Merge branch 'eliminate-composite'

Conflicts:
	pixman/pixman-sse2.c

11 files changed, 750 insertions, 1008 deletions

diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index 26f7267c..557301e3 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -383,7 +383,7 @@ pixman_blt_neon (uint32_t *src_bits,
     }
 }
 
-static const pixman_fast_path_t arm_neon_fast_path_array[] =
+static const pixman_fast_path_t arm_neon_fast_paths[] =
 {
     PIXMAN_STD_FAST_PATH (SRC,  r5g6b5,   null,     r5g6b5,   neon_composite_src_0565_0565),
     PIXMAN_STD_FAST_PATH (SRC,  b5g6r5,   null,     b5g6r5,   neon_composite_src_0565_0565),
@@ -435,41 +435,6 @@ static const pixman_fast_path_t arm_neon_fast_path_array[] =
     { PIXMAN_OP_NONE },
 };
 
-const pixman_fast_path_t *const arm_neon_fast_paths = arm_neon_fast_path_array;
-
-static void
-arm_neon_composite (pixman_implementation_t *imp,
-                    pixman_op_t              op,
-                    pixman_image_t *         src,
-                    pixman_image_t *         mask,
-                    pixman_image_t *         dest,
-                    int32_t                  src_x,
-                    int32_t                  src_y,
-                    int32_t                  mask_x,
-                    int32_t                  mask_y,
-                    int32_t                  dest_x,
-                    int32_t                  dest_y,
-                    int32_t                  width,
-                    int32_t                  height)
-{
-    if (_pixman_run_fast_path (arm_neon_fast_paths, imp,
-                               op, src, mask, dest,
-                               src_x, src_y,
-                               mask_x, mask_y,
-                               dest_x, dest_y,
-                               width, height))
-    {
-	return;
-    }
-
-    _pixman_implementation_composite (imp->delegate, op,
-                                      src, mask, dest,
-                                      src_x, src_y,
-                                      mask_x, mask_y,
-                                      dest_x, dest_y,
-                                      width, height);
-}
-
 static pixman_bool_t
 arm_neon_blt (pixman_implementation_t *imp,
               uint32_t *               src_bits,
@@ -551,12 +516,12 @@ pixman_implementation_t *
 _pixman_implementation_create_arm_neon (void)
 {
     pixman_implementation_t *general = _pixman_implementation_create_fast_path ();
-    pixman_implementation_t *imp = _pixman_implementation_create (general);
+    pixman_implementation_t *imp =
+	_pixman_implementation_create (general, arm_neon_fast_paths);
 
     imp->combine_32[PIXMAN_OP_OVER] = neon_combine_over_u;
     imp->combine_32[PIXMAN_OP_ADD] = neon_combine_add_u;
 
-    imp->composite = arm_neon_composite;
     imp->blt = arm_neon_blt;
     imp->fill = arm_neon_fill;
 
diff --git a/pixman/pixman-arm-simd.c b/pixman/pixman-arm-simd.c
index dd8dc5c0..09a28883 100644
--- a/pixman/pixman-arm-simd.c
+++ b/pixman/pixman-arm-simd.c
@@ -419,7 +419,7 @@ arm_composite_over_n_8_8888 (pixman_implementation_t * impl,
     }
 }
 
-static const pixman_fast_path_t arm_simd_fast_path_array[] =
+static const pixman_fast_path_t arm_simd_fast_paths[] =
 {
     PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null, a8r8g8b8, arm_composite_over_8888_8888),
     PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null, x8r8g8b8, arm_composite_over_8888_8888),
@@ -438,48 +438,11 @@ static const pixman_fast_path_t arm_simd_fast_path_array[] =
     { PIXMAN_OP_NONE },
 };
 
-const pixman_fast_path_t *const arm_simd_fast_paths = arm_simd_fast_path_array;
-
-static void
-arm_simd_composite (pixman_implementation_t *imp,
-                    pixman_op_t              op,
-                    pixman_image_t *         src,
-                    pixman_image_t *         mask,
-                    pixman_image_t *         dest,
-                    int32_t                  src_x,
-                    int32_t                  src_y,
-                    int32_t                  mask_x,
-                    int32_t                  mask_y,
-                    int32_t                  dest_x,
-                    int32_t                  dest_y,
-                    int32_t                  width,
-                    int32_t                  height)
-{
-    if (_pixman_run_fast_path (arm_simd_fast_paths, imp,
-                               op, src, mask, dest,
-                               src_x, src_y,
-                               mask_x, mask_y,
-                               dest_x, dest_y,
-                               width, height))
-    {
-	return;
-    }
-
-    _pixman_implementation_composite (imp->delegate, op,
-                                      src, mask, dest,
-                                      src_x, src_y,
-                                      mask_x, mask_y,
-                                      dest_x, dest_y,
-                                      width, height);
-}
-
 pixman_implementation_t *
 _pixman_implementation_create_arm_simd (void)
 {
     pixman_implementation_t *general = _pixman_implementation_create_fast_path ();
-    pixman_implementation_t *imp = _pixman_implementation_create (general);
-
-    imp->composite = arm_simd_composite;
+    pixman_implementation_t *imp = _pixman_implementation_create (general, arm_simd_fast_paths);
 
     return imp;
 }
diff --git a/pixman/pixman-fast-path.c b/pixman/pixman-fast-path.c
index 170e9d69..4d26b0fa 100644
--- a/pixman/pixman-fast-path.c
+++ b/pixman/pixman-fast-path.c
@@ -1033,7 +1033,7 @@ fast_composite_add_n_8_8 (pixman_implementation_t *imp,
 #define UPDATE_BITMASK(n) ((n) << 1)
 #endif
 
-#define TEST_BIT(p, n)						\
+#define TEST_BIT(p, n)					\
     (*((p) + ((n) >> 5)) & CREATE_BITMASK ((n) & 31))
 #define SET_BIT(p, n)							\
     do { *((p) + ((n) >> 5)) |= CREATE_BITMASK ((n) & 31); } while (0);
@@ -1345,64 +1345,6 @@ fast_composite_src_8888_x888 (pixman_implementation_t *imp,
     }
 }
 
-static const pixman_fast_path_t c_fast_paths[] =
-{
-    PIXMAN_STD_FAST_PATH (OVER, solid, a8, r5g6b5, fast_composite_over_n_8_0565),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a8, b5g6r5, fast_composite_over_n_8_0565),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a8, r8g8b8, fast_composite_over_n_8_0888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a8, b8g8r8, fast_composite_over_n_8_0888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a8, a8r8g8b8, fast_composite_over_n_8_8888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a8, x8r8g8b8, fast_composite_over_n_8_8888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a8, a8b8g8r8, fast_composite_over_n_8_8888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a8, x8b8g8r8, fast_composite_over_n_8_8888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a1, a8r8g8b8, fast_composite_over_n_1_8888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a1, x8r8g8b8, fast_composite_over_n_1_8888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a1, a8b8g8r8, fast_composite_over_n_1_8888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a1, x8b8g8r8, fast_composite_over_n_1_8888),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a1, r5g6b5,   fast_composite_over_n_1_0565),
-    PIXMAN_STD_FAST_PATH (OVER, solid, a1, b5g6r5,   fast_composite_over_n_1_0565),
-    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8r8g8b8, a8r8g8b8, fast_composite_over_n_8888_8888_ca),
-    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8r8g8b8, x8r8g8b8, fast_composite_over_n_8888_8888_ca),
-    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8r8g8b8, r5g6b5, fast_composite_over_n_8888_0565_ca),
-    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8b8g8r8, a8b8g8r8, fast_composite_over_n_8888_8888_ca),
-    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8b8g8r8, x8b8g8r8, fast_composite_over_n_8888_8888_ca),
-    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8b8g8r8, b5g6r5, fast_composite_over_n_8888_0565_ca),
-    PIXMAN_STD_FAST_PATH (OVER, x8r8g8b8, a8, x8r8g8b8, fast_composite_over_x888_8_8888),
-    PIXMAN_STD_FAST_PATH (OVER, x8r8g8b8, a8, a8r8g8b8, fast_composite_over_x888_8_8888),
-    PIXMAN_STD_FAST_PATH (OVER, x8b8g8r8, a8, x8b8g8r8, fast_composite_over_x888_8_8888),
-    PIXMAN_STD_FAST_PATH (OVER, x8b8g8r8, a8, a8b8g8r8, fast_composite_over_x888_8_8888),
-    PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null, a8r8g8b8, fast_composite_over_8888_8888),
-    PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null, x8r8g8b8, fast_composite_over_8888_8888),
-    PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null, r5g6b5, fast_composite_over_8888_0565),
-    PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null, a8b8g8r8, fast_composite_over_8888_8888),
-    PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null, x8b8g8r8, fast_composite_over_8888_8888),
-    PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null, b5g6r5, fast_composite_over_8888_0565),
-    PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, null, a8r8g8b8, fast_composite_add_8888_8888),
-    PIXMAN_STD_FAST_PATH (ADD, a8b8g8r8, null, a8b8g8r8, fast_composite_add_8888_8888),
-    PIXMAN_STD_FAST_PATH (ADD, a8, null, a8, fast_composite_add_8000_8000),
-    PIXMAN_STD_FAST_PATH (ADD, a1, null, a1, fast_composite_add_1000_1000),
-    PIXMAN_STD_FAST_PATH_CA (ADD, solid, a8r8g8b8, a8r8g8b8, fast_composite_add_n_8888_8888_ca),
-    PIXMAN_STD_FAST_PATH (ADD, solid, a8, a8, fast_composite_add_n_8_8),
-    PIXMAN_STD_FAST_PATH (SRC, solid, null, a8r8g8b8, fast_composite_solid_fill),
-    PIXMAN_STD_FAST_PATH (SRC, solid, null, x8r8g8b8, fast_composite_solid_fill),
-    PIXMAN_STD_FAST_PATH (SRC, solid, null, a8b8g8r8, fast_composite_solid_fill),
-    PIXMAN_STD_FAST_PATH (SRC, solid, null, x8b8g8r8, fast_composite_solid_fill),
-    PIXMAN_STD_FAST_PATH (SRC, solid, null, a8, fast_composite_solid_fill),
-    PIXMAN_STD_FAST_PATH (SRC, solid, null, r5g6b5, fast_composite_solid_fill),
-    PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, x8r8g8b8, fast_composite_src_8888_x888),
-    PIXMAN_STD_FAST_PATH (SRC, x8r8g8b8, null, x8r8g8b8, fast_composite_src_8888_x888),
-    PIXMAN_STD_FAST_PATH (SRC, a8b8g8r8, null, x8b8g8r8, fast_composite_src_8888_x888),
-    PIXMAN_STD_FAST_PATH (SRC, x8b8g8r8, null, x8b8g8r8, fast_composite_src_8888_x888),
-    PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, r5g6b5, fast_composite_src_x888_0565),
-    PIXMAN_STD_FAST_PATH (SRC, x8r8g8b8, null, r5g6b5, fast_composite_src_x888_0565),
-    PIXMAN_STD_FAST_PATH (SRC, a8b8g8r8, null, b5g6r5, fast_composite_src_x888_0565),
-    PIXMAN_STD_FAST_PATH (SRC, x8b8g8r8, null, b5g6r5, fast_composite_src_x888_0565),
-    PIXMAN_STD_FAST_PATH (IN, a8, null, a8, fast_composite_in_8_8),
-    PIXMAN_STD_FAST_PATH (IN, solid, a8, a8, fast_composite_in_n_8_8),
-
-    { PIXMAN_OP_NONE },
-};
-
 static force_inline pixman_bool_t
 repeat (pixman_repeat_t repeat, int *c, int size)
 {
@@ -1591,67 +1533,98 @@ fast_composite_scaled_nearest (pixman_implementation_t *imp,
     }
 }
 
