Merge remote-tracking branch 'origin/released'

* origin/released:
  fontconfig libX11 libXdmcp libxcb xkeyboard-config mesa pixman xserver git update 19 aug 2013

Conflicts:
	fontconfig/src/fccache.c

10 files changed, 381 insertions, 5 deletions

diff --git a/pixman/RELEASING b/pixman/RELEASING
index fbe15813d..657857de2 100644
--- a/pixman/RELEASING
+++ b/pixman/RELEASING
@@ -55,3 +55,5 @@ Here are the steps to follow to create a new pixman release:
    You must use "--tags" here; otherwise the new tag will not
    be pushed out.
 
+8) Change the topic of the #cairo IRC channel on freenode to advertise
+   the new version.
diff --git a/pixman/pixman/pixman-fast-path.c b/pixman/pixman/pixman-fast-path.c
index 3982dce8b..2608268d9 100644
--- a/pixman/pixman/pixman-fast-path.c
+++ b/pixman/pixman/pixman-fast-path.c
@@ -2261,6 +2261,237 @@ fast_write_back_r5g6b5 (pixman_iter_t *iter)
     }
 }
 
+typedef struct
+{
+    int		y;
+    uint64_t *	buffer;
+} line_t;
+
+typedef struct
+{
+    line_t		line0;
+    line_t		line1;
+    pixman_fixed_t	y;
+    pixman_fixed_t	x;
+    uint64_t		data[1];
+} bilinear_info_t;
+
+static void
+fetch_horizontal (bits_image_t *image, line_t *line,
+		  int y, pixman_fixed_t x, pixman_fixed_t ux, int n)
+{
+    uint32_t *bits = image->bits + y * image->rowstride;
+    int i;
+
+    for (i = 0; i < n; ++i)
+    {
+	int x0 = pixman_fixed_to_int (x);
+	int x1 = x0 + 1;
+	int32_t dist_x;
+
+	uint32_t left = *(bits + x0);
+	uint32_t right = *(bits + x1);
+
+	dist_x = pixman_fixed_to_bilinear_weight (x);
+	dist_x <<= (8 - BILINEAR_INTERPOLATION_BITS);
+
+#if SIZEOF_LONG <= 4
+	{
+	    uint32_t lag, rag, ag;
+	    uint32_t lrb, rrb, rb;
+
+	    lag = (left & 0xff00ff00) >> 8;
+	    rag = (right & 0xff00ff00) >> 8;
+	    ag = (lag << 8) + dist_x * (rag - lag);
+
+	    lrb = (left & 0x00ff00ff);
+	    rrb = (right & 0x00ff00ff);
+	    rb = (lrb << 8) + dist_x * (rrb - lrb);
+
+	    *((uint32_t *)(line->buffer + i)) = ag;
+	    *((uint32_t *)(line->buffer + i) + 1) = rb;
+	}
+#else
+	{
+	    uint64_t lagrb, ragrb;
+	    uint32_t lag, rag;
+	    uint32_t lrb, rrb;
+
+	    lag = (left & 0xff00ff00);
+	    lrb = (left & 0x00ff00ff);
+	    rag = (right & 0xff00ff00);
+	    rrb = (right & 0x00ff00ff);
+	    lagrb = (((uint64_t)lag) << 24) | lrb;
+	    ragrb = (((uint64_t)rag) << 24) | rrb;
+
+	    line->buffer[i] = (lagrb << 8) + dist_x * (ragrb - lagrb);
+	}
+#endif
+
+	x += ux;
+    }
+
+    line->y = y;
+}
+
+static uint32_t *
+fast_fetch_bilinear_cover (pixman_iter_t *iter, const uint32_t *mask)
+{
+    pixman_fixed_t fx, ux;
+    bilinear_info_t *info = iter->data;
+    line_t *line0, *line1;
+    int y0, y1;
+    int32_t dist_y;
+    int i;
+
+    fx = info->x;
+    ux = iter->image->common.transform->matrix[0][0];
+
+    y0 = pixman_fixed_to_int (info->y);
+    y1 = y0 + 1;
+    dist_y = pixman_fixed_to_bilinear_weight (info->y);
+    dist_y <<= (8 - BILINEAR_INTERPOLATION_BITS);
+
+    line0 = &info->line0;
+    line1 = &info->line1;
+
+    if (line0->y != y0 || line1->y != y1)
+    {
+	if (line0->y == y1 || line1->y == y0)
+	{
+	    line_t tmp = *line0;
+	    *line0 = *line1;
+	    *line1 = tmp;
+	}
+
+	if (line0->y != y0)
+	{
+	    fetch_horizontal (
+		&iter->image->bits, line0, y0, fx, ux, iter->width);
+	}
+
+	if (line1->y != y1)
+	{
+	    fetch_horizontal (
+		&iter->image->bits, line1, y1, fx, ux, iter->width);
+	}
+    }
+
+    for (i = 0; i < iter->width; ++i)
