-
Notifications
You must be signed in to change notification settings - Fork 0
/
Sharpen_avx512.cpp
330 lines (298 loc) · 13.1 KB
/
Sharpen_avx512.cpp
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
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
//
// FFT3DFilter plugin for Avisynth 2.6 - 3D Frequency Domain filter
// AVX512 version of SSE filtering functions
//
// Derived from C version of function. (Copyright(C)2004-2006 A.G.Balakhnin aka Fizick, [email protected], http://avisynth.org.ru)
// Copyright(C) 2018 Daniel Klíma aka Klimax
//
// This program is free software; you can redistribute it and/or modify
// it under the terms of the GNU General Public License version 2 as published by
// the Free Software Foundation.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program; if not, write to the Free Software
// Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
//
//-----------------------------------------------------------------------------------------
//
#include "windows.h"
#include "fftwlite.h"
#include <intrin.h>
#include "Sharpen.h"
// since v1.7 we use outpitch instead of outwidth
//-------------------------------------------------------------------------------------------
//
void SharpenFilter::Sharpen_AVX512() noexcept
{
const int outwidth8 = outwidth - outwidth % 8;
if (sharpen != 0 && dehalo == 0)
{
const __m512 sharpen8 = _mm512_set1_ps(sharpen);
const __m512 sigmaSquaredSharpenMax8 = _mm512_set1_ps(sigmaSquaredSharpenMax);
const __m512 sigmaSquaredSharpenMin8 = _mm512_set1_ps(sigmaSquaredSharpenMin);
for (int block = start_block; block < blocks; block++)
{
for (int h = 0; h < bh; h++) // middle
{
for (int w = 0; w < outwidth8; w = w + 8)
{
const __m512 cur = _mm512_load_ps(outcur[w]);
const __m512 r1 = _mm512_mul_ps(cur, cur);
__m512 r2 = _mm512_shuffle_ps(r1, r1, _MM_SHUFFLE(0, 3, 0, 1));
const __m512 psd8 = _mm512_add_ps(r1, r2); //psd1, -, psd2, -
__m512 sharp8 = _mm512_castps256_ps512(_mm256_load_ps(&wsharpen[w]));
sharp8 = _mm512_permutexvar_ps(_mm512_set_epi32(7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0, 0), sharp8);
r2 = _mm512_mul_ps(sharpen8, sharp8);
__m512 r3 = _mm512_mul_ps(psd8, sigmaSquaredSharpenMax8);
__m512 r4 = _mm512_add_ps(psd8, sigmaSquaredSharpenMin8);
const __m512 r5 = _mm512_add_ps(psd8, sigmaSquaredSharpenMax8);
r4 = _mm512_mul_ps(r4, r5);
r3 = _mm512_div_ps(r3, r4);
r3 = _mm512_sqrt_ps(r3);
r2 = _mm512_mul_ps(r2, r3);
r2 = _mm512_add_ps(r2, _mm512_set1_ps(1.0f));
r2 = _mm512_moveldup_ps(r2);
_mm512_store_ps(outcur[w], _mm512_mul_ps(r2, cur));
}
_mm_prefetch((const char*)(outcur + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(wsharpen + outpitch), _MM_HINT_T0);
outcur += outpitch;
wsharpen += outpitch;
}
wsharpen -= outpitch * bh;
}
}
else if (sharpen == 0 && dehalo != 0)
{
const __m512 dehalo8 = _mm512_set1_ps(dehalo);
const __m512 ht2n8 = _mm512_set1_ps(ht2n);
for (int block = start_block; block < blocks; block++)
{
for (int h = 0; h < bh; h++) // middle
{
for (int w = 0; w < outwidth8; w = w + 8)
