OLD | NEW |
1 /* | 1 /* |
2 * Copyright (c) 2013 The WebM project authors. All Rights Reserved. | 2 * Copyright (c) 2014 The WebM project authors. All Rights Reserved. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license | 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 | 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 | 6 * tree. An additional intellectual property rights grant can be found |
7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
9 */ | 9 */ |
10 | 10 |
| 11 #include <arm_neon.h> |
| 12 |
11 #include "./vp9_rtcd.h" | 13 #include "./vp9_rtcd.h" |
| 14 #include "./vpx_config.h" |
12 #include "vpx/vpx_integer.h" | 15 #include "vpx/vpx_integer.h" |
13 | 16 |
14 void vp9_lpf_horizontal_8_dual_neon(uint8_t *s, int p /* pitch */, | 17 static INLINE void vp9_loop_filter_neon_16( |
| 18 uint8x16_t qblimit, // blimit |
| 19 uint8x16_t qlimit, // limit |
| 20 uint8x16_t qthresh, // thresh |
| 21 uint8x16_t q3, // p3 |
| 22 uint8x16_t q4, // p2 |
| 23 uint8x16_t q5, // p1 |
| 24 uint8x16_t q6, // p0 |
| 25 uint8x16_t q7, // q0 |
| 26 uint8x16_t q8, // q1 |
| 27 uint8x16_t q9, // q2 |
| 28 uint8x16_t q10, // q3 |
| 29 uint8x16_t *q5r, // p1 |
| 30 uint8x16_t *q6r, // p0 |
| 31 uint8x16_t *q7r, // q0 |
| 32 uint8x16_t *q8r) { // q1 |
| 33 uint8x16_t q1u8, q2u8, q11u8, q12u8, q13u8, q14u8, q15u8; |
| 34 int16x8_t q2s16, q11s16; |
| 35 uint16x8_t q4u16; |
| 36 int8x16_t q0s8, q1s8, q2s8, q11s8, q12s8, q13s8; |
| 37 int8x8_t d2s8, d3s8; |
| 38 |
| 39 q11u8 = vabdq_u8(q3, q4); |
| 40 q12u8 = vabdq_u8(q4, q5); |
| 41 q13u8 = vabdq_u8(q5, q6); |
| 42 q14u8 = vabdq_u8(q8, q7); |
| 43 q3 = vabdq_u8(q9, q8); |
| 44 q4 = vabdq_u8(q10, q9); |
| 45 |
| 46 q11u8 = vmaxq_u8(q11u8, q12u8); |
| 47 q12u8 = vmaxq_u8(q13u8, q14u8); |
| 48 q3 = vmaxq_u8(q3, q4); |
| 49 q15u8 = vmaxq_u8(q11u8, q12u8); |
| 50 |
| 51 q9 = vabdq_u8(q6, q7); |
| 52 |
| 53 // vp8_hevmask |
| 54 q13u8 = vcgtq_u8(q13u8, qthresh); |
| 55 q14u8 = vcgtq_u8(q14u8, qthresh); |
| 56 q15u8 = vmaxq_u8(q15u8, q3); |
| 57 |
| 58 q2u8 = vabdq_u8(q5, q8); |
| 59 q9 = vqaddq_u8(q9, q9); |
| 60 |
| 61 q15u8 = vcgeq_u8(qlimit, q15u8); |
| 62 |
| 63 // vp8_filter() function |
| 64 // convert to signed |
| 65 q10 = vdupq_n_u8(0x80); |
| 66 q8 = veorq_u8(q8, q10); |
| 67 q7 = veorq_u8(q7, q10); |
| 68 q6 = veorq_u8(q6, q10); |
| 69 q5 = veorq_u8(q5, q10); |
| 70 |
| 71 q2u8 = vshrq_n_u8(q2u8, 1); |
| 72 q9 = vqaddq_u8(q9, q2u8); |
| 73 |
| 74 q2s16 = vsubl_s8(vget_low_s8(vreinterpretq_s8_u8(q7)), |
