1;
2;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
3;
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.
9;
10
11
12    EXPORT |vp8_memcpy_partial_neon|
13
14    ARM
15    REQUIRE8
16    PRESERVE8
17
18    AREA ||.text||, CODE, READONLY, ALIGN=2
19;=========================================
20;this is not a full memcpy function!!!
21;void vp8_memcpy_partial_neon(unsigned char *dst_ptr, unsigned char *src_ptr,
22;                             int sz);
23|vp8_memcpy_partial_neon| PROC
24    vpush               {d8-d15}
25    ;pld                [r1]                        ;preload pred data
26    ;pld                [r1, #128]
27    ;pld                [r1, #256]
28    ;pld                [r1, #384]
29
30    mov             r12, r2, lsr #8                 ;copy 256 bytes data at one time
31
32memcpy_neon_loop
33    vld1.8          {q0, q1}, [r1]!                 ;load src data
34    subs            r12, r12, #1
35    vld1.8          {q2, q3}, [r1]!
36    vst1.8          {q0, q1}, [r0]!                 ;copy to dst_ptr
37    vld1.8          {q4, q5}, [r1]!
38    vst1.8          {q2, q3}, [r0]!
39    vld1.8          {q6, q7}, [r1]!
40    vst1.8          {q4, q5}, [r0]!
41    vld1.8          {q8, q9}, [r1]!
42    vst1.8          {q6, q7}, [r0]!
43    vld1.8          {q10, q11}, [r1]!
44    vst1.8          {q8, q9}, [r0]!
45    vld1.8          {q12, q13}, [r1]!
46    vst1.8          {q10, q11}, [r0]!
47    vld1.8          {q14, q15}, [r1]!
48    vst1.8          {q12, q13}, [r0]!
49    vst1.8          {q14, q15}, [r0]!
50
51    ;pld                [r1]                        ;preload pred data -- need to adjust for real device
52    ;pld                [r1, #128]
53    ;pld                [r1, #256]
54    ;pld                [r1, #384]
55
56    bne             memcpy_neon_loop
57
58    ands            r3, r2, #0xff                   ;extra copy
59    beq             done_copy_neon_loop
60
61extra_copy_neon_loop
62    vld1.8          {q0}, [r1]!                 ;load src data
63    subs            r3, r3, #16
64    vst1.8          {q0}, [r0]!
65    bne             extra_copy_neon_loop
66
67done_copy_neon_loop
68    vpop            {d8-d15}
69    bx              lr
70    ENDP
71
72    END
73