-
Notifications
You must be signed in to change notification settings - Fork 0
/
Conversions-avx2.cpp
373 lines (349 loc) · 15.5 KB
/
Conversions-avx2.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
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
/*
FFT3DFilter plugin for Avisynth 2.6 - 3D Frequency Domain filter
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 "fft3dfilter.h"
#include <intrin.h>
//
//-----------------------------------------------------------------------------------------
// make destination frame plane from overlaped blocks
// use synthesis windows wsynxl, wsynxr, wsynyl, wsynyr
void FFT3DFilter::DecodeOverlapPlane_AVX2(const float *__restrict inp0, BYTE *__restrict dstp0) noexcept
{
int w(0);
BYTE *__restrict dstp = dstp0;// + (hrest/2)*coverpitch + wrest/2; // centered
const float *__restrict inp = inp0;
const int dbwow = bw - ow;
const int dbwow8 = dbwow - dbwow % 8;
const int d2bwow = dbwow - ow;
const int d2bwow8 = d2bwow - d2bwow % 8;
const int ow8 = ow - ow % 8;
const int ow4 = ow - ow % 4;
const __m256 norm8 = _mm256_set1_ps(norm);
const __m128i planebase4 = _mm_set1_epi32(planeBase);
const __m256i planebase8 = _mm256_set1_epi32(planeBase);
const int xoffset = bh * bw - dbwow;
const int yoffset = bw * nox*bh - bw * (bh - oh); // vertical offset of same block (overlap)
// first top big non-overlapped) part
{
for (int h = 0; h < bh - oh; h++)
{
inp = inp0 + h * bw;
__m256 input = _mm256_load_ps(&inp[0]);
for (w = 0; w < dbwow8; w = w + 8) // first half line of first block
{ // Copy each byte from float array to dest with windows
const __m256 r1 = _mm256_mul_ps(input, norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < dbwow; w++) // first half line of first block
{ // Copy each byte from float array to dest with windows
dstp[w] = (BYTE)min(255, max(0, (int)(inp[w] * norm) + planeBase));
}
inp += dbwow;
dstp += dbwow;
for (int ihx = 1; ihx < nox; ihx++) // middle horizontal half-blocks
{
input = _mm256_load_ps(&inp[0]);
for (w = 0; w < ow8; w = w + 8) // half line of block
{
__m256 r1 = _mm256_mul_ps(input, _mm256_load_ps(&wsynxr[w]));
const __m256 r3 = _mm256_mul_ps(_mm256_load_ps(&inp[w + xoffset]), _mm256_load_ps(&wsynxl[w]));
r1 = _mm256_add_ps(r1, r3);
r1 = _mm256_mul_ps(r1, norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < ow; w++) // half line of block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w] * wsynxr[w] + inp[w + xoffset] * wsynxl[w])*norm) + planeBase)); // overlapped Copy
}
inp += xoffset + ow;
dstp += ow;
for (w = 0; w < d2bwow8; w = w + 8) // last half line of last block
{
const __m256 r1 = _mm256_mul_ps(_mm256_load_ps(&inp[w]), norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < d2bwow; w++) // first half line of first block
{
dstp[w] = (BYTE)min(255, max(0, (inp[w] * norm) + planeBase)); // Copy each byte from float array to dest with windows
}
inp += d2bwow;
dstp += d2bwow;
}
input = _mm256_load_ps(&inp[0]);
for (w = 0; w < ow8; w = w + 8) // first half line of first block
{ // Copy each byte from float array to dest with windows
const __m256 r1 = _mm256_mul_ps(input, norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < ow; w++) // last half line of last block
{
dstp[w] = (BYTE)min(255, max(0, (inp[w] * norm) + planeBase));
}
inp += ow;
dstp += ow;
dstp += (coverpitch - coverwidth); // Add the pitch of one line (in bytes) to the dest image.
