1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
|
/*
* Copyright (c) 2018, Alliance for Open Media. All rights reserved
*
* This source code is subject to the terms of the BSD 2 Clause License and
* the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
* was not distributed with this source code in the LICENSE file, you can
* obtain it at www.aomedia.org/license/software. If the Alliance for Open
* Media Patent License 1.0 was not distributed with this source code in the
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
#include <assert.h>
#include <smmintrin.h>
#include "config/aom_config.h"
#include "aom_ports/mem.h"
#include "aom/aom_integer.h"
#include "aom_dsp/x86/synonyms.h"
static INLINE int64_t summary_all_sse4(const __m128i *sum_all) {
int64_t sum;
const __m128i sum0 = _mm_cvtepu32_epi64(*sum_all);
const __m128i sum1 = _mm_cvtepu32_epi64(_mm_srli_si128(*sum_all, 8));
const __m128i sum_2x64 = _mm_add_epi64(sum0, sum1);
const __m128i sum_1x64 = _mm_add_epi64(sum_2x64, _mm_srli_si128(sum_2x64, 8));
xx_storel_64(&sum, sum_1x64);
return sum;
}
static INLINE void sse_w16_sse4_1(__m128i *sum, const uint8_t *a,
const uint8_t *b) {
const __m128i v_a0 = xx_loadu_128(a);
const __m128i v_b0 = xx_loadu_128(b);
const __m128i v_a00_w = _mm_cvtepu8_epi16(v_a0);
const __m128i v_a01_w = _mm_cvtepu8_epi16(_mm_srli_si128(v_a0, 8));
const __m128i v_b00_w = _mm_cvtepu8_epi16(v_b0);
const __m128i v_b01_w = _mm_cvtepu8_epi16(_mm_srli_si128(v_b0, 8));
const __m128i v_d00_w = _mm_sub_epi16(v_a00_w, v_b00_w);
const __m128i v_d01_w = _mm_sub_epi16(v_a01_w, v_b01_w);
*sum = _mm_add_epi32(*sum, _mm_madd_epi16(v_d00_w, v_d00_w));
*sum = _mm_add_epi32(*sum, _mm_madd_epi16(v_d01_w, v_d01_w));
}
int64_t aom_sse_sse4_1(const uint8_t *a, int a_stride, const uint8_t *b,
int b_stride, int width, int height) {
int y = 0;
int64_t sse = 0;
__m128i sum = _mm_setzero_si128();
switch (width) {
case 4:
do {
const __m128i v_a0 = xx_loadl_32(a);
const __m128i v_a1 = xx_loadl_32(a + a_stride);
const __m128i v_b0 = xx_loadl_32(b);
const __m128i v_b1 = xx_loadl_32(b + b_stride);
const __m128i v_a_w = _mm_cvtepu8_epi16(_mm_unpacklo_epi32(v_a0, v_a1));
const __m128i v_b_w = _mm_cvtepu8_epi16(_mm_unpacklo_epi32(v_b0, v_b1));
const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
sum = _mm_add_epi32(sum, _mm_madd_epi16(v_d_w, v_d_w));
a += a_stride << 1;
b += b_stride << 1;
y += 2;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 8:
do {
const __m128i v_a0 = xx_loadl_64(a);
const __m128i v_b0 = xx_loadl_64(b);
const __m128i v_a_w = _mm_cvtepu8_epi16(v_a0);
const __m128i v_b_w = _mm_cvtepu8_epi16(v_b0);
const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
sum = _mm_add_epi32(sum, _mm_madd_epi16(v_d_w, v_d_w));
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 16:
do {
sse_w16_sse4_1(&sum, a, b);
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 32:
do {
sse_w16_sse4_1(&sum, a, b);
sse_w16_sse4_1(&sum, a + 16, b + 16);
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 64:
do {
sse_w16_sse4_1(&sum, a, b);
sse_w16_sse4_1(&sum, a + 16 * 1, b + 16 * 1);
sse_w16_sse4_1(&sum, a + 16 * 2, b + 16 * 2);
sse_w16_sse4_1(&sum, a + 16 * 3, b + 16 * 3);
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 128:
do {
sse_w16_sse4_1(&sum, a, b);
sse_w16_sse4_1(&sum, a + 16 * 1, b + 16 * 1);
