1 /*
2 * Copyright (c) 2017 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 #ifndef VPX_VPX_DSP_ARM_SUM_NEON_H_
12 #define VPX_VPX_DSP_ARM_SUM_NEON_H_
13
14 #include <arm_neon.h>
15
16 #include "./vpx_config.h"
17 #include "vpx/vpx_integer.h"
18
horizontal_add_int16x8(const int16x8_t a)19 static INLINE int32_t horizontal_add_int16x8(const int16x8_t a) {
20 #if defined(__aarch64__)
21 return vaddlvq_s16(a);
22 #else
23 const int32x4_t b = vpaddlq_s16(a);
24 const int64x2_t c = vpaddlq_s32(b);
25 const int32x2_t d = vadd_s32(vreinterpret_s32_s64(vget_low_s64(c)),
26 vreinterpret_s32_s64(vget_high_s64(c)));
27 return vget_lane_s32(d, 0);
28 #endif
29 }
30
horizontal_add_uint16x8(const uint16x8_t a)31 static INLINE uint32_t horizontal_add_uint16x8(const uint16x8_t a) {
32 #if defined(__aarch64__)
33 return vaddlvq_u16(a);
34 #else
35 const uint32x4_t b = vpaddlq_u16(a);
36 const uint64x2_t c = vpaddlq_u32(b);
37 const uint32x2_t d = vadd_u32(vreinterpret_u32_u64(vget_low_u64(c)),
38 vreinterpret_u32_u64(vget_high_u64(c)));
39 return vget_lane_u32(d, 0);
40 #endif
41 }
42
horizontal_add_int32x2(const int32x2_t a)43 static INLINE int32_t horizontal_add_int32x2(const int32x2_t a) {
44 #if defined(__aarch64__)
45 return vaddv_s32(a);
46 #else
47 return vget_lane_s32(a, 0) + vget_lane_s32(a, 1);
48 #endif
49 }
50
horizontal_add_uint32x2(const uint32x2_t a)51 static INLINE uint32_t horizontal_add_uint32x2(const uint32x2_t a) {
52 #if defined(__aarch64__)
53 return vaddv_u32(a);
54 #else
55 return vget_lane_u32(a, 0) + vget_lane_u32(a, 1);
56 #endif
57 }
58
horizontal_add_int32x4(const int32x4_t a)59 static INLINE int32_t horizontal_add_int32x4(const int32x4_t a) {
60 #if defined(__aarch64__)
61 return vaddvq_s32(a);
62 #else
63 const int64x2_t b = vpaddlq_s32(a);
64 const int32x2_t c = vadd_s32(vreinterpret_s32_s64(vget_low_s64(b)),
65 vreinterpret_s32_s64(vget_high_s64(b)));
66 return vget_lane_s32(c, 0);
67 #endif
68 }
69
horizontal_add_uint32x4(const uint32x4_t a)70 static INLINE uint32_t horizontal_add_uint32x4(const uint32x4_t a) {
71 #if defined(__aarch64__)
72 return vaddvq_u32(a);
73 #else
74 const uint64x2_t b = vpaddlq_u32(a);
75 const uint32x2_t c = vadd_u32(vreinterpret_u32_u64(vget_low_u64(b)),
76 vreinterpret_u32_u64(vget_high_u64(b)));
77 return vget_lane_u32(c, 0);
78 #endif
79 }
80 #endif // VPX_VPX_DSP_ARM_SUM_NEON_H_
81