2 ; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
4 ; Use of this source code is governed by a BSD-style license
5 ; that can be found in the LICENSE file in the root of the source
6 ; tree. An additional intellectual property rights grant can be found
7 ; in the file PATENTS. All contributing project authors may
8 ; be found in the AUTHORS file in the root of the source tree.
11 EXPORT |vp8_subtract_b_neon|
12 EXPORT |vp8_subtract_mby_neon|
13 EXPORT |vp8_subtract_mbuv_neon|
15 INCLUDE asm_enc_offsets.asm
21 AREA ||.text||, CODE, READONLY, ALIGN=2
23 ;void vp8_subtract_b_neon(BLOCK *be, BLOCKD *bd, int pitch)
24 |vp8_subtract_b_neon| PROC
28 ldr r3, [r0, #vp8_block_base_src]
29 ldr r4, [r0, #vp8_block_src]
30 ldr r5, [r0, #vp8_block_src_diff]
32 ldr r6, [r0, #vp8_block_src_stride]
33 add r3, r3, r4 ; src = *base_src + src
34 ldr r7, [r1, #vp8_blockd_predictor]
36 vld1.8 {d0}, [r3], r6 ;load src
37 vld1.8 {d1}, [r7], r2 ;load pred
52 vst1.16 {d20}, [r5], r2 ;store diff
53 vst1.16 {d22}, [r5], r2
54 vst1.16 {d24}, [r5], r2
55 vst1.16 {d26}, [r5], r2
63 ;==========================================
64 ;void vp8_subtract_mby_neon(short *diff, unsigned char *src, unsigned char *pred, int stride)
65 |vp8_subtract_mby_neon| PROC
69 vld1.8 {q0}, [r1], r3 ;load src
70 vld1.8 {q1}, [r2]! ;load pred
84 vsubl.u8 q14, d12, d14
85 vsubl.u8 q15, d13, d15
87 vst1.16 {q8}, [r0]! ;store diff
102 ;=================================
103 ;void vp8_subtract_mbuv_neon(short *diff, unsigned char *usrc, unsigned char *vsrc, unsigned char *pred, int stride)
104 |vp8_subtract_mbuv_neon| PROC
108 add r0, r0, #512 ; short *udiff = diff + 256;
109 add r3, r3, #256 ; unsigned char *upred = pred + 256;
111 vld1.8 {d0}, [r1], r12 ;load src
112 vld1.8 {d1}, [r3]! ;load pred
113 vld1.8 {d2}, [r1], r12
115 vld1.8 {d4}, [r1], r12
117 vld1.8 {d6}, [r1], r12
119 vld1.8 {d8}, [r1], r12
121 vld1.8 {d10}, [r1], r12
123 vld1.8 {d12}, [r1], r12
125 vld1.8 {d14}, [r1], r12
133 vsubl.u8 q13, d10, d11
134 vsubl.u8 q14, d12, d13
135 vsubl.u8 q15, d14, d15
137 vst1.16 {q8}, [r0]! ;store diff
147 vld1.8 {d0}, [r2], r12 ;load src
148 vld1.8 {d1}, [r3]! ;load pred
149 vld1.8 {d2}, [r2], r12
151 vld1.8 {d4}, [r2], r12
153 vld1.8 {d6}, [r2], r12
155 vld1.8 {d8}, [r2], r12
157 vld1.8 {d10}, [r2], r12
159 vld1.8 {d12}, [r2], r12
161 vld1.8 {d14}, [r2], r12
169 vsubl.u8 q13, d10, d11
170 vsubl.u8 q14, d12, d13
171 vsubl.u8 q15, d14, d15
173 vst1.16 {q8}, [r0]! ;store diff