-static void
-fast_path_composite (pixman_implementation_t *imp,
-                     pixman_op_t              op,
-                     pixman_image_t *         src,
-                     pixman_image_t *         mask,
-                     pixman_image_t *         dest,
-                     int32_t                  src_x,
-                     int32_t                  src_y,
-                     int32_t                  mask_x,
-                     int32_t                  mask_y,
-                     int32_t                  dest_x,
-                     int32_t                  dest_y,
-                     int32_t                  width,
-                     int32_t                  height)
+static const pixman_fast_path_t c_fast_paths[] =
 {
-    if (src->type == BITS
-        && src->common.transform
-        && !mask
-        && (op == PIXMAN_OP_SRC || op == PIXMAN_OP_OVER)
-        && !src->common.alpha_map && !dest->common.alpha_map
-        && (src->common.filter == PIXMAN_FILTER_NEAREST)
-	&& (dest->bits.format == PIXMAN_a8r8g8b8 || dest->bits.format == PIXMAN_x8r8g8b8)
-	&& (src->bits.format == PIXMAN_a8r8g8b8 || src->bits.format == PIXMAN_x8r8g8b8)
-        && !src->bits.read_func && !src->bits.write_func
-        && !dest->bits.read_func && !dest->bits.write_func)
-    {
-	/* ensure that the transform matrix only has a scale */
-	if (src->common.transform->matrix[0][1] == 0 &&
-	    src->common.transform->matrix[1][0] == 0 &&
-	    src->common.transform->matrix[2][0] == 0 &&
-	    src->common.transform->matrix[2][1] == 0 &&
-	    src->common.transform->matrix[2][2] == pixman_fixed_1)
-	{
-	    _pixman_walk_composite_region (imp, op,
-	                                   src, mask, dest,
-	                                   src_x, src_y,
-	                                   mask_x, mask_y,
-	                                   dest_x, dest_y,
-	                                   width, height,
-	                                   fast_composite_scaled_nearest);
-	    return;
-	}
-    }
+    PIXMAN_STD_FAST_PATH (OVER, solid, a8, r5g6b5, fast_composite_over_n_8_0565),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a8, b5g6r5, fast_composite_over_n_8_0565),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a8, r8g8b8, fast_composite_over_n_8_0888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a8, b8g8r8, fast_composite_over_n_8_0888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a8, a8r8g8b8, fast_composite_over_n_8_8888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a8, x8r8g8b8, fast_composite_over_n_8_8888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a8, a8b8g8r8, fast_composite_over_n_8_8888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a8, x8b8g8r8, fast_composite_over_n_8_8888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a1, a8r8g8b8, fast_composite_over_n_1_8888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a1, x8r8g8b8, fast_composite_over_n_1_8888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a1, a8b8g8r8, fast_composite_over_n_1_8888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a1, x8b8g8r8, fast_composite_over_n_1_8888),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a1, r5g6b5,   fast_composite_over_n_1_0565),
+    PIXMAN_STD_FAST_PATH (OVER, solid, a1, b5g6r5,   fast_composite_over_n_1_0565),
+    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8r8g8b8, a8r8g8b8, fast_composite_over_n_8888_8888_ca),
+    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8r8g8b8, x8r8g8b8, fast_composite_over_n_8888_8888_ca),
+    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8r8g8b8, r5g6b5, fast_composite_over_n_8888_0565_ca),
+    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8b8g8r8, a8b8g8r8, fast_composite_over_n_8888_8888_ca),
+    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8b8g8r8, x8b8g8r8, fast_composite_over_n_8888_8888_ca),
+    PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8b8g8r8, b5g6r5, fast_composite_over_n_8888_0565_ca),
+    PIXMAN_STD_FAST_PATH (OVER, x8r8g8b8, a8, x8r8g8b8, fast_composite_over_x888_8_8888),
+    PIXMAN_STD_FAST_PATH (OVER, x8r8g8b8, a8, a8r8g8b8, fast_composite_over_x888_8_8888),
+    PIXMAN_STD_FAST_PATH (OVER, x8b8g8r8, a8, x8b8g8r8, fast_composite_over_x888_8_8888),
+    PIXMAN_STD_FAST_PATH (OVER, x8b8g8r8, a8, a8b8g8r8, fast_composite_over_x888_8_8888),
+    PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null, a8r8g8b8, fast_composite_over_8888_8888),
+    PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null, x8r8g8b8, fast_composite_over_8888_8888),
+    PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null, r5g6b5, fast_composite_over_8888_0565),
+    PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null, a8b8g8r8, fast_composite_over_8888_8888),
+    PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null, x8b8g8r8, fast_composite_over_8888_8888),
+    PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null, b5g6r5, fast_composite_over_8888_0565),
+    PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, null, a8r8g8b8, fast_composite_add_8888_8888),
+    PIXMAN_STD_FAST_PATH (ADD, a8b8g8r8, null, a8b8g8r8, fast_composite_add_8888_8888),
+    PIXMAN_STD_FAST_PATH (ADD, a8, null, a8, fast_composite_add_8000_8000),
+    PIXMAN_STD_FAST_PATH (ADD, a1, null, a1, fast_composite_add_1000_1000),
+    PIXMAN_STD_FAST_PATH_CA (ADD, solid, a8r8g8b8, a8r8g8b8, fast_composite_add_n_8888_8888_ca),
+    PIXMAN_STD_FAST_PATH (ADD, solid, a8, a8, fast_composite_add_n_8_8),
+    PIXMAN_STD_FAST_PATH (SRC, solid, null, a8r8g8b8, fast_composite_solid_fill),
+    PIXMAN_STD_FAST_PATH (SRC, solid, null, x8r8g8b8, fast_composite_solid_fill),
+    PIXMAN_STD_FAST_PATH (SRC, solid, null, a8b8g8r8, fast_composite_solid_fill),
+    PIXMAN_STD_FAST_PATH (SRC, solid, null, x8b8g8r8, fast_composite_solid_fill),
+    PIXMAN_STD_FAST_PATH (SRC, solid, null, a8, fast_composite_solid_fill),
+    PIXMAN_STD_FAST_PATH (SRC, solid, null, r5g6b5, fast_composite_solid_fill),
+    PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, x8r8g8b8, fast_composite_src_8888_x888),
+    PIXMAN_STD_FAST_PATH (SRC, x8r8g8b8, null, x8r8g8b8, fast_composite_src_8888_x888),
+    PIXMAN_STD_FAST_PATH (SRC, a8b8g8r8, null, x8b8g8r8, fast_composite_src_8888_x888),
+    PIXMAN_STD_FAST_PATH (SRC, x8b8g8r8, null, x8b8g8r8, fast_composite_src_8888_x888),
+    PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, r5g6b5, fast_composite_src_x888_0565),
+    PIXMAN_STD_FAST_PATH (SRC, x8r8g8b8, null, r5g6b5, fast_composite_src_x888_0565),
+    PIXMAN_STD_FAST_PATH (SRC, a8b8g8r8, null, b5g6r5, fast_composite_src_x888_0565),
+    PIXMAN_STD_FAST_PATH (SRC, x8b8g8r8, null, b5g6r5, fast_composite_src_x888_0565),
+    PIXMAN_STD_FAST_PATH (IN, a8, null, a8, fast_composite_in_8_8),
+    PIXMAN_STD_FAST_PATH (IN, solid, a8, a8, fast_composite_in_n_8_8),
 
-    if (_pixman_run_fast_path (c_fast_paths, imp,
-                               op, src, mask, dest,
-                               src_x, src_y,
-                               mask_x, mask_y,
-                               dest_x, dest_y,
-                               width, height))
-    {
-	return;
+#define SCALED_NEAREST_FLAGS						\
+    (FAST_PATH_SCALE_TRANSFORM	|					\
+     FAST_PATH_NO_ALPHA_MAP	|					\
+     FAST_PATH_NEAREST_FILTER	|					\
+     FAST_PATH_NO_ACCESSORS	|					\
+     FAST_PATH_NO_WIDE_FORMAT)
+
+#define NEAREST_FAST_PATH(op,s,d)		\
+    {   PIXMAN_OP_ ## op,			\
+	PIXMAN_ ## s, SCALED_NEAREST_FLAGS,	\
+	PIXMAN_null, 0,				\
+	PIXMAN_ ## d, FAST_PATH_STD_DEST_FLAGS,	\
+	fast_composite_scaled_nearest,		\
     }
 
-    _pixman_implementation_composite (imp->delegate, op,
-                                      src, mask, dest,
-                                      src_x, src_y,
-                                      mask_x, mask_y,
-                                      dest_x, dest_y,
-                                      width, height);
-}
+    NEAREST_FAST_PATH (SRC, x8r8g8b8, x8r8g8b8),
+    NEAREST_FAST_PATH (SRC, a8r8g8b8, x8r8g8b8),
+    NEAREST_FAST_PATH (SRC, x8b8g8r8, x8b8g8r8),
+    NEAREST_FAST_PATH (SRC, a8b8g8r8, x8b8g8r8),
+
+    NEAREST_FAST_PATH (SRC, x8r8g8b8, a8r8g8b8),
+    NEAREST_FAST_PATH (SRC, a8r8g8b8, a8r8g8b8),
+    NEAREST_FAST_PATH (SRC, x8b8g8r8, a8b8g8r8),
+    NEAREST_FAST_PATH (SRC, a8b8g8r8, a8b8g8r8),
+
+    NEAREST_FAST_PATH (OVER, x8r8g8b8, x8r8g8b8),
+    NEAREST_FAST_PATH (OVER, a8r8g8b8, x8r8g8b8),
+    NEAREST_FAST_PATH (OVER, x8b8g8r8, x8b8g8r8),
+    NEAREST_FAST_PATH (OVER, a8b8g8r8, x8b8g8r8),
+
+    NEAREST_FAST_PATH (OVER, x8r8g8b8, a8r8g8b8),
+    NEAREST_FAST_PATH (OVER, a8r8g8b8, a8r8g8b8),
+    NEAREST_FAST_PATH (OVER, x8b8g8r8, a8b8g8r8),
+    NEAREST_FAST_PATH (OVER, a8b8g8r8, a8b8g8r8),
+
+    {   PIXMAN_OP_NONE	},
+};
 
 static void
 pixman_fill8 (uint32_t *bits,
@@ -1764,9 +1737,8 @@ pixman_implementation_t *
 _pixman_implementation_create_fast_path (void)
 {
     pixman_implementation_t *general = _pixman_implementation_create_general ();
-    pixman_implementation_t *imp = _pixman_implementation_create (general);
+    pixman_implementation_t *imp = _pixman_implementation_create (general, c_fast_paths);
 