| 75 vget_low_s8(vreinterpretq_s8_u8(q6))); |
| 76 q11s16 = vsubl_s8(vget_high_s8(vreinterpretq_s8_u8(q7)), |
| 77 vget_high_s8(vreinterpretq_s8_u8(q6))); |
| 78 |
| 79 q9 = vcgeq_u8(qblimit, q9); |
| 80 |
| 81 q1s8 = vqsubq_s8(vreinterpretq_s8_u8(q5), |
| 82 vreinterpretq_s8_u8(q8)); |
| 83 |
| 84 q14u8 = vorrq_u8(q13u8, q14u8); |
| 85 |
| 86 q4u16 = vdupq_n_u16(3); |
| 87 q2s16 = vmulq_s16(q2s16, vreinterpretq_s16_u16(q4u16)); |
| 88 q11s16 = vmulq_s16(q11s16, vreinterpretq_s16_u16(q4u16)); |
| 89 |
| 90 q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q14u8); |
| 91 q15u8 = vandq_u8(q15u8, q9); |
| 92 |
| 93 q1s8 = vreinterpretq_s8_u8(q1u8); |
| 94 q2s16 = vaddw_s8(q2s16, vget_low_s8(q1s8)); |
| 95 q11s16 = vaddw_s8(q11s16, vget_high_s8(q1s8)); |
| 96 |
| 97 q4 = vdupq_n_u8(3); |
| 98 q9 = vdupq_n_u8(4); |
| 99 // vp8_filter = clamp(vp8_filter + 3 * ( qs0 - ps0)) |
| 100 d2s8 = vqmovn_s16(q2s16); |
| 101 d3s8 = vqmovn_s16(q11s16); |
| 102 q1s8 = vcombine_s8(d2s8, d3s8); |
| 103 q1u8 = vandq_u8(vreinterpretq_u8_s8(q1s8), q15u8); |
| 104 q1s8 = vreinterpretq_s8_u8(q1u8); |
| 105 |
| 106 q2s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q4)); |
| 107 q1s8 = vqaddq_s8(q1s8, vreinterpretq_s8_u8(q9)); |
| 108 q2s8 = vshrq_n_s8(q2s8, 3); |
| 109 q1s8 = vshrq_n_s8(q1s8, 3); |
| 110 |
| 111 q11s8 = vqaddq_s8(vreinterpretq_s8_u8(q6), q2s8); |
| 112 q0s8 = vqsubq_s8(vreinterpretq_s8_u8(q7), q1s8); |
| 113 |
| 114 q1s8 = vrshrq_n_s8(q1s8, 1); |
| 115 q1s8 = vbicq_s8(q1s8, vreinterpretq_s8_u8(q14u8)); |
| 116 |
| 117 q13s8 = vqaddq_s8(vreinterpretq_s8_u8(q5), q1s8); |
| 118 q12s8 = vqsubq_s8(vreinterpretq_s8_u8(q8), q1s8); |
| 119 |
| 120 *q8r = veorq_u8(vreinterpretq_u8_s8(q12s8), q10); |
| 121 *q7r = veorq_u8(vreinterpretq_u8_s8(q0s8), q10); |
| 122 *q6r = veorq_u8(vreinterpretq_u8_s8(q11s8), q10); |
| 123 *q5r = veorq_u8(vreinterpretq_u8_s8(q13s8), q10); |
| 124 return; |
| 125 } |
| 126 |
| 127 void vp9_lpf_horizontal_4_dual_neon(uint8_t *s, int p /* pitch */, |
15 const uint8_t *blimit0, | 128 const uint8_t *blimit0, |
16 const uint8_t *limit0, | 129 const uint8_t *limit0, |
17 const uint8_t *thresh0, | 130 const uint8_t *thresh0, |
18 const uint8_t *blimit1, | 131 const uint8_t *blimit1, |
19 const uint8_t *limit1, | 132 const uint8_t *limit1, |
20 const uint8_t *thresh1) { | 133 const uint8_t *thresh1) { |
21 vp9_lpf_horizontal_8(s, p, blimit0, limit0, thresh0, 1); | 134 uint8x8_t dblimit0, dlimit0, dthresh0, dblimit1, dlimit1, dthresh1; |