}
}
for (int ihy = 1; ihy < noy; ihy += 1) // middle vertical
{
for (int h = 0; h < oh; h++) // top overlapped part
{
inp = inp0 + (ihy - 1)*(yoffset + (bh - oh)*bw) + (bh - oh)*bw + h * bw;
const float wsynyrh = wsynyr[h] * norm; // remove from cycle for speed
const float wsynylh = wsynyl[h] * norm;
__m256 input = _mm256_load_ps(&inp[0]);
const __m128 wsynyrh4 = _mm_broadcast_ss(&wsynyrh), wsynylh4 = _mm_broadcast_ss(&wsynylh);
const __m256 wsynyrh8 = _mm256_broadcastss_ps(wsynyrh4), wsynylh8 = _mm256_broadcastss_ps(wsynylh4);
for (w = 0; w < dbwow8; w = w + 8) // first half line of first block
{
__m256 r1 = _mm256_mul_ps(input, wsynyrh8);
const __m256 r3 = _mm256_mul_ps(_mm256_load_ps(&inp[w + yoffset]), wsynylh8);
r1 = _mm256_add_ps(r1, r3);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < dbwow; w++) // first half line of first block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w] * wsynyrh + inp[w + yoffset] * wsynylh)) + planeBase)); //
}
inp += dbwow;
dstp += dbwow;
for (int ihx = 1; ihx < nox; ihx++) // middle blocks
{
for (w = 0; w < ow4; w = w + 4) // half overlapped line of block ; AVX processing fo some reason fails
{
__m128 r1 = _mm_mul_ps(_mm_load_ps(&inp[w]), _mm_load_ps(&wsynxr[w]));
const __m128 r3 = _mm_mul_ps(_mm_load_ps(&inp[w + xoffset]), _mm_load_ps(&wsynxl[w]));
__m128 r6 = _mm_mul_ps(_mm_load_ps(&inp[w + yoffset]), _mm_load_ps(&wsynxr[w]));
const __m128 r8 = _mm_mul_ps(_mm_load_ps(&inp[w + xoffset + yoffset]), _mm_load_ps(&wsynxl[w]));
r1 = _mm_add_ps(r1, r3);
r6 = _mm_add_ps(r6, r8);
r1 = _mm_mul_ps(r1, wsynyrh4);
r6 = _mm_mul_ps(r6, wsynylh4);
r1 = _mm_add_ps(r1, r6);
__m128i r1i = _mm_add_epi32(_mm_cvtps_epi32(r1), planebase4);
r1i = _mm_packus_epi32(r1i, _mm_setzero_si128());
r1i = _mm_packus_epi16(r1i, _mm_setzero_si128());
const int out_t = _mm_cvtsi128_si32(r1i);
memcpy(&dstp[w], &out_t, 4);
}
for (; w < ow; w++) // half overlapped line of block
{
dstp[w] = (BYTE)min(255, max(0, (((inp[w] * wsynxr[w] + inp[w + xoffset] * wsynxl[w])*wsynyrh
+ (inp[w + yoffset] * wsynxr[w] + inp[w + xoffset + yoffset] * wsynxl[w])*wsynylh)) + planeBase)); // x overlapped
}
inp += xoffset + ow;
dstp += ow;
for (w = 0; w < d2bwow8; w = w + 8) // last half line of last block
{
__m256 r1 = _mm256_mul_ps(_mm256_load_ps(&inp[w]), wsynyrh8);
const __m256 r2 = _mm256_mul_ps(_mm256_load_ps(&inp[w + yoffset]), wsynylh8);
r1 = _mm256_add_ps(r1, r2);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < d2bwow; w++) // double minus - half non-overlapped line of block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w] * wsynyrh + inp[w + yoffset] * wsynylh)) + planeBase));
}
inp += d2bwow;
dstp += d2bwow;
}
input = _mm256_load_ps(&inp[0]);
for (w = 0; w < ow8; w = w + 8) // last half line of last block
{
__m256 r1 = _mm256_mul_ps(input, wsynyrh8);
const __m256 r3 = _mm256_mul_ps(_mm256_load_ps(&inp[w + yoffset]), wsynylh8);
r1 = _mm256_add_ps(r1, r3);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < ow; w++) // last half line of last block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w] * wsynyrh + inp[w + yoffset] * wsynylh)) + planeBase));
}
inp += ow;
dstp += ow;
dstp += (coverpitch - coverwidth); // Add the pitch of one line (in bytes) to the source image.