-    imp->composite = fast_path_composite;
     imp->fill = fast_path_fill;
 
     return imp;
diff --git a/pixman/pixman-general.c b/pixman/pixman-general.c
index c96a3f97..bddf79aa 100644
--- a/pixman/pixman-general.c
+++ b/pixman/pixman-general.c
@@ -264,26 +264,11 @@ general_composite_rect  (pixman_implementation_t *imp,
 	free (scanline_buffer);
 }
 
-static void
-general_composite (pixman_implementation_t * imp,
-                   pixman_op_t               op,
-                   pixman_image_t *          src,
-                   pixman_image_t *          mask,
-                   pixman_image_t *          dest,
-                   int32_t                   src_x,
-                   int32_t                   src_y,
-                   int32_t                   mask_x,
-                   int32_t                   mask_y,
-                   int32_t                   dest_x,
-                   int32_t                   dest_y,
-                   int32_t                   width,
-                   int32_t                   height)
+static const pixman_fast_path_t general_fast_path[] =
 {
-    _pixman_walk_composite_region (imp, op, src, mask, dest, src_x, src_y,
-                                   mask_x, mask_y, dest_x, dest_y,
-				   width, height,
-                                   general_composite_rect);
-}
+    { PIXMAN_OP_any, PIXMAN_any, 0, PIXMAN_any,	0, PIXMAN_any, 0, general_composite_rect },
+    { PIXMAN_OP_NONE }
+};
 
 static pixman_bool_t
 general_blt (pixman_implementation_t *imp,
@@ -322,12 +307,11 @@ general_fill (pixman_implementation_t *imp,
 pixman_implementation_t *
 _pixman_implementation_create_general (void)
 {
-    pixman_implementation_t *imp = _pixman_implementation_create (NULL);
+    pixman_implementation_t *imp = _pixman_implementation_create (NULL, general_fast_path);
 
     _pixman_setup_combiner_functions_32 (imp);
     _pixman_setup_combiner_functions_64 (imp);
 
-    imp->composite = general_composite;
     imp->blt = general_blt;
     imp->fill = general_fill;
 
diff --git a/pixman/pixman-implementation.c b/pixman/pixman-implementation.c
index 6488332b..bc3749ef 100644
--- a/pixman/pixman-implementation.c
+++ b/pixman/pixman-implementation.c
@@ -28,30 +28,6 @@
 #include "pixman-private.h"
 
 static void
-delegate_composite (pixman_implementation_t * imp,
-                    pixman_op_t               op,
-                    pixman_image_t *          src,
-                    pixman_image_t *          mask,
-                    pixman_image_t *          dest,
-                    int32_t                   src_x,
-                    int32_t                   src_y,
-                    int32_t                   mask_x,
-                    int32_t                   mask_y,
-                    int32_t                   dest_x,
-                    int32_t                   dest_y,
-                    int32_t                   width,
-                    int32_t                   height)
-{
-    _pixman_implementation_composite (imp->delegate,
-                                      op,
-                                      src, mask, dest,
-                                      src_x, src_y,
-                                      mask_x, mask_y,
-                                      dest_x, dest_y,
-                                      width, height);
-}
-
-static void
 delegate_combine_32 (pixman_implementation_t * imp,
                      pixman_op_t               op,
                      uint32_t *                dest,
@@ -136,7 +112,8 @@ delegate_fill (pixman_implementation_t *imp,
 }
 
 pixman_implementation_t *
-_pixman_implementation_create (pixman_implementation_t *delegate)
+_pixman_implementation_create (pixman_implementation_t *delegate,
+			       const pixman_fast_path_t *fast_paths)
 {
     pixman_implementation_t *imp = malloc (sizeof (pixman_implementation_t));
     pixman_implementation_t *d;
@@ -145,6 +122,8 @@ _pixman_implementation_create (pixman_implementation_t *delegate)
     if (!imp)
 	return NULL;
 
+    assert (fast_paths);
+
     /* Make sure the whole delegate chain has the right toplevel */
     imp->delegate = delegate;
     for (d = imp; d != NULL; d = d->delegate)
@@ -152,7 +131,6 @@ _pixman_implementation_create (pixman_implementation_t *delegate)
 
     /* Fill out function pointers with ones that just delegate
      */
-    imp->composite = delegate_composite;
     imp->blt = delegate_blt;
     imp->fill = delegate_fill;
 
@@ -164,6 +142,8 @@ _pixman_implementation_create (pixman_implementation_t *delegate)
 	imp->combine_64_ca[i] = delegate_combine_64_ca;
     }
 
+    imp->fast_paths = fast_paths;
+    
     return imp;
 }
 
@@ -211,27 +191,6 @@ _pixman_implementation_combine_64_ca (pixman_implementation_t * imp,
     (*imp->combine_64_ca[op]) (imp, op, dest, src, mask, width);
 }
 
-void
-_pixman_implementation_composite (pixman_implementation_t * imp,
-                                  pixman_op_t               op,
-                                  pixman_image_t *          src,
-                                  pixman_image_t *          mask,
-                                  pixman_image_t *          dest,
-                                  int32_t                   src_x,
-                                  int32_t                   src_y,
-                                  int32_t                   mask_x,
-                                  int32_t                   mask_y,
-                                  int32_t                   dest_x,
-                                  int32_t                   dest_y,
-                                  int32_t                   width,
-                                  int32_t                   height)
-{
-    (*imp->composite) (imp, op,
-		       src, mask, dest,
-		       src_x, src_y, mask_x, mask_y, dest_x, dest_y,
-		       width, height);
-}
-
 pixman_bool_t
 _pixman_implementation_blt (pixman_implementation_t * imp,
                             uint32_t *                src_bits,
diff --git a/pixman/pixman-mmx.c b/pixman/pixman-mmx.c
index a4affa57..e084e7ff 100644
--- a/pixman/pixman-mmx.c
+++ b/pixman/pixman-mmx.c
@@ -3288,37 +3288,6 @@ static const pixman_fast_path_t mmx_fast_paths[] =
     { PIXMAN_OP_NONE },
 };
 
-static void
-mmx_composite (pixman_implementation_t *imp,
-               pixman_op_t              op,
-               pixman_image_t *         src,
-               pixman_image_t *         mask,
-               pixman_image_t *         dest,
-               int32_t                  src_x,
-               int32_t                  src_y,
-               int32_t                  mask_x,
-               int32_t                  mask_y,
-               int32_t                  dest_x,
-               int32_t                  dest_y,
-               int32_t                  width,
-               int32_t                  height)
-{
-    if (_pixman_run_fast_path (mmx_fast_paths, imp,
-                               op, src, mask, dest,
-                               src_x, src_y,
-                               mask_x, mask_y,
-                               dest_x, dest_y,
-                               width, height))
-    {
-	return;
-    }
-
-    _pixman_implementation_composite (imp->delegate,
-                                      op, src, mask, dest, src_x, src_y,
-                                      mask_x, mask_y, dest_x, dest_y,
-                                      width, height);
-}
-
 static pixman_bool_t
 mmx_blt (pixman_implementation_t *imp,
          uint32_t *               src_bits,
@@ -3372,7 +3341,7 @@ pixman_implementation_t *
 _pixman_implementation_create_mmx (void)
 {
     pixman_implementation_t *general = _pixman_implementation_create_fast_path ();
-    pixman_implementation_t *imp = _pixman_implementation_create (general);
+    pixman_implementation_t *imp = _pixman_implementation_create (general, mmx_fast_paths);
 
     imp->combine_32[PIXMAN_OP_OVER] = mmx_combine_over_u;
     imp->combine_32[PIXMAN_OP_OVER_REVERSE] = mmx_combine_over_reverse_u;
@@ -3398,7 +3367,6 @@ _pixman_implementation_create_mmx (void)
     imp->combine_32_ca[PIXMAN_OP_XOR] = mmx_combine_xor_ca;
     imp->combine_32_ca[PIXMAN_OP_ADD] = mmx_combine_add_ca;
 
-    imp->composite = mmx_composite;
     imp->blt = mmx_blt;
     imp->fill = mmx_fill;
 
diff --git a/pixman/pixman-private.h b/pixman/pixman-private.h
index 4b5af087..048d030e 100644
--- a/pixman/pixman-private.h
+++ b/pixman/pixman-private.h
@@ -372,7 +372,6 @@ pixman_rasterize_edges_accessors (pixman_image_t *image,
 /*
  * Implementations
  */
-
 typedef struct pixman_implementation_t pixman_implementation_t;
 
 typedef void (*pixman_combine_32_func_t) (pixman_implementation_t *imp,
@@ -428,23 +427,36 @@ typedef pixman_bool_t (*pixman_fill_func_t) (pixman_implementation_t *imp,
 void _pixman_setup_combiner_functions_32 (pixman_implementation_t *imp);
 void _pixman_setup_combiner_functions_64 (pixman_implementation_t *imp);
 
-struct pixman_implementation_t
+typedef struct
 {
-    pixman_implementation_t *toplevel;
-    pixman_implementation_t *delegate;
+    pixman_op_t             op;
+    pixman_format_code_t    src_format;
+    uint32_t		    src_flags;
+    pixman_format_code_t    mask_format;
+    uint32_t		    mask_flags;
+    pixman_format_code_t    dest_format;
+    uint32_t		    dest_flags;
+    pixman_composite_func_t func;
+} pixman_fast_path_t;
 
-    pixman_composite_func_t  composite;
-    pixman_blt_func_t        blt;
-    pixman_fill_func_t       fill;
+struct pixman_implementation_t
+{
+    pixman_implementation_t *	toplevel;
+    pixman_implementation_t *	delegate;
+    const pixman_fast_path_t *	fast_paths;
+    
+    pixman_blt_func_t		blt;
+    pixman_fill_func_t		fill;
 