22 vp9_lpf_horizontal_8(s + 8, p, blimit1, limit1, thresh1, 1); | 135 uint8x16_t qblimit, qlimit, qthresh; |
| 136 uint8x16_t q3u8, q4u8, q5u8, q6u8, q7u8, q8u8, q9u8, q10u8; |
| 137 |
| 138 dblimit0 = vld1_u8(blimit0); |
| 139 dlimit0 = vld1_u8(limit0); |
| 140 dthresh0 = vld1_u8(thresh0); |
| 141 dblimit1 = vld1_u8(blimit1); |
| 142 dlimit1 = vld1_u8(limit1); |
| 143 dthresh1 = vld1_u8(thresh1); |
| 144 qblimit = vcombine_u8(dblimit0, dblimit1); |
| 145 qlimit = vcombine_u8(dlimit0, dlimit1); |
| 146 qthresh = vcombine_u8(dthresh0, dthresh1); |
| 147 |
| 148 s -= (p << 2); |
| 149 |
| 150 q3u8 = vld1q_u8(s); |
| 151 s += p; |
| 152 q4u8 = vld1q_u8(s); |
| 153 s += p; |
| 154 q5u8 = vld1q_u8(s); |
| 155 s += p; |
| 156 q6u8 = vld1q_u8(s); |
| 157 s += p; |
| 158 q7u8 = vld1q_u8(s); |
| 159 s += p; |
| 160 q8u8 = vld1q_u8(s); |
| 161 s += p; |
| 162 q9u8 = vld1q_u8(s); |
| 163 s += p; |
| 164 q10u8 = vld1q_u8(s); |
| 165 |
| 166 vp9_loop_filter_neon_16(qblimit, qlimit, qthresh, |
| 167 q3u8, q4u8, q5u8, q6u8, q7u8, q8u8, q9u8, q10u8, |
| 168 &q5u8, &q6u8, &q7u8, &q8u8); |
| 169 |
| 170 s -= (p * 5); |
| 171 vst1q_u8(s, q5u8); |
| 172 s += p; |
| 173 vst1q_u8(s, q6u8); |
| 174 s += p; |
| 175 vst1q_u8(s, q7u8); |
| 176 s += p; |
| 177 vst1q_u8(s, q8u8); |
| 178 return; |
23 } | 179 } |
24 | |
25 void vp9_lpf_vertical_4_dual_neon(uint8_t *s, int p, | |
26 const uint8_t *blimit0, | |
27 const uint8_t *limit0, | |
28 const uint8_t *thresh0, | |
29 const uint8_t *blimit1, | |
30 const uint8_t *limit1, | |
31 const uint8_t *thresh1) { | |
32 vp9_lpf_vertical_4_neon(s, p, blimit0, limit0, thresh0, 1); | |
33 vp9_lpf_vertical_4_neon(s + 8 * p, p, blimit1, limit1, thresh1, 1); | |
34 } | |
35 | |
36 void vp9_lpf_vertical_8_dual_neon(uint8_t *s, int p, | |
37 const uint8_t *blimit0, | |
38 const uint8_t *limit0, | |
39 const uint8_t *thresh0, | |
40 const uint8_t *blimit1, | |
41 const uint8_t *limit1, | |
42 const uint8_t *thresh1) { | |
43 vp9_lpf_vertical_8_neon(s, p, blimit0, limit0, thresh0, 1); | |
44 vp9_lpf_vertical_8_neon(s + 8 * p, p, blimit1, limit1, thresh1, 1); | |
45 } | |
46 | |
47 void vp9_lpf_vertical_16_dual_neon(uint8_t *s, int p, | |
48 const uint8_t *blimit, | |
49 const uint8_t *limit, | |
50 const uint8_t *thresh) { | |
51 vp9_lpf_vertical_16_neon(s, p, blimit, limit, thresh); | |
52 vp9_lpf_vertical_16_neon(s + 8 * p, p, blimit, limit, thresh); | |
53 } | |
OLD | NEW |