-
Notifications
You must be signed in to change notification settings - Fork 21
/
simd_utils_riscv_int.h
379 lines (325 loc) · 10.2 KB
/
simd_utils_riscv_int.h
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
374
375
376
377
378
379
/*
* Project : SIMD_Utils
* Version : 0.2.5
* Author : JishinMaster
* Licence : BSD-2
*/
#pragma once
#include <fenv.h>
#include <math.h>
#include <riscv_vector.h>
#include <stdint.h>
#include <stdlib.h>
#include <string.h>
static inline void copys_vec(int32_t *src, int32_t *dst, int len)
{
size_t i;
int32_t *src_tmp = src;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
VSTORE_INT(dst_tmp, VLOAD_INT(src_tmp, i), i);
src_tmp += i;
dst_tmp += i;
}
}
static inline void subs_vec(int32_t *src1, int32_t *src2, int32_t *dst, int len)
{
size_t i;
int32_t *src1_tmp = src1;
int32_t *src2_tmp = src2;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va, vb;
va = VLOAD_INT(src1_tmp, i);
vb = VLOAD_INT(src2_tmp, i);
VSTORE_INT(dst_tmp, VSUB_INT(va, vb, i), i);
src1_tmp += i;
src2_tmp += i;
dst_tmp += i;
}
}
static inline void adds_vec(int32_t *src1, int32_t *src2, int32_t *dst, int len)
{
size_t i;
int32_t *src1_tmp = src1;
int32_t *src2_tmp = src2;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va, vb;
va = VLOAD_INT(src1_tmp, i);
vb = VLOAD_INT(src2_tmp, i);
VSTORE_INT(dst_tmp, VADD_INT(va, vb, i), i);
src1_tmp += i;
src2_tmp += i;
dst_tmp += i;
}
}
static inline void addcs_vec(int32_t *src, int32_t value, int32_t *dst, int len)
{
size_t i;
int32_t *src_tmp = src;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va;
va = VLOAD_INT(src_tmp, i);
VSTORE_INT(dst_tmp, VADD1_INT(va, value, i), i);
src_tmp += i;
dst_tmp += i;
}
}
static inline void muls_vec(int32_t *src1, int32_t *src2, int32_t *dst, int len)
{
size_t i;
int32_t *src1_tmp = src1;
int32_t *src2_tmp = src2;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va, vb;
va = VLOAD_INT(src1_tmp, i);
vb = VLOAD_INT(src2_tmp, i);
VSTORE_INT(dst_tmp, VMUL_INT(va, vb, i), i);
src1_tmp += i;
src2_tmp += i;
dst_tmp += i;
}
}
static inline void mulcs_vec(int32_t *src, int32_t value, int32_t *dst, int len)
{
size_t i;
int32_t *src_tmp = src;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va;
va = VLOAD_INT(src_tmp, i);
VSTORE_INT(dst_tmp, VMUL1_INT(va, value, i), i);
src_tmp += i;
dst_tmp += i;
}
}
static inline void vectorSlopes_vec(int32_t *dst, int len, int32_t offset, int32_t slope)
{
size_t i;
int32_t *dst_tmp = dst;
int32_t coef_max[MAX_ELTS32];
// to be improved!