-    pixman_combine_32_func_t combine_32[PIXMAN_N_OPERATORS];
-    pixman_combine_32_func_t combine_32_ca[PIXMAN_N_OPERATORS];
-    pixman_combine_64_func_t combine_64[PIXMAN_N_OPERATORS];
-    pixman_combine_64_func_t combine_64_ca[PIXMAN_N_OPERATORS];
+    pixman_combine_32_func_t	combine_32[PIXMAN_N_OPERATORS];
+    pixman_combine_32_func_t	combine_32_ca[PIXMAN_N_OPERATORS];
+    pixman_combine_64_func_t	combine_64[PIXMAN_N_OPERATORS];
+    pixman_combine_64_func_t	combine_64_ca[PIXMAN_N_OPERATORS];
 };
 
 pixman_implementation_t *
-_pixman_implementation_create (pixman_implementation_t *delegate);
+_pixman_implementation_create (pixman_implementation_t *delegate,
+			       const pixman_fast_path_t *fast_paths);
 
 void
 _pixman_implementation_combine_32 (pixman_implementation_t *imp,
@@ -474,20 +486,6 @@ _pixman_implementation_combine_64_ca (pixman_implementation_t *imp,
                                       const uint64_t *         src,
                                       const uint64_t *         mask,
                                       int                      width);
-void
-_pixman_implementation_composite (pixman_implementation_t *imp,
-                                  pixman_op_t              op,
-                                  pixman_image_t *         src,
-                                  pixman_image_t *         mask,
-                                  pixman_image_t *         dest,
-                                  int32_t                  src_x,
-                                  int32_t                  src_y,
-                                  int32_t                  mask_x,
-                                  int32_t                  mask_y,
-                                  int32_t                  dest_x,
-                                  int32_t                  dest_y,
-                                  int32_t                  width,
-                                  int32_t                  height);
 
 pixman_bool_t
 _pixman_implementation_blt (pixman_implementation_t *imp,
@@ -564,17 +562,22 @@ _pixman_choose_implementation (void);
 #define PIXMAN_pixbuf		PIXMAN_FORMAT (0, 2, 0, 0, 0, 0)
 #define PIXMAN_rpixbuf		PIXMAN_FORMAT (0, 3, 0, 0, 0, 0)
 #define PIXMAN_unknown		PIXMAN_FORMAT (0, 4, 0, 0, 0, 0)
-
-#define FAST_PATH_ID_TRANSFORM			(1 << 0)
-#define FAST_PATH_NO_ALPHA_MAP			(1 << 1)
-#define FAST_PATH_NO_CONVOLUTION_FILTER		(1 << 2)
-#define FAST_PATH_NO_PAD_REPEAT			(1 << 3)
-#define FAST_PATH_NO_REFLECT_REPEAT		(1 << 4)
-#define FAST_PATH_NO_ACCESSORS			(1 << 5)
-#define FAST_PATH_NO_WIDE_FORMAT		(1 << 6)
-#define FAST_PATH_reserved			(1 << 7)
-#define FAST_PATH_COMPONENT_ALPHA		(1 << 8)
-#define FAST_PATH_UNIFIED_ALPHA			(1 << 9)
+#define PIXMAN_any		PIXMAN_FORMAT (0, 5, 0, 0, 0, 0)	
+
+#define PIXMAN_OP_any		(PIXMAN_N_OPERATORS + 1)
+
+#define FAST_PATH_ID_TRANSFORM			(1 <<  0)
+#define FAST_PATH_NO_ALPHA_MAP			(1 <<  1)
+#define FAST_PATH_NO_CONVOLUTION_FILTER		(1 <<  2)
+#define FAST_PATH_NO_PAD_REPEAT			(1 <<  3)
+#define FAST_PATH_NO_REFLECT_REPEAT		(1 <<  4)
+#define FAST_PATH_NO_ACCESSORS			(1 <<  5)
+#define FAST_PATH_NO_WIDE_FORMAT		(1 <<  6)
+#define FAST_PATH_COVERS_CLIP			(1 <<  7)
+#define FAST_PATH_COMPONENT_ALPHA		(1 <<  8)
+#define FAST_PATH_UNIFIED_ALPHA			(1 <<  9)
+#define FAST_PATH_SCALE_TRANSFORM		(1 << 10)
+#define FAST_PATH_NEAREST_FILTER		(1 << 11)
 
 #define _FAST_PATH_STANDARD_FLAGS					\
     (FAST_PATH_ID_TRANSFORM		|				\
@@ -583,7 +586,8 @@ _pixman_choose_implementation (void);
      FAST_PATH_NO_PAD_REPEAT		|				\
      FAST_PATH_NO_REFLECT_REPEAT	|				\
      FAST_PATH_NO_ACCESSORS		|				\
-     FAST_PATH_NO_WIDE_FORMAT)
+     FAST_PATH_NO_WIDE_FORMAT		|				\
+     FAST_PATH_COVERS_CLIP)
 
 #define FAST_PATH_STD_SRC_FLAGS						\
     _FAST_PATH_STANDARD_FLAGS
@@ -597,18 +601,6 @@ _pixman_choose_implementation (void);
     (FAST_PATH_NO_ACCESSORS		|				\
      FAST_PATH_NO_WIDE_FORMAT)
 
-typedef struct
-{
-    pixman_op_t             op;
-    pixman_format_code_t    src_format;
-    uint32_t		    src_flags;
-    pixman_format_code_t    mask_format;
-    uint32_t		    mask_flags;
-    pixman_format_code_t    dest_format;
-    uint32_t		    dest_flags;
-    pixman_composite_func_t func;
-} pixman_fast_path_t;
-
 #define FAST_PATH(op, src, src_flags, mask, mask_flags, dest, dest_flags, func) \
     PIXMAN_OP_ ## op,							\
     PIXMAN_ ## src,							\
@@ -649,38 +641,6 @@ pixman_bool_t
 pixman_addition_overflows_int (unsigned int a, unsigned int b);
 
 /* Compositing utilities */
-pixman_bool_t
-_pixman_run_fast_path (const pixman_fast_path_t *paths,
-                       pixman_implementation_t * imp,
-                       pixman_op_t               op,
-                       pixman_image_t *          src,
-                       pixman_image_t *          mask,
-                       pixman_image_t *          dest,
-                       int32_t                   src_x,
-                       int32_t                   src_y,
-                       int32_t                   mask_x,
-                       int32_t                   mask_y,
-                       int32_t                   dest_x,
-                       int32_t                   dest_y,
-                       int32_t                   width,
-                       int32_t                   height);
-
-void
-_pixman_walk_composite_region (pixman_implementation_t *imp,
-                               pixman_op_t              op,
-                               pixman_image_t *         src_image,
-                               pixman_image_t *         mask_image,
-                               pixman_image_t *         dst_image,
-                               int32_t                  src_x,
-                               int32_t                  src_y,
-                               int32_t                  mask_x,
-                               int32_t                  mask_y,
-                               int32_t                  dest_x,
-                               int32_t                  dest_y,
-                               int32_t                  width,
-                               int32_t                  height,
-                               pixman_composite_func_t  composite_rect);
-
 void
 pixman_expand (uint64_t *           dst,
                const uint32_t *     src,
diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
index f69de0b3..946e7ba3 100644
--- a/pixman/pixman-sse2.c
+++ b/pixman/pixman-sse2.c
@@ -5842,61 +5842,6 @@ static const pixman_fast_path_t sse2_fast_paths[] =
     { PIXMAN_OP_NONE },
 };
 
-/*
- * Work around GCC bug causing crashes in Mozilla with SSE2
- *
- * When using -msse, gcc generates movdqa instructions assuming that
- * the stack is 16 byte aligned. Unfortunately some applications, such
- * as Mozilla and Mono, end up aligning the stack to 4 bytes, which
- * causes the movdqa instructions to fail.
- *
- * The __force_align_arg_pointer__ makes gcc generate a prologue that
- * realigns the stack pointer to 16 bytes.
- *
- * On x86-64 this is not necessary because the standard ABI already
- * calls for a 16 byte aligned stack.
- *
- * See https://bugs.freedesktop.org/show_bug.cgi?id=15693
- */
-#if defined(__GNUC__) && !defined(__x86_64__) && !defined(__amd64__)
-__attribute__((__force_align_arg_pointer__))
-#endif
-static void
-sse2_composite (pixman_implementation_t *imp,
-                pixman_op_t              op,
-                pixman_image_t *         src,
-                pixman_image_t *         mask,
-                pixman_image_t *         dest,
-                int32_t                  src_x,
-                int32_t                  src_y,
-                int32_t                  mask_x,
-                int32_t                  mask_y,
-                int32_t                  dest_x,
-                int32_t                  dest_y,
-                int32_t                  width,
-                int32_t                  height)
-{
-    if (_pixman_run_fast_path (sse2_fast_paths, imp,
-                               op, src, mask, dest,
-                               src_x, src_y,
-                               mask_x, mask_y,
-                               dest_x, dest_y,
-                               width, height))
-    {
-	return;
-    }
-
-    _pixman_implementation_composite (imp->delegate, op,
-                                      src, mask, dest,
-                                      src_x, src_y,
-                                      mask_x, mask_y,
-                                      dest_x, dest_y,
-                                      width, height);
-}
-
-#if defined(__GNUC__) && !defined(__x86_64__) && !defined(__amd64__)
-__attribute__((__force_align_arg_pointer__))
-#endif
 static pixman_bool_t
 sse2_blt (pixman_implementation_t *imp,
           uint32_t *               src_bits,
@@ -5960,7 +5905,7 @@ _pixman_implementation_create_sse2 (void)
 #else
     pixman_implementation_t *fallback = _pixman_implementation_create_fast_path ();
 #endif
-    pixman_implementation_t *imp = _pixman_implementation_create (fallback);
+    pixman_implementation_t *imp = _pixman_implementation_create (fallback, sse2_fast_paths);
 
     /* SSE2 constants */
     mask_565_r  = create_mask_2x32_128 (0x00f80000, 0x00f80000);
@@ -6018,7 +5963,6 @@ _pixman_implementation_create_sse2 (void)
     imp->combine_32_ca[PIXMAN_OP_XOR] = sse2_combine_xor_ca;
     imp->combine_32_ca[PIXMAN_OP_ADD] = sse2_combine_add_ca;
 
-    imp->composite = sse2_composite;
     imp->blt = sse2_blt;
     imp->fill = sse2_fill;
 
diff --git a/pixman/pixman-utils.c b/pixman/pixman-utils.c
index ec54c2d9..3ef88b75 100644
--- a/pixman/pixman-utils.c
+++ b/pixman/pixman-utils.c
@@ -30,208 +30,6 @@
 
