pixman: Branch 'master' - 12 commits
Siarhei Siamashka
siamashka at kemper.freedesktop.org
Fri Dec 3 07:00:57 PST 2010
pixman/pixman-arm-common.h | 15 -
pixman/pixman-arm-neon-asm.S | 473 ++++++++++++++++++++++++++++++++++---------
pixman/pixman-arm-neon-asm.h | 5
pixman/pixman-arm-neon.c | 42 +++
pixman/pixman-arm-simd.c | 4
5 files changed, 430 insertions(+), 109 deletions(-)
New commits:
commit 733f68912f4a44c24ad3973049a7e1d98f4c6ea8
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Mon Nov 29 09:11:29 2010 +0200
ARM: added 'neon_composite_in_n_8' fast path
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 2c0fd37..cf014fa 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -1427,6 +1427,58 @@ generate_composite_function \
/******************************************************************************/
+.macro pixman_composite_in_n_8_process_pixblock_head
+ /* expecting source data in {d0, d1, d2, d3} */
+ /* and destination data in {d4, d5, d6, d7} */
+ vmull.u8 q8, d4, d3
+ vmull.u8 q9, d5, d3
+ vmull.u8 q10, d6, d3
+ vmull.u8 q11, d7, d3
+.endm
+
+.macro pixman_composite_in_n_8_process_pixblock_tail
+ vrshr.u16 q14, q8, #8
+ vrshr.u16 q15, q9, #8
+ vrshr.u16 q12, q10, #8
+ vrshr.u16 q13, q11, #8
+ vraddhn.u16 d28, q8, q14
+ vraddhn.u16 d29, q9, q15
+ vraddhn.u16 d30, q10, q12
+ vraddhn.u16 d31, q11, q13
+.endm
+
+.macro pixman_composite_in_n_8_process_pixblock_tail_head
+ pixman_composite_in_n_8_process_pixblock_tail
+ vld1.8 {d4, d5, d6, d7}, [DST_R, :128]!
+ cache_preload 32, 32
+ pixman_composite_in_n_8_process_pixblock_head
+ vst1.8 {d28, d29, d30, d31}, [DST_W, :128]!
+.endm
+
+.macro pixman_composite_in_n_8_init
+ add DUMMY, sp, #ARGS_STACK_OFFSET
+ vld1.32 {d3[0]}, [DUMMY]
+ vdup.8 d3, d3[3]
+.endm
+
+.macro pixman_composite_in_n_8_cleanup
+.endm
+
+generate_composite_function \
+ pixman_composite_in_n_8_asm_neon, 0, 0, 8, \
+ FLAG_DST_READWRITE, \
+ 32, /* number of pixels, processed in a single block */ \
+ 5, /* prefetch distance */ \
+ pixman_composite_in_n_8_init, \
+ pixman_composite_in_n_8_cleanup, \
+ pixman_composite_in_n_8_process_pixblock_head, \
+ pixman_composite_in_n_8_process_pixblock_tail, \
+ pixman_composite_in_n_8_process_pixblock_tail_head, \
+ 28, /* dst_w_basereg */ \
+ 4, /* dst_r_basereg */ \
+ 0, /* src_basereg */ \
+ 24 /* mask_basereg */
+
.macro pixman_composite_add_n_8_8_process_pixblock_head
/* expecting source data in {d8, d9, d10, d11} */
/* d8 - blue, d9 - green, d10 - red, d11 - alpha */
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index 1a3741c..e3eca2b 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -69,6 +69,8 @@ PIXMAN_ARM_BIND_FAST_PATH_N_DST (SKIP_ZERO_SRC, neon, over_n_8888,
uint32_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_N_DST (SKIP_ZERO_SRC, neon, over_reverse_n_8888,
uint32_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_N_DST (0, neon, in_n_8,
+ uint8_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, neon, over_n_8_0565,
uint8_t, 1, uint16_t, 1)
@@ -298,6 +300,7 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
PIXMAN_STD_FAST_PATH (ADD, a8, null, a8, neon_composite_add_8_8),
PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, null, a8r8g8b8, neon_composite_add_8888_8888),
PIXMAN_STD_FAST_PATH (ADD, a8b8g8r8, null, a8b8g8r8, neon_composite_add_8888_8888),
+ PIXMAN_STD_FAST_PATH (IN, solid, null, a8, neon_composite_in_n_8),
PIXMAN_STD_FAST_PATH (OVER_REVERSE, solid, null, a8r8g8b8, neon_composite_over_reverse_n_8888),
PIXMAN_STD_FAST_PATH (OVER_REVERSE, solid, null, a8b8g8r8, neon_composite_over_reverse_n_8888),
PIXMAN_STD_FAST_PATH (OUT_REVERSE, a8, null, r5g6b5, neon_composite_out_reverse_8_0565),
commit af7a69d90ea2b43a4e850870727723d719f09a1c
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Mon Nov 29 09:00:46 2010 +0200
ARM: added flags parameter to some asm fast path wrapper macros
Not all types of operations can be skipped when having transparent
solid source or transparent solid mask. Add an extra flags parameter
for providing this information to the wrappers.
diff --git a/pixman/pixman-arm-common.h b/pixman/pixman-arm-common.h
index 2cff6c8..66f448d 100644
--- a/pixman/pixman-arm-common.h
+++ b/pixman/pixman-arm-common.h
@@ -47,6 +47,9 @@
* or mask), the corresponding stride argument is unused.
