pixman: Branch 'master' - 4 commits
Søren Sandmann Pedersen
sandmann at kemper.freedesktop.org
Thu Jul 22 05:00:38 PDT 2010
pixman/pixman-fast-path.c | 38 ++++++
pixman/pixman-sse2.c | 291 +++++++++++++++++++++++++++++++++++++++++++++-
2 files changed, 328 insertions(+), 1 deletion(-)
New commits:
commit bf125fbbb701788d5d9ed9ff368bb4fe9d9c895e
Author: Søren Sandmann Pedersen <ssp at redhat.com>
Date: Sun Apr 25 20:25:50 2010 -0400
[sse2] Add sse2_composite_add_n_8()
This shows up when epiphany displays the "ImageTest" on
glimr.rubyforge.org/cake/canvas.html
diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
index 822fffe..3dd7967 100644
--- a/pixman/pixman-sse2.c
+++ b/pixman/pixman-sse2.c
@@ -5386,6 +5386,103 @@ sse2_composite_add_n_8_8 (pixman_implementation_t *imp,
_mm_empty ();
}
+/* -------------------------------------------------------------------------
+ * composite_add_n_8_8
+ */
+
+static void
+sse2_composite_add_n_8 (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)
+{
+ uint8_t *dst_line, *dst;
+ int dst_stride;
+ int32_t w;
+ uint32_t src;
+
+ __m128i xmm_src;
+
+ PIXMAN_IMAGE_GET_LINE (
+ dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
+
+ src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+ src >>= 24;
+
+ if (src == 0x00)
+ return;
+
+ if (src == 0xff)
+ {
+ pixman_fill (dst_image->bits.bits, dst_image->bits.rowstride,
+ 8, dest_x, dest_y, width, height, 0xff);
+
+ return;
+ }
+
+ src = (src << 24) | (src << 16) | (src << 8) | src;
+ xmm_src = _mm_set_epi32 (src, src, src, src);
+
+ while (height--)
+ {
+ dst = dst_line;
+ dst_line += dst_stride;
+ w = width;
+
+ /* call prefetch hint to optimize cache load*/
+ cache_prefetch ((__m128i*)dst);
+
+ while (w && ((unsigned long)dst & 15))
+ {
+ *dst = (uint8_t)_mm_cvtsi64_si32 (
+ _mm_adds_pu8 (
+ _mm_movepi64_pi64 (xmm_src),
+ _mm_cvtsi32_si64 (*dst)));
+
+ w--;
+ dst++;
+ }
+
+ /* call prefetch hint to optimize cache load*/
+ cache_prefetch ((__m128i*)dst);
+
+ while (w >= 16)
+ {
+ /* fill cache line with next memory */
+ cache_prefetch_next ((__m128i*)dst);
+
+ save_128_aligned (
+ (__m128i*)dst, _mm_adds_epu8 (xmm_src, load_128_aligned ((__m128i*)dst)));
+
+ dst += 16;
+ w -= 16;
+ }
+
+ while (w)
+ {
+ *dst = (uint8_t)_mm_cvtsi64_si32 (
+ _mm_adds_pu8 (
+ _mm_movepi64_pi64 (xmm_src),
+ _mm_cvtsi32_si64 (*dst)));
+
+ w--;
+ dst++;
+ }
+ }
+
+ _mm_empty ();
+}
+
/* ----------------------------------------------------------------------
* composite_add_8000_8000
*/
@@ -6309,6 +6406,7 @@ static const pixman_fast_path_t sse2_fast_paths[] =
PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, null, a8r8g8b8, sse2_composite_add_8888_8888),
PIXMAN_STD_FAST_PATH (ADD, a8b8g8r8, null, a8b8g8r8, sse2_composite_add_8888_8888),
