pixman: Branch 'master' - 2 commits
Siarhei Siamashka
siamashka at kemper.freedesktop.org
Mon Dec 6 18:54:38 PST 2010
pixman/pixman-arm-neon-asm.S | 57 +++++++++++++++++++++++++++++++++++++++++++
pixman/pixman-arm-neon.c | 5 +++
pixman/pixman-general.c | 6 ++--
3 files changed, 65 insertions(+), 3 deletions(-)
New commits:
commit 3d094997b1820719d15cec7dc633ed37e1912bfc
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Tue Nov 30 00:31:06 2010 +0200
Fix for potential unaligned memory accesses
The temporary scanline buffer allocated on stack was declared
as uint8_t array. As a result, the compiler was free to select
any arbitrary alignment for it (even though there is typically
no reason to use really weird alignments here and the stack is
normally at least 4 bytes aligned on most platforms). Having
improper alignment is non-portable and can impact performance
or even make the code misbehave depending on the target platform.
Using uint64_t type for this array should ensure that any possible
memory accesses done by pixman code are going to be handled correctly
(pixman-combine64.c can access this buffer via uint64_t * pointer).
Some alignment related problem was reported in:
http://lists.freedesktop.org/archives/pixman/2010-November/000747.html
diff --git a/pixman/pixman-general.c b/pixman/pixman-general.c
index 4d234a0..8130f16 100644
--- a/pixman/pixman-general.c
+++ b/pixman/pixman-general.c
@@ -56,8 +56,8 @@ general_composite_rect (pixman_implementation_t *imp,
int32_t width,
int32_t height)
{
- uint8_t stack_scanline_buffer[SCANLINE_BUFFER_LENGTH * 3];
- uint8_t *scanline_buffer = stack_scanline_buffer;
+ uint64_t stack_scanline_buffer[(SCANLINE_BUFFER_LENGTH * 3 + 7) / 8];
+ uint8_t *scanline_buffer = (uint8_t *) stack_scanline_buffer;
uint8_t *src_buffer, *mask_buffer, *dest_buffer;
fetch_scanline_t fetch_src = NULL, fetch_mask = NULL, fetch_dest = NULL;
pixman_combine_32_func_t compose;
@@ -255,7 +255,7 @@ general_composite_rect (pixman_implementation_t *imp,
}
}
- if (scanline_buffer != stack_scanline_buffer)
+ if (scanline_buffer != (uint8_t *) stack_scanline_buffer)
free (scanline_buffer);
}
commit 985e59a82fa5e644cb6516dc174ab3f79f1448df
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Thu Nov 25 02:28:29 2010 +0200
ARM: added 'neon_src_rpixbuf_8888' fast path
With this optimization added, pixman assisted conversion from
non-premultiplied to premultiplied alpha format is now fully
NEON optimized (both with and without R/B color components
swapping in the process).
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index cf014fa..4dddde1 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -2072,6 +2072,63 @@ generate_composite_function \
/******************************************************************************/
+.macro pixman_composite_src_rpixbuf_8888_process_pixblock_head
+ vmull.u8 q8, d3, d0
+ vmull.u8 q9, d3, d1
+ vmull.u8 q10, d3, d2
+.endm
+
+.macro pixman_composite_src_rpixbuf_8888_process_pixblock_tail
+ vrshr.u16 q11, q8, #8
+ vswp d3, d31
+ vrshr.u16 q12, q9, #8
+ vrshr.u16 q13, q10, #8
+ vraddhn.u16 d28, q11, q8
+ vraddhn.u16 d29, q12, q9
+ vraddhn.u16 d30, q13, q10
+.endm
+
+.macro pixman_composite_src_rpixbuf_8888_process_pixblock_tail_head
+ vrshr.u16 q11, q8, #8
+ vswp d3, d31
+ vrshr.u16 q12, q9, #8
+ vrshr.u16 q13, q10, #8
+ fetch_src_pixblock
+ vraddhn.u16 d28, q11, q8
+ PF add PF_X, PF_X, #8
+ PF tst PF_CTL, #0xF
+ PF addne PF_X, PF_X, #8
+ PF subne PF_CTL, PF_CTL, #1
+ vraddhn.u16 d29, q12, q9
+ vraddhn.u16 d30, q13, q10
+ vmull.u8 q8, d3, d0
+ vmull.u8 q9, d3, d1
+ vmull.u8 q10, d3, d2
+ vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
+ PF cmp PF_X, ORIG_W
+ PF pld, [PF_SRC, PF_X, lsl #src_bpp_shift]
+ PF subge PF_X, PF_X, ORIG_W
+ PF subges PF_CTL, PF_CTL, #0x10
+ PF ldrgeb DUMMY, [PF_SRC, SRC_STRIDE, lsl #src_bpp_shift]!
+.endm
+
+generate_composite_function \
+ pixman_composite_src_rpixbuf_8888_asm_neon, 32, 0, 32, \
+ FLAG_DST_WRITEONLY | FLAG_DEINTERLEAVE_32BPP, \
+ 8, /* number of pixels, processed in a single block */ \
+ 10, /* prefetch distance */ \
+ default_init, \
+ default_cleanup, \
+ pixman_composite_src_rpixbuf_8888_process_pixblock_head, \
+ pixman_composite_src_rpixbuf_8888_process_pixblock_tail, \
+ pixman_composite_src_rpixbuf_8888_process_pixblock_tail_head, \
+ 28, /* dst_w_basereg */ \
+ 0, /* dst_r_basereg */ \
+ 0, /* src_basereg */ \
+ 0 /* mask_basereg */
+
+/******************************************************************************/
+
.macro pixman_composite_over_0565_8_0565_process_pixblock_head
/* mask is in d15 */
convert_0565_to_x888 q4, d2, d1, d0
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index e3eca2b..c28c481 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -52,6 +52,8 @@ PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, src_0888_0565_rev,
uint8_t, 3, uint16_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, src_pixbuf_8888,
uint32_t, 1, uint32_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, src_rpixbuf_8888,
+ uint32_t, 1, uint32_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, add_8_8,
uint8_t, 1, uint8_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, add_8888_8888,
@@ -249,6 +251,9 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
PIXMAN_STD_FAST_PATH (SRC, b8g8r8, null, x8r8g8b8, neon_composite_src_0888_8888_rev),
PIXMAN_STD_FAST_PATH (SRC, b8g8r8, null, r5g6b5, neon_composite_src_0888_0565_rev),
PIXMAN_STD_FAST_PATH (SRC, pixbuf, pixbuf, a8r8g8b8, neon_composite_src_pixbuf_8888),
+ PIXMAN_STD_FAST_PATH (SRC, pixbuf, pixbuf, a8b8g8r8, neon_composite_src_rpixbuf_8888),
+ PIXMAN_STD_FAST_PATH (SRC, rpixbuf, rpixbuf, a8r8g8b8, neon_composite_src_rpixbuf_8888),
+ PIXMAN_STD_FAST_PATH (SRC, rpixbuf, rpixbuf, a8b8g8r8, neon_composite_src_pixbuf_8888),
PIXMAN_STD_FAST_PATH (OVER, solid, a8, a8, neon_composite_over_n_8_8),
PIXMAN_STD_FAST_PATH (OVER, solid, a8, r5g6b5, neon_composite_over_n_8_0565),
PIXMAN_STD_FAST_PATH (OVER, solid, a8, b5g6r5, neon_composite_over_n_8_0565),
More information about the xorg-commit
mailing list