pixman/pixman-fast-path.c | 292 ++++++++++++++++++++++++++++++++++++++++++++++
pixman/pixman-image.c | 17 ++
pixman/pixman-private.h | 3
test/affine-test.c | 34 ++++-
4 files changed, 342 insertions(+), 4 deletions(-)
New commits:
commit 8d359b00c5bb9960c3c584a7f77a943c0ce61368
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Fri Jul 30 18:37:51 2010 +0300
C fast paths for a simple 90/270 degrees rotation
Depending on CPU architecture, performance is in the range of 1.5 to 4 times
slower than simple nonrotated copy (which would be an ideal case, perfectly
utilizing memory bandwidth), but still is more than 7 times faster if
compared to general path.
This implementation sets a performance baseline for rotation. The use
of SIMD instructions may further improve memory bandwidth utilization.
diff --git a/pixman/pixman-fast-path.c b/pixman/pixman-fast-path.c
index eb09715..771ec9d 100644
--- a/pixman/pixman-fast-path.c
+++ b/pixman/pixman-fast-path.c
@@ -1616,6 +1616,272 @@ fast_composite_scaled_nearest (pixman_implementation_t *imp,
}
}
+#define CACHE_LINE_SIZE 64
+
+#define FAST_SIMPLE_ROTATE(suffix, pix_type) \
+ \
+static void \
+blt_rotated_90_trivial_##suffix (pix_type *dst, \
+ int dst_stride, \
+ const pix_type *src, \
+ int src_stride, \
+ int w, \
+ int h) \
+{ \
+ int x, y; \
+ for (y = 0; y < h; y++) \
+ { \
+ const pix_type *s = src + (h - y - 1); \
+ pix_type *d = dst + dst_stride * y; \
+ for (x = 0; x < w; x++) \
+ { \
+ *d++ = *s; \
+ s += src_stride; \
+ } \
+ } \
+} \
+ \
+static void \
+blt_rotated_270_trivial_##suffix (pix_type *dst, \
+ int dst_stride, \
+ const pix_type *src, \
+ int src_stride, \
+ int w, \
+ int h) \
+{ \
+ int x, y; \
+ for (y = 0; y < h; y++) \
+ { \
+ const pix_type *s = src + src_stride * (w - 1) + y; \
+ pix_type *d = dst + dst_stride * y; \
+ for (x = 0; x < w; x++) \
+ { \
+ *d++ = *s; \
+ s -= src_stride; \
+ } \
+ } \
+} \
+ \
+static void \
+blt_rotated_90_##suffix (pix_type *dst, \
+ int dst_stride, \
+ const pix_type *src, \
+ int src_stride, \
+ int W, \
+ int H) \
+{ \
+ int x; \
+ int leading_pixels = 0, trailing_pixels = 0; \
+ const int TILE_SIZE = CACHE_LINE_SIZE / sizeof(pix_type); \
+ \
+ /* \
+ * split processing into handling destination as TILE_SIZExH cache line \
+ * aligned vertical stripes (optimistically assuming that destination \
+ * stride is a multiple of cache line, if not - it will be just a bit \
+ * slower) \
+ */ \
+ \
+ if ((uintptr_t)dst & (CACHE_LINE_SIZE - 1)) \
+ { \
+ leading_pixels = TILE_SIZE - (((uintptr_t)dst & \
+ (CACHE_LINE_SIZE - 1)) / sizeof(pix_type)); \
+ if (leading_pixels > W) \
+ leading_pixels = W; \
+ \
+ /* unaligned leading part NxH (where N < TILE_SIZE) */ \
+ blt_rotated_90_trivial_##suffix ( \
+ dst, \
+ dst_stride, \
+ src, \
+ src_stride, \
+ leading_pixels, \
+ H); \
+ \
+ dst += leading_pixels; \