PIXMAN_STD_FAST_PATH (ADD, solid, a8, a8, sse2_composite_add_n_8_8),
+ PIXMAN_STD_FAST_PATH (ADD, solid, null, a8, sse2_composite_add_n_8),
/* PIXMAN_OP_SRC */
PIXMAN_STD_FAST_PATH (SRC, solid, a8, a8r8g8b8, sse2_composite_src_n_8_8888),
commit 16ae3285e6601ea177637dddd20d2857d13decac
Author: Søren Sandmann Pedersen <ssp at redhat.com>
Date: Sun Apr 25 19:54:28 2010 -0400
[sse2] Add sse2_composite_in_n_8()
This shows up when epiphany displays the "ImageTest" on
glimr.rubyforge.org/cake/canvas.html
diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
index 2c9f5eb..822fffe 100644
--- a/pixman/pixman-sse2.c
+++ b/pixman/pixman-sse2.c
@@ -5066,6 +5066,112 @@ sse2_composite_in_n_8_8 (pixman_implementation_t *imp,
_mm_empty ();
}
+/* -----------------------------------------------------------------------
+ * composite_in_n_8
+ */
+
+static void
+sse2_composite_in_n_8 (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)
+{
+ uint8_t *dst_line, *dst;
+ int dst_stride;
+ uint32_t d;
+ uint32_t src;
+ int32_t w;
+
+ __m128i xmm_alpha;
+ __m128i xmm_dst, xmm_dst_lo, xmm_dst_hi;
+
+ PIXMAN_IMAGE_GET_LINE (
+ dst_image, dest_x, dest_y, uint8_t, dst_stride, dst_line, 1);
+
+ src = _pixman_image_get_solid (src_image, dst_image->bits.format);
+
+ xmm_alpha = expand_alpha_1x128 (expand_pixel_32_1x128 (src));
+
+ src = src >> 24;
+
+ if (src == 0xff)
+ return;
+
+ if (src == 0x00)
+ {
+ pixman_fill (dst_image->bits.bits, dst_image->bits.rowstride,
+ 8, dest_x, dest_y, width, height, src);
+
+ return;
+ }
+
+ while (height--)
+ {
+ dst = dst_line;
+ dst_line += dst_stride;
+ w = width;
+
+ /* call prefetch hint to optimize cache load*/
+ cache_prefetch ((__m128i*)dst);
+
+ while (w && ((unsigned long)dst & 15))
+ {
+ d = (uint32_t) *dst;
+
+ *dst++ = (uint8_t) pack_1x64_32 (
+ pix_multiply_1x64 (
+ _mm_movepi64_pi64 (xmm_alpha),
+ unpack_32_1x64 (d)));
+ w--;
+ }
+
+ /* call prefetch hint to optimize cache load*/
+ cache_prefetch ((__m128i*)dst);
+
+ while (w >= 16)
+ {
+ /* fill cache line with next memory */
+ cache_prefetch_next ((__m128i*)dst);
+
+ xmm_dst = load_128_aligned ((__m128i*)dst);
+
+ unpack_128_2x128 (xmm_dst, &xmm_dst_lo, &xmm_dst_hi);
+
+ pix_multiply_2x128 (&xmm_alpha, &xmm_alpha,
+ &xmm_dst_lo, &xmm_dst_hi,
+ &xmm_dst_lo, &xmm_dst_hi);
+
+ save_128_aligned (
+ (__m128i*)dst, pack_2x128_128 (xmm_dst_lo, xmm_dst_hi));
+
+ dst += 16;
+ w -= 16;
+ }
+
+ while (w)
+ {
+ d = (uint32_t) *dst;
+
+ *dst++ = (uint8_t) pack_1x64_32 (
+ pix_multiply_1x64 (
+ _mm_movepi64_pi64 (xmm_alpha),
+ unpack_32_1x64 (d)));
+ w--;
+ }
+ }
+
+ _mm_empty ();
+}
+
/* ---------------------------------------------------------------------------
* composite_in_8_8
*/
@@ -6192,7 +6298,7 @@ static const pixman_fast_path_t sse2_fast_paths[] =
