diff options
Diffstat (limited to 'pixman')
| -rw-r--r-- | pixman/pixman/pixman-arm-common.h | 15 | ||||
| -rw-r--r-- | pixman/pixman/pixman-arm-neon-asm.S | 530 | ||||
| -rw-r--r-- | pixman/pixman/pixman-arm-neon-asm.h | 5 | ||||
| -rw-r--r-- | pixman/pixman/pixman-arm-neon.c | 47 | ||||
| -rw-r--r-- | pixman/pixman/pixman-arm-simd.c | 4 | ||||
| -rw-r--r-- | pixman/pixman/pixman-general.c | 6 |
6 files changed, 495 insertions, 112 deletions
diff --git a/pixman/pixman/pixman-arm-common.h b/pixman/pixman/pixman-arm-common.h index 08f34cc20..ede63a629 100644 --- a/pixman/pixman/pixman-arm-common.h +++ b/pixman/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/pixman-arm-neon-asm.S b/pixman/pixman/pixman-arm-neon-asm.S index 0229bedfa..51533d42f 100644 --- a/pixman/pixman/pixman-arm-neon-asm.S +++ b/pixman/pixman/pixman-arm-neon-asm.S @@ -791,61 +791,112 @@ 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
-.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
+.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
+ 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_8888_8_0565_process_pixblock_tail
+ /* 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]!
+.macro pixman_composite_over_8888_8_0565_process_pixblock_tail_head
vld1.16 {d4, d5}, [DST_R, :128]!
- vld1.8 {d24}, [MASK]!
+ 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
+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
@@ -877,32 +928,32 @@ 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
/******************************************************************************/
-/* 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
- vld1.8 {d24}, [MASK]!
- pixman_composite_over_n_8_0565_process_pixblock_head
- vst1.16 {d28, d29}, [DST_W, :128]!
+.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_8_0565_asm_neon, 32, 8, 16, \
+ 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 */ \
- 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_n_0565_init, \
+ pixman_composite_over_8888_n_0565_cleanup, \
+ 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 */ \
@@ -1171,7 +1222,7 @@ 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
@@ -1203,6 +1254,74 @@ 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
+ 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
+.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
@@ -1273,7 +1392,7 @@ 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
@@ -1308,6 +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 */
@@ -1337,7 +1508,7 @@ 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
@@ -1394,7 +1565,7 @@ 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
@@ -1423,34 +1594,50 @@ 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]!
- vld4.8 {d24, d25, d26, d27}, [MASK]!
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 \
@@ -1476,6 +1663,78 @@ 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_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_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} */
@@ -1519,7 +1778,7 @@ 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
@@ -1590,7 +1849,7 @@ 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
@@ -1632,7 +1891,7 @@ 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
@@ -1813,6 +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
@@ -1849,7 +2165,7 @@ 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]!
@@ -1875,6 +2191,34 @@ 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
@@ -1901,7 +2245,7 @@ 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/pixman-arm-neon-asm.h b/pixman/pixman/pixman-arm-neon-asm.h index 1d8a31c1e..6e3d583f5 100644 --- a/pixman/pixman/pixman-arm-neon-asm.h +++ b/pixman/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
diff --git a/pixman/pixman/pixman-arm-neon.c b/pixman/pixman/pixman-arm-neon.c index 858bb071e..e89cc3215 100644 --- a/pixman/pixman/pixman-arm-neon.c +++ b/pixman/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,
@@ -63,29 +65,43 @@ 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_DST (0, neon, in_n_8,
+ uint8_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, add_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 (SKIP_ZERO_SRC, neon, add_n_8_8,
uint8_t, 1, uint8_t, 1)
+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 (SKIP_ZERO_MASK, neon, over_8888_n_0565,
+ uint32_t, 1, uint16_t, 1)
+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 (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,
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,
@@ -235,6 +251,10 @@ 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),
PIXMAN_STD_FAST_PATH (OVER, solid, a8, a8r8g8b8, neon_composite_over_n_8_8888),
@@ -250,6 +270,10 @@ 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, 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),
@@ -268,13 +292,20 @@ 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),
+ 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),
+ 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),
diff --git a/pixman/pixman/pixman-arm-simd.c b/pixman/pixman/pixman-arm-simd.c index 7f789ceab..e03dfe020 100644 --- a/pixman/pixman/pixman-arm-simd.c +++ b/pixman/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,
diff --git a/pixman/pixman/pixman-general.c b/pixman/pixman/pixman-general.c index 105125359..5a639051b 100644 --- a/pixman/pixman/pixman-general.c +++ b/pixman/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);
}
|