+ src += leading_pixels * src_stride; \
+ W -= leading_pixels; \
+ } \
+ \
+ if ((uintptr_t)(dst + W) & (CACHE_LINE_SIZE - 1)) \
+ { \
+ trailing_pixels = (((uintptr_t)(dst + W) & \
+ (CACHE_LINE_SIZE - 1)) / sizeof(pix_type)); \
+ if (trailing_pixels > W) \
+ trailing_pixels = W; \
+ W -= trailing_pixels; \
+ } \
+ \
+ for (x = 0; x < W; x += TILE_SIZE) \
+ { \
+ /* aligned middle part TILE_SIZExH */ \
+ blt_rotated_90_trivial_##suffix ( \
+ dst + x, \
+ dst_stride, \
+ src + src_stride * x, \
+ src_stride, \
+ TILE_SIZE, \
+ H); \
+ } \
+ \
+ if (trailing_pixels) \
+ { \
+ /* unaligned trailing part NxH (where N < TILE_SIZE) */ \
+ blt_rotated_90_trivial_##suffix ( \
+ dst + W, \
+ dst_stride, \
+ src + W * src_stride, \
+ src_stride, \
+ trailing_pixels, \
+ H); \
+ } \
+} \
+ \
+static void \
+blt_rotated_270_##suffix (pix_type *dst, \
+ int dst_stride, \
+ const pix_type *src, \
+ int src_stride, \
+ int W, \
+ int H) \
+{ \
+ int x; \
+ int leading_pixels = 0, trailing_pixels = 0; \
+ const int TILE_SIZE = CACHE_LINE_SIZE / sizeof(pix_type); \
+ \
+ /* \
+ * split processing into handling destination as TILE_SIZExH cache line \
+ * aligned vertical stripes (optimistically assuming that destination \
+ * stride is a multiple of cache line, if not - it will be just a bit \
+ * slower) \
+ */ \
+ \
+ if ((uintptr_t)dst & (CACHE_LINE_SIZE - 1)) \
+ { \
+ leading_pixels = TILE_SIZE - (((uintptr_t)dst & \
+ (CACHE_LINE_SIZE - 1)) / sizeof(pix_type)); \
+ if (leading_pixels > W) \
+ leading_pixels = W; \
+ \
+ /* unaligned leading part NxH (where N < TILE_SIZE) */ \
+ blt_rotated_270_trivial_##suffix ( \
+ dst, \
+ dst_stride, \
+ src + src_stride * (W - leading_pixels), \
+ src_stride, \
+ leading_pixels, \
+ H); \
+ \
+ dst += leading_pixels; \
+ W -= leading_pixels; \
+ } \
+ \
+ if ((uintptr_t)(dst + W) & (CACHE_LINE_SIZE - 1)) \
+ { \
+ trailing_pixels = (((uintptr_t)(dst + W) & \
+ (CACHE_LINE_SIZE - 1)) / sizeof(pix_type)); \
+ if (trailing_pixels > W) \
+ trailing_pixels = W; \
+ W -= trailing_pixels; \
+ src += trailing_pixels * src_stride; \
+ } \
+ \
+ for (x = 0; x < W; x += TILE_SIZE) \
+ { \
+ /* aligned middle part TILE_SIZExH */ \
+ blt_rotated_270_trivial_##suffix ( \
+ dst + x, \
+ dst_stride, \
+ src + src_stride * (W - x - TILE_SIZE), \
+ src_stride, \
+ TILE_SIZE, \
+ H); \
+ } \
+ \
+ if (trailing_pixels) \
+ { \
+ /* unaligned trailing part NxH (where N < TILE_SIZE) */ \
+ blt_rotated_270_trivial_##suffix ( \
+ dst + W, \
+ dst_stride, \
+ src - trailing_pixels * src_stride, \
+ src_stride, \
+ trailing_pixels, \
+ H); \
+ } \
+} \
+ \
+static void \
+fast_composite_rotate_90_##suffix (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) \
+{ \
+ pix_type *dst_line; \
+ pix_type *src_line; \
+ int dst_stride, src_stride; \
+ int src_x_t, src_y_t; \
+ \
+ PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, pix_type, \