+    {
+#if SIZEOF_LONG <= 4
+	uint32_t ta, tr, tg, tb;
+	uint32_t ba, br, bg, bb;
+	uint32_t tag, trb;
+	uint32_t bag, brb;
+	uint32_t a, r, g, b;
+
+	tag = *((uint32_t *)(line0->buffer + i));
+	trb = *((uint32_t *)(line0->buffer + i) + 1);
+	bag = *((uint32_t *)(line1->buffer + i));
+	brb = *((uint32_t *)(line1->buffer + i) + 1);
+
+	ta = tag >> 16;
+	ba = bag >> 16;
+	a = (ta << 8) + dist_y * (ba - ta);
+
+	tr = trb >> 16;
+	br = brb >> 16;
+	r = (tr << 8) + dist_y * (br - tr);
+
+	tg = tag & 0xffff;
+	bg = bag & 0xffff;
+	g = (tg << 8) + dist_y * (bg - tg);
+	
+	tb = trb & 0xffff;
+	bb = brb & 0xffff;
+	b = (tb << 8) + dist_y * (bb - tb);
+
+	a = (a <<  8) & 0xff000000;
+	r = (r <<  0) & 0x00ff0000;
+	g = (g >>  8) & 0x0000ff00;
+	b = (b >> 16) & 0x000000ff;
+#else
+	uint64_t top = line0->buffer[i];
+	uint64_t bot = line1->buffer[i];
+	uint64_t tar = (top & 0xffff0000ffff0000ULL) >> 16;
+	uint64_t bar = (bot & 0xffff0000ffff0000ULL) >> 16;
+	uint64_t tgb = (top & 0x0000ffff0000ffffULL);
+	uint64_t bgb = (bot & 0x0000ffff0000ffffULL);
+	uint64_t ar, gb;
+	uint32_t a, r, g, b;
+
+	ar = (tar << 8) + dist_y * (bar - tar);
+	gb = (tgb << 8) + dist_y * (bgb - tgb);
+
+	a = ((ar >> 24) & 0xff000000);
+	r = ((ar >>  0) & 0x00ff0000);
+	g = ((gb >> 40) & 0x0000ff00);
+	b = ((gb >> 16) & 0x000000ff);
+#endif
+
+	iter->buffer[i] = a | r | g | b;
+    }
+
+    info->y += iter->image->common.transform->matrix[1][1];
+
+    return iter->buffer;
+}
+
+static void
+bilinear_cover_iter_fini (pixman_iter_t *iter)
+{
+    free (iter->data);
+}
+
+static void
+fast_bilinear_cover_iter_init (pixman_iter_t *iter, const pixman_iter_info_t *iter_info)
+{
+    int width = iter->width;
+    bilinear_info_t *info;
+    pixman_vector_t v;
+
+    /* Reference point is the center of the pixel */
+    v.vector[0] = pixman_int_to_fixed (iter->x) + pixman_fixed_1 / 2;
+    v.vector[1] = pixman_int_to_fixed (iter->y) + pixman_fixed_1 / 2;
+    v.vector[2] = pixman_fixed_1;
+
+    if (!pixman_transform_point_3d (iter->image->common.transform, &v))
+	goto fail;
+
+    info = malloc (sizeof (*info) + (2 * width - 1) * sizeof (uint64_t));
+    if (!info)
+	goto fail;
+
+    info->x = v.vector[0] - pixman_fixed_1 / 2;
+    info->y = v.vector[1] - pixman_fixed_1 / 2;
+
+    /* It is safe to set the y coordinates to -1 initially
+     * because COVER_CLIP_BILINEAR ensures that we will only
+     * be asked to fetch lines in the [0, height) interval
+     */
+    info->line0.y = -1;
+    info->line0.buffer = &(info->data[0]);
+    info->line1.y = -1;
+    info->line1.buffer = &(info->data[width]);
+
+    iter->get_scanline = fast_fetch_bilinear_cover;
+    iter->fini = bilinear_cover_iter_fini;
+
+    iter->data = info;
+    return;
+
+fail:
+    /* Something went wrong, either a bad matrix or OOM; in such cases,
+     * we don't guarantee any particular rendering.
+     */
+    _pixman_log_error (
+	FUNC, "Allocation failure or bad matrix, skipping rendering\n");
+    
+    iter->get_scanline = _pixman_iter_get_scanline_noop;
+    iter->fini = bilinear_cover_iter_fini;
+}
+
 #define IMAGE_FLAGS							\
     (FAST_PATH_STANDARD_FLAGS | FAST_PATH_ID_TRANSFORM |		\
      FAST_PATH_BITS_IMAGE | FAST_PATH_SAMPLES_COVER_CLIP_NEAREST)
@@ -2280,6 +2511,16 @@ static const pixman_iter_info_t fast_iters[] =
       _pixman_iter_init_bits_stride,
       fast_dest_fetch_noop, fast_write_back_r5g6b5 },
 
