From: Siarhei Siamashka <siarhei.siamashka at nokia.com>
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.
---
pixman/pixman-fast-path.c | 292 +++++++++++++++++++++++++++++++++++++++++++++
1 files changed, 292 insertions(+), 0 deletions(-)
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 },
};
--
1.7.3.4