forked from simdjson/simdjson
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathjsonminifier.cpp
More file actions
469 lines (439 loc) · 22.4 KB
/
jsonminifier.cpp
File metadata and controls
469 lines (439 loc) · 22.4 KB
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
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
#include "simdjson/portability.h"
#include <cstdint>
#ifndef __AVX2__
namespace simdjson {
static uint8_t jump_table[256 * 3] = {
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0,
1, 1, 0, 1, 0, 0, 1, 0, 0, 1, 1, 0, 1, 1, 0, 1, 0, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 0, 0,
1, 1, 1, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0,
1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0,
1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 0, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0,
1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0,
1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0,
1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0,
1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0,
1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0,
1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1,
1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1, 0, 1, 1,
};
size_t json_minify(const unsigned char *bytes, size_t how_many,
unsigned char *out) {
size_t i = 0, pos = 0;
uint8_t quote = 0;
uint8_t nonescape = 1;
while (i < how_many) {
unsigned char c = bytes[i];
uint8_t *meta = jump_table + 3 * c;
quote = quote ^ (meta[0] & nonescape);
out[pos] = c;
pos += meta[2] | quote;
i += 1;
nonescape = (~nonescape) | (meta[1]);
}
return pos;
}
} // namespace simdjson
#else
#include "simdprune_tables.h"
#include <cstring>
namespace simdjson {
// a straightforward comparison of a mask against input.
static uint64_t cmp_mask_against_input_mini(__m256i input_lo, __m256i input_hi,
__m256i mask) {
__m256i cmp_res_0 = _mm256_cmpeq_epi8(input_lo, mask);
uint64_t res_0 = static_cast<uint32_t>(_mm256_movemask_epi8(cmp_res_0));
__m256i cmp_res_1 = _mm256_cmpeq_epi8(input_hi, mask);
uint64_t res_1 = _mm256_movemask_epi8(cmp_res_1);
return res_0 | (res_1 << 32);
}
// Write up to 16 bytes, only the bytes corresponding to a 1-bit are written
// out. credit: Anime Tosho
static __m128i skinnycleanm128(__m128i x, int mask) {
int mask1 = mask & 0xFF;
int mask2 = (mask >> 8) & 0xFF;
__m128i shufmask = _mm_castps_si128(
_mm_loadh_pi(_mm_castsi128_ps(_mm_loadl_epi64(
(const __m128i *)(thintable_epi8 + mask1))),
(const __m64 *)(thintable_epi8 + mask2)));
shufmask =
_mm_add_epi8(shufmask, _mm_set_epi32(0x08080808, 0x08080808, 0, 0));
__m128i pruned = _mm_shuffle_epi8(x, shufmask);
intptr_t popx2 = BitsSetTable256mul2[mask1];
__m128i compactmask =
_mm_loadu_si128((const __m128i *)(pshufb_combine_table + popx2 * 8));
return _mm_shuffle_epi8(pruned, compactmask);
}
// take input from buf and remove useless whitespace, input and output can be
// the same, result is null terminated, return the string length (minus the null
// termination)
size_t json_minify(const uint8_t *buf, size_t len, uint8_t *out) {
// Useful constant masks
const uint64_t even_bits = 0x5555555555555555ULL;
const uint64_t odd_bits = ~even_bits;
uint8_t *initout(out);
uint64_t prev_iter_ends_odd_backslash =
0ULL; // either 0 or 1, but a 64-bit value
uint64_t prev_iter_inside_quote = 0ULL; // either all zeros or all ones
size_t idx = 0;
if (len >= 64) {
size_t avx_len = len - 63;
for (; idx < avx_len; idx += 64) {
__m256i input_lo =
_mm256_loadu_si256(reinterpret_cast<const __m256i *>(buf + idx + 0));
__m256i input_hi =
_mm256_loadu_si256(reinterpret_cast<const __m256i *>(buf + idx + 32));
uint64_t bs_bits = cmp_mask_against_input_mini(input_lo, input_hi,