PIXMAN_STD_FAST_PATH (OVER, rpixbuf, rpixbuf, b5g6r5, sse2_composite_over_pixbuf_0565),
PIXMAN_STD_FAST_PATH (OVER, x8r8g8b8, null, x8r8g8b8, sse2_composite_copy_area),
PIXMAN_STD_FAST_PATH (OVER, x8b8g8r8, null, x8b8g8r8, sse2_composite_copy_area),
-
+
/* PIXMAN_OP_OVER_REVERSE */
PIXMAN_STD_FAST_PATH (OVER_REVERSE, solid, null, a8r8g8b8, sse2_composite_over_reverse_n_8888),
PIXMAN_STD_FAST_PATH (OVER_REVERSE, solid, null, a8b8g8r8, sse2_composite_over_reverse_n_8888),
@@ -6223,6 +6329,7 @@ static const pixman_fast_path_t sse2_fast_paths[] =
/* PIXMAN_OP_IN */
PIXMAN_STD_FAST_PATH (IN, a8, null, a8, sse2_composite_in_8_8),
PIXMAN_STD_FAST_PATH (IN, solid, a8, a8, sse2_composite_in_n_8_8),
+ PIXMAN_STD_FAST_PATH (IN, solid, null, a8, sse2_composite_in_n_8),
{ PIXMAN_OP_NONE },
};
commit e0b430a13ee4619bd6d82c4ebff8a401a254e9bc
Author: Søren Sandmann Pedersen <ssp at redhat.com>
Date: Tue Jul 13 00:31:35 2010 -0400
[sse2] Add sse2_composite_src_x888_8888()
This operation shows up when Firefox displays
http://dougx.net/plunder/plunder.html
diff --git a/pixman/pixman-sse2.c b/pixman/pixman-sse2.c
index d5349d7..2c9f5eb 100644
--- a/pixman/pixman-sse2.c
+++ b/pixman/pixman-sse2.c
@@ -3517,6 +3517,88 @@ sse2_composite_over_8888_n_8888 (pixman_implementation_t *imp,
_mm_empty ();
}
+/*---------------------------------------------------------------------
+ * composite_over_8888_n_8888
+ */
+
+static void
+sse2_composite_src_x888_8888 (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)
+{
+ uint32_t *dst_line, *dst;
+ uint32_t *src_line, *src;
+ int32_t w;
+ int dst_stride, src_stride;
+
+
+ PIXMAN_IMAGE_GET_LINE (
+ dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+ PIXMAN_IMAGE_GET_LINE (
+ src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+
+ while (height--)
+ {
+ dst = dst_line;
+ dst_line += dst_stride;
+ src = src_line;
+ src_line += src_stride;
+ w = width;
+
+ /* call prefetch hint to optimize cache load*/
+ cache_prefetch ((__m128i*)src);
+
+ while (w && (unsigned long)dst & 15)
+ {
+ *dst++ = *src++ | 0xff000000;
+ w--;
+ }
+
+ /* call prefetch hint to optimize cache load*/
+ cache_prefetch ((__m128i*)src);
+
+ while (w >= 16)
+ {
+ __m128i xmm_src1, xmm_src2, xmm_src3, xmm_src4;
+
+ /* fill cache line with next memory */
+ cache_prefetch_next ((__m128i*)src);
+
+ xmm_src1 = load_128_unaligned ((__m128i*)src + 0);
+ xmm_src2 = load_128_unaligned ((__m128i*)src + 1);
+ xmm_src3 = load_128_unaligned ((__m128i*)src + 2);
+ xmm_src4 = load_128_unaligned ((__m128i*)src + 3);
+
+ save_128_aligned ((__m128i*)dst + 0, _mm_or_si128 (xmm_src1, mask_ff000000));
+ save_128_aligned ((__m128i*)dst + 1, _mm_or_si128 (xmm_src2, mask_ff000000));
+ save_128_aligned ((__m128i*)dst + 2, _mm_or_si128 (xmm_src3, mask_ff000000));