 #include "pixman-private.h"
 
-/*
- * Computing composite region
- */
-static inline pixman_bool_t
-clip_general_image (pixman_region32_t * region,
-                    pixman_region32_t * clip,
-                    int                 dx,
-                    int                 dy)
-{
-    if (pixman_region32_n_rects (region) == 1 &&
-        pixman_region32_n_rects (clip) == 1)
-    {
-	pixman_box32_t *  rbox = pixman_region32_rectangles (region, NULL);
-	pixman_box32_t *  cbox = pixman_region32_rectangles (clip, NULL);
-	int v;
-
-	if (rbox->x1 < (v = cbox->x1 + dx))
-	    rbox->x1 = v;
-	if (rbox->x2 > (v = cbox->x2 + dx))
-	    rbox->x2 = v;
-	if (rbox->y1 < (v = cbox->y1 + dy))
-	    rbox->y1 = v;
-	if (rbox->y2 > (v = cbox->y2 + dy))
-	    rbox->y2 = v;
-	if (rbox->x1 >= rbox->x2 || rbox->y1 >= rbox->y2)
-	{
-	    pixman_region32_init (region);
-	    return FALSE;
-	}
-    }
-    else if (!pixman_region32_not_empty (clip))
-    {
-	return FALSE;
-    }
-    else
-    {
-	if (dx || dy)
-	    pixman_region32_translate (region, -dx, -dy);
-
-	if (!pixman_region32_intersect (region, region, clip))
-	    return FALSE;
-
-	if (dx || dy)
-	    pixman_region32_translate (region, dx, dy);
-    }
-
-    return pixman_region32_not_empty (region);
-}
-
-static inline pixman_bool_t
-clip_source_image (pixman_region32_t * region,
-                   pixman_image_t *    image,
-                   int                 dx,
-                   int                 dy)
-{
-    /* Source clips are ignored, unless they are explicitly turned on
-     * and the clip in question was set by an X client. (Because if
-     * the clip was not set by a client, then it is a hierarchy
-     * clip and those should always be ignored for sources).
-     */
-    if (!image->common.clip_sources || !image->common.client_clip)
-	return TRUE;
-
-    return clip_general_image (region,
-                               &image->common.clip_region,
-                               dx, dy);
-}
-
-/*
- * returns FALSE if the final region is empty.  Indistinguishable from
- * an allocation failure, but rendering ignores those anyways.
- */
-static pixman_bool_t
-pixman_compute_composite_region32 (pixman_region32_t * region,
-                                   pixman_image_t *    src_image,
-                                   pixman_image_t *    mask_image,
-                                   pixman_image_t *    dst_image,
-                                   int32_t             src_x,
-                                   int32_t             src_y,
-                                   int32_t             mask_x,
-                                   int32_t             mask_y,
-                                   int32_t             dest_x,
-                                   int32_t             dest_y,
-                                   int32_t             width,
-                                   int32_t             height)
-{
-    region->extents.x1 = dest_x;
-    region->extents.x2 = dest_x + width;
-    region->extents.y1 = dest_y;
-    region->extents.y2 = dest_y + height;
-
-    region->extents.x1 = MAX (region->extents.x1, 0);
-    region->extents.y1 = MAX (region->extents.y1, 0);
-    region->extents.x2 = MIN (region->extents.x2, dst_image->bits.width);
-    region->extents.y2 = MIN (region->extents.y2, dst_image->bits.height);
-
-    region->data = 0;
-
-    /* Check for empty operation */
-    if (region->extents.x1 >= region->extents.x2 ||
-        region->extents.y1 >= region->extents.y2)
-    {
-	pixman_region32_init (region);
-	return FALSE;
-    }
-
-    if (dst_image->common.have_clip_region)
-    {
-	if (!clip_general_image (region, &dst_image->common.clip_region, 0, 0))
-	{
-	    pixman_region32_fini (region);
-	    return FALSE;
-	}
-    }
-
-    if (dst_image->common.alpha_map && dst_image->common.alpha_map->common.have_clip_region)
-    {
-	if (!clip_general_image (region, &dst_image->common.alpha_map->common.clip_region,
-	                         -dst_image->common.alpha_origin_x,
-	                         -dst_image->common.alpha_origin_y))
-	{
-	    pixman_region32_fini (region);
-	    return FALSE;
-	}
-    }
-
-    /* clip against src */
-    if (src_image->common.have_clip_region)
-    {
-	if (!clip_source_image (region, src_image, dest_x - src_x, dest_y - src_y))
-	{
-	    pixman_region32_fini (region);
-	    return FALSE;
-	}
-    }
-    if (src_image->common.alpha_map && src_image->common.alpha_map->common.have_clip_region)
-    {
-	if (!clip_source_image (region, (pixman_image_t *)src_image->common.alpha_map,
-	                        dest_x - (src_x - src_image->common.alpha_origin_x),
-	                        dest_y - (src_y - src_image->common.alpha_origin_y)))
-	{
-	    pixman_region32_fini (region);
-	    return FALSE;
-	}
-    }
-    /* clip against mask */
-    if (mask_image && mask_image->common.have_clip_region)
-    {
-	if (!clip_source_image (region, mask_image, dest_x - mask_x, dest_y - mask_y))
-	{
-	    pixman_region32_fini (region);
-	    return FALSE;
-	}
-	if (mask_image->common.alpha_map && mask_image->common.alpha_map->common.have_clip_region)
-	{
-	    if (!clip_source_image (region, (pixman_image_t *)mask_image->common.alpha_map,
-	                            dest_x - (mask_x - mask_image->common.alpha_origin_x),
-	                            dest_y - (mask_y - mask_image->common.alpha_origin_y)))
-	    {
-		pixman_region32_fini (region);
-		return FALSE;
-	    }
-	}
-    }
-
-    return TRUE;
-}
-
-PIXMAN_EXPORT pixman_bool_t
-pixman_compute_composite_region (pixman_region16_t * region,
-                                 pixman_image_t *    src_image,
-                                 pixman_image_t *    mask_image,
-                                 pixman_image_t *    dst_image,
-                                 int16_t             src_x,
-                                 int16_t             src_y,
-                                 int16_t             mask_x,
-                                 int16_t             mask_y,
-                                 int16_t             dest_x,
-                                 int16_t             dest_y,
-                                 uint16_t            width,
-                                 uint16_t            height)
-{
-    pixman_region32_t r32;
-    pixman_bool_t retval;
-
-    pixman_region32_init (&r32);
-
-    retval = pixman_compute_composite_region32 (
-	&r32, src_image, mask_image, dst_image,
-	src_x, src_y, mask_x, mask_y, dest_x, dest_y,
-	width, height);
-
-    if (retval)
-    {
-	if (!pixman_region16_copy_from_region32 (region, &r32))
-	    retval = FALSE;
-    }
-
-    pixman_region32_fini (&r32);
-    return retval;
-}
-
 pixman_bool_t
 pixman_multiply_overflows_int (unsigned int a,
                                unsigned int b)
@@ -369,335 +167,6 @@ pixman_contract (uint32_t *      dst,
     }
 }
 