+    { PIXMAN_a8r8g8b8,
+      (FAST_PATH_STANDARD_FLAGS			|
+       FAST_PATH_SCALE_TRANSFORM		|
+       FAST_PATH_BILINEAR_FILTER		|
+       FAST_PATH_SAMPLES_COVER_CLIP_BILINEAR),
+      ITER_NARROW | ITER_SRC,
+      fast_bilinear_cover_iter_init,
+      NULL, NULL
+    },
+
     { PIXMAN_null },
 };
 
diff --git a/pixman/pixman/pixman-general.c b/pixman/pixman/pixman-general.c
index 4da5da5e2..6310bff9d 100644
--- a/pixman/pixman/pixman-general.c
+++ b/pixman/pixman/pixman-general.c
@@ -208,6 +208,13 @@ general_composite_rect  (pixman_implementation_t *imp,
 	dest_iter.write_back (&dest_iter);
     }
 
+    if (src_iter.fini)
+	src_iter.fini (&src_iter);
+    if (mask_iter.fini)
+	mask_iter.fini (&mask_iter);
+    if (dest_iter.fini)
+	dest_iter.fini (&dest_iter);
+    
     if (scanline_buffer != (uint8_t *) stack_scanline_buffer)
 	free (scanline_buffer);
 }
diff --git a/pixman/pixman/pixman-image.c b/pixman/pixman/pixman-image.c
index 4f9c2f966..1ff1a4974 100644
--- a/pixman/pixman/pixman-image.c
+++ b/pixman/pixman/pixman-image.c
@@ -926,6 +926,9 @@ _pixman_image_get_solid (pixman_implementation_t *imp,
 	    ITER_NARROW | ITER_SRC, image->common.flags);
 	
 	result = *iter.get_scanline (&iter, NULL);
+
+	if (iter.fini)
+	    iter.fini (&iter);
     }
 
     /* If necessary, convert RGB <--> BGR. */
diff --git a/pixman/pixman/pixman-implementation.c b/pixman/pixman/pixman-implementation.c
index 7f1332abf..f82c915dc 100644
--- a/pixman/pixman/pixman-implementation.c
+++ b/pixman/pixman/pixman-implementation.c
@@ -313,6 +313,7 @@ _pixman_implementation_iter_init (pixman_implementation_t *imp,
     iter->height = height;
     iter->iter_flags = iter_flags;
     iter->image_flags = image_flags;
+    iter->fini = NULL;
 
     if (!iter->image)
     {
diff --git a/pixman/pixman/pixman-private.h b/pixman/pixman/pixman-private.h
index af4a0b6e0..964660508 100644
--- a/pixman/pixman/pixman-private.h
+++ b/pixman/pixman/pixman-private.h
@@ -209,6 +209,7 @@ union pixman_image
 typedef struct pixman_iter_t pixman_iter_t;
 typedef uint32_t *(* pixman_iter_get_scanline_t) (pixman_iter_t *iter, const uint32_t *mask);
 typedef void      (* pixman_iter_write_back_t)   (pixman_iter_t *iter);
+typedef void	  (* pixman_iter_fini_t)	 (pixman_iter_t *iter);
 
 typedef enum
 {
@@ -255,6 +256,7 @@ struct pixman_iter_t
     /* These function pointers are initialized by the implementation */
     pixman_iter_get_scanline_t	get_scanline;
     pixman_iter_write_back_t	write_back;
+    pixman_iter_fini_t          fini;
 
