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