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 (file)
--- 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 (file)
--- 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),