-static void
-walk_region_internal (pixman_implementation_t *imp,
-                      pixman_op_t              op,
-                      pixman_image_t *         src_image,
-                      pixman_image_t *         mask_image,
-                      pixman_image_t *         dst_image,
-                      int32_t                  src_x,
-                      int32_t                  src_y,
-                      int32_t                  mask_x,
-                      int32_t                  mask_y,
-                      int32_t                  dest_x,
-                      int32_t                  dest_y,
-                      int32_t                  width,
-                      int32_t                  height,
-                      pixman_bool_t            src_repeat,
-                      pixman_bool_t            mask_repeat,
-                      pixman_region32_t *      region,
-                      pixman_composite_func_t  composite_rect)
-{
-    int n;
-    const pixman_box32_t *pbox;
-    int w, h, w_this, h_this;
-    int x_msk, y_msk, x_src, y_src, x_dst, y_dst;
-
-    pbox = pixman_region32_rectangles (region, &n);
-    while (n--)
-    {
-	h = pbox->y2 - pbox->y1;
-	y_src = pbox->y1 - dest_y + src_y;
-	y_msk = pbox->y1 - dest_y + mask_y;
-	y_dst = pbox->y1;
-
-	while (h)
-	{
-	    h_this = h;
-	    w = pbox->x2 - pbox->x1;
-	    x_src = pbox->x1 - dest_x + src_x;
-	    x_msk = pbox->x1 - dest_x + mask_x;
-	    x_dst = pbox->x1;
-
-	    if (mask_repeat)
-	    {
-		y_msk = MOD (y_msk, mask_image->bits.height);
-		if (h_this > mask_image->bits.height - y_msk)
-		    h_this = mask_image->bits.height - y_msk;
-	    }
-
-	    if (src_repeat)
-	    {
-		y_src = MOD (y_src, src_image->bits.height);
-		if (h_this > src_image->bits.height - y_src)
-		    h_this = src_image->bits.height - y_src;
-	    }
-
-	    while (w)
-	    {
-		w_this = w;
-
-		if (mask_repeat)
-		{
-		    x_msk = MOD (x_msk, mask_image->bits.width);
-		    if (w_this > mask_image->bits.width - x_msk)
-			w_this = mask_image->bits.width - x_msk;
-		}
-
-		if (src_repeat)
-		{
-		    x_src = MOD (x_src, src_image->bits.width);
-		    if (w_this > src_image->bits.width - x_src)
-			w_this = src_image->bits.width - x_src;
-		}
-
-		(*composite_rect) (imp, op,
-				   src_image, mask_image, dst_image,
-				   x_src, y_src, x_msk, y_msk, x_dst, y_dst,
-				   w_this, h_this);
-		w -= w_this;
-
-		x_src += w_this;
-		x_msk += w_this;
-		x_dst += w_this;
-	    }
-
-	    h -= h_this;
-	    y_src += h_this;
-	    y_msk += h_this;
-	    y_dst += h_this;
-	}
-
-	pbox++;
-    }
-}
-
-void
-_pixman_walk_composite_region (pixman_implementation_t *imp,
-                               pixman_op_t              op,
-                               pixman_image_t *         src_image,
-                               pixman_image_t *         mask_image,
-                               pixman_image_t *         dst_image,
-                               int32_t                  src_x,
-                               int32_t                  src_y,
-                               int32_t                  mask_x,
-                               int32_t                  mask_y,
-                               int32_t                  dest_x,
-                               int32_t                  dest_y,
-                               int32_t                  width,
-                               int32_t                  height,
-                               pixman_composite_func_t  composite_rect)
-{
-    pixman_region32_t region;
-
-    pixman_region32_init (&region);
-
-    if (pixman_compute_composite_region32 (
-            &region, src_image, mask_image, dst_image,
-            src_x, src_y, mask_x, mask_y, dest_x, dest_y,
-            width, height))
-    {
-	walk_region_internal (imp, op,
-	                      src_image, mask_image, dst_image,
-	                      src_x, src_y, mask_x, mask_y, dest_x, dest_y,
-	                      width, height, FALSE, FALSE,
-	                      &region,
-	                      composite_rect);
-
-	pixman_region32_fini (&region);
-    }
-}
-
-static void
-get_image_info (pixman_image_t       *image,
-		pixman_format_code_t *code,
-		uint32_t	     *flags)
-{
-    *flags = 0;
-    
-    if (!image)
-    {
-	*code = PIXMAN_null;
-    }
-    else
-    {
-	if (!image->common.transform)
-	    *flags |= FAST_PATH_ID_TRANSFORM;
-
-	if (!image->common.alpha_map)
-	    *flags |= FAST_PATH_NO_ALPHA_MAP;
-
-	if (image->common.filter != PIXMAN_FILTER_CONVOLUTION)
-	    *flags |= FAST_PATH_NO_CONVOLUTION_FILTER;
-
-	if (image->common.repeat != PIXMAN_REPEAT_PAD)
-	    *flags |= FAST_PATH_NO_PAD_REPEAT;
-
-	if (image->common.repeat != PIXMAN_REPEAT_REFLECT)
-	    *flags |= FAST_PATH_NO_REFLECT_REPEAT;
-
-	*flags |= (FAST_PATH_NO_ACCESSORS | FAST_PATH_NO_WIDE_FORMAT);
-	if (image->type == BITS)
-	{
-	    if (image->bits.read_func || image->bits.write_func)
-		*flags &= ~FAST_PATH_NO_ACCESSORS;
-
-	    if (PIXMAN_FORMAT_IS_WIDE (image->bits.format))
-		*flags &= ~FAST_PATH_NO_WIDE_FORMAT;
-	}
-
-	if (image->common.component_alpha)
-	    *flags |= FAST_PATH_COMPONENT_ALPHA;
-	else
-	    *flags |= FAST_PATH_UNIFIED_ALPHA;
-
-	if (_pixman_image_is_solid (image))
-	    *code = PIXMAN_solid;
-	else if (image->common.type == BITS)
-	    *code = image->bits.format;
-	else
-	    *code = PIXMAN_unknown;
-    }
-}
-
-static force_inline pixman_bool_t
-image_covers (pixman_image_t *image,
-              pixman_box32_t *extents,
-              int             x,
-              int             y)
-{
-    if (image->common.type == BITS &&
-	image->common.repeat == PIXMAN_REPEAT_NONE)
-    {
-	if (x > extents->x1 || y > extents->y1 ||
-	    x + image->bits.width < extents->x2 ||
-	    y + image->bits.height < extents->y2)
-	{
-	    return FALSE;
-	}
-    }
-
-    return TRUE;
-}
-
-static force_inline pixman_bool_t
-sources_cover (pixman_image_t *src,
-	       pixman_image_t *mask,
-	       pixman_box32_t *extents,
-	       int             src_x,
-	       int             src_y,
-	       int             mask_x,
-	       int             mask_y,
-	       int             dest_x,
-	       int             dest_y)
-{
-    if (!image_covers (src, extents, dest_x - src_x, dest_y - src_y))
-	return FALSE;
-
-    if (!mask)
-	return TRUE;
-
-    if (!image_covers (mask, extents, dest_x - mask_x, dest_y - mask_y))
-	return FALSE;
-
-    return TRUE;
-}
-
-pixman_bool_t
-_pixman_run_fast_path (const pixman_fast_path_t *paths,
-                       pixman_implementation_t * imp,
-                       pixman_op_t               op,
-                       pixman_image_t *          src,
-                       pixman_image_t *          mask,
-                       pixman_image_t *          dest,
-                       int32_t                   src_x,
-                       int32_t                   src_y,
-                       int32_t                   mask_x,
-                       int32_t                   mask_y,
-                       int32_t                   dest_x,
-                       int32_t                   dest_y,
-                       int32_t                   width,
-                       int32_t                   height)
-{
-    pixman_format_code_t src_format, mask_format, dest_format;
-    uint32_t src_flags, mask_flags, dest_flags;
-    pixman_composite_func_t func;
-    const pixman_fast_path_t *info;
-    pixman_bool_t result;
-
-    get_image_info (src,  &src_format,  &src_flags);
-    get_image_info (mask, &mask_format, &mask_flags);
-    get_image_info (dest, &dest_format, &dest_flags);
-    
-    /* Check for pixbufs */
-    if ((mask_format == PIXMAN_a8r8g8b8 || mask_format == PIXMAN_a8b8g8r8) &&
-	(src->type == BITS && src->bits.bits == mask->bits.bits)	   &&
-	(src->common.repeat == mask->common.repeat)			   &&
-	(src_x == mask_x && src_y == mask_y))
-    {
-	if (src_format == PIXMAN_x8b8g8r8)
-	    src_format = mask_format = PIXMAN_pixbuf;
-	else if (src_format == PIXMAN_x8r8g8b8)
-	    src_format = mask_format = PIXMAN_rpixbuf;
-    }
-
-    func = NULL;
-    for (info = paths; info->op != PIXMAN_OP_NONE; ++info)
-    {
-	if (info->op == op					&&
-	    (info->src_format == src_format)			&&
-	    (info->src_flags & src_flags) == info->src_flags	&&
-	    (info->mask_format == mask_format)			&&
-	    (info->mask_flags & mask_flags) == info->mask_flags	&&
-	    (info->dest_format == dest_format)			&&
-	    (info->dest_flags & dest_flags) == info->dest_flags)
-	{
-	    func = info->func;
-	    break;
-	}
-    }
-
-    result = FALSE;
-    
-    if (func)
-    {
-	pixman_region32_t region;
-	pixman_region32_init (&region);
-
-	if (pixman_compute_composite_region32 (
-	        &region, src, mask, dest,
-	        src_x, src_y, mask_x, mask_y, dest_x, dest_y, width, height))
-	{
-	    pixman_box32_t *extents = pixman_region32_extents (&region);
-
-	    if (sources_cover (
-		    src, mask, extents,
-		    src_x, src_y, mask_x, mask_y, dest_x, dest_y))
-	    {
-		pixman_bool_t src_repeat, mask_repeat;
-
-		src_repeat =
-		    src->type == BITS					&&
-		    src_flags & FAST_PATH_ID_TRANSFORM			&&
-		    src->common.repeat == PIXMAN_REPEAT_NORMAL		&&
-		    src_format != PIXMAN_solid;
-		
-		mask_repeat =
-		    mask						&&
-		    mask->type == BITS					&&
-		    mask_flags & FAST_PATH_ID_TRANSFORM			&&
-		    mask->common.repeat == PIXMAN_REPEAT_NORMAL		&&
-		    mask_format != PIXMAN_solid;
-		
-		walk_region_internal (imp, op,
-		                      src, mask, dest,
-		                      src_x, src_y, mask_x, mask_y,
-		                      dest_x, dest_y,
-		                      width, height,
-		                      src_repeat, mask_repeat,
-		                      &region,
-		                      func);
-
-		result = TRUE;
-	    }
-
-	    pixman_region32_fini (&region);
-	}
-    }
-
-    return result;
-}
-
 #define N_TMP_BOXES (16)
 
 pixman_bool_t
diff --git a/pixman/pixman-vmx.c b/pixman/pixman-vmx.c
index 06325a7c..e811cf73 100644
--- a/pixman/pixman-vmx.c
+++ b/pixman/pixman-vmx.c
@@ -1607,11 +1607,16 @@ vmx_combine_add_ca (pixman_implementation_t *imp,
     }
 }
 
+static const pixman_fast_path_t vmx_fast_paths[] =
+{
+    {   PIXMAN_OP_NONE	},
+};
+
 pixman_implementation_t *
 _pixman_implementation_create_vmx (void)
 {
     pixman_implementation_t *fast = _pixman_implementation_create_fast_path ();
-    pixman_implementation_t *imp = _pixman_implementation_create (fast);
+    pixman_implementation_t *imp = _pixman_implementation_create (fast, vmx_fast_paths);
 
