Add dest iteratorsmore-iters

Notes on this / FIXME:

- This adds another 565 iterator, but we already have one. The new one
  is likely worse, so it should just go away.

- Maybe noop variants should be generated

- Maybe they should also be used as COVER_CLIP_NEAREST source
  fetchers.

Finally, maybe the macros from pixman-access.c should be moved to
pixman-inline.h so that they can be used here. That would make it
possible to generate fast path iters for basically any format.

1 files changed, 89 insertions, 7 deletions

diff --git a/pixman/pixman-fast-path-iters.c b/pixman/pixman-fast-path-iters.c
index dd72b11b..bb8acd87 100644
--- a/pixman/pixman-fast-path-iters.c
+++ b/pixman/pixman-fast-path-iters.c
@@ -952,29 +952,53 @@ bits_image_fetch_nearest_affine (pixman_image_t * image,
 }
 
 static force_inline uint32_t
-convert_a8r8g8b8 (const uint8_t *row, int x)
+convert_from_a8r8g8b8 (const uint8_t *row, int x)
 {
     return *(((uint32_t *)row) + x);
 }
 
 static force_inline uint32_t
-convert_x8r8g8b8 (const uint8_t *row, int x)
+convert_from_x8r8g8b8 (const uint8_t *row, int x)
 {
     return *(((uint32_t *)row) + x);
 }
 
 static force_inline uint32_t
-convert_a8 (const uint8_t *row, int x)
+convert_from_a8 (const uint8_t *row, int x)
 {
     return *(row + x) << 24;
 }
 
 static force_inline uint32_t
-convert_r5g6b5 (const uint8_t *row, int x)
+convert_from_r5g6b5 (const uint8_t *row, int x)
 {
     return convert_0565_to_0888 (*((uint16_t *)row + x));
 }
 
+static force_inline void
+convert_to_r5g6b5 (uint8_t *row, int x, uint32_t pixel)
+{
+    *((uint16_t *)row + x) = convert_8888_to_0565 (pixel);
+}
+
+static force_inline void
+convert_to_a8r8g8b8 (uint8_t *row, int x, uint32_t pixel)
+{
+    *(((uint32_t *)row) + x) = pixel;
+}
+
+static force_inline void
+convert_to_x8r8g8b8 (uint8_t *row, int x, uint32_t pixel)
+{
+    *(((uint32_t *)row) + x) = pixel;
+}
+
+static force_inline void
+convert_to_a8 (uint8_t *row, int x, uint32_t pixel)
+{
+    *(row + x) = pixel >> 24;
+}
+
 #define MAKE_SEPARABLE_CONVOLUTION_FETCHER(name, format, repeat_mode)  \
     static uint32_t *							\
     bits_image_fetch_separable_convolution_affine_ ## name (pixman_iter_t   *iter, \
@@ -985,7 +1009,7 @@ convert_r5g6b5 (const uint8_t *row, int x)
 	    iter->x, iter->y++,                                         \
 	    iter->width,                                                \
 	    iter->buffer, mask,                                         \
-	    convert_ ## format,                                         \
+	    convert_from_ ## format,					\
 	    PIXMAN_ ## format,                                          \
 	    repeat_mode);                                               \
 									\
@@ -1001,7 +1025,7 @@ convert_r5g6b5 (const uint8_t *row, int x)
 					  iter->x, iter->y++,		\
 					  iter->width,			\
 					  iter->buffer, mask,		\
-					  convert_ ## format,		\
+					  convert_from_ ## format,	\
 					  PIXMAN_ ## format,		\
 					  repeat_mode);			\
 	return iter->buffer;						\
@@ -1016,7 +1040,7 @@ convert_r5g6b5 (const uint8_t *row, int x)
 					 iter->x, iter->y++,		\
 					 iter->width,			\
 					 iter->buffer, mask,		\
-					 convert_ ## format,		\
+					 convert_from_ ## format,	\
 					 PIXMAN_ ## format,		\
 					 repeat_mode);			\
 	return iter->buffer;						\
@@ -1220,6 +1244,39 @@ bits_image_fetch_untransformed_float (pixman_iter_t * iter,
     return buffer;
 }
 
+#define MAKE_DEST_ITER(format)                                         \
+    static uint32_t *                                                  \
+    fetch_ ## format (pixman_iter_t *iter, const uint32_t *m)          \
+    {                                                                  \
+	uint32_t *buffer = iter->buffer;                                \
+	uint8_t *bits = (uint8_t *)iter->bits;				\
+	uint32_t mask = PIXMAN_FORMAT_A (PIXMAN_ ## format)? 0 : 0xff000000; \
+	int i;                                                          \
+									\
+	for (i = 0; i < iter->width; ++i)                               \
+	    buffer[i] = mask | convert_from_ ## format (bits, i);	\
+									\
+	return iter->buffer;                                            \
+    }									\
+    									\
+    static void								\
+    write_back_ ## format (pixman_iter_t *iter)				\
+    {									\
+	uint32_t *buffer = iter->buffer;                                \
+	uint8_t *bits = (uint8_t *)iter->bits;				\
+	int i;                                                          \
+									\
+	for (i = 0; i < iter->width; ++i)                               \
+	    convert_to_ ## format (bits, i, buffer[i]);			\
+									\
+	iter->bits += iter->stride;					\
+    }
+
+MAKE_DEST_ITER (a8r8g8b8)
+MAKE_DEST_ITER (x8r8g8b8)
+MAKE_DEST_ITER (a8)
+MAKE_DEST_ITER (r5g6b5)
+
 #define IMAGE_FLAGS							\
     (FAST_PATH_STANDARD_FLAGS | FAST_PATH_ID_TRANSFORM |		\
      FAST_PATH_BITS_IMAGE | FAST_PATH_SAMPLES_COVER_CLIP_NEAREST)
@@ -1239,6 +1296,31 @@ static const pixman_iter_info_t fast_iters[] =
       _pixman_iter_init_bits_stride,
       fast_dest_fetch_noop, fast_write_back_r5g6b5 },
 
+    { PIXMAN_a8, FAST_PATH_STD_DEST_FLAGS,
+      ITER_NARROW | ITER_DEST,
+      _pixman_iter_init_bits_stride,
+      fetch_a8, write_back_a8,
+    },
+    
+    { PIXMAN_a8r8g8b8, FAST_PATH_STD_DEST_FLAGS,
+      ITER_NARROW | ITER_DEST,
+      _pixman_iter_init_bits_stride,
+      fetch_a8r8g8b8, write_back_a8r8g8b8,
+    },
+    
+    { PIXMAN_x8r8g8b8, FAST_PATH_STD_DEST_FLAGS,
+      ITER_NARROW | ITER_DEST,
+      _pixman_iter_init_bits_stride,
+      fetch_x8r8g8b8, write_back_x8r8g8b8,
+    },
+
+    /* FIXME: We already have a better version of this */
+    { PIXMAN_r5g6b5, FAST_PATH_STD_DEST_FLAGS,
+      ITER_NARROW | ITER_DEST,
+      _pixman_iter_init_bits_stride,
+      fetch_r5g6b5, write_back_r5g6b5,
+    },
+    
     { PIXMAN_any,
       (FAST_PATH_NO_ALPHA_MAP			|
        FAST_PATH_ID_TRANSFORM			|