*/
+#define SKIP_ZERO_SRC 1
+#define SKIP_ZERO_MASK 2
+
#define PIXMAN_ARM_BIND_FAST_PATH_SRC_DST(cputype, name, \
src_type, src_cnt, \
dst_type, dst_cnt) \
@@ -87,7 +90,7 @@ cputype##_composite_##name (pixman_implementation_t *imp, \
src_line, src_stride); \
}
-#define PIXMAN_ARM_BIND_FAST_PATH_N_DST(cputype, name, \
+#define PIXMAN_ARM_BIND_FAST_PATH_N_DST(flags, cputype, name, \
dst_type, dst_cnt) \
void \
pixman_composite_##name##_asm_##cputype (int32_t w, \
@@ -117,7 +120,7 @@ cputype##_composite_##name (pixman_implementation_t *imp, \
\
src = _pixman_image_get_solid (src_image, dst_image->bits.format); \
\
- if (src == 0) \
+ if ((flags & SKIP_ZERO_SRC) && src == 0) \
return; \
\
PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, dst_type, \
@@ -128,7 +131,7 @@ cputype##_composite_##name (pixman_implementation_t *imp, \
src); \
}
-#define PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST(cputype, name, \
+#define PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST(flags, cputype, name, \
mask_type, mask_cnt, \
dst_type, dst_cnt) \
void \
@@ -163,7 +166,7 @@ cputype##_composite_##name (pixman_implementation_t *imp, \
\
src = _pixman_image_get_solid (src_image, dst_image->bits.format); \
\
- if (src == 0) \
+ if ((flags & SKIP_ZERO_SRC) && src == 0) \
return; \
\
PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, dst_type, \
@@ -177,7 +180,7 @@ cputype##_composite_##name (pixman_implementation_t *imp, \
mask_line, mask_stride); \
}
-#define PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST(cputype, name, \
+#define PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST(flags, cputype, name, \
src_type, src_cnt, \
dst_type, dst_cnt) \
void \
@@ -211,7 +214,7 @@ cputype##_composite_##name (pixman_implementation_t *imp, \
\
mask = _pixman_image_get_solid (mask_image, dst_image->bits.format);\
\
- if (mask == 0) \
+ if ((flags & SKIP_ZERO_MASK) && mask == 0) \
return; \
\
PIXMAN_IMAGE_GET_LINE (dst_image, dest_x, dest_y, dst_type, \
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index f0dc111..1a3741c 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -63,33 +63,33 @@ PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, over_8888_8888,
PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (neon, out_reverse_8_0565,
uint8_t, 1, uint16_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_DST (neon, over_n_0565,
+PIXMAN_ARM_BIND_FAST_PATH_N_DST (SKIP_ZERO_SRC, neon, over_n_0565,
uint16_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_DST (neon, over_n_8888,
+PIXMAN_ARM_BIND_FAST_PATH_N_DST (SKIP_ZERO_SRC, neon, over_n_8888,
uint32_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_DST (neon, over_reverse_n_8888,
+PIXMAN_ARM_BIND_FAST_PATH_N_DST (SKIP_ZERO_SRC, neon, over_reverse_n_8888,
uint32_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8_0565,
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, neon, over_n_8_0565,
uint8_t, 1, uint16_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8_8888,
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, neon, over_n_8_8888,
uint8_t, 1, uint32_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8888_8888_ca,
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, neon, over_n_8888_8888_ca,
uint32_t, 1, uint32_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8_8,
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, neon, over_n_8_8,
uint8_t, 1, uint8_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, add_n_8_8,
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, neon, add_n_8_8,
uint8_t, 1, uint8_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, add_n_8_8888,
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, neon, add_n_8_8888,
uint8_t, 1, uint32_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_8888_n_8888,
+PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (SKIP_ZERO_MASK, neon, over_8888_n_8888,
uint32_t, 1, uint32_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_8888_n_0565,
+PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (SKIP_ZERO_MASK, neon, over_8888_n_0565,
uint32_t, 1, uint16_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_0565_n_0565,
+PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (SKIP_ZERO_MASK, neon, over_0565_n_0565,
uint16_t, 1, uint16_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, add_8888_n_8888,
+PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (SKIP_ZERO_MASK, neon, add_8888_n_8888,
uint32_t, 1, uint32_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, add_8_8_8,
diff --git a/pixman/pixman-arm-simd.c b/pixman/pixman-arm-simd.c
index 3b05007..dc2f471 100644
--- a/pixman/pixman-arm-simd.c
+++ b/pixman/pixman-arm-simd.c
@@ -381,10 +381,10 @@ PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (armv6, add_8_8,
PIXMAN_ARM_BIND_FAST_PATH_SRC_DST (armv6, over_8888_8888,
uint32_t, 1, uint32_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (armv6, over_8888_n_8888,
+PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (SKIP_ZERO_MASK, armv6, over_8888_n_8888,
uint32_t, 1, uint32_t, 1)
-PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (armv6, over_n_8_8888,
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (SKIP_ZERO_SRC, armv6, over_n_8_8888,
uint8_t, 1, uint32_t, 1)
PIXMAN_ARM_BIND_SCALED_NEAREST_SRC_DST (armv6, 0565_0565, SRC,
commit f6843e3797eea7e4aed7614b1086f5cefc06c0f9
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Mon Nov 29 03:31:32 2010 +0200
ARM: added 'neon_composite_add_8888_n_8888' fast path
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index dd6f2c5..2c0fd37 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -1657,6 +1657,32 @@ generate_composite_function \
/******************************************************************************/
+.macro pixman_composite_add_8888_n_8888_init
+ add DUMMY, sp, #(ARGS_STACK_OFFSET + 8)
+ vld1.32 {d27[0]}, [DUMMY]
+ vdup.8 d27, d27[3]
+.endm
+
+.macro pixman_composite_add_8888_n_8888_cleanup
+.endm
+
+generate_composite_function \
+ pixman_composite_add_8888_n_8888_asm_neon, 32, 0, 32, \
+ FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
+ 8, /* number of pixels, processed in a single block */ \
+ 5, /* prefetch distance */ \
+ pixman_composite_add_8888_n_8888_init, \
+ pixman_composite_add_8888_n_8888_cleanup, \
+ pixman_composite_add_8888_8888_8888_process_pixblock_head, \
+ pixman_composite_add_8888_8888_8888_process_pixblock_tail, \
+ pixman_composite_add_8888_8888_8888_process_pixblock_tail_head, \
+ 28, /* dst_w_basereg */ \
+ 4, /* dst_r_basereg */ \
+ 0, /* src_basereg */ \