{
const __m512 cur = _mm512_load_ps(outcur[w]);
const __m512 r1 = _mm512_mul_ps(cur, cur);
__m512 r2 = _mm512_shuffle_ps(r1, r1, _MM_SHUFFLE(0, 3, 0, 1));
const __m512 psd8 = _mm512_add_ps(r1, r2); //psd1, -, psd2, -
__m512 wdehalo8 = _mm512_castps256_ps512(_mm256_load_ps(&wdehalo[w]));
wdehalo8 = _mm512_permutexvar_ps(_mm512_set_epi32(7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0, 0), wdehalo8);
__m512 r3 = _mm512_add_ps(psd8, ht2n8);
__m512 r4 = _mm512_mul_ps(psd8, dehalo8);
r4 = _mm512_mul_ps(r4, wdehalo8);
r4 = _mm512_add_ps(r3, r4);
r3 = _mm512_div_ps(r3, r4);
r2 = _mm512_moveldup_ps(r3);
_mm512_store_ps(outcur[w], _mm512_mul_ps(r2, cur));
}
_mm_prefetch((const char*)(outcur + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(wdehalo + outpitch), _MM_HINT_T0);
outcur += outpitch;
wdehalo += outpitch;
}
wdehalo -= outpitch * bh;
}
}
else if (sharpen != 0 && dehalo != 0)
{
const __m512 sharpen8 = _mm512_set1_ps(sharpen);
const __m512 sigmaSquaredSharpenMax8 = _mm512_set1_ps(sigmaSquaredSharpenMax);
const __m512 sigmaSquaredSharpenMin8 = _mm512_set1_ps(sigmaSquaredSharpenMin);
const __m512 dehalo8 = _mm512_set1_ps(dehalo);
const __m512 ht2n8 = _mm512_set1_ps(ht2n);
for (int block = start_block; block < blocks; block++)
{
for (int h = 0; h < bh; h++) // middle
{
for (int w = 0; w < outwidth8; w = w + 8)
{
const __m512 cur = _mm512_load_ps(outcur[w]);
const __m512 r1 = _mm512_mul_ps(cur, cur);
__m512 r2 = _mm512_shuffle_ps(r1, r1, _MM_SHUFFLE(0, 3, 0, 1));
const __m512 psd8 = _mm512_add_ps(r1, r2); //psd1, -, psd2, -
__m512 sharp8 = _mm512_castps256_ps512(_mm256_load_ps(&wsharpen[w]));
sharp8 = _mm512_permutexvar_ps(_mm512_set_epi32(7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0, 0), sharp8);
r2 = _mm512_mul_ps(sharpen8, sharp8);
__m512 r3 = _mm512_mul_ps(psd8, sigmaSquaredSharpenMax8);
__m512 r4 = _mm512_add_ps(psd8, sigmaSquaredSharpenMin8);
const __m512 r5 = _mm512_add_ps(psd8, sigmaSquaredSharpenMax8);
r4 = _mm512_mul_ps(r4, r5);
r3 = _mm512_div_ps(r3, r4);
r3 = _mm512_sqrt_ps(r3);
r2 = _mm512_mul_ps(r2, r3);
r2 = _mm512_add_ps(r2, _mm512_set1_ps(1.0f));
__m512 wdehalo8 = _mm512_castps256_ps512(_mm256_load_ps(&wdehalo[w]));
wdehalo8 = _mm512_permutexvar_ps(_mm512_set_epi32(7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0, 0), wdehalo8);
r3 = _mm512_add_ps(psd8, ht2n8);
r4 = _mm512_mul_ps(psd8, dehalo8);
r4 = _mm512_mul_ps(r4, wdehalo8);
r4 = _mm512_add_ps(r3, r4);
r3 = _mm512_div_ps(r3, r4);
r2 = _mm512_mul_ps(r2, r3);
r2 = _mm512_moveldup_ps(r2);
_mm512_store_ps(outcur[w], _mm512_mul_ps(r2, cur));
}
_mm_prefetch((const char*)(outcur + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(wsharpen + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(wdehalo + outpitch), _MM_HINT_T0);
outcur += outpitch;
wsharpen += outpitch;
wdehalo += outpitch;
}
wsharpen -= outpitch * bh;
wdehalo -= outpitch * bh;
}
}
}
//-------------------------------------------------------------------------------------------
//
