media/libvpx/vp8/encoder/arm/neon/vp8_memcpy_neon.asm@b8a032363ba2
media/libvpx/vp8/encoder/arm/neon/vp8_memcpy_neon.asm
Thu, 22 Jan 2015 13:21:57 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Thu, 22 Jan 2015 13:21:57 +0100
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
- permissions
- -rw-r--r--
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
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 ;
12 EXPORT |vp8_memcpy_partial_neon|
14 ARM
15 REQUIRE8
16 PRESERVE8
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 ;pld [r1] ;preload pred data
25 ;pld [r1, #128]
26 ;pld [r1, #256]
27 ;pld [r1, #384]
29 mov r12, r2, lsr #8 ;copy 256 bytes data at one time
31 memcpy_neon_loop
32 vld1.8 {q0, q1}, [r1]! ;load src data
33 subs r12, r12, #1
34 vld1.8 {q2, q3}, [r1]!
35 vst1.8 {q0, q1}, [r0]! ;copy to dst_ptr
36 vld1.8 {q4, q5}, [r1]!
37 vst1.8 {q2, q3}, [r0]!
38 vld1.8 {q6, q7}, [r1]!
39 vst1.8 {q4, q5}, [r0]!
40 vld1.8 {q8, q9}, [r1]!
41 vst1.8 {q6, q7}, [r0]!
42 vld1.8 {q10, q11}, [r1]!
43 vst1.8 {q8, q9}, [r0]!
44 vld1.8 {q12, q13}, [r1]!
45 vst1.8 {q10, q11}, [r0]!
46 vld1.8 {q14, q15}, [r1]!
47 vst1.8 {q12, q13}, [r0]!
48 vst1.8 {q14, q15}, [r0]!
50 ;pld [r1] ;preload pred data -- need to adjust for real device
51 ;pld [r1, #128]
52 ;pld [r1, #256]
53 ;pld [r1, #384]
55 bne memcpy_neon_loop
57 ands r3, r2, #0xff ;extra copy
58 beq done_copy_neon_loop
60 extra_copy_neon_loop
61 vld1.8 {q0}, [r1]! ;load src data
62 subs r3, r3, #16
63 vst1.8 {q0}, [r0]!
64 bne extra_copy_neon_loop
66 done_copy_neon_loop
67 bx lr
68 ENDP
70 END