+ 27 /* mask_basereg */
+
+/******************************************************************************/
+
.macro pixman_composite_out_reverse_8888_n_8888_process_pixblock_head
/* expecting source data in {d0, d1, d2, d3} */
/* destination data in {d4, d5, d6, d7} */
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index 5ad58bd..f0dc111 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -89,6 +89,8 @@ PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_8888_n_0565,
uint32_t, 1, uint16_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_0565_n_0565,
uint16_t, 1, uint16_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, add_8888_n_8888,
+ uint32_t, 1, uint32_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, add_8_8_8,
uint8_t, 1, uint8_t, 1, uint8_t, 1)
@@ -291,6 +293,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, a8, a8r8g8b8, neon_composite_add_8888_8_8888),
PIXMAN_STD_FAST_PATH (ADD, a8b8g8r8, a8, a8b8g8r8, neon_composite_add_8888_8_8888),
PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, a8r8g8b8, a8r8g8b8, neon_composite_add_8888_8888_8888),
+ PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, solid, a8r8g8b8, neon_composite_add_8888_n_8888),
+ PIXMAN_STD_FAST_PATH (ADD, a8b8g8r8, solid, a8b8g8r8, neon_composite_add_8888_n_8888),
PIXMAN_STD_FAST_PATH (ADD, a8, null, a8, neon_composite_add_8_8),
PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, null, a8r8g8b8, neon_composite_add_8888_8888),
PIXMAN_STD_FAST_PATH (ADD, a8b8g8r8, null, a8b8g8r8, neon_composite_add_8888_8888),
commit b066b520dfaf0a9f4d1bc9a73c789091e9ce7cc8
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Mon Nov 29 02:38:52 2010 +0200
ARM: added 'neon_composite_add_n_8_8888' fast path
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 829ef84..dd6f2c5 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -1628,6 +1628,35 @@ generate_composite_function \
/******************************************************************************/
+.macro pixman_composite_add_n_8_8888_init
+ add DUMMY, sp, #ARGS_STACK_OFFSET
+ vld1.32 {d3[0]}, [DUMMY]
+ vdup.8 d0, d3[0]
+ vdup.8 d1, d3[1]
+ vdup.8 d2, d3[2]
+ vdup.8 d3, d3[3]
+.endm
+
+.macro pixman_composite_add_n_8_8888_cleanup
+.endm
+
+generate_composite_function \
+ pixman_composite_add_n_8_8888_asm_neon, 0, 8, 32, \
+ FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
+ 8, /* number of pixels, processed in a single block */ \
+ 5, /* prefetch distance */ \
+ pixman_composite_add_n_8_8888_init, \
+ pixman_composite_add_n_8_8888_cleanup, \
+ pixman_composite_add_8888_8888_8888_process_pixblock_head, \
+ pixman_composite_add_8888_8888_8888_process_pixblock_tail, \
+ pixman_composite_add_8888_8888_8888_process_pixblock_tail_head, \
+ 28, /* dst_w_basereg */ \
+ 4, /* dst_r_basereg */ \
+ 0, /* src_basereg */ \
+ 27 /* mask_basereg */
+
+/******************************************************************************/
+
.macro pixman_composite_out_reverse_8888_n_8888_process_pixblock_head
/* expecting source data in {d0, d1, d2, d3} */
/* destination data in {d4, d5, d6, d7} */
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index eaf9787..5ad58bd 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -80,6 +80,8 @@ PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8_8,
uint8_t, 1, uint8_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, add_n_8_8,
uint8_t, 1, uint8_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, add_n_8_8888,
+ uint8_t, 1, uint32_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_8888_n_8888,
uint32_t, 1, uint32_t, 1)
@@ -281,6 +283,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
PIXMAN_STD_FAST_PATH (OVER, x8r8g8b8, null, a8r8g8b8, neon_composite_src_x888_8888),
PIXMAN_STD_FAST_PATH (OVER, x8b8g8r8, null, a8b8g8r8, neon_composite_src_x888_8888),
PIXMAN_STD_FAST_PATH (ADD, solid, a8, a8, neon_composite_add_n_8_8),
+ PIXMAN_STD_FAST_PATH (ADD, solid, a8, a8r8g8b8, neon_composite_add_n_8_8888),
+ PIXMAN_STD_FAST_PATH (ADD, solid, a8, a8b8g8r8, neon_composite_add_n_8_8888),
PIXMAN_STD_FAST_PATH (ADD, a8, a8, a8, neon_composite_add_8_8_8),
PIXMAN_STD_FAST_PATH (ADD, r5g6b5, a8, r5g6b5, neon_composite_add_0565_8_0565),
PIXMAN_STD_FAST_PATH (ADD, b5g6r5, a8, b5g6r5, neon_composite_add_0565_8_0565),
commit 1fba7790367d7b726d05a33bbbcebe10b9280a31
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Mon Nov 29 02:10:22 2010 +0200
ARM: better NEON instructions scheduling for add_8888_8888_8888
Provides a minor performance improvement by using pipelining and hiding
instructions latencies. Also do not clobber d0-d3 registers (source
image pixels) while doing calculations in order to allow the use of
the same macro for add_n_8_8888 fast path later.
Benchmark from ARM Cortex-A8 @500MHz:
== before ==
add_8888_8888_8888 = L1: 95.94 L2: 42.27 M: 25.60 (121.09%)
HT: 14.54 VT: 13.13 R: 12.77 RT: 4.49 (48Kops/s)
add_8888_8_8888 = L1: 104.51 L2: 57.81 M: 36.06 (106.62%)
HT: 19.24 VT: 16.45 R: 14.71 RT: 4.80 (51Kops/s)
== after ==
add_8888_8888_8888 = L1: 106.66 L2: 47.82 M: 27.32 (129.30%)
HT: 15.44 VT: 13.96 R: 12.86 RT: 4.48 (48Kops/s)
add_8888_8_8888 = L1: 107.72 L2: 61.02 M: 38.26 (113.16%)
HT: 19.48 VT: 16.72 R: 14.82 RT: 4.80 (51Kops/s)
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 11ef166..829ef84 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -1542,34 +1542,50 @@ generate_composite_function \
/* expecting source data in {d0, d1, d2, d3} */
/* destination data in {d4, d5, d6, d7} */
/* mask in {d24, d25, d26, d27} */
- vmull.u8 q8, d27, d0
- vmull.u8 q9, d27, d1
+ vmull.u8 q8, d27, d0
+ vmull.u8 q9, d27, d1
vmull.u8 q10, d27, d2
vmull.u8 q11, d27, d3
- vrshr.u16 q0, q8, #8
- vrshr.u16 q1, q9, #8
- vrshr.u16 q12, q10, #8
- vrshr.u16 q13, q11, #8
- vraddhn.u16 d0, q0, q8
- vraddhn.u16 d1, q1, q9
- vraddhn.u16 d2, q12, q10
- vraddhn.u16 d3, q13, q11
- vqadd.u8 q14, q0, q2
- vqadd.u8 q15, q1, q3
+ /* 1 cycle bubble */
+ vrsra.u16 q8, q8, #8
+ vrsra.u16 q9, q9, #8
+ vrsra.u16 q10, q10, #8
+ vrsra.u16 q11, q11, #8
.endm
.macro pixman_composite_add_8888_8888_8888_process_pixblock_tail
+ /* 2 cycle bubble */
+ vrshrn.u16 d28, q8, #8
+ vrshrn.u16 d29, q9, #8
+ vrshrn.u16 d30, q10, #8
+ vrshrn.u16 d31, q11, #8
+ vqadd.u8 q14, q2, q14
+ /* 1 cycle bubble */
+ vqadd.u8 q15, q3, q15
.endm
-/* TODO: expand macros and do better instructions scheduling */
.macro pixman_composite_add_8888_8888_8888_process_pixblock_tail_head
- pixman_composite_add_8888_8888_8888_process_pixblock_tail
- vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
- vld4.8 {d4, d5, d6, d7}, [DST_R, :128]!
