From d8820360f79d07e03c91ecd201880bc6b1706f19 Mon Sep 17 00:00:00 2001 From: Siarhei Siamashka Date: Thu, 23 Sep 2010 22:28:55 +0300 Subject: [PATCH] ARM: added 'neon_composite_add_0565_8_0565' fast path == before == add_0565_8_0565 = L1: 14.05 L2: 14.03 M: 11.57 ( 12.94%) HT: 8.31 VT: 8.10 R: 7.47 RT: 3.64 ( 42Kops/s) == after == add_0565_8_0565 = L1: 123.36 L2: 94.70 M: 74.36 ( 83.15%) HT: 31.17 VT: 23.97 R: 21.06 RT: 6.42 ( 70Kops/s) --- pixman/pixman-arm-neon-asm.S | 52 ++++++++++++++++++++++++++++++++++++ pixman/pixman-arm-neon.c | 4 +++ 2 files changed, 56 insertions(+) diff --git a/pixman/pixman-arm-neon-asm.S b/pixman/pixman-arm-neon-asm.S index a2a5354..accd051 100644 --- a/pixman/pixman-arm-neon-asm.S +++ b/pixman/pixman-arm-neon-asm.S @@ -1872,3 +1872,55 @@ generate_composite_function \ 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 + convert_0565_to_x888 q5, d6, d5, d4 + /* source pixel data is in {d0, d1, d2, XX} */ + /* destination pixel data is in {d4, d5, d6, XX} */ + vmull.u8 q6, d15, d2 + vmull.u8 q5, d15, d1 + vmull.u8 q4, d15, d0 + vrshr.u16 q12, q6, #8 + vrshr.u16 q11, q5, #8 + vrshr.u16 q10, q4, #8 + vraddhn.u16 d2, q6, q12 + vraddhn.u16 d1, q5, q11 + vraddhn.u16 d0, q4, q10 +.endm + +.macro pixman_composite_add_0565_8_0565_process_pixblock_tail + vqadd.u8 q0, q0, q2 + vqadd.u8 q1, q1, q3 + /* 32bpp result is in {d0, d1, d2, XX} */ + convert_8888_to_0565 d2, d1, d0, q14, q15, q3 +.endm + +/* TODO: expand macros and do better instructions scheduling */ +.macro pixman_composite_add_0565_8_0565_process_pixblock_tail_head + vld1.8 {d15}, [MASK]! + pixman_composite_add_0565_8_0565_process_pixblock_tail + vld1.16 {d8, d9}, [SRC]! + vld1.16 {d10, d11}, [DST_R, :128]! + cache_preload 8, 8 + pixman_composite_add_0565_8_0565_process_pixblock_head + vst1.16 {d28, d29}, [DST_W, :128]! +.endm + +generate_composite_function \ + pixman_composite_add_0565_8_0565_asm_neon, 16, 8, 16, \ + FLAG_DST_READWRITE, \ + 8, /* number of pixels, processed in a single block */ \ + 5, /* prefetch distance */ \ + default_init_need_all_regs, \ + default_cleanup_need_all_regs, \ + pixman_composite_add_0565_8_0565_process_pixblock_head, \ + pixman_composite_add_0565_8_0565_process_pixblock_tail, \ + pixman_composite_add_0565_8_0565_process_pixblock_tail_head, \ + 28, /* dst_w_basereg */ \ + 10, /* dst_r_basereg */ \ + 8, /* src_basereg */ \ + 15 /* mask_basereg */ diff --git a/pixman/pixman-arm-neon.c b/pixman/pixman-arm-neon.c index cdaa29a..385987d 100644 --- a/pixman/pixman-arm-neon.c +++ b/pixman/pixman-arm-neon.c @@ -82,6 +82,8 @@ PIXMAN_ARM_BIND_FAST_PATH_SRC_N_DST (neon, over_8888_n_8888, 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_8888_8888, uint32_t, 1, uint32_t, 1, uint32_t, 1) PIXMAN_ARM_BIND_FAST_PATH_SRC_MASK_DST (neon, over_8888_8_8888, @@ -256,6 +258,8 @@ static const pixman_fast_path_t arm_neon_fast_paths[] = 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, 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, 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), -- 2.34.1