|
1 /* |
|
2 * Copyright 2012 The LibYuv 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 #include "libyuv/basic_types.h" |
|
12 |
|
13 #ifdef __cplusplus |
|
14 namespace libyuv { |
|
15 extern "C" { |
|
16 #endif |
|
17 |
|
18 #if !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__) |
|
19 |
|
20 uint32 SumSquareError_NEON(const uint8* src_a, const uint8* src_b, int count) { |
|
21 volatile uint32 sse; |
|
22 asm volatile ( |
|
23 "vmov.u8 q8, #0 \n" |
|
24 "vmov.u8 q10, #0 \n" |
|
25 "vmov.u8 q9, #0 \n" |
|
26 "vmov.u8 q11, #0 \n" |
|
27 |
|
28 ".p2align 2 \n" |
|
29 "1: \n" |
|
30 "vld1.8 {q0}, [%0]! \n" |
|
31 "vld1.8 {q1}, [%1]! \n" |
|
32 "subs %2, %2, #16 \n" |
|
33 "vsubl.u8 q2, d0, d2 \n" |
|
34 "vsubl.u8 q3, d1, d3 \n" |
|
35 "vmlal.s16 q8, d4, d4 \n" |
|
36 "vmlal.s16 q9, d6, d6 \n" |
|
37 "vmlal.s16 q10, d5, d5 \n" |
|
38 "vmlal.s16 q11, d7, d7 \n" |
|
39 "bgt 1b \n" |
|
40 |
|
41 "vadd.u32 q8, q8, q9 \n" |
|
42 "vadd.u32 q10, q10, q11 \n" |
|
43 "vadd.u32 q11, q8, q10 \n" |
|
44 "vpaddl.u32 q1, q11 \n" |
|
45 "vadd.u64 d0, d2, d3 \n" |
|
46 "vmov.32 %3, d0[0] \n" |
|
47 : "+r"(src_a), |
|
48 "+r"(src_b), |
|
49 "+r"(count), |
|
50 "=r"(sse) |
|
51 : |
|
52 : "memory", "cc", "q0", "q1", "q2", "q3", "q8", "q9", "q10", "q11"); |
|
53 return sse; |
|
54 } |
|
55 |
|
56 #endif // __ARM_NEON__ |
|
57 |
|
58 #ifdef __cplusplus |
|
59 } // extern "C" |
|
60 } // namespace libyuv |
|
61 #endif |