void SharpenFilter::Sharpen_degrid_AVX512() noexcept
{
const int outwidth8 = outwidth - outwidth % 8;
if (sharpen != 0 && dehalo == 0)
{
const __m512 sharpen8 = _mm512_set1_ps(sharpen);
const __m512 sigmaSquaredSharpenMax8 = _mm512_set1_ps(sigmaSquaredSharpenMax);
const __m512 sigmaSquaredSharpenMin8 = _mm512_set1_ps(sigmaSquaredSharpenMin);
for (int block = start_block; block < blocks; block++)
{
const __m512 gridfraction8 = _mm512_set1_ps(degrid * outcur[0][0] / gridsample[0][0]);
for (int h = 0; h < bh; h++) // middle
{
for (int w = 0; w < outwidth8; w = w + 8)
{
const __m512 gridcorrection = _mm512_mul_ps(gridfraction8, _mm512_load_ps(gridsample[w])); //gridcorrection
__m512 cur = _mm512_load_ps(outcur[w]);
cur = _mm512_sub_ps(cur, gridcorrection);
const __m512 r1 = _mm512_mul_ps(cur, cur);
__m512 r2 = _mm512_shuffle_ps(r1, r1, _MM_SHUFFLE(0, 3, 0, 1));
const __m512 psd8 = _mm512_add_ps(r1, r2); //psd1, -, psd2, -
__m512 sharp8 = _mm512_castps256_ps512(_mm256_load_ps(&wsharpen[w]));
sharp8 = _mm512_permutexvar_ps(_mm512_set_epi32(7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0, 0), sharp8);
r2 = _mm512_mul_ps(sharpen8, sharp8);
__m512 r3 = _mm512_mul_ps(psd8, sigmaSquaredSharpenMax8);
__m512 r4 = _mm512_add_ps(psd8, sigmaSquaredSharpenMin8);
const __m512 r5 = _mm512_add_ps(psd8, sigmaSquaredSharpenMax8);
r4 = _mm512_mul_ps(r4, r5);
r3 = _mm512_div_ps(r3, r4);
r3 = _mm512_sqrt_ps(r3);
r2 = _mm512_mul_ps(r2, r3);
r2 = _mm512_add_ps(r2, _mm512_set1_ps(1.0f));
r2 = _mm512_moveldup_ps(r2);
r2 = _mm512_mul_ps(r2, cur);
_mm512_store_ps(outcur[w], _mm512_add_ps(r2, gridcorrection));
}
_mm_prefetch((const char*)(outcur + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(gridsample + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(wsharpen + outpitch), _MM_HINT_T0);
outcur += outpitch;
wsharpen += outpitch;
gridsample += outpitch;
}
wsharpen -= outpitch * bh;
gridsample -= outpitch * bh; // restore pointer to only valid first block - bug fixed in v1.8.1
}
}
if (sharpen == 0 && dehalo != 0)
{
const __m512 dehalo8 = _mm512_set1_ps(dehalo);
const __m512 ht2n8 = _mm512_set1_ps(ht2n);
for (int block = start_block; block < blocks; block++)
{
const __m512 gridfraction8 = _mm512_set1_ps(degrid * outcur[0][0] / gridsample[0][0]);
for (int h = 0; h < bh; h++) // middle
{
for (int w = 0; w < outwidth8; w = w + 8)
{
const __m512 gridcorrection = _mm512_mul_ps(gridfraction8, _mm512_load_ps(gridsample[w])); //gridcorrection
__m512 cur = _mm512_load_ps(outcur[w]);
cur = _mm512_sub_ps(cur, gridcorrection);
const __m512 r1 = _mm512_mul_ps(cur, cur);
__m512 r2 = _mm512_shuffle_ps(r1, r1, _MM_SHUFFLE(0, 3, 0, 1));
const __m512 psd8 = _mm512_add_ps(r1, r2); //psd1, -, psd2, -
__m512 wdehalo8 = _mm512_castps256_ps512(_mm256_load_ps(&wdehalo[w]));
wdehalo8 = _mm512_permutexvar_ps(_mm512_set_epi32(7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0, 0), wdehalo8);
__m512 r3 = _mm512_add_ps(psd8, ht2n8);
__m512 r4 = _mm512_mul_ps(psd8, dehalo8);
r4 = _mm512_mul_ps(r4, wdehalo8);
