+
 #endif
 
diff --git a/gst/videoscale/vs_scanline.c b/gst/videoscale/vs_scanline.c
index 7cebef521..d524e73b2 100644
--- a/gst/videoscale/vs_scanline.c
+++ b/gst/videoscale/vs_scanline.c
@@ -95,6 +95,76 @@ vs_scanline_merge_linear_Y (uint8_t * dest, uint8_t * src1, uint8_t * src2,
   oil_merge_linear_u8 (dest, src1, src2, &value, n);
 }
 
+void
+vs_scanline_downsample_Y16 (uint8_t * dest, uint8_t * src, int n)
+{
+  int i;
+  uint16_t *d = (uint16_t *) dest, *s = (uint16_t *) src;
+
+  for (i = 0; i < n; i++) {
+    d[i] = (s[i * 2] + s[i * 2 + 1]) / 2;
+  }
+}
+
+void
+vs_scanline_resample_nearest_Y16 (uint8_t * dest, uint8_t * src, int src_width,
+    int n, int *accumulator, int increment)
+{
+  int acc = *accumulator;
+  int i;
+  int j;
+  int x;
+  uint16_t *d = (uint16_t *) dest, *s = (uint16_t *) src;
+
+  for (i = 0; i < n; i++) {
+    j = acc >> 16;
+    x = acc & 0xffff;
+    d[i] = (x < 32768 || j + 1 >= src_width) ? s[j] : s[j + 1];
+
+    acc += increment;
+  }
+
+  *accumulator = acc;
+}
+
+#include <glib.h>
+void
+vs_scanline_resample_linear_Y16 (uint8_t * dest, uint8_t * src, int src_width,
+    int n, int *accumulator, int increment)
+{
+  int acc = *accumulator;
+  int i;
+  int j;
+  int x;
+  uint16_t *d = (uint16_t *) dest, *s = (uint16_t *) src;
+
+  for (i = 0; i < n; i++) {
+    j = acc >> 16;
+    x = acc & 0xffff;
+
+    if (j + 1 < src_width)
+      d[i] = (s[j] * (65536 - x) + s[j + 1] * x) >> 16;
+    else
+      d[i] = s[j];
+
+    acc += increment;
+  }
+
+  *accumulator = acc;
+}
+
+void
+vs_scanline_merge_linear_Y16 (uint8_t * dest, uint8_t * src1, uint8_t * src2,
+    int n, int x)
+{
+  int i;
+  uint16_t *d = (uint16_t *) dest, *s1 = (uint16_t *) src1, *s2 =
+      (uint16_t *) src2;
+
+  for (i = 0; i < n; i++) {
+    d[i] = (s1[i] * (65536 - x) + s2[i] * x) >> 16;
+  }
+}
 
 /* RGBA */
 
diff --git a/gst/videoscale/vs_scanline.h b/gst/videoscale/vs_scanline.h
index 655a17d99..2e0b17960 100644
--- a/gst/videoscale/vs_scanline.h
+++ b/gst/videoscale/vs_scanline.h
@@ -65,6 +65,10 @@ void vs_scanline_resample_nearest_RGB555 (uint8_t *dest, uint8_t *src, int src_w
 void vs_scanline_resample_linear_RGB555 (uint8_t *dest, uint8_t *src, int src_width, int n, int *accumulator, int increment);
 void vs_scanline_merge_linear_RGB555 (uint8_t *dest, uint8_t *src1, uint8_t *src2, int n, int x);
 
+void vs_scanline_downsample_Y16 (uint8_t *dest, uint8_t *src, int n);
+void vs_scanline_resample_nearest_Y16 (uint8_t *dest, uint8_t *src, int n, int src_width, int *accumulator, int increment);
+void vs_scanline_resample_linear_Y16 (uint8_t *dest, uint8_t *src, int n, int src_width, int *accumulator, int increment);
+void vs_scanline_merge_linear_Y16 (uint8_t *dest, uint8_t *src1, uint8_t *src2, int n, int x);
 
 #endif