+ dst_stride, dst_line, 1); \
+ src_x_t = -src_y + pixman_fixed_to_int ( \
+ src_image->common.transform->matrix[0][2] + \
+ pixman_fixed_1 / 2 - pixman_fixed_e) - height;\
+ src_y_t = src_x + pixman_fixed_to_int ( \
+ src_image->common.transform->matrix[1][2] + \
+ pixman_fixed_1 / 2 - pixman_fixed_e); \
+ PIXMAN_IMAGE_GET_LINE (src_image, src_x_t, src_y_t, pix_type, \
+ src_stride, src_line, 1); \
+ blt_rotated_90_##suffix (dst_line, dst_stride, src_line, src_stride, \
+ width, height); \
+} \
+ \
+static void \
+fast_composite_rotate_270_##suffix (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) \
+{ \
+ pix_type *dst_line; \
+ pix_type *src_line; \
+ int dst_stride, src_stride; \
+ int src_x_t, src_y_t; \
+ \
+ PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, pix_type, \
+ dst_stride, dst_line, 1); \
+ src_x_t = src_y + pixman_fixed_to_int ( \
+ src_image->common.transform->matrix[0][2] + \
+ pixman_fixed_1 / 2 - pixman_fixed_e); \
+ src_y_t = -src_x + pixman_fixed_to_int ( \
+ src_image->common.transform->matrix[1][2] + \
+ pixman_fixed_1 / 2 - pixman_fixed_e) - width; \
+ PIXMAN_IMAGE_GET_LINE (src_image, src_x_t, src_y_t, pix_type, \
+ src_stride, src_line, 1); \
+ blt_rotated_270_##suffix (dst_line, dst_stride, src_line, src_stride, \
+ width, height); \
+}
+
+FAST_SIMPLE_ROTATE (8, uint8_t)
+FAST_SIMPLE_ROTATE (565, uint16_t)
+FAST_SIMPLE_ROTATE (8888, uint32_t)
+
static const pixman_fast_path_t c_fast_paths[] =
{
PIXMAN_STD_FAST_PATH (OVER, solid, a8, r5g6b5, fast_composite_over_n_8_0565),
@@ -1734,6 +2000,32 @@ static const pixman_fast_path_t c_fast_paths[] =
NEAREST_FAST_PATH (OVER, x8b8g8r8, a8b8g8r8),
NEAREST_FAST_PATH (OVER, a8b8g8r8, a8b8g8r8),
+#define SIMPLE_ROTATE_FLAGS(angle) \
+ (FAST_PATH_ROTATE_ ## angle ## _TRANSFORM | \
+ FAST_PATH_NEAREST_FILTER | \
+ FAST_PATH_SAMPLES_COVER_CLIP | \
+ FAST_PATH_STANDARD_FLAGS)
+
+#define SIMPLE_ROTATE_FAST_PATH(op,s,d,suffix) \
+ { PIXMAN_OP_ ## op, \
+ PIXMAN_ ## s, SIMPLE_ROTATE_FLAGS (90), \
+ PIXMAN_null, 0, \
+ PIXMAN_ ## d, FAST_PATH_STD_DEST_FLAGS, \
+ fast_composite_rotate_90_##suffix, \
+ }, \
+ { PIXMAN_OP_ ## op, \
+ PIXMAN_ ## s, SIMPLE_ROTATE_FLAGS (270), \
+ PIXMAN_null, 0, \
+ PIXMAN_ ## d, FAST_PATH_STD_DEST_FLAGS, \
+ fast_composite_rotate_270_##suffix, \
+ }
+
+ SIMPLE_ROTATE_FAST_PATH (SRC, a8r8g8b8, a8r8g8b8, 8888),
+ SIMPLE_ROTATE_FAST_PATH (SRC, a8r8g8b8, x8r8g8b8, 8888),
+ SIMPLE_ROTATE_FAST_PATH (SRC, x8r8g8b8, x8r8g8b8, 8888),
+ SIMPLE_ROTATE_FAST_PATH (SRC, r5g6b5, r5g6b5, 565),
+ SIMPLE_ROTATE_FAST_PATH (SRC, a8, a8, 8),
+
{ PIXMAN_OP_NONE },
};
commit e0c7948c970b816f323a6402241ca70fa855c12c
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Thu Jul 29 17:58:13 2010 +0300
New flags for 90/180/270 rotation
These flags are set when the transform is a simple nonscaled 90/180/270
degrees rotation.