     /* These fields are scratch data that implementations can use */
     void *			data;
diff --git a/pixman/test/Makefile.sources b/pixman/test/Makefile.sources
index b5fc740f3..2fabdb574 100644
--- a/pixman/test/Makefile.sources
+++ b/pixman/test/Makefile.sources
@@ -33,6 +33,7 @@ OTHERPROGRAMS =                 \
 	lowlevel-blt-bench	\
 	radial-perf-test	\
         check-formats           \
+	scaling-bench		\
 	$(NULL)
 
 # Utility functions
diff --git a/pixman/test/alpha-loop.c b/pixman/test/alpha-loop.c
index eca761537..4d4384d00 100644
--- a/pixman/test/alpha-loop.c
+++ b/pixman/test/alpha-loop.c
@@ -8,6 +8,7 @@
 int
 main (int argc, char **argv)
 {
+    pixman_image_t *a, *d, *s;
     uint8_t *alpha;
     uint32_t *src, *dest;
 
@@ -17,9 +18,9 @@ main (int argc, char **argv)
     src = (uint32_t *)make_random_bytes (WIDTH * HEIGHT * 4);
     dest = (uint32_t *)make_random_bytes (WIDTH * HEIGHT * 4);
 
-    pixman_image_t *a = pixman_image_create_bits (PIXMAN_a8, WIDTH, HEIGHT, (uint32_t *)alpha, WIDTH);
-    pixman_image_t *d = pixman_image_create_bits (PIXMAN_a8r8g8b8, WIDTH, HEIGHT, dest, WIDTH * 4);
-    pixman_image_t *s = pixman_image_create_bits (PIXMAN_a2r10g10b10, WIDTH, HEIGHT, src, WIDTH * 4);
+    a = pixman_image_create_bits (PIXMAN_a8, WIDTH, HEIGHT, (uint32_t *)alpha, WIDTH);
+    d = pixman_image_create_bits (PIXMAN_a8r8g8b8, WIDTH, HEIGHT, dest, WIDTH * 4);
+    s = pixman_image_create_bits (PIXMAN_a2r10g10b10, WIDTH, HEIGHT, src, WIDTH * 4);
 
     fail_after (5, "Infinite loop detected: 5 seconds without progress\n");
 
diff --git a/pixman/test/matrix-test.c b/pixman/test/matrix-test.c
index 8437dd291..0a5f203f5 100644
--- a/pixman/test/matrix-test.c
+++ b/pixman/test/matrix-test.c
@@ -70,6 +70,53 @@ pixman_bool_t does_it_fit_fixed_48_16 (__float128 x)
 
 #endif
 
+static inline uint32_t
+byteswap32 (uint32_t x)
+{
+    return ((x & ((uint32_t)0xFF << 24)) >> 24) |
+           ((x & ((uint32_t)0xFF << 16)) >>  8) |
+           ((x & ((uint32_t)0xFF <<  8)) <<  8) |
+           ((x & ((uint32_t)0xFF <<  0)) << 24);
+}
+
+static inline uint64_t
+byteswap64 (uint64_t x)
+{
+    return ((x & ((uint64_t)0xFF << 56)) >> 56) |
+           ((x & ((uint64_t)0xFF << 48)) >> 40) |
+           ((x & ((uint64_t)0xFF << 40)) >> 24) |
+           ((x & ((uint64_t)0xFF << 32)) >>  8) |
+           ((x & ((uint64_t)0xFF << 24)) <<  8) |
+           ((x & ((uint64_t)0xFF << 16)) << 24) |
+           ((x & ((uint64_t)0xFF <<  8)) << 40) |
+           ((x & ((uint64_t)0xFF <<  0)) << 56);
+}
+
+static void
+byteswap_transform (pixman_transform_t *t)
+{
+    int i, j;
+
+    if (is_little_endian ())
+        return;
+
+    for (i = 0; i < 3; i++)
+        for (j = 0; j < 3; j++)
+            t->matrix[i][j] = byteswap32 (t->matrix[i][j]);
+}
+
+static void
+byteswap_vector_48_16 (pixman_vector_48_16_t *v)
+{
+    int i;
+
+    if (is_little_endian ())
+        return;
+
+    for (i = 0; i < 3; i++)
+        v->v[i] = byteswap64 (v->v[i]);
+}
+
 uint32_t
 test_matrix (int testnum, int verbose)
 {
@@ -90,6 +137,8 @@ test_matrix (int testnum, int verbose)
 #endif
         prng_randmemset (&ti, sizeof(ti), 0);
         prng_randmemset (&vi, sizeof(vi), 0);
+        byteswap_transform (&ti);
+        byteswap_vector_48_16 (&vi);
 