r4 = _mm512_add_ps(r3, r4);
r3 = _mm512_div_ps(r3, r4);
r2 = _mm512_moveldup_ps(r3);
r2 = _mm512_mul_ps(r2, cur);
_mm512_store_ps(outcur[w], _mm512_add_ps(r2, gridcorrection));
}
_mm_prefetch((const char*)(outcur + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(gridsample + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(wdehalo + outpitch), _MM_HINT_T0);
outcur += outpitch;
wdehalo += outpitch;
gridsample += outpitch;
}
wdehalo -= outpitch * bh;
gridsample -= outpitch * bh; // restore pointer to only valid first block - bug fixed in v1.8.1
}
}
if (sharpen != 0 && dehalo != 0)
{
const __m512 sharpen8 = _mm512_set1_ps(sharpen);
const __m512 sigmaSquaredSharpenMax8 = _mm512_set1_ps(sigmaSquaredSharpenMax);
const __m512 sigmaSquaredSharpenMin8 = _mm512_set1_ps(sigmaSquaredSharpenMin);
const __m512 dehalo8 = _mm512_set1_ps(dehalo);
const __m512 ht2n8 = _mm512_set1_ps(ht2n);
for (int block = start_block; block < blocks; block++)
{
const __m512 gridfraction8 = _mm512_set1_ps(degrid * outcur[0][0] / gridsample[0][0]);
for (int h = 0; h < bh; h++) // middle
{
for (int w = 0; w < outwidth8; w = w + 8)
{
const __m512 gridcorrection = _mm512_mul_ps(gridfraction8, _mm512_load_ps(gridsample[w])); //gridcorrection
__m512 cur = _mm512_load_ps(outcur[w]);
cur = _mm512_sub_ps(cur, gridcorrection);
const __m512 r1 = _mm512_mul_ps(cur, cur);
__m512 r2 = _mm512_shuffle_ps(r1, r1, _MM_SHUFFLE(0, 3, 0, 1));
const __m512 psd8 = _mm512_add_ps(r1, r2); //psd1, -, psd2, -
__m512 sharp8 = _mm512_castps256_ps512(_mm256_load_ps(&wsharpen[w]));
sharp8 = _mm512_permutexvar_ps(_mm512_set_epi32(7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0, 0), sharp8);
r2 = _mm512_mul_ps(sharpen8, sharp8);
__m512 r3 = _mm512_mul_ps(psd8, sigmaSquaredSharpenMax8);
__m512 r4 = _mm512_add_ps(psd8, sigmaSquaredSharpenMin8);
const __m512 r5 = _mm512_add_ps(psd8, sigmaSquaredSharpenMax8);
r4 = _mm512_mul_ps(r4, r5);
r3 = _mm512_div_ps(r3, r4);
r3 = _mm512_sqrt_ps(r3);
r2 = _mm512_mul_ps(r2, r3);
r2 = _mm512_add_ps(r2, _mm512_set1_ps(1.0f));
__m512 wdehalo8 = _mm512_castps256_ps512(_mm256_load_ps(&wdehalo[w]));
wdehalo8 = _mm512_permutexvar_ps(_mm512_set_epi32(7, 7, 6, 6, 5, 5, 4, 4, 3, 3, 2, 2, 1, 1, 0, 0), wdehalo8);
r3 = _mm512_add_ps(psd8, ht2n8);
r4 = _mm512_mul_ps(psd8, dehalo8);
r4 = _mm512_mul_ps(r4, wdehalo8);
r4 = _mm512_add_ps(r3, r4);
r3 = _mm512_div_ps(r3, r4);
r2 = _mm512_mul_ps(r2, r3);
r2 = _mm512_moveldup_ps(r2);
r2 = _mm512_mul_ps(r2, cur);
_mm512_store_ps(outcur[w], _mm512_add_ps(r2, gridcorrection));
}
_mm_prefetch((const char*)(outcur + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(gridsample + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(wsharpen + outpitch), _MM_HINT_T0);
_mm_prefetch((const char*)(wdehalo + outpitch), _MM_HINT_T0);
outcur += outpitch;
wsharpen += outpitch;
wdehalo += outpitch;
gridsample += outpitch;
}
wsharpen -= outpitch * bh;
wdehalo -= outpitch * bh;
gridsample -= outpitch * bh; // restore pointer to only valid first block - bug fixed in v1.8.1
}
}
}