     /* Set up function pointers */
 
diff --git a/pixman/pixman.c b/pixman/pixman.c
index 9ab875d7..6a260ed4 100644
--- a/pixman/pixman.c
+++ b/pixman/pixman.c
@@ -117,43 +117,554 @@ apply_workaround (pixman_image_t *image,
 		  int *           save_dx,
 		  int *           save_dy)
 {
-    /* Some X servers generate images that point to the
-     * wrong place in memory, but then set the clip region
-     * to point to the right place. Because of an old bug
-     * in pixman, this would actually work.
-     *
-     * Here we try and undo the damage
+    if (image && image->common.need_workaround)
+    {
+	/* Some X servers generate images that point to the
+	 * wrong place in memory, but then set the clip region
+	 * to point to the right place. Because of an old bug
+	 * in pixman, this would actually work.
+	 *
+	 * Here we try and undo the damage
+	 */
+	int bpp = PIXMAN_FORMAT_BPP (image->bits.format) / 8;
+	pixman_box32_t *extents;
+	uint8_t *t;
+	int dx, dy;
+	
+	extents = pixman_region32_extents (&(image->common.clip_region));
+	dx = extents->x1;
+	dy = extents->y1;
+	
+	*save_bits = image->bits.bits;
+	
+	*x -= dx;
+	*y -= dy;
+	pixman_region32_translate (&(image->common.clip_region), -dx, -dy);
+	
+	t = (uint8_t *)image->bits.bits;
+	t += dy * image->bits.rowstride * 4 + dx * bpp;
+	image->bits.bits = (uint32_t *)t;
+	
+	*save_dx = dx;
+	*save_dy = dy;
+    }
+}
+
+static void
+unapply_workaround (pixman_image_t *image, uint32_t *bits, int dx, int dy)
+{
+    if (image && image->common.need_workaround)
+    {
+	image->bits.bits = bits;
+	pixman_region32_translate (&image->common.clip_region, dx, dy);
+    }
+}
+
+/*
+ * Computing composite region
+ */
+static inline pixman_bool_t
+clip_general_image (pixman_region32_t * region,
+                    pixman_region32_t * clip,
+                    int                 dx,
+                    int                 dy)
+{
+    if (pixman_region32_n_rects (region) == 1 &&
+        pixman_region32_n_rects (clip) == 1)
+    {
+	pixman_box32_t *  rbox = pixman_region32_rectangles (region, NULL);
+	pixman_box32_t *  cbox = pixman_region32_rectangles (clip, NULL);
+	int v;
+
+	if (rbox->x1 < (v = cbox->x1 + dx))
+	    rbox->x1 = v;
+	if (rbox->x2 > (v = cbox->x2 + dx))
+	    rbox->x2 = v;
+	if (rbox->y1 < (v = cbox->y1 + dy))
+	    rbox->y1 = v;
+	if (rbox->y2 > (v = cbox->y2 + dy))
+	    rbox->y2 = v;
+	if (rbox->x1 >= rbox->x2 || rbox->y1 >= rbox->y2)
+	{
+	    pixman_region32_init (region);
+	    return FALSE;
+	}
+    }
+    else if (!pixman_region32_not_empty (clip))
+    {
+	return FALSE;
+    }
+    else
+    {
+	if (dx || dy)
+	    pixman_region32_translate (region, -dx, -dy);
+
+	if (!pixman_region32_intersect (region, region, clip))
+	    return FALSE;
+
+	if (dx || dy)
+	    pixman_region32_translate (region, dx, dy);
+    }
+
+    return pixman_region32_not_empty (region);
+}
+
+static inline pixman_bool_t
+clip_source_image (pixman_region32_t * region,
+                   pixman_image_t *    image,
+                   int                 dx,
+                   int                 dy)
+{
+    /* Source clips are ignored, unless they are explicitly turned on
+     * and the clip in question was set by an X client. (Because if
+     * the clip was not set by a client, then it is a hierarchy
+     * clip and those should always be ignored for sources).
      */
-    int bpp = PIXMAN_FORMAT_BPP (image->bits.format) / 8;
-    pixman_box32_t *extents;
-    uint8_t *t;
-    int dx, dy;
+    if (!image->common.clip_sources || !image->common.client_clip)
+	return TRUE;
+
+    return clip_general_image (region,
+                               &image->common.clip_region,
+                               dx, dy);
+}
+
+/*
+ * returns FALSE if the final region is empty.  Indistinguishable from
+ * an allocation failure, but rendering ignores those anyways.
+ */
+static pixman_bool_t
+pixman_compute_composite_region32 (pixman_region32_t * region,
+                                   pixman_image_t *    src_image,
+                                   pixman_image_t *    mask_image,
+                                   pixman_image_t *    dst_image,
+                                   int32_t             src_x,
+                                   int32_t             src_y,
+                                   int32_t             mask_x,
+                                   int32_t             mask_y,
+                                   int32_t             dest_x,
+                                   int32_t             dest_y,
+                                   int32_t             width,
+                                   int32_t             height)
+{
+    region->extents.x1 = dest_x;
+    region->extents.x2 = dest_x + width;
+    region->extents.y1 = dest_y;
+    region->extents.y2 = dest_y + height;
 
-    extents = pixman_region32_extents (&(image->common.clip_region));
-    dx = extents->x1;
-    dy = extents->y1;
+    region->extents.x1 = MAX (region->extents.x1, 0);
+    region->extents.y1 = MAX (region->extents.y1, 0);
+    region->extents.x2 = MIN (region->extents.x2, dst_image->bits.width);
+    region->extents.y2 = MIN (region->extents.y2, dst_image->bits.height);
 
-    *save_bits = image->bits.bits;
+    region->data = 0;
 
-    *x -= dx;
-    *y -= dy;
-    pixman_region32_translate (&(image->common.clip_region), -dx, -dy);
+    /* Check for empty operation */
+    if (region->extents.x1 >= region->extents.x2 ||
+        region->extents.y1 >= region->extents.y2)
+    {
+	pixman_region32_init (region);
+	return FALSE;
+    }
+
+    if (dst_image->common.have_clip_region)
+    {
+	if (!clip_general_image (region, &dst_image->common.clip_region, 0, 0))
+	{
+	    pixman_region32_fini (region);
+	    return FALSE;
+	}
+    }
+
+    if (dst_image->common.alpha_map && dst_image->common.alpha_map->common.have_clip_region)
+    {
+	if (!clip_general_image (region, &dst_image->common.alpha_map->common.clip_region,
+	                         -dst_image->common.alpha_origin_x,
+	                         -dst_image->common.alpha_origin_y))
+	{
+	    pixman_region32_fini (region);
+	    return FALSE;
+	}
+    }
+
+    /* clip against src */
+    if (src_image->common.have_clip_region)
+    {
+	if (!clip_source_image (region, src_image, dest_x - src_x, dest_y - src_y))
+	{
+	    pixman_region32_fini (region);
+	    return FALSE;
+	}
+    }
+    if (src_image->common.alpha_map && src_image->common.alpha_map->common.have_clip_region)
+    {
+	if (!clip_source_image (region, (pixman_image_t *)src_image->common.alpha_map,
+	                        dest_x - (src_x - src_image->common.alpha_origin_x),
+	                        dest_y - (src_y - src_image->common.alpha_origin_y)))
+	{
+	    pixman_region32_fini (region);
+	    return FALSE;
+	}
+    }
+    /* clip against mask */
+    if (mask_image && mask_image->common.have_clip_region)
+    {
+	if (!clip_source_image (region, mask_image, dest_x - mask_x, dest_y - mask_y))
+	{
+	    pixman_region32_fini (region);
+	    return FALSE;
+	}
+	if (mask_image->common.alpha_map && mask_image->common.alpha_map->common.have_clip_region)
+	{
+	    if (!clip_source_image (region, (pixman_image_t *)mask_image->common.alpha_map,
+	                            dest_x - (mask_x - mask_image->common.alpha_origin_x),
+	                            dest_y - (mask_y - mask_image->common.alpha_origin_y)))
+	    {
+		pixman_region32_fini (region);
+		return FALSE;
+	    }
+	}
+    }
 
-    t = (uint8_t *)image->bits.bits;
-    t += dy * image->bits.rowstride * 4 + dx * bpp;
-    image->bits.bits = (uint32_t *)t;
+    return TRUE;
+}
 
-    *save_dx = dx;
-    *save_dy = dy;
+static void
+walk_region_internal (pixman_implementation_t *imp,
+                      pixman_op_t              op,
+                      pixman_image_t *         src_image,
+                      pixman_image_t *         mask_image,
+                      pixman_image_t *         dst_image,
+                      int32_t                  src_x,
+                      int32_t                  src_y,
+                      int32_t                  mask_x,
+                      int32_t                  mask_y,
+                      int32_t                  dest_x,
+                      int32_t                  dest_y,
+                      int32_t                  width,
+                      int32_t                  height,
+                      pixman_bool_t            src_repeat,
+                      pixman_bool_t            mask_repeat,
+                      pixman_region32_t *      region,
+                      pixman_composite_func_t  composite_rect)
+{
+    int w, h, w_this, h_this;
+    int x_msk, y_msk, x_src, y_src, x_dst, y_dst;
+    int src_dy = src_y - dest_y;
+    int src_dx = src_x - dest_x;
+    int mask_dy = mask_y - dest_y;
+    int mask_dx = mask_x - dest_x;
+    const pixman_box32_t *pbox;
+    int n;
+
+    pbox = pixman_region32_rectangles (region, &n);
+
+    /* Fast path for non-repeating sources */
+    if (!src_repeat && !mask_repeat)
+    {
+       while (n--)
+       {
+           (*composite_rect) (imp, op,
+                              src_image, mask_image, dst_image,
+                              pbox->x1 + src_dx,
+                              pbox->y1 + src_dy,
+                              pbox->x1 + mask_dx,
+                              pbox->y1 + mask_dy,
+                              pbox->x1,
+                              pbox->y1,
+                              pbox->x2 - pbox->x1,
+                              pbox->y2 - pbox->y1);
+           
+           pbox++;
+       }
+
+       return;
+    }
+    
+    while (n--)
+    {
+	h = pbox->y2 - pbox->y1;
+	y_src = pbox->y1 + src_dy;
+	y_msk = pbox->y1 + mask_dy;
+	y_dst = pbox->y1;
+
+	while (h)
+	{
+	    h_this = h;
+	    w = pbox->x2 - pbox->x1;
+	    x_src = pbox->x1 + src_dx;
+	    x_msk = pbox->x1 + mask_dx;
+	    x_dst = pbox->x1;
+
+	    if (mask_repeat)
+	    {
+		y_msk = MOD (y_msk, mask_image->bits.height);
+		if (h_this > mask_image->bits.height - y_msk)
+		    h_this = mask_image->bits.height - y_msk;
+	    }
+
+	    if (src_repeat)
+	    {
+		y_src = MOD (y_src, src_image->bits.height);
+		if (h_this > src_image->bits.height - y_src)
+		    h_this = src_image->bits.height - y_src;
+	    }
+
+	    while (w)
+	    {
+		w_this = w;
+
+		if (mask_repeat)
+		{
+		    x_msk = MOD (x_msk, mask_image->bits.width);
+		    if (w_this > mask_image->bits.width - x_msk)
+			w_this = mask_image->bits.width - x_msk;
+		}
+
+		if (src_repeat)
+		{
+		    x_src = MOD (x_src, src_image->bits.width);
+		    if (w_this > src_image->bits.width - x_src)
+			w_this = src_image->bits.width - x_src;
+		}
+
+		(*composite_rect) (imp, op,
+				   src_image, mask_image, dst_image,
+				   x_src, y_src, x_msk, y_msk, x_dst, y_dst,
+				   w_this, h_this);
+		w -= w_this;
+
+		x_src += w_this;
+		x_msk += w_this;
+		x_dst += w_this;
+	    }
+
+	    h -= h_this;
+	    y_src += h_this;
+	    y_msk += h_this;
+	    y_dst += h_this;
+	}
+
+	pbox++;
+    }
 }
 