- fetch_mask_pixblock
fetch_src_pixblock
+ vrshrn.u16 d28, q8, #8
+ fetch_mask_pixblock
+ vrshrn.u16 d29, q9, #8
+ vmull.u8 q8, d27, d0
+ vrshrn.u16 d30, q10, #8
+ vmull.u8 q9, d27, d1
+ vrshrn.u16 d31, q11, #8
+ vmull.u8 q10, d27, d2
+ vqadd.u8 q14, q2, q14
+ vmull.u8 q11, d27, d3
+ vqadd.u8 q15, q3, q15
+ vrsra.u16 q8, q8, #8
+ vld4.8 {d4, d5, d6, d7}, [DST_R, :128]!
+ vrsra.u16 q9, q9, #8
+ vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
+ vrsra.u16 q10, q10, #8
+
cache_preload 8, 8
- pixman_composite_add_8888_8888_8888_process_pixblock_head
+
+ vrsra.u16 q11, q11, #8
.endm
generate_composite_function \
commit c3f48b6aa2f9354af02ffc8c938ec6753fdcbde3
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Sun Nov 28 22:05:53 2010 +0200
ARM: added 'neon_composite_add_8888_8_8888' fast path
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 81c0a34..11ef166 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -1595,6 +1595,23 @@ generate_composite_function_single_scanline \
/******************************************************************************/
+generate_composite_function \
+ pixman_composite_add_8888_8_8888_asm_neon, 32, 8, 32, \
+ FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
+ 8, /* number of pixels, processed in a single block */ \
+ 5, /* prefetch distance */ \
+ default_init, \
+ default_cleanup, \
+ pixman_composite_add_8888_8888_8888_process_pixblock_head, \
+ pixman_composite_add_8888_8888_8888_process_pixblock_tail, \
+ pixman_composite_add_8888_8888_8888_process_pixblock_tail_head, \
+ 28, /* dst_w_basereg */ \
+ 4, /* dst_r_basereg */ \
+ 0, /* src_basereg */ \
+ 27 /* mask_basereg */
+
+/******************************************************************************/
+
.macro pixman_composite_out_reverse_8888_n_8888_process_pixblock_head
/* expecting source data in {d0, d1, d2, d3} */
/* destination data in {d4, d5, d6, d7} */
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index b01c3e0..eaf9787 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -92,6 +92,8 @@ PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, add_8_8_8,
uint8_t, 1, uint8_t, 1, uint8_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, add_0565_8_0565,
uint16_t, 1, uint8_t, 1, uint16_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, add_8888_8_8888,
+ uint32_t, 1, uint8_t, 1, uint32_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, add_8888_8888_8888,
uint32_t, 1, uint32_t, 1, uint32_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, over_8888_8_8888,
@@ -282,6 +284,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
PIXMAN_STD_FAST_PATH (ADD, a8, a8, a8, neon_composite_add_8_8_8),
PIXMAN_STD_FAST_PATH (ADD, r5g6b5, a8, r5g6b5, neon_composite_add_0565_8_0565),
PIXMAN_STD_FAST_PATH (ADD, b5g6r5, a8, b5g6r5, neon_composite_add_0565_8_0565),
+ PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, a8, a8r8g8b8, neon_composite_add_8888_8_8888),
+ PIXMAN_STD_FAST_PATH (ADD, a8b8g8r8, a8, a8b8g8r8, neon_composite_add_8888_8_8888),
PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, a8r8g8b8, a8r8g8b8, neon_composite_add_8888_8888_8888),
PIXMAN_STD_FAST_PATH (ADD, a8, null, a8, neon_composite_add_8_8),
PIXMAN_STD_FAST_PATH (ADD, a8r8g8b8, null, a8r8g8b8, neon_composite_add_8888_8888),
commit 6d2f7f981b52b41f4321071c325babcf792bd666
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Sat Nov 27 15:53:54 2010 +0200
ARM: added 'neon_composite_over_0565_n_0565' fast path
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 4175144..81c0a34 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -1994,6 +1994,34 @@ generate_composite_function \
/******************************************************************************/
+.macro pixman_composite_over_0565_n_0565_init
+ add DUMMY, sp, #(ARGS_STACK_OFFSET + 8)
+ vpush {d8-d15}
+ vld1.32 {d15[0]}, [DUMMY]
+ vdup.8 d15, d15[3]
+.endm
+
+.macro pixman_composite_over_0565_n_0565_cleanup
+ vpop {d8-d15}
+.endm
+
+generate_composite_function \
+ pixman_composite_over_0565_n_0565_asm_neon, 16, 0, 16, \
+ FLAG_DST_READWRITE, \
+ 8, /* number of pixels, processed in a single block */ \
+ 5, /* prefetch distance */ \
+ pixman_composite_over_0565_n_0565_init, \
+ pixman_composite_over_0565_n_0565_cleanup, \
+ pixman_composite_over_0565_8_0565_process_pixblock_head, \
+ pixman_composite_over_0565_8_0565_process_pixblock_tail, \
+ pixman_composite_over_0565_8_0565_process_pixblock_tail_head, \
+ 28, /* dst_w_basereg */ \
+ 10, /* dst_r_basereg */ \
+ 8, /* src_basereg */ \
+ 15 /* mask_basereg */
+
+/******************************************************************************/
+
.macro pixman_composite_add_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 8156bbb..b01c3e0 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -85,6 +85,8 @@ PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_8888_n_8888,
uint32_t, 1, uint32_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_8888_n_0565,
uint32_t, 1, uint16_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_0565_n_0565,
+ uint16_t, 1, uint16_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, add_8_8_8,
uint8_t, 1, uint8_t, 1, uint8_t, 1)
@@ -257,6 +259,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, solid, x8r8g8b8, neon_composite_over_8888_n_8888),
PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, solid, r5g6b5, neon_composite_over_8888_n_0565),
PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, solid, b5g6r5, neon_composite_over_8888_n_0565),
+ PIXMAN_STD_FAST_PATH (OVER, r5g6b5, solid, r5g6b5, neon_composite_over_0565_n_0565),
+ PIXMAN_STD_FAST_PATH (OVER, b5g6r5, solid, b5g6r5, neon_composite_over_0565_n_0565),
PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, a8, a8r8g8b8, neon_composite_over_8888_8_8888),
PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, a8, x8r8g8b8, neon_composite_over_8888_8_8888),
PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, a8, a8b8g8r8, neon_composite_over_8888_8_8888),
commit 3990931bf6197eff1cec06cf24bce53ddf9a539a
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Sat Nov 27 04:47:39 2010 +0200
ARM: reuse common NEON code for over_{n_8|8888_n|8888_8}_0565
Renamed suppementary macros from 'over_n_8_0565' to 'over_8888_8_0565',
because they can actually support all variants of this operation:
over_8888_8_0565/over_n_8_0565/over_8888_n_0565.