for (int s = 0; s < MAX_ELTS32; s++) {
coef_max[s] = (int32_t) (s) *slope;
}
i = VSETVL32(len);
V_ELT_INT coef = VLOAD_INT(coef_max, i);
V_ELT_INT slope_vec = VLOAD1_INT((int32_t) (i) *slope, i);
V_ELT_INT curVal = VADD1_INT(coef, offset, i);
VSTORE_INT(dst_tmp, curVal, i);
dst_tmp += i;
len -= i;
for (; (i = VSETVL32(len)) > 0; len -= i) {
curVal = VADD_INT(curVal, slope_vec, i);
VSTORE_INT(dst_tmp, curVal, i);
dst_tmp += i;
}
}
static inline void threshold_gt_s_vec(int32_t *src, int32_t *dst, int len, float value)
{
size_t i;
int32_t *src_tmp = src;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va;
va = VLOAD_INT(src_tmp, i);
VSTORE_INT(dst_tmp, VMIN1_INT(va, value, i), i);
src_tmp += i;
dst_tmp += i;
}
}
static inline void threshold_lt_s_vec(int32_t *src, int32_t *dst, int len, float value)
{
size_t i;
int32_t *src_tmp = src;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va;
va = VLOAD_INT(src_tmp, i);
VSTORE_INT(dst_tmp, VMAX1_INT(va, value, i), i);
src_tmp += i;
dst_tmp += i;
}
}
static inline void threshold_gtabs_s_vec(int32_t *src, int32_t *dst, int len, int32_t value)
{
size_t i;
int32_t *src_tmp = src;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT src_tmp_vec = VLOAD_INT(src_tmp, i);
V_ELT_BOOL32 srcsup0 = VGT1_INT_BOOL(src_tmp_vec, 0, i);
V_ELT_INT src_tmp_neg_vec = VNEG_INT(src_tmp_vec, i);
V_ELT_INT src_abs = VMERGE_INT(srcsup0, src_tmp_vec, src_tmp_neg_vec, i);
V_ELT_BOOL32 eqmask = VEQ_INT_BOOL(src_abs, src_tmp_vec, i);
V_ELT_INT min = VMIN1_INT(src_tmp_vec, value, i);
V_ELT_INT max = VMAX1_INT(src_tmp_vec, -value, i);
V_ELT_INT dst_tmp_vec = VMERGE_INT(eqmask, min, max, i);
VSTORE_INT(dst_tmp, dst_tmp_vec, i);
src_tmp += i;
dst_tmp += i;
}
}
static inline void threshold_ltabs_s_vec(int32_t *src, int32_t *dst, int len, int32_t value)
{
size_t i;
int32_t *src_tmp = src;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT src_tmp_vec = VLOAD_INT(src_tmp, i);
V_ELT_BOOL32 srcsup0 = VGT1_INT_BOOL(src_tmp_vec, 0, i);
V_ELT_INT src_tmp_neg_vec = VNEG_INT(src_tmp_vec, i);
V_ELT_INT src_abs = VMERGE_INT(srcsup0, src_tmp_vec, src_tmp_neg_vec, i);
V_ELT_BOOL32 eqmask = VEQ_INT_BOOL(src_abs, src_tmp_vec, i);
V_ELT_INT max = VMAX1_INT(src_tmp_vec, value, i);
V_ELT_INT min = VMIN1_INT(src_tmp_vec, -value, i);
V_ELT_INT dst_tmp_vec = VMERGE_INT(eqmask, max, min, i);
VSTORE_INT(dst_tmp, dst_tmp_vec, i);
src_tmp += i;
dst_tmp += i;
}
}
static inline void minmaxs_vec(int32_t *src, int len, int32_t *min_value, int32_t *max_value)
{
size_t i, i_last;
int32_t *src_tmp = src;
i = VSETVL32(len);
vint32m1_t min0 = VLOAD_INTHH(src_tmp, 1); // or vfmv_v_f_f32m1
vint32m1_t max0 = min0;
V_ELT_INT minv, maxv, v1;
minv = VLOAD_INT(src_tmp, i);
maxv = minv;
for (; (i = VSETVL32(len)) > 0; len -= i) {
v1 = VLOAD_INT(src_tmp, i);
minv = VMIN_INT(v1, minv, i);
maxv = VMAX_INT(v1, maxv, i);
src_tmp += i;
i_last = i;
}
min0 = VREDMIN_INT(min0, minv, min0, i_last);
max0 = VREDMAX_INT(max0, maxv, max0, i_last);
VSTORE_INTHH(min_value, min0, 1);
VSTORE_INTHH(max_value, max0, 1);
}
static inline void maxeverys_vec(int32_t *src1, int32_t *src2, int32_t *dst, int len)
{
size_t i;
