1/*
2 *  Copyright (c) 2014 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#include <arm_neon.h>
12#include "vpx_ports/arm.h"
13
14#ifdef VPX_INCOMPATIBLE_GCC
15#include "./vp8_rtcd.h"
16void vp8_short_walsh4x4_neon(int16_t *input, int16_t *output, int pitch) {
17  vp8_short_walsh4x4_c(input, output, pitch);
18}
19#else
20void vp8_short_walsh4x4_neon(int16_t *input, int16_t *output, int pitch) {
21  uint16x4_t d16u16;
22  int16x8_t q0s16, q1s16;
23  int16x4_t dEmptys16, d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16;
24  int32x4_t qEmptys32, q0s32, q1s32, q2s32, q3s32, q8s32;
25  int32x4_t q9s32, q10s32, q11s32, q15s32;
26  uint32x4_t q8u32, q9u32, q10u32, q11u32;
27  int16x4x2_t v2tmp0, v2tmp1;
28  int32x2x2_t v2tmp2, v2tmp3;
29
30  dEmptys16 = vdup_n_s16(0);
31  qEmptys32 = vdupq_n_s32(0);
32  q15s32 = vdupq_n_s32(3);
33
34  d0s16 = vld1_s16(input);
35  input += pitch / 2;
36  d1s16 = vld1_s16(input);
37  input += pitch / 2;
38  d2s16 = vld1_s16(input);
39  input += pitch / 2;
40  d3s16 = vld1_s16(input);
41
42  v2tmp2 = vtrn_s32(vreinterpret_s32_s16(d0s16), vreinterpret_s32_s16(d2s16));
43  v2tmp3 = vtrn_s32(vreinterpret_s32_s16(d1s16), vreinterpret_s32_s16(d3s16));
44  v2tmp0 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[0]),   // d0
45                    vreinterpret_s16_s32(v2tmp3.val[0]));  // d1
46  v2tmp1 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[1]),   // d2
47                    vreinterpret_s16_s32(v2tmp3.val[1]));  // d3
48
49  d4s16 = vadd_s16(v2tmp0.val[0], v2tmp1.val[0]);
50  d5s16 = vadd_s16(v2tmp0.val[1], v2tmp1.val[1]);
51  d6s16 = vsub_s16(v2tmp0.val[1], v2tmp1.val[1]);
52  d7s16 = vsub_s16(v2tmp0.val[0], v2tmp1.val[0]);
53
54  d4s16 = vshl_n_s16(d4s16, 2);
55  d5s16 = vshl_n_s16(d5s16, 2);
56  d6s16 = vshl_n_s16(d6s16, 2);
57  d7s16 = vshl_n_s16(d7s16, 2);
58
59  d16u16 = vceq_s16(d4s16, dEmptys16);
60  d16u16 = vmvn_u16(d16u16);
61
62  d0s16 = vadd_s16(d4s16, d5s16);
63  d3s16 = vsub_s16(d4s16, d5s16);
64  d1s16 = vadd_s16(d7s16, d6s16);
65  d2s16 = vsub_s16(d7s16, d6s16);
66
67  d0s16 = vsub_s16(d0s16, vreinterpret_s16_u16(d16u16));
68
69  // Second for-loop
70  v2tmp2 = vtrn_s32(vreinterpret_s32_s16(d1s16), vreinterpret_s32_s16(d3s16));
71  v2tmp3 = vtrn_s32(vreinterpret_s32_s16(d0s16), vreinterpret_s32_s16(d2s16));
72  v2tmp0 = vtrn_s16(vreinterpret_s16_s32(v2tmp3.val[1]),   // d2
73                    vreinterpret_s16_s32(v2tmp2.val[1]));  // d3
74  v2tmp1 = vtrn_s16(vreinterpret_s16_s32(v2tmp3.val[0]),   // d0
75                    vreinterpret_s16_s32(v2tmp2.val[0]));  // d1
76
77  q8s32 = vaddl_s16(v2tmp1.val[0], v2tmp0.val[0]);
78  q9s32 = vaddl_s16(v2tmp1.val[1], v2tmp0.val[1]);
79  q10s32 = vsubl_s16(v2tmp1.val[1], v2tmp0.val[1]);
80  q11s32 = vsubl_s16(v2tmp1.val[0], v2tmp0.val[0]);
81
82  q0s32 = vaddq_s32(q8s32, q9s32);
83  q1s32 = vaddq_s32(q11s32, q10s32);
84  q2s32 = vsubq_s32(q11s32, q10s32);
85  q3s32 = vsubq_s32(q8s32, q9s32);
86
87  q8u32 = vcltq_s32(q0s32, qEmptys32);
88  q9u32 = vcltq_s32(q1s32, qEmptys32);
89  q10u32 = vcltq_s32(q2s32, qEmptys32);
90  q11u32 = vcltq_s32(q3s32, qEmptys32);
91
92  q8s32 = vreinterpretq_s32_u32(q8u32);
93  q9s32 = vreinterpretq_s32_u32(q9u32);
94  q10s32 = vreinterpretq_s32_u32(q10u32);
95  q11s32 = vreinterpretq_s32_u32(q11u32);
96
97  q0s32 = vsubq_s32(q0s32, q8s32);
98  q1s32 = vsubq_s32(q1s32, q9s32);
99  q2s32 = vsubq_s32(q2s32, q10s32);
100  q3s32 = vsubq_s32(q3s32, q11s32);
101
102  q8s32 = vaddq_s32(q0s32, q15s32);
103  q9s32 = vaddq_s32(q1s32, q15s32);
104  q10s32 = vaddq_s32(q2s32, q15s32);
105  q11s32 = vaddq_s32(q3s32, q15s32);
106
107  d0s16 = vshrn_n_s32(q8s32, 3);
108  d1s16 = vshrn_n_s32(q9s32, 3);
109  d2s16 = vshrn_n_s32(q10s32, 3);
110  d3s16 = vshrn_n_s32(q11s32, 3);
111
112  q0s16 = vcombine_s16(d0s16, d1s16);
113  q1s16 = vcombine_s16(d2s16, d3s16);
114
115  vst1q_s16(output, q0s16);
116  vst1q_s16(output + 8, q1s16);
117  return;
118}
119#endif  // VPX_INCOMPATIBLE_GCC
120