Also 'over_8888_8_0565' now uses more optimized common code instead of its
own variant, improving performance a bit. Even though this operation is
still memory bandwidth limited, scaled variants of these fast paths may
put more stress on CPU later.
Benchmarked on ARM Cortex-A8 @500MHz:
== before ==
over_8888_8_0565 = L1: 67.10 L2: 53.82 M: 44.70 (105.17%)
HT: 18.73 VT: 16.91 R: 14.25 RT: 4.80 (52Kops/s)
== after ==
over_8888_8_0565 = L1: 77.83 L2: 58.14 M: 44.82 (105.52%)
HT: 20.58 VT: 17.44 R: 15.05 RT: 4.88 (52Kops/s)
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 3e52a49..4175144 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -791,7 +791,7 @@ generate_composite_function \
/******************************************************************************/
-.macro pixman_composite_over_n_8_0565_process_pixblock_head
+.macro pixman_composite_over_8888_8_0565_process_pixblock_head
vmull.u8 q0, d24, d8 /* IN for SRC pixels (part1) */
vmull.u8 q1, d24, d9
vmull.u8 q6, d24, d10
@@ -816,7 +816,7 @@ generate_composite_function \
vmull.u8 q10, d3, d30
.endm
-.macro pixman_composite_over_n_8_0565_process_pixblock_tail
+.macro pixman_composite_over_8888_8_0565_process_pixblock_tail
/* 3 cycle bubble (after vmull.u8) */
vrshr.u16 q13, q8, #8
vrshr.u16 q11, q9, #8
@@ -835,7 +835,7 @@ generate_composite_function \
vsri.u16 q14, q9, #11
.endm
-.macro pixman_composite_over_n_8_0565_process_pixblock_tail_head
+.macro pixman_composite_over_8888_8_0565_process_pixblock_tail_head
vld1.16 {d4, d5}, [DST_R, :128]!
vshrn.u16 d6, q2, #8
fetch_mask_pixblock
@@ -880,6 +880,23 @@ generate_composite_function \
vmull.u8 q10, d3, d30
.endm
+generate_composite_function \
+ pixman_composite_over_8888_8_0565_asm_neon, 32, 8, 16, \
+ FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
+ 8, /* number of pixels, processed in a single block */ \
+ 5, /* prefetch distance */ \
+ default_init_need_all_regs, \
+ default_cleanup_need_all_regs, \
+ pixman_composite_over_8888_8_0565_process_pixblock_head, \
+ pixman_composite_over_8888_8_0565_process_pixblock_tail, \
+ pixman_composite_over_8888_8_0565_process_pixblock_tail_head, \
+ 28, /* dst_w_basereg */ \
+ 4, /* dst_r_basereg */ \
+ 8, /* src_basereg */ \
+ 24 /* mask_basereg */
+
+/******************************************************************************/
+
/*
* This function needs a special initialization of solid mask.
* Solid source pixel data is fetched from stack at ARGS_STACK_OFFSET
@@ -911,9 +928,9 @@ generate_composite_function \
5, /* prefetch distance */ \
pixman_composite_over_n_8_0565_init, \
pixman_composite_over_n_8_0565_cleanup, \
- pixman_composite_over_n_8_0565_process_pixblock_head, \
- pixman_composite_over_n_8_0565_process_pixblock_tail, \
- pixman_composite_over_n_8_0565_process_pixblock_tail_head
+ pixman_composite_over_8888_8_0565_process_pixblock_head, \
+ pixman_composite_over_8888_8_0565_process_pixblock_tail, \
+ pixman_composite_over_8888_8_0565_process_pixblock_tail_head
/******************************************************************************/
@@ -935,36 +952,8 @@ generate_composite_function \
5, /* prefetch distance */ \
pixman_composite_over_8888_n_0565_init, \
pixman_composite_over_8888_n_0565_cleanup, \
- pixman_composite_over_n_8_0565_process_pixblock_head, \
- pixman_composite_over_n_8_0565_process_pixblock_tail, \
- pixman_composite_over_n_8_0565_process_pixblock_tail_head, \
- 28, /* dst_w_basereg */ \
- 4, /* dst_r_basereg */ \
- 8, /* src_basereg */ \
- 24 /* mask_basereg */
-
-/******************************************************************************/
-
-/* TODO: expand macros and do better instructions scheduling */
-.macro pixman_composite_over_8888_8_0565_process_pixblock_tail_head
- vld1.16 {d4, d5}, [DST_R, :128]!
- pixman_composite_over_n_8_0565_process_pixblock_tail
- fetch_src_pixblock
- cache_preload 8, 8
- fetch_mask_pixblock
- pixman_composite_over_n_8_0565_process_pixblock_head
- vst1.16 {d28, d29}, [DST_W, :128]!
-.endm
-
-generate_composite_function \
- pixman_composite_over_8888_8_0565_asm_neon, 32, 8, 16, \
- FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
- 8, /* number of pixels, processed in a single block */ \
- 5, /* prefetch distance */ \
- default_init_need_all_regs, \
- default_cleanup_need_all_regs, \
- pixman_composite_over_n_8_0565_process_pixblock_head, \
- pixman_composite_over_n_8_0565_process_pixblock_tail, \
+ pixman_composite_over_8888_8_0565_process_pixblock_head, \
+ pixman_composite_over_8888_8_0565_process_pixblock_tail, \
pixman_composite_over_8888_8_0565_process_pixblock_tail_head, \
28, /* dst_w_basereg */ \
4, /* dst_r_basereg */ \
commit a7c36681c0c1955ff9110b81f1789e56abb10a95
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Sat Nov 27 03:53:12 2010 +0200
ARM: added 'neon_composite_over_8888_n_0565' fast path
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index ffffc1c..3e52a49 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -917,6 +917,34 @@ generate_composite_function \
/******************************************************************************/
+.macro pixman_composite_over_8888_n_0565_init
+ add DUMMY, sp, #(ARGS_STACK_OFFSET + 8)
+ vpush {d8-d15}
+ vld1.32 {d24[0]}, [DUMMY]
+ vdup.8 d24, d24[3]
+.endm
+
+.macro pixman_composite_over_8888_n_0565_cleanup
+ vpop {d8-d15}
+.endm
+
+generate_composite_function \
+ pixman_composite_over_8888_n_0565_asm_neon, 32, 0, 16, \
+ FLAG_DST_READWRITE | FLAG_DEINTERLEAVE_32BPP, \
+ 8, /* number of pixels, processed in a single block */ \
+ 5, /* prefetch distance */ \
+ pixman_composite_over_8888_n_0565_init, \
+ pixman_composite_over_8888_n_0565_cleanup, \
+ pixman_composite_over_n_8_0565_process_pixblock_head, \
+ pixman_composite_over_n_8_0565_process_pixblock_tail, \
+ pixman_composite_over_n_8_0565_process_pixblock_tail_head, \
+ 28, /* dst_w_basereg */ \
+ 4, /* dst_r_basereg */ \
+ 8, /* src_basereg */ \
+ 24 /* mask_basereg */
+
+/******************************************************************************/
+
/* TODO: expand macros and do better instructions scheduling */
.macro pixman_composite_over_8888_8_0565_process_pixblock_tail_head
vld1.16 {d4, d5}, [DST_R, :128]!