sse_w16_sse4_1(&sum, a + 16 * 2, b + 16 * 2);
sse_w16_sse4_1(&sum, a + 16 * 3, b + 16 * 3);
sse_w16_sse4_1(&sum, a + 16 * 4, b + 16 * 4);
sse_w16_sse4_1(&sum, a + 16 * 5, b + 16 * 5);
sse_w16_sse4_1(&sum, a + 16 * 6, b + 16 * 6);
sse_w16_sse4_1(&sum, a + 16 * 7, b + 16 * 7);
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
default: break;
}
return sse;
}
static INLINE void highbd_sse_w8_sse4_1(__m128i *sum, const uint16_t *a,
const uint16_t *b) {
const __m128i v_a_w = xx_loadu_128(a);
const __m128i v_b_w = xx_loadu_128(b);
const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
*sum = _mm_add_epi32(*sum, _mm_madd_epi16(v_d_w, v_d_w));
}
int64_t aom_highbd_sse_sse4_1(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride, int width,
int height) {
int32_t y = 0;
int64_t sse = 0;
uint16_t *a = CONVERT_TO_SHORTPTR(a8);
uint16_t *b = CONVERT_TO_SHORTPTR(b8);
__m128i sum = _mm_setzero_si128();
switch (width) {
case 4:
do {
const __m128i v_a0 = xx_loadl_64(a);
const __m128i v_a1 = xx_loadl_64(a + a_stride);
const __m128i v_b0 = xx_loadl_64(b);
const __m128i v_b1 = xx_loadl_64(b + b_stride);
const __m128i v_a_w = _mm_unpacklo_epi64(v_a0, v_a1);
const __m128i v_b_w = _mm_unpacklo_epi64(v_b0, v_b1);
const __m128i v_d_w = _mm_sub_epi16(v_a_w, v_b_w);
sum = _mm_add_epi32(sum, _mm_madd_epi16(v_d_w, v_d_w));
a += a_stride << 1;
b += b_stride << 1;
y += 2;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 8:
do {
highbd_sse_w8_sse4_1(&sum, a, b);
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 16:
do {
highbd_sse_w8_sse4_1(&sum, a, b);
highbd_sse_w8_sse4_1(&sum, a + 8, b + 8);
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 32:
do {
highbd_sse_w8_sse4_1(&sum, a, b);
highbd_sse_w8_sse4_1(&sum, a + 8 * 1, b + 8 * 1);
highbd_sse_w8_sse4_1(&sum, a + 8 * 2, b + 8 * 2);
highbd_sse_w8_sse4_1(&sum, a + 8 * 3, b + 8 * 3);
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 64:
do {
highbd_sse_w8_sse4_1(&sum, a, b);
highbd_sse_w8_sse4_1(&sum, a + 8 * 1, b + 8 * 1);
highbd_sse_w8_sse4_1(&sum, a + 8 * 2, b + 8 * 2);
highbd_sse_w8_sse4_1(&sum, a + 8 * 3, b + 8 * 3);
highbd_sse_w8_sse4_1(&sum, a + 8 * 4, b + 8 * 4);
highbd_sse_w8_sse4_1(&sum, a + 8 * 5, b + 8 * 5);
highbd_sse_w8_sse4_1(&sum, a + 8 * 6, b + 8 * 6);
highbd_sse_w8_sse4_1(&sum, a + 8 * 7, b + 8 * 7);
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
case 128:
do {
highbd_sse_w8_sse4_1(&sum, a, b);
highbd_sse_w8_sse4_1(&sum, a + 8 * 1, b + 8 * 1);
highbd_sse_w8_sse4_1(&sum, a + 8 * 2, b + 8 * 2);
highbd_sse_w8_sse4_1(&sum, a + 8 * 3, b + 8 * 3);
highbd_sse_w8_sse4_1(&sum, a + 8 * 4, b + 8 * 4);
highbd_sse_w8_sse4_1(&sum, a + 8 * 5, b + 8 * 5);
highbd_sse_w8_sse4_1(&sum, a + 8 * 6, b + 8 * 6);
highbd_sse_w8_sse4_1(&sum, a + 8 * 7, b + 8 * 7);
highbd_sse_w8_sse4_1(&sum, a + 8 * 8, b + 8 * 8);
highbd_sse_w8_sse4_1(&sum, a + 8 * 9, b + 8 * 9);
highbd_sse_w8_sse4_1(&sum, a + 8 * 10, b + 8 * 10);
highbd_sse_w8_sse4_1(&sum, a + 8 * 11, b + 8 * 11);
highbd_sse_w8_sse4_1(&sum, a + 8 * 12, b + 8 * 12);
highbd_sse_w8_sse4_1(&sum, a + 8 * 13, b + 8 * 13);
highbd_sse_w8_sse4_1(&sum, a + 8 * 14, b + 8 * 14);
highbd_sse_w8_sse4_1(&sum, a + 8 * 15, b + 8 * 15);
a += a_stride;
b += b_stride;
y += 1;
} while (y < height);
sse = summary_all_sse4(&sum);
break;
default: break;
}
return sse;
}
|