diff --git a/pixman/pixman-image.c b/pixman/pixman-image.c
index a72299b..e91d87c 100644
--- a/pixman/pixman-image.c
+++ b/pixman/pixman-image.c
@@ -211,8 +211,25 @@ compute_image_info (pixman_image_t *image)
if (image->common.transform->matrix[0][1] == 0 &&
image->common.transform->matrix[1][0] == 0)
{
+ if (image->common.transform->matrix[0][0] == -pixman_fixed_1 &&
+ image->common.transform->matrix[1][1] == -pixman_fixed_1)
+ {
+ flags |= FAST_PATH_ROTATE_180_TRANSFORM;
+ }
flags |= FAST_PATH_SCALE_TRANSFORM;
}
+ else if (image->common.transform->matrix[0][0] == 0 &&
+ image->common.transform->matrix[1][1] == 0)
+ {
+ pixman_fixed_t m01 = image->common.transform->matrix[0][1];
+ if (m01 == -image->common.transform->matrix[1][0])
+ {
+ if (m01 == -pixman_fixed_1)
+ flags |= FAST_PATH_ROTATE_90_TRANSFORM;
+ else if (m01 == pixman_fixed_1)
+ flags |= FAST_PATH_ROTATE_270_TRANSFORM;
+ }
+ }
}
if (image->common.transform->matrix[0][0] > 0)
diff --git a/pixman/pixman-private.h b/pixman/pixman-private.h
index f5d0ba1..1473dc4 100644
--- a/pixman/pixman-private.h
+++ b/pixman/pixman-private.h
@@ -611,6 +611,9 @@ _pixman_iter_get_scanline_noop (pixman_iter_t *iter, const uint32_t *mask);
#define FAST_PATH_AFFINE_TRANSFORM (1 << 18)
#define FAST_PATH_Y_UNIT_ZERO (1 << 19)
#define FAST_PATH_BILINEAR_FILTER (1 << 20)
+#define FAST_PATH_ROTATE_90_TRANSFORM (1 << 21)
+#define FAST_PATH_ROTATE_180_TRANSFORM (1 << 22)
+#define FAST_PATH_ROTATE_270_TRANSFORM (1 << 23)
#define FAST_PATH_PAD_REPEAT \
(FAST_PATH_NO_NONE_REPEAT | \
commit 3b68c295fd45297a631569b35608364dbcb6d452
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Tue Oct 26 15:40:01 2010 +0300
test: affine-test updated to stress 90/180/270 degrees rotation more
diff --git a/test/affine-test.c b/test/affine-test.c
index f7f058a..b7a1fa6 100644
--- a/test/affine-test.c
+++ b/test/affine-test.c
@@ -112,10 +112,36 @@ test_composite (int testnum,
if (lcg_rand_n (4) > 0)
{
- int c = lcg_rand_N (2 * 65536) - 65536;
- int s = lcg_rand_N (2 * 65536) - 65536;
-
+ int c, s, tx = 0, ty = 0;
+ switch (lcg_rand_n (4))
+ {
+ case 0:
+ /* 90 degrees */
+ c = 0;
+ s = pixman_fixed_1;
+ tx = pixman_int_to_fixed (MAX_SRC_HEIGHT);
+ break;
+ case 1:
+ /* 180 degrees */
+ c = -pixman_fixed_1;
+ s = 0;
+ tx = pixman_int_to_fixed (MAX_SRC_WIDTH);
+ ty = pixman_int_to_fixed (MAX_SRC_HEIGHT);
+ break;
+ case 2:
+ /* 270 degrees */
+ c = 0;
+ s = -pixman_fixed_1;
+ ty = pixman_int_to_fixed (MAX_SRC_WIDTH);
+ break;
+ default:
+ /* arbitrary rotation */
+ c = lcg_rand_N (2 * 65536) - 65536;
+ s = lcg_rand_N (2 * 65536) - 65536;
+ break;
+ }
pixman_transform_rotate (&transform, NULL, c, s);
+ pixman_transform_translate (&transform, NULL, tx, ty);
}
pixman_image_set_transform (src_img, &transform);
@@ -256,6 +282,6 @@ main (int argc, const char *argv[])
{
pixman_disable_out_of_bounds_workaround ();
- return fuzzer_test_main ("affine", 8000000, 0x46EC3C6A,
+ return fuzzer_test_main ("affine", 8000000, 0x4B5D1852,
test_composite, argc, argv);
}