}
// middle vertical non-ovelapped part
for (int h = 0; h < (bh - oh - oh); h++)
{
inp = inp0 + (ihy - 1)*(yoffset + (bh - oh)*bw) + (bh)*bw + h * bw + yoffset;
for (w = 0; w < dbwow; w++) // first half line of first block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w])*norm) + planeBase));
}
inp += dbwow;
dstp += dbwow;
for (int ihx = 1; ihx < nox; ihx++) // middle blocks
{
__m256 input = _mm256_load_ps(&inp[0]);
for (w = 0; w < ow8; w = w + 8) // half line of block
{
__m256 r1 = _mm256_mul_ps(input, _mm256_load_ps(&wsynxr[w]));
const __m256 r3 = _mm256_mul_ps(_mm256_load_ps(&inp[w + xoffset]), _mm256_load_ps(&wsynxl[w]));
r1 = _mm256_add_ps(r1, r3);
r1 = _mm256_mul_ps(r1, norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < ow; w++) // half overlapped line of block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w] * wsynxr[w] + inp[w + xoffset] * wsynxl[w])*norm) + planeBase)); // x overlapped
}
inp += xoffset + ow;
dstp += ow;
for (w = 0; w < d2bwow8; w = w + 8) // last half line of last block
{
const __m256 r1 = _mm256_mul_ps(_mm256_load_ps(&inp[w]), norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < d2bwow; w++) // half non-overlapped line of block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w])*norm) + planeBase));
}
inp += d2bwow;
dstp += d2bwow;
}
__m256 input = _mm256_load_ps(&inp[0]);
for (w = 0; w < ow8; w = w + 8) // first half line of first block
{ // Copy each byte from float array to dest with windows
const __m256 r1 = _mm256_mul_ps(input, norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < ow; w++) // last half line of last block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w])*norm) + planeBase));
}
inp += ow;
dstp += ow;
dstp += (coverpitch - coverwidth); // Add the pitch of one line (in bytes) to the source image.
}
}
const int ihy = noy; // last bottom part
{
for (int h = 0; h < oh; h++)
{
inp = inp0 + (ihy - 1)*(yoffset + (bh - oh)*bw) + (bh - oh)*bw + h * bw;
__m256 input = _mm256_load_ps(&inp[0]);
for (w = 0; w < dbwow8; w = w + 8) // first half line of first block
{ // Copy each byte from float array to dest with windows
const __m256 r1 = _mm256_mul_ps(input, norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < dbwow; w++) // first half line of first block
{
dstp[w] = (BYTE)min(255, max(0, (inp[w] * norm) + planeBase));
}
inp += dbwow;
dstp += dbwow;
for (int ihx = 1; ihx < nox; ihx++) // middle blocks
{
input = _mm256_load_ps(&inp[0]);
for (w = 0; w < ow8; w = w + 8) // half line of block
{
__m256 r1 = _mm256_mul_ps(input, _mm256_load_ps(&wsynxr[w]));
const __m256 r3 = _mm256_mul_ps(_mm256_load_ps(&inp[w + xoffset]), _mm256_load_ps(&wsynxl[w]));
r1 = _mm256_add_ps(r1, r3);
r1 = _mm256_mul_ps(r1, norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < ow; w++) // half line of block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w] * wsynxr[w] + inp[w + xoffset] * wsynxl[w])*norm) + planeBase)); // overlapped Copy
}
inp += xoffset + ow;
dstp += ow;
for (w = 0; w < d2bwow8; w = w + 8) // last half line of last block
{
const __m256 r1 = _mm256_mul_ps(_mm256_load_ps(&inp[w]), norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < d2bwow; w++) // half line of block
{
dstp[w] = (BYTE)min(255, max(0, ((inp[w])*norm) + planeBase));
}
inp += d2bwow;
dstp += d2bwow;
}
input = _mm256_load_ps(&inp[0]);
for (w = 0; w < ow8; w = w + 8) // first half line of first block
{ // Copy each byte from float array to dest with windows
const __m256 r1 = _mm256_mul_ps(input, norm8);
__m256i r1i = _mm256_add_epi32(_mm256_cvtps_epi32(r1), planebase8);
r1i = _mm256_packus_epi32(r1i, _mm256_setzero_si256());
r1i = _mm256_packus_epi16(r1i, _mm256_setzero_si256());
input = _mm256_load_ps(&inp[w + 8]);
_mm_storel_epi64((__m128i*)&dstp[w], _mm256_castsi256_si128(r1i));
}
for (; w < ow; w++) // last half line of last block
{
dstp[w] = (BYTE)min(255, max(0, (inp[w] * norm) + planeBase));
}
inp += ow;
dstp += ow;
dstp += (coverpitch - coverwidth); // Add the pitch of one line (in bytes) to the source image.
}
}
}