2 * Copyright (c) 2014 The WebM project authors. All Rights Reserved.
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.
13 #if (__GNUC__ == 4 && (__GNUC_MINOR__ == 6))
14 #warning Using GCC 4.6 is not recommended
15 // Some versions of gcc4.6 do not correctly process this function. When built
16 // with any gcc4.6, use the C code.
17 #include "./vp8_rtcd.h"
18 void vp8_short_walsh4x4_neon(
22 vp8_short_walsh4x4_c(input, output, pitch);
25 void vp8_short_walsh4x4_neon(
30 int16x8_t q0s16, q1s16;
31 int16x4_t dEmptys16, d0s16, d1s16, d2s16, d3s16, d4s16, d5s16, d6s16, d7s16;
32 int32x4_t qEmptys32, q0s32, q1s32, q2s32, q3s32, q8s32;
33 int32x4_t q9s32, q10s32, q11s32, q15s32;
34 uint32x4_t q8u32, q9u32, q10u32, q11u32;
35 int16x4x2_t v2tmp0, v2tmp1;
36 int32x2x2_t v2tmp2, v2tmp3;
38 dEmptys16 = vdup_n_s16(0);
39 qEmptys32 = vdupq_n_s32(0);
40 q15s32 = vdupq_n_s32(3);
42 d0s16 = vld1_s16(input);
44 d1s16 = vld1_s16(input);
46 d2s16 = vld1_s16(input);
48 d3s16 = vld1_s16(input);
50 v2tmp2 = vtrn_s32(vreinterpret_s32_s16(d0s16),
51 vreinterpret_s32_s16(d2s16));
52 v2tmp3 = vtrn_s32(vreinterpret_s32_s16(d1s16),
53 vreinterpret_s32_s16(d3s16));
54 v2tmp0 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[0]), // d0
55 vreinterpret_s16_s32(v2tmp3.val[0])); // d1
56 v2tmp1 = vtrn_s16(vreinterpret_s16_s32(v2tmp2.val[1]), // d2
57 vreinterpret_s16_s32(v2tmp3.val[1])); // d3
59 d4s16 = vadd_s16(v2tmp0.val[0], v2tmp1.val[0]);
60 d5s16 = vadd_s16(v2tmp0.val[1], v2tmp1.val[1]);
61 d6s16 = vsub_s16(v2tmp0.val[1], v2tmp1.val[1]);
62 d7s16 = vsub_s16(v2tmp0.val[0], v2tmp1.val[0]);
64 d4s16 = vshl_n_s16(d4s16, 2);
65 d5s16 = vshl_n_s16(d5s16, 2);
66 d6s16 = vshl_n_s16(d6s16, 2);
67 d7s16 = vshl_n_s16(d7s16, 2);
69 d16u16 = vceq_s16(d4s16, dEmptys16);
70 d16u16 = vmvn_u16(d16u16);
72 d0s16 = vadd_s16(d4s16, d5s16);
73 d3s16 = vsub_s16(d4s16, d5s16);
74 d1s16 = vadd_s16(d7s16, d6s16);
75 d2s16 = vsub_s16(d7s16, d6s16);
77 d0s16 = vsub_s16(d0s16, vreinterpret_s16_u16(d16u16));
80 v2tmp2 = vtrn_s32(vreinterpret_s32_s16(d1s16),
81 vreinterpret_s32_s16(d3s16));
82 v2tmp3 = vtrn_s32(vreinterpret_s32_s16(d0s16),
83 vreinterpret_s32_s16(d2s16));
84 v2tmp0 = vtrn_s16(vreinterpret_s16_s32(v2tmp3.val[1]), // d2
85 vreinterpret_s16_s32(v2tmp2.val[1])); // d3
86 v2tmp1 = vtrn_s16(vreinterpret_s16_s32(v2tmp3.val[0]), // d0
87 vreinterpret_s16_s32(v2tmp2.val[0])); // d1
89 q8s32 = vaddl_s16(v2tmp1.val[0], v2tmp0.val[0]);
90 q9s32 = vaddl_s16(v2tmp1.val[1], v2tmp0.val[1]);
91 q10s32 = vsubl_s16(v2tmp1.val[1], v2tmp0.val[1]);
92 q11s32 = vsubl_s16(v2tmp1.val[0], v2tmp0.val[0]);
94 q0s32 = vaddq_s32(q8s32, q9s32);
95 q1s32 = vaddq_s32(q11s32, q10s32);
96 q2s32 = vsubq_s32(q11s32, q10s32);
97 q3s32 = vsubq_s32(q8s32, q9s32);
99 q8u32 = vcltq_s32(q0s32, qEmptys32);
100 q9u32 = vcltq_s32(q1s32, qEmptys32);
101 q10u32 = vcltq_s32(q2s32, qEmptys32);
102 q11u32 = vcltq_s32(q3s32, qEmptys32);
104 q8s32 = vreinterpretq_s32_u32(q8u32);
105 q9s32 = vreinterpretq_s32_u32(q9u32);
106 q10s32 = vreinterpretq_s32_u32(q10u32);
107 q11s32 = vreinterpretq_s32_u32(q11u32);
109 q0s32 = vsubq_s32(q0s32, q8s32);
110 q1s32 = vsubq_s32(q1s32, q9s32);
111 q2s32 = vsubq_s32(q2s32, q10s32);
112 q3s32 = vsubq_s32(q3s32, q11s32);
114 q8s32 = vaddq_s32(q0s32, q15s32);
115 q9s32 = vaddq_s32(q1s32, q15s32);
116 q10s32 = vaddq_s32(q2s32, q15s32);
117 q11s32 = vaddq_s32(q3s32, q15s32);
119 d0s16 = vshrn_n_s32(q8s32, 3);
120 d1s16 = vshrn_n_s32(q9s32, 3);
121 d2s16 = vshrn_n_s32(q10s32, 3);
122 d3s16 = vshrn_n_s32(q11s32, 3);
124 q0s16 = vcombine_s16(d0s16, d1s16);
125 q1s16 = vcombine_s16(d2s16, d3s16);
127 vst1q_s16(output, q0s16);
128 vst1q_s16(output + 8, q1s16);
131 #endif // (__GNUC__ == 4 && (__GNUC_MINOR__ == 6))