 static void
-unapply_workaround (pixman_image_t *image, uint32_t *bits, int dx, int dy)
+get_image_info (pixman_image_t       *image,
+		pixman_format_code_t *code,
+		uint32_t	     *flags)
+{
+    *flags = 0;
+    
+    if (!image->common.transform)
+    {
+	*flags |= FAST_PATH_ID_TRANSFORM;
+    }
+    else
+    {
+	if (image->common.transform->matrix[0][1] == 0 &&
+	    image->common.transform->matrix[1][0] == 0 &&
+	    image->common.transform->matrix[2][0] == 0 &&
+	    image->common.transform->matrix[2][1] == 0 &&
+	    image->common.transform->matrix[2][2] == pixman_fixed_1)
+	{
+	    *flags |= FAST_PATH_SCALE_TRANSFORM;
+	}
+    }
+    
+    if (!image->common.alpha_map)
+	*flags |= FAST_PATH_NO_ALPHA_MAP;
+    
+    if (image->common.filter != PIXMAN_FILTER_CONVOLUTION)
+    {
+	*flags |= FAST_PATH_NO_CONVOLUTION_FILTER;
+	
+	if (image->common.filter == PIXMAN_FILTER_NEAREST)
+	    *flags |= FAST_PATH_NEAREST_FILTER;
+    }
+    
+    if (image->common.repeat != PIXMAN_REPEAT_PAD)
+	*flags |= FAST_PATH_NO_PAD_REPEAT;
+    
+    if (image->common.repeat != PIXMAN_REPEAT_REFLECT)
+	*flags |= FAST_PATH_NO_REFLECT_REPEAT;
+    
+    *flags |= (FAST_PATH_NO_ACCESSORS | FAST_PATH_NO_WIDE_FORMAT);
+    if (image->type == BITS)
+    {
+	if (image->bits.read_func || image->bits.write_func)
+	    *flags &= ~FAST_PATH_NO_ACCESSORS;
+	
+	if (PIXMAN_FORMAT_IS_WIDE (image->bits.format))
+	    *flags &= ~FAST_PATH_NO_WIDE_FORMAT;
+    }
+    
+    if (image->common.component_alpha)
+	*flags |= FAST_PATH_COMPONENT_ALPHA;
+    else
+	*flags |= FAST_PATH_UNIFIED_ALPHA;
+    
+    if (_pixman_image_is_solid (image))
+	*code = PIXMAN_solid;
+    else if (image->common.type == BITS)
+	*code = image->bits.format;
+    else
+	*code = PIXMAN_unknown;
+}
+
+static force_inline pixman_bool_t
+image_covers (pixman_image_t *image,
+              pixman_box32_t *extents,
+              int             x,
+              int             y)
 {
-    image->bits.bits = bits;
-    pixman_region32_translate (&image->common.clip_region, dx, dy);
+    if (image->common.type == BITS &&
+	image->common.repeat == PIXMAN_REPEAT_NONE)
+    {
+	if (x > extents->x1 || y > extents->y1 ||
+	    x + image->bits.width < extents->x2 ||
+	    y + image->bits.height < extents->y2)
+	{
+	    return FALSE;
+	}
+    }
+
+    return TRUE;
+}
+
+static void
+do_composite (pixman_implementation_t *imp,
+	      pixman_op_t	       op,
+	      pixman_image_t	      *src,
+	      pixman_image_t	      *mask,
+	      pixman_image_t	      *dest,
+	      int		       src_x,
+	      int		       src_y,
+	      int		       mask_x,
+	      int		       mask_y,
+	      int		       dest_x,
+	      int		       dest_y,
+	      int		       width,
+	      int		       height)
+{
+    pixman_format_code_t src_format, mask_format, dest_format;
+    uint32_t src_flags, mask_flags, dest_flags;
+    pixman_bool_t src_repeat, mask_repeat;
+    pixman_region32_t region;
+    pixman_box32_t *extents;
+
+    get_image_info (src,  &src_format,  &src_flags);
+    if (mask)
+    {
+	get_image_info (mask, &mask_format, &mask_flags);
+    }
+    else
+    {
+	mask_format = PIXMAN_null;
+	mask_flags = 0;
+    }
+    get_image_info (dest, &dest_format, &dest_flags);
+    
+    /* Check for pixbufs */
+    if ((mask_format == PIXMAN_a8r8g8b8 || mask_format == PIXMAN_a8b8g8r8) &&
+	(src->type == BITS && src->bits.bits == mask->bits.bits)	   &&
+	(src->common.repeat == mask->common.repeat)			   &&
+	(src_x == mask_x && src_y == mask_y))
+    {
+	if (src_format == PIXMAN_x8b8g8r8)
+	    src_format = mask_format = PIXMAN_pixbuf;
+	else if (src_format == PIXMAN_x8r8g8b8)
+	    src_format = mask_format = PIXMAN_rpixbuf;
+    }
+	    
+    src_repeat =
+	src->type == BITS					&&
+	src_flags & FAST_PATH_ID_TRANSFORM			&&
+	src->common.repeat == PIXMAN_REPEAT_NORMAL		&&
+	src_format != PIXMAN_solid;
+    
+    mask_repeat =
+	mask							&&
+	mask->type == BITS					&&
+	mask_flags & FAST_PATH_ID_TRANSFORM			&&
+	mask->common.repeat == PIXMAN_REPEAT_NORMAL		&&
+	mask_format != PIXMAN_solid;
+    
+    pixman_region32_init (&region);
+    
+    if (!pixman_compute_composite_region32 (
+	    &region, src, mask, dest,
+	    src_x, src_y, mask_x, mask_y, dest_x, dest_y, width, height))
+    {
+	return;
+    }
+    
+    extents = pixman_region32_extents (&region);
+    
+    if (image_covers (src, extents, dest_x - src_x, dest_y - src_y))
+	src_flags |= FAST_PATH_COVERS_CLIP;
+    
+    if (mask && image_covers (mask, extents, dest_x - mask_x, dest_y - mask_y))
+	mask_flags |= FAST_PATH_COVERS_CLIP;
+	    
+    while (imp)
+    {
+	const pixman_fast_path_t *info;
+	    
+	for (info = imp->fast_paths; info->op != PIXMAN_OP_NONE; ++info)
+	{
+	    if ((info->op == op || info->op == PIXMAN_OP_any)		&&
+		/* src */
+		((info->src_format == src_format) ||
+		 (info->src_format == PIXMAN_any))			&&
+		(info->src_flags & src_flags) == info->src_flags	&&
+		/* mask */
+		((info->mask_format == mask_format) ||
+		 (info->mask_format == PIXMAN_any))			&&
+		(info->mask_flags & mask_flags) == info->mask_flags	&&
+		/* dest */
+		((info->dest_format == dest_format) ||
+		 (info->dest_format == PIXMAN_any))			&&
+		(info->dest_flags & dest_flags) == info->dest_flags)
+	    {
+		walk_region_internal (imp, op,
+				      src, mask, dest,
+				      src_x, src_y, mask_x, mask_y,
+				      dest_x, dest_y,
+				      width, height,
+				      src_repeat, mask_repeat,
+				      &region,
+				      info->func);
+		
+		goto done;
+	    }
+	}
+	
+	imp = imp->delegate;
+    }
+
+done:
+    pixman_region32_fini (&region);
 }
 
+/*
+ * Work around GCC bug causing crashes in Mozilla with SSE2
+ *
+ * When using -msse, gcc generates movdqa instructions assuming that
+ * the stack is 16 byte aligned. Unfortunately some applications, such
+ * as Mozilla and Mono, end up aligning the stack to 4 bytes, which
+ * causes the movdqa instructions to fail.
+ *
+ * The __force_align_arg_pointer__ makes gcc generate a prologue that
+ * realigns the stack pointer to 16 bytes.
+ *
+ * On x86-64 this is not necessary because the standard ABI already
+ * calls for a 16 byte aligned stack.
+ *
+ * See https://bugs.freedesktop.org/show_bug.cgi?id=15693
+ */
+#if defined (USE_SSE2) && defined(__GNUC__) && !defined(__x86_64__) && !defined(__amd64__)
+__attribute__((__force_align_arg_pointer__))
+#endif
 PIXMAN_EXPORT void
 pixman_image_composite (pixman_op_t      op,
                         pixman_image_t * src,
@@ -192,6 +703,7 @@ pixman_image_composite32 (pixman_op_t      op,
     int mask_dx, mask_dy;
     uint32_t *dest_bits;
     int dest_dx, dest_dy;
+    pixman_bool_t need_workaround;
 
     _pixman_image_validate (src);
     if (mask)
@@ -214,26 +726,34 @@ pixman_image_composite32 (pixman_op_t      op,
     if (!imp)
 	imp = _pixman_choose_implementation ();
 
-    if (src->common.need_workaround)
+    need_workaround =
+	(src->common.need_workaround)			||
+	(mask && mask->common.need_workaround)		||
+	(dest->common.need_workaround);
+   
+    if (need_workaround)
+    {
 	apply_workaround (src, &src_x, &src_y, &src_bits, &src_dx, &src_dy);
-    if (mask && mask->common.need_workaround)
 	apply_workaround (mask, &mask_x, &mask_y, &mask_bits, &mask_dx, &mask_dy);
-    if (dest->common.need_workaround)
 	apply_workaround (dest, &dest_x, &dest_y, &dest_bits, &dest_dx, &dest_dy);
+    }
 
-    _pixman_implementation_composite (imp, op,
-                                      src, mask, dest,
-                                      src_x, src_y,
-                                      mask_x, mask_y,
-                                      dest_x, dest_y,
-                                      width, height);
-
-    if (src->common.need_workaround)
-	unapply_workaround (src, src_bits, src_dx, src_dy);
-    if (mask && mask->common.need_workaround)
-	unapply_workaround (mask, mask_bits, mask_dx, mask_dy);
-    if (dest->common.need_workaround)
-	unapply_workaround (dest, dest_bits, dest_dx, dest_dy);
+    do_composite (imp, op,
+		  src, mask, dest,
+		  src_x, src_y,
+		  mask_x, mask_y,
+		  dest_x, dest_y,
+		  width, height);
+    
+    if (need_workaround)
+    {
+	if (src->common.need_workaround)
+	    unapply_workaround (src, src_bits, src_dx, src_dy);
+	if (mask && mask->common.need_workaround)
+	    unapply_workaround (mask, mask_bits, mask_dx, mask_dy);
+	if (dest->common.need_workaround)
+	    unapply_workaround (dest, dest_bits, dest_dx, dest_dy);
+    }
 }
 
 PIXMAN_EXPORT pixman_bool_t
@@ -598,3 +1118,36 @@ pixman_format_supported_destination (pixman_format_code_t format)
     return pixman_format_supported_source (format);
 }
 
+PIXMAN_EXPORT pixman_bool_t
+pixman_compute_composite_region (pixman_region16_t * region,
+                                 pixman_image_t *    src_image,
+                                 pixman_image_t *    mask_image,
+                                 pixman_image_t *    dst_image,
+                                 int16_t             src_x,
+                                 int16_t             src_y,
+                                 int16_t             mask_x,
+                                 int16_t             mask_y,
+                                 int16_t             dest_x,
+                                 int16_t             dest_y,
+                                 uint16_t            width,
+                                 uint16_t            height)
+{
+    pixman_region32_t r32;
+    pixman_bool_t retval;
+
+    pixman_region32_init (&r32);
+
+    retval = pixman_compute_composite_region32 (
+	&r32, src_image, mask_image, dst_image,
+	src_x, src_y, mask_x, mask_y, dest_x, dest_y,
+	width, height);
+
+    if (retval)
+    {
+	if (!pixman_region16_copy_from_region32 (region, &r32))
+	    retval = FALSE;
+    }
+
+    pixman_region32_fini (&r32);
+    return retval;
+}