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index 72ef75e..8156bbb 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -83,6 +83,8 @@ PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, add_n_8_8,
PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_8888_n_8888,
uint32_t, 1, uint32_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_8888_n_0565,
+ uint32_t, 1, uint16_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, add_8_8_8,
uint8_t, 1, uint8_t, 1, uint8_t, 1)
@@ -253,6 +255,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
PIXMAN_STD_FAST_PATH_CA (OVER, solid, a8b8g8r8, x8b8g8r8, neon_composite_over_n_8888_8888_ca),
PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, solid, a8r8g8b8, neon_composite_over_8888_n_8888),
PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, solid, x8r8g8b8, neon_composite_over_8888_n_8888),
+ PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, solid, r5g6b5, neon_composite_over_8888_n_0565),
+ PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, solid, b5g6r5, neon_composite_over_8888_n_0565),
PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, a8, a8r8g8b8, neon_composite_over_8888_8_8888),
PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, a8, x8r8g8b8, neon_composite_over_8888_8_8888),
PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, a8, a8b8g8r8, neon_composite_over_8888_8_8888),
commit e6814837a6ccd3e4db329e0131eaf2055d2c864b
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Fri Nov 26 17:06:58 2010 +0200
ARM: better NEON instructions scheduling for over_n_8_0565
Code rearranged to get better instructions scheduling for ARM Cortex-A8/A9.
Now it is ~30% faster for the pixel data in L1 cache and makes better use
of memory bandwidth when running at lower clock frequencies (ex. 500MHz).
Also register d24 (pixels from the mask image) is now not clobbered by
supplementary macros, which allows to reuse them for the other variants
of compositing operations later.
Benchmark from ARM Cortex-A8 @500MHz:
== before ==
over_n_8_0565 = L1: 63.90 L2: 63.15 M: 60.97 ( 73.53%)
HT: 28.89 VT: 24.14 R: 21.33 RT: 6.78 ( 67Kops/s)
== after ==
over_n_8_0565 = L1: 82.64 L2: 75.19 M: 71.52 ( 84.14%)
HT: 30.49 VT: 25.56 R: 22.36 RT: 6.89 ( 68Kops/s)
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 155a236..ffffc1c 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -792,58 +792,92 @@ generate_composite_function \
/******************************************************************************/
.macro pixman_composite_over_n_8_0565_process_pixblock_head
- /* in */
- vmull.u8 q0, d24, d8
- vmull.u8 q1, d24, d9
- vmull.u8 q6, d24, d10
- vmull.u8 q7, d24, d11
- vrshr.u16 q10, q0, #8
- vrshr.u16 q11, q1, #8
- vrshr.u16 q12, q6, #8
- vrshr.u16 q13, q7, #8
- vraddhn.u16 d0, q0, q10
- vraddhn.u16 d1, q1, q11
- vraddhn.u16 d2, q6, q12
- vraddhn.u16 d3, q7, q13
-
- vshrn.u16 d6, q2, #8
- vshrn.u16 d7, q2, #3
- vsli.u16 q2, q2, #5
- vsri.u8 d6, d6, #5
- vmvn.8 d3, d3
- vsri.u8 d7, d7, #6
- vshrn.u16 d30, q2, #2
- /* now do alpha blending */
- vmull.u8 q10, d3, d6
- vmull.u8 q11, d3, d7
- vmull.u8 q12, d3, d30
- vrshr.u16 q13, q10, #8
- vrshr.u16 q3, q11, #8
- vrshr.u16 q15, q12, #8
- vraddhn.u16 d20, q10, q13
- vraddhn.u16 d23, q11, q3
- vraddhn.u16 d22, q12, q15
+ vmull.u8 q0, d24, d8 /* IN for SRC pixels (part1) */
+ vmull.u8 q1, d24, d9
+ vmull.u8 q6, d24, d10
+ vmull.u8 q7, d24, d11
+ vshrn.u16 d6, q2, #8 /* convert DST_R data to 32-bpp (part1) */
+ vshrn.u16 d7, q2, #3
+ vsli.u16 q2, q2, #5
+ vrshr.u16 q8, q0, #8 /* IN for SRC pixels (part2) */
+ vrshr.u16 q9, q1, #8
+ vrshr.u16 q10, q6, #8
+ vrshr.u16 q11, q7, #8
+ vraddhn.u16 d0, q0, q8
+ vraddhn.u16 d1, q1, q9
+ vraddhn.u16 d2, q6, q10
+ vraddhn.u16 d3, q7, q11
+ vsri.u8 d6, d6, #5 /* convert DST_R data to 32-bpp (part2) */
+ vsri.u8 d7, d7, #6
+ vmvn.8 d3, d3
+ vshrn.u16 d30, q2, #2
+ vmull.u8 q8, d3, d6 /* now do alpha blending */
+ vmull.u8 q9, d3, d7
+ vmull.u8 q10, d3, d30
.endm
.macro pixman_composite_over_n_8_0565_process_pixblock_tail
- vqadd.u8 d16, d2, d20
- vqadd.u8 q9, q0, q11
- /* convert to r5g6b5 */
- vshll.u8 q14, d16, #8
- vshll.u8 q8, d19, #8
- vshll.u8 q9, d18, #8
- vsri.u16 q14, q8, #5
- vsri.u16 q14, q9, #11
+ /* 3 cycle bubble (after vmull.u8) */
+ vrshr.u16 q13, q8, #8
+ vrshr.u16 q11, q9, #8
+ vrshr.u16 q15, q10, #8
+ vraddhn.u16 d16, q8, q13
+ vraddhn.u16 d27, q9, q11
+ vraddhn.u16 d26, q10, q15
+ vqadd.u8 d16, d2, d16
+ /* 1 cycle bubble */
+ vqadd.u8 q9, q0, q13
+ vshll.u8 q14, d16, #8 /* convert to 16bpp */
+ vshll.u8 q8, d19, #8
+ vshll.u8 q9, d18, #8
+ vsri.u16 q14, q8, #5
+ /* 1 cycle bubble */
+ vsri.u16 q14, q9, #11
.endm
-/* TODO: expand macros and do better instructions scheduling */
.macro pixman_composite_over_n_8_0565_process_pixblock_tail_head
- pixman_composite_over_n_8_0565_process_pixblock_tail
- vst1.16 {d28, d29}, [DST_W, :128]!
vld1.16 {d4, d5}, [DST_R, :128]!
