[Pixman] [PATCH 2/3] ARM: added 'neon_composite_src_x888_8888' fast path

[Siarhei Siamashka]

From: Siarhei Siamashka <siarhei.siamashka at nokia.com>

This fast path function improves performance of 'gnome-system-monitor'
cairo-perf trace.

Benchmark from ARM Cortex-A8 @720MHz

before:

[ # ]  backend                         test   min(s) median(s) stddev. count
[  0]    image         gnome-system-monitor   68.838   68.899   0.05%    5/6

after:

[ # ]  backend                         test   min(s) median(s) stddev. count
[  0]    image         gnome-system-monitor   53.336   53.384   0.09%    6/6
---
 pixman/pixman-arm-neon-asm.S |   38 ++++++++++++++++++++++++++++++++++++++
 pixman/pixman-arm-neon.c     |    4 ++++
 2 files changed, 42 insertions(+), 0 deletions(-)

diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S
index 2db4da8..42ac1cb 100644
--- a/pixman/pixman-arm-neon-asm.S
+++ b/pixman/pixman-arm-neon-asm.S
@@ -947,6 +947,44 @@ generate_composite_function \
 
 /******************************************************************************/
 
+.macro pixman_composite_src_x888_8888_process_pixblock_head
+    vorr     q0, q0, q2
+    vorr     q1, q1, q2
+.endm
+
+.macro pixman_composite_src_x888_8888_process_pixblock_tail
+.endm
+
+.macro pixman_composite_src_x888_8888_process_pixblock_tail_head
+    vst1.32 {d0, d1, d2, d3}, [DST_W, :128]!
+    vld1.32 {d0, d1, d2, d3}, [SRC]!
+    vorr     q0, q0, q2
+    vorr     q1, q1, q2
+    cache_preload 8, 8
+.endm
+
+.macro pixman_composite_src_x888_8888_init
+    vmov.u8  q2, #0xFF
+    vshl.u32 q2, q2, #24
+.endm
+
+generate_composite_function \
+    pixman_composite_src_x888_8888_asm_neon, 32, 0, 32, \
+    FLAG_DST_WRITEONLY, \
+    8, /* number of pixels, processed in a single block */ \
+    10, /* prefetch distance */ \
+    pixman_composite_src_x888_8888_init, \
+    default_cleanup, \
+    pixman_composite_src_x888_8888_process_pixblock_head, \
+    pixman_composite_src_x888_8888_process_pixblock_tail, \
+    pixman_composite_src_x888_8888_process_pixblock_tail_head, \
+    0, /* dst_w_basereg */ \
+    0, /* dst_r_basereg */ \
+    0, /* src_basereg   */ \
+    0  /* mask_basereg  */
+
+/******************************************************************************/
+
 .macro pixman_composite_over_n_8_8888_process_pixblock_head
     /* expecting deinterleaved 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 00b5c35..12d92a2 100644
--- a/pixman/pixman-arm-neon.c
+++ b/pixman/pixman-arm-neon.c
@@ -251,6 +251,7 @@ neon_composite_##name (pixman_implementation_t *imp,                    \
 
 
 BIND_SRC_NULL_DST(src_8888_8888, uint32_t, 1, uint32_t, 1)
+BIND_SRC_NULL_DST(src_x888_8888, uint32_t, 1, uint32_t, 1)
 BIND_SRC_NULL_DST(src_0565_0565, uint16_t, 1, uint16_t, 1)
 BIND_SRC_NULL_DST(src_0888_0888, uint8_t, 3, uint8_t, 3)
 BIND_SRC_NULL_DST(src_8888_0565, uint32_t, 1, uint16_t, 1)
@@ -400,6 +401,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
     PIXMAN_STD_FAST_PATH (SRC,  x8r8g8b8, null,     x8r8g8b8, neon_composite_src_8888_8888),
     PIXMAN_STD_FAST_PATH (SRC,  a8b8g8r8, null,     x8b8g8r8, neon_composite_src_8888_8888),
     PIXMAN_STD_FAST_PATH (SRC,  x8b8g8r8, null,     x8b8g8r8, neon_composite_src_8888_8888),
+    PIXMAN_STD_FAST_PATH (SRC,  a8r8g8b8, null,     a8r8g8b8, neon_composite_src_8888_8888),
+    PIXMAN_STD_FAST_PATH (SRC,  x8r8g8b8, null,     a8r8g8b8, neon_composite_src_x888_8888),
     PIXMAN_STD_FAST_PATH (SRC,  r8g8b8,   null,     r8g8b8,   neon_composite_src_0888_0888),
     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),
@@ -428,6 +431,7 @@ static const pixman_fast_path_t arm_neon_fast_paths[] =
     PIXMAN_STD_FAST_PATH (OVER, a8r8g8b8, null,     x8r8g8b8, neon_composite_over_8888_8888),
     PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null,     a8b8g8r8, neon_composite_over_8888_8888),
     PIXMAN_STD_FAST_PATH (OVER, a8b8g8r8, null,     x8b8g8r8, neon_composite_over_8888_8888),
+    PIXMAN_STD_FAST_PATH (OVER, x8r8g8b8, null,     a8r8g8b8, neon_composite_src_x888_8888),
     PIXMAN_STD_FAST_PATH (ADD,  solid,    a8,       a8,       neon_composite_add_n_8_8),
     PIXMAN_STD_FAST_PATH (ADD,  a8,       a8,       a8,       neon_composite_add_8_8_8),
     PIXMAN_STD_FAST_PATH (ADD,  a8r8g8b8, a8r8g8b8, a8r8g8b8, neon_composite_add_8888_8888_8888),
-- 
1.6.4.4