+ save_128_aligned ((__m128i*)dst + 3, _mm_or_si128 (xmm_src4, mask_ff000000));
+
+ dst += 16;
+ src += 16;
+ w -= 16;
+ }
+
+ while (w)
+ {
+ *dst++ = *src++ | 0xff000000;
+ w--;
+ }
+ }
+
+ _mm_empty ();
+}
+
/* ---------------------------------------------------------------------
* composite_over_x888_n_8888
*/
@@ -6127,6 +6209,8 @@ static const pixman_fast_path_t sse2_fast_paths[] =
PIXMAN_STD_FAST_PATH (SRC, solid, a8, x8r8g8b8, sse2_composite_src_n_8_8888),
PIXMAN_STD_FAST_PATH (SRC, solid, a8, a8b8g8r8, sse2_composite_src_n_8_8888),
PIXMAN_STD_FAST_PATH (SRC, solid, a8, x8b8g8r8, sse2_composite_src_n_8_8888),
+ PIXMAN_STD_FAST_PATH (SRC, x8r8g8b8, null, a8r8g8b8, sse2_composite_src_x888_8888),
+ PIXMAN_STD_FAST_PATH (SRC, x8b8g8r8, null, a8b8g8r8, sse2_composite_src_x888_8888),
PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, a8r8g8b8, sse2_composite_copy_area),
PIXMAN_STD_FAST_PATH (SRC, a8b8g8r8, null, a8b8g8r8, sse2_composite_copy_area),
PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, x8r8g8b8, sse2_composite_copy_area),
commit 16bae8347529c1c976e6d7af90e0fb5811605af1
Author: Søren Sandmann Pedersen <ssp at redhat.com>
Date: Tue Jul 13 00:08:10 2010 -0400
[fast] Add fast_composite_src_x888_8888()
This shows up on when Firefox displays http://dougx.net/plunder/plunder.html
diff --git a/pixman/pixman-fast-path.c b/pixman/pixman-fast-path.c
index c65dfb1..6ed1580 100644
--- a/pixman/pixman-fast-path.c
+++ b/pixman/pixman-fast-path.c
@@ -724,6 +724,42 @@ fast_composite_over_8888_8888 (pixman_implementation_t *imp,
}
}
+static void
+fast_composite_src_x888_8888 (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)
+{
+ uint32_t *dst_line, *dst;
+ uint32_t *src_line, *src;
+ int dst_stride, src_stride;
+ int32_t w;
+
+ PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, uint32_t, dst_stride, dst_line, 1);
+ PIXMAN_IMAGE_GET_LINE (src_image, src_x, src_y, uint32_t, src_stride, src_line, 1);
+
+ while (height--)
+ {
+ dst = dst_line;
+ dst_line += dst_stride;
+ src = src_line;
+ src_line += src_stride;
+ w = width;
+
+ while (w--)
+ *dst++ = (*src++) | 0xff000000;
+ }
+}
+
#if 0
static void
fast_composite_over_8888_0888 (pixman_implementation_t *imp,
@@ -1805,6 +1841,8 @@ static const pixman_fast_path_t c_fast_paths[] =
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, x8r8g8b8, null, a8r8g8b8, fast_composite_src_x888_8888),
+ PIXMAN_STD_FAST_PATH (SRC, x8b8g8r8, null, a8b8g8r8, fast_composite_src_x888_8888),
PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, x8r8g8b8, fast_composite_src_memcpy),
PIXMAN_STD_FAST_PATH (SRC, a8r8g8b8, null, a8r8g8b8, fast_composite_src_memcpy),
PIXMAN_STD_FAST_PATH (SRC, x8r8g8b8, null, x8r8g8b8, fast_composite_src_memcpy),
More information about the xorg-commit
mailing list