+ vshrn.u16 d6, q2, #8
fetch_mask_pixblock
+ vshrn.u16 d7, q2, #3
+ fetch_src_pixblock
+ vmull.u8 q6, d24, d10
+ vrshr.u16 q13, q8, #8
+ vrshr.u16 q11, q9, #8
+ vrshr.u16 q15, q10, #8
+ vraddhn.u16 d16, q8, q13
+ vraddhn.u16 d27, q9, q11
+ vraddhn.u16 d26, q10, q15
+ vqadd.u8 d16, d2, d16
+ vmull.u8 q1, d24, d9
+ vqadd.u8 q9, q0, q13
+ vshll.u8 q14, d16, #8
+ vmull.u8 q0, d24, d8
+ vshll.u8 q8, d19, #8
+ vshll.u8 q9, d18, #8
+ vsri.u16 q14, q8, #5
+ vmull.u8 q7, d24, d11
+ vsri.u16 q14, q9, #11
+
cache_preload 8, 8
- pixman_composite_over_n_8_0565_process_pixblock_head
+
+ vsli.u16 q2, q2, #5
+ vrshr.u16 q8, q0, #8
+ vrshr.u16 q9, q1, #8
+ vrshr.u16 q10, q6, #8
+ vrshr.u16 q11, q7, #8
+ vraddhn.u16 d0, q0, q8
+ vraddhn.u16 d1, q1, q9
+ vraddhn.u16 d2, q6, q10
+ vraddhn.u16 d3, q7, q11
+ vsri.u8 d6, d6, #5
+ vsri.u8 d7, d7, #6
+ vmvn.8 d3, d3
+ vshrn.u16 d30, q2, #2
+ vst1.16 {d28, d29}, [DST_W, :128]!
+ vmull.u8 q8, d3, d6
+ vmull.u8 q9, d3, d7
+ vmull.u8 q10, d3, d30
.endm
/*
commit 3be86a92ccab240859062a541cdb871d81c9501a
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Sun Nov 28 21:45:06 2010 +0200
ARM: introduced 'fetch_mask_pixblock' macro to simplify code
This macro hides the implementation details of pixels fetching
for the mask image just like 'fetch_src_pixblock' does for the
source image. This provides more possibilities for reusing the
same code blocks in different compositing functions.
This patch does not introduce any functional changes and the
resulting code in the compiled object file is exactly the same.
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index a3875ee..155a236 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -841,7 +841,7 @@ generate_composite_function \
pixman_composite_over_n_8_0565_process_pixblock_tail
vst1.16 {d28, d29}, [DST_W, :128]!
vld1.16 {d4, d5}, [DST_R, :128]!
- vld1.8 {d24}, [MASK]!
+ fetch_mask_pixblock
cache_preload 8, 8
pixman_composite_over_n_8_0565_process_pixblock_head
.endm
@@ -889,7 +889,7 @@ generate_composite_function \
pixman_composite_over_n_8_0565_process_pixblock_tail
fetch_src_pixblock
cache_preload 8, 8
- vld1.8 {d24}, [MASK]!
+ fetch_mask_pixblock
pixman_composite_over_n_8_0565_process_pixblock_head
vst1.16 {d28, d29}, [DST_W, :128]!
.endm
@@ -1171,7 +1171,7 @@ generate_composite_function \
pixman_composite_over_n_8_8888_process_pixblock_tail
vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
vld4.8 {d4, d5, d6, d7}, [DST_R, :128]!
- vld1.8 {d24}, [MASK]!
+ fetch_mask_pixblock
cache_preload 8, 8
pixman_composite_over_n_8_8888_process_pixblock_head
.endm
@@ -1241,7 +1241,7 @@ generate_composite_function \
.macro pixman_composite_over_n_8_8_process_pixblock_tail_head
vld1.8 {d4, d5, d6, d7}, [DST_R, :128]!
pixman_composite_over_n_8_8_process_pixblock_tail
- vld1.8 {d24, d25, d26, d27}, [MASK]!
+ fetch_mask_pixblock
cache_preload 32, 32
vst1.8 {d28, d29, d30, d31}, [DST_W, :128]!
pixman_composite_over_n_8_8_process_pixblock_head
@@ -1341,7 +1341,7 @@ generate_composite_function \
vraddhn.u16 d29, q15, q9
vraddhn.u16 d30, q6, q10
vraddhn.u16 d31, q7, q11
- vld4.8 {d24, d25, d26, d27}, [MASK]!
+ fetch_mask_pixblock
vqadd.u8 q14, q0, q14
vqadd.u8 q15, q1, q15
cache_preload 8, 8
@@ -1405,7 +1405,7 @@ generate_composite_function \
pixman_composite_add_n_8_8_process_pixblock_tail
vst1.8 {d28, d29, d30, d31}, [DST_W, :128]!
vld1.8 {d4, d5, d6, d7}, [DST_R, :128]!
- vld1.8 {d24, d25, d26, d27}, [MASK]!
+ fetch_mask_pixblock
cache_preload 32, 32
pixman_composite_add_n_8_8_process_pixblock_head
.endm
@@ -1462,7 +1462,7 @@ generate_composite_function \
pixman_composite_add_8_8_8_process_pixblock_tail
vst1.8 {d28, d29, d30, d31}, [DST_W, :128]!
vld1.8 {d4, d5, d6, d7}, [DST_R, :128]!
- vld1.8 {d24, d25, d26, d27}, [MASK]!
+ fetch_mask_pixblock
fetch_src_pixblock
cache_preload 32, 32
pixman_composite_add_8_8_8_process_pixblock_head
@@ -1515,7 +1515,7 @@ generate_composite_function \
pixman_composite_add_8888_8888_8888_process_pixblock_tail
vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
vld4.8 {d4, d5, d6, d7}, [DST_R, :128]!
- vld4.8 {d24, d25, d26, d27}, [MASK]!
+ fetch_mask_pixblock
fetch_src_pixblock
cache_preload 8, 8
pixman_composite_add_8888_8888_8888_process_pixblock_head
@@ -1587,7 +1587,7 @@ generate_composite_function_single_scanline \
pixman_composite_out_reverse_8888_n_8888_process_pixblock_tail
fetch_src_pixblock
cache_preload 8, 8
- vld4.8 {d12, d13, d14, d15}, [MASK]!