         for (j = 0; j < 3; j++)
         {
@@ -132,8 +181,6 @@ test_matrix (int testnum, int verbose)
         else
             transform_ok = pixman_transform_point_31_16 (&ti, &vi, &result_i);
 
-        crc32 = compute_crc32 (crc32, &result_i, sizeof(result_i));
-
 #ifdef HAVE_FLOAT128
         /* compare with a reference 128-bit floating point implementation */
         for (j = 0; j < 3; j++)
@@ -173,6 +220,8 @@ test_matrix (int testnum, int verbose)
             }
         }
 #endif
+        byteswap_vector_48_16 (&result_i);
+        crc32 = compute_crc32 (crc32, &result_i, sizeof (result_i));
     }
     return crc32;
 }
diff --git a/pixman/test/scaling-bench.c b/pixman/test/scaling-bench.c
new file mode 100644
index 000000000..b39adeff5
--- /dev/null
+++ b/pixman/test/scaling-bench.c
@@ -0,0 +1,69 @@
+#include <stdlib.h>
+#include "utils.h"
+
+#define SOURCE_WIDTH 320
+#define SOURCE_HEIGHT 240
+
+static pixman_image_t *
+make_source (void)
+{
+    size_t n_bytes = (SOURCE_WIDTH + 2) * (SOURCE_HEIGHT + 2) * 4;
+    uint32_t *data = malloc (n_bytes);
+    pixman_image_t *source;
+
+    prng_randmemset (data, n_bytes, 0);
+    
+    source = pixman_image_create_bits (
+	PIXMAN_a8r8g8b8, SOURCE_WIDTH + 2, SOURCE_HEIGHT + 2,
+	data,
+	(SOURCE_WIDTH + 2) * 4);
+
+    pixman_image_set_filter (source, PIXMAN_FILTER_BILINEAR, NULL, 0);
+
+    return source;
+}
+
+int
+main ()
+{
+    double scale;
+    pixman_image_t *src;
+
+    prng_srand (23874);
+    
+    src = make_source ();
+    printf ("# %-6s %-22s   %-14s %-12s\n",
+	    "ratio",
+	    "resolutions",
+	    "time / ms",
+	    "time per pixel / ns");
+    for (scale = 0.1; scale < 10.005; scale += 0.01)
+    {
+	int dest_width = SOURCE_WIDTH * scale + 0.5;
+	int dest_height = SOURCE_HEIGHT * scale + 0.5;
+	pixman_fixed_t s = (1 / scale) * 65536.0 + 0.5;
+	pixman_transform_t transform;
+	pixman_image_t *dest;
+	double t1, t2;
+
+	pixman_transform_init_scale (&transform, s, s);
+	pixman_image_set_transform (src, &transform);
+	
+	dest = pixman_image_create_bits (
+	    PIXMAN_a8r8g8b8, dest_width, dest_height, NULL, -1);
+
+	t1 = gettime();
+	pixman_image_composite (
+	    PIXMAN_OP_OVER, src, NULL, dest,
+	    scale, scale, 0, 0, 0, 0, dest_width, dest_height);
+	t2 = gettime();
+	
+	printf ("%6.2f : %4dx%-4d => %4dx%-4d : %12.4f : %12.4f\n",
+		scale, SOURCE_WIDTH, SOURCE_HEIGHT, dest_width, dest_height,
+		(t2 - t1) * 1000, ((t2 - t1) / (dest_width * dest_height)) * 1000000000);
+
+	pixman_image_unref (dest);
+    }
+
+    return 0;
+}