int32_t *src1_tmp = src1;
int32_t *src2_tmp = src2;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va, vb;
va = VLOAD_INT(src1_tmp, i);
vb = VLOAD_INT(src2_tmp, i);
VSTORE_INT(dst_tmp, VMAX_INT(va, vb, i), i);
src1_tmp += i;
src2_tmp += i;
dst_tmp += i;
}
}
static inline void mineverys_vec(int32_t *src1, int32_t *src2, int32_t *dst, int len)
{
size_t i;
int32_t *src1_tmp = src1;
int32_t *src2_tmp = src2;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va, vb;
va = VLOAD_INT(src1_tmp, i);
vb = VLOAD_INT(src2_tmp, i);
VSTORE_INT(dst_tmp, VMIN_INT(va, vb, i), i);
src1_tmp += i;
src2_tmp += i;
dst_tmp += i;
}
}
static inline void threshold_ltval_gtval_s_vec(int32_t *src, int32_t *dst, int len, int32_t ltlevel, int32_t ltvalue, int32_t gtlevel, int32_t gtvalue)
{
size_t i;
int32_t *src_tmp = src;
int32_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_INT va = VLOAD_INT(src_tmp, i);
V_ELT_BOOL32 lt_mask = VLT1_INT_BOOL(va, ltlevel, i);
V_ELT_BOOL32 gt_mask = VGT1_INT_BOOL(va, gtlevel, i);
V_ELT_INT tmp = VMERGE1_INT(lt_mask, va, ltvalue, i);
tmp = VMERGE1_INT(gt_mask, tmp, gtvalue, i);
VSTORE_INT(dst_tmp, tmp, i);
src_tmp += i;
dst_tmp += i;
}
}
static inline void flips_vec(int32_t *src, int32_t *dst, int len)
{
size_t i;
i = VSETVL32(len);
int vec_size = VSETVL32(4096);
int32_t *src_tmp = src + len - i;
int32_t *dst_tmp = dst;
#if MAX_ELTS32 == 32
// max vector size is 1024bits, but could be less (128bits on C906 core)
uint32_t index[32] = {31, 30, 29, 28, 27, 26, 25, 24, 23, 22, 21, 20,
19, 18, 17, 16, 15, 14, 13, 12, 11, 10, 9, 8, 7,
6, 5, 4, 3, 2, 1, 0};
#else
uint32_t index[MAX_ELTS32];
for (int l = 0; l < MAX_ELTS32; l++)
index[l] = MAX_ELTS32 - l - 1;
#endif
V_ELT_UINT index_vec = VLOAD_UINT(index + MAX_ELTS32 - vec_size, i);
V_ELT_INT a, b;
for (; (i = VSETVL32(len)) >= vec_size; len -= i) {
a = VLOAD_INT(src_tmp, i);
b = VGATHER_INT(a, index_vec, i);
VSTORE_INT(dst_tmp, b, i);
src_tmp -= i;
dst_tmp += i;
}
if (i) {
src_tmp = src;
index_vec = VLOAD_UINT(index + MAX_ELTS32 - i, i);
a = VLOAD_INT(src_tmp, i);
b = VGATHER_INT(a, index_vec, i);
VSTORE_INT(dst_tmp, b, i);
}
}
// Could it be improved? would need proper latency & cpi
static inline void sum16s32s_vec(int16_t *src, int len, int32_t *dst, int scale_factor)
{
size_t i;
int16_t *src_tmp = src;
int16_t scale = 1 << scale_factor;
*dst = 0;
i = VSETVL16(len);
vint32m1_t tmp = VLOAD1_INTHH(0, i);
for (; (i = VSETVL16(len)) > 0; len -= i) {
V_ELT_SHORT va = VLOAD_SHORT(src_tmp, i);
tmp = VREDSUMW_SHORT(tmp, va, tmp, i);
src_tmp += i;
}
VSTORE_INTHH(dst, tmp, 1);
*dst /= scale;
}
static inline void absdiff16s_vec(int16_t *src1, int16_t *src2, int16_t *dst, int len)
{
size_t i;
int16_t *src1_tmp = src1;
int16_t *src2_tmp = src2;
int16_t *dst_tmp = dst;
for (; (i = VSETVL32(len)) > 0; len -= i) {
V_ELT_SHORT va, vb, vc;
va = VLOAD_SHORT(src1_tmp, i);
vb = VLOAD_SHORT(src2_tmp, i);
V_ELT_BOOL16 cmp = VGT_SHORT_BOOL(va, vb, i);
V_ELT_SHORT difab = VSUB_SHORT(va, vb, i);
V_ELT_SHORT difba = VSUB_SHORT(vb, va, i);
vc = VMERGE_SHORT(cmp, difba, difab, i);
VSTORE_SHORT(dst_tmp, vc, i);
src1_tmp += i;
src2_tmp += i;
dst_tmp += i;
}
}