+ fetch_mask_pixblock
pixman_composite_out_reverse_8888_n_8888_process_pixblock_head
vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
.endm
@@ -1658,7 +1658,7 @@ generate_composite_function \
pixman_composite_over_8888_n_8888_process_pixblock_tail
fetch_src_pixblock
cache_preload 8, 8
- vld4.8 {d12, d13, d14, d15}, [MASK]!
+ fetch_mask_pixblock
pixman_composite_over_8888_n_8888_process_pixblock_head
vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
.endm
@@ -1700,7 +1700,7 @@ generate_composite_function_single_scanline \
pixman_composite_over_8888_n_8888_process_pixblock_tail
fetch_src_pixblock
cache_preload 8, 8
- vld1.8 {d15}, [MASK]!
+ fetch_mask_pixblock
pixman_composite_over_8888_n_8888_process_pixblock_head
vst4.8 {d28, d29, d30, d31}, [DST_W, :128]!
.endm
@@ -1917,7 +1917,7 @@ generate_composite_function \
/* TODO: expand macros and do better instructions scheduling */
.macro pixman_composite_over_0565_8_0565_process_pixblock_tail_head
- vld1.8 {d15}, [MASK]!
+ fetch_mask_pixblock
pixman_composite_over_0565_8_0565_process_pixblock_tail
fetch_src_pixblock
vld1.16 {d10, d11}, [DST_R, :128]!
@@ -1969,7 +1969,7 @@ generate_composite_function \
/* TODO: expand macros and do better instructions scheduling */
.macro pixman_composite_add_0565_8_0565_process_pixblock_tail_head
- vld1.8 {d15}, [MASK]!
+ fetch_mask_pixblock
pixman_composite_add_0565_8_0565_process_pixblock_tail
fetch_src_pixblock
vld1.16 {d10, d11}, [DST_R, :128]!
diff --git a/pixman/pixman-arm-neon-asm.h b/pixman/pixman-arm-neon-asm.h
index c75bdc3..24fa361 100644
--- a/pixman/pixman-arm-neon-asm.h
+++ b/pixman/pixman-arm-neon-asm.h
@@ -431,6 +431,11 @@
.endif
.endm
+.macro fetch_mask_pixblock
+ pixld pixblock_size, mask_bpp, \
+ (mask_basereg - pixblock_size * mask_bpp / 64), MASK
+.endm
+
/*
* Macro which is used to process leading pixels until destination
* pointer is properly aligned (at 16 bytes boundary). When destination
commit 98d08b37f17a3379d0ceff8bb7de8f943873fbd8
Author: Siarhei Siamashka <siarhei.siamashka at nokia.com>
Date: Fri Nov 26 08:55:49 2010 +0200
ARM: added 'neon_composite_over_n_8_8' fast path
diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 91ec27d..a3875ee 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -1203,6 +1203,74 @@ generate_composite_function \
/******************************************************************************/
+.macro pixman_composite_over_n_8_8_process_pixblock_head
+ vmull.u8 q0, d24, d8
+ vmull.u8 q1, d25, d8
+ vmull.u8 q6, d26, d8
+ vmull.u8 q7, d27, d8
+ vrshr.u16 q10, q0, #8
+ vrshr.u16 q11, q1, #8
+ vrshr.u16 q12, q6, #8
+ vrshr.u16 q13, q7, #8
+ vraddhn.u16 d0, q0, q10
+ vraddhn.u16 d1, q1, q11
+ vraddhn.u16 d2, q6, q12
+ vraddhn.u16 d3, q7, q13
+ vmvn.8 q12, q0
+ vmvn.8 q13, q1
+ vmull.u8 q8, d24, d4
+ vmull.u8 q9, d25, d5
+ vmull.u8 q10, d26, d6
+ vmull.u8 q11, d27, d7
+.endm
+
+.macro pixman_composite_over_n_8_8_process_pixblock_tail
+ vrshr.u16 q14, q8, #8
+ vrshr.u16 q15, q9, #8
+ vrshr.u16 q12, q10, #8
+ vrshr.u16 q13, q11, #8
+ vraddhn.u16 d28, q14, q8
+ vraddhn.u16 d29, q15, q9
+ vraddhn.u16 d30, q12, q10
+ vraddhn.u16 d31, q13, q11
+ vqadd.u8 q14, q0, q14
+ vqadd.u8 q15, q1, q15
+.endm
+
+/* TODO: expand macros and do better instructions scheduling */
+.macro pixman_composite_over_n_8_8_process_pixblock_tail_head
+ vld1.8 {d4, d5, d6, d7}, [DST_R, :128]!
+ pixman_composite_over_n_8_8_process_pixblock_tail
+ vld1.8 {d24, d25, d26, d27}, [MASK]!
+ cache_preload 32, 32
+ vst1.8 {d28, d29, d30, d31}, [DST_W, :128]!
+ pixman_composite_over_n_8_8_process_pixblock_head
+.endm
+
+.macro pixman_composite_over_n_8_8_init
+ add DUMMY, sp, #ARGS_STACK_OFFSET
+ vpush {d8-d15}
+ vld1.32 {d8[0]}, [DUMMY]
+ vdup.8 d8, d8[3]
+.endm
+
+.macro pixman_composite_over_n_8_8_cleanup
+ vpop {d8-d15}
+.endm
+
+generate_composite_function \
+ pixman_composite_over_n_8_8_asm_neon, 0, 8, 8, \
+ FLAG_DST_READWRITE, \
+ 32, /* number of pixels, processed in a single block */ \
+ 5, /* prefetch distance */ \
+ pixman_composite_over_n_8_8_init, \
+ pixman_composite_over_n_8_8_cleanup, \
+ pixman_composite_over_n_8_8_process_pixblock_head, \
+ pixman_composite_over_n_8_8_process_pixblock_tail, \
+ pixman_composite_over_n_8_8_process_pixblock_tail_head
+
+/******************************************************************************/
+
.macro pixman_composite_over_n_8888_8888_ca_process_pixblock_head
/*
* 'combine_mask_ca' replacement
diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c
index 2f82069..72ef75e 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -76,6 +76,8 @@ PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8_8888,
uint8_t, 1, uint32_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8888_8888_ca,
uint32_t, 1, uint32_t, 1)
+PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, over_n_8_8,
+ uint8_t, 1, uint8_t, 1)
PIXMAN_ARM_BIND_FAST_PATH_N_MASK_DST (neon, add_n_8_8,
uint8_t, 1, uint8_t, 1)
@@ -235,6 +237,7 @@ 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 (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),
PIXMAN_STD_FAST_PATH (OVER, solid, a8, a8r8g8b8, neon_composite_over_n_8_8888),
More information about the xorg-commit
mailing list