_mm256_set1_epi8('\\'));
uint64_t start_edges = bs_bits & ~(bs_bits << 1);
uint64_t even_start_mask = even_bits ^ prev_iter_ends_odd_backslash;
uint64_t even_starts = start_edges & even_start_mask;
uint64_t odd_starts = start_edges & ~even_start_mask;
uint64_t even_carries = bs_bits + even_starts;
uint64_t odd_carries;
bool iter_ends_odd_backslash =
add_overflow(bs_bits, odd_starts, &odd_carries);
odd_carries |= prev_iter_ends_odd_backslash;
prev_iter_ends_odd_backslash = iter_ends_odd_backslash ? 0x1ULL : 0x0ULL;
uint64_t even_carry_ends = even_carries & ~bs_bits;
uint64_t odd_carry_ends = odd_carries & ~bs_bits;
uint64_t even_start_odd_end = even_carry_ends & odd_bits;
uint64_t odd_start_even_end = odd_carry_ends & even_bits;
uint64_t odd_ends = even_start_odd_end | odd_start_even_end;
uint64_t quote_bits = cmp_mask_against_input_mini(input_lo, input_hi,
_mm256_set1_epi8('"'));
quote_bits = quote_bits & ~odd_ends;
uint64_t quote_mask = _mm_cvtsi128_si64(_mm_clmulepi64_si128(
_mm_set_epi64x(0ULL, quote_bits), _mm_set1_epi8(0xFF), 0));
quote_mask ^= prev_iter_inside_quote;
prev_iter_inside_quote = static_cast<uint64_t>(
static_cast<int64_t>(quote_mask) >>
63); // might be undefined behavior, should be fully defined in C++20,
// ok according to John Regher from Utah University
const __m256i low_nibble_mask = _mm256_setr_epi8(
// 0 9 a b c d
16, 0, 0, 0, 0, 0, 0, 0, 0, 8, 12, 1, 2, 9, 0, 0, 16, 0, 0, 0, 0, 0,
0, 0, 0, 8, 12, 1, 2, 9, 0, 0);
const __m256i high_nibble_mask = _mm256_setr_epi8(
// 0 2 3 5 7
8, 0, 18, 4, 0, 1, 0, 1, 0, 0, 0, 3, 2, 1, 0, 0, 8, 0, 18, 4, 0, 1, 0,
1, 0, 0, 0, 3, 2, 1, 0, 0);
__m256i whitespace_shufti_mask = _mm256_set1_epi8(0x18);
__m256i v_lo = _mm256_and_si256(
_mm256_shuffle_epi8(low_nibble_mask, input_lo),
_mm256_shuffle_epi8(high_nibble_mask,
_mm256_and_si256(_mm256_srli_epi32(input_lo, 4),
_mm256_set1_epi8(0x7f))));
__m256i v_hi = _mm256_and_si256(
_mm256_shuffle_epi8(low_nibble_mask, input_hi),
_mm256_shuffle_epi8(high_nibble_mask,
_mm256_and_si256(_mm256_srli_epi32(input_hi, 4),
_mm256_set1_epi8(0x7f))));
__m256i tmp_ws_lo = _mm256_cmpeq_epi8(
_mm256_and_si256(v_lo, whitespace_shufti_mask), _mm256_set1_epi8(0));
__m256i tmp_ws_hi = _mm256_cmpeq_epi8(
_mm256_and_si256(v_hi, whitespace_shufti_mask), _mm256_set1_epi8(0));
uint64_t ws_res_0 =
static_cast<uint32_t>(_mm256_movemask_epi8(tmp_ws_lo));
uint64_t ws_res_1 = _mm256_movemask_epi8(tmp_ws_hi);
uint64_t whitespace = ~(ws_res_0 | (ws_res_1 << 32));
whitespace &= ~quote_mask;
uint64_t non_whitespace = ~whitespace;
__m128i x1 = _mm256_extracti128_si256(input_lo, 0);
__m128i x2 = _mm256_extracti128_si256(input_lo, 1);
__m128i x3 = _mm256_extracti128_si256(input_hi, 0);
__m128i x4 = _mm256_extracti128_si256(input_hi, 1);
int mask1 = non_whitespace & 0xFFFF;
int mask2 = (non_whitespace >> 16) & 0xFFFF;
int mask3 = (non_whitespace >> 32) & 0xFFFF;
int mask4 = (non_whitespace >> 48) & 0xFFFF;
x1 = skinnycleanm128(x1, mask1);
x2 = skinnycleanm128(x2, mask2);
x3 = skinnycleanm128(x3, mask3);
x4 = skinnycleanm128(x4, mask4);
int pop1 = hamming(non_whitespace & 0xFFFF);
int pop2 = hamming(non_whitespace & UINT64_C(0xFFFFFFFF));
int pop3 = hamming(non_whitespace) & UINT64_C(0xFFFFFFFFFFFF));
int pop4 = hamming(non_whitespace);
_mm_storeu_si128(reinterpret_cast<__m128i *>(out), x1);
_mm_storeu_si128(reinterpret_cast<__m128i *>(out + pop1), x2);
_mm_storeu_si128(reinterpret_cast<__m128i *>(out + pop2), x3);
_mm_storeu_si128(reinterpret_cast<__m128i *>(out + pop3), x4);
out += pop4;
}
}
// we finish off the job... copying and pasting the code is not ideal here,
// but it gets the job done.
if (idx < len) {
uint8_t buffer[64];
memset(buffer, 0, 64);
memcpy(buffer, buf + idx, len - idx);
__m256i input_lo =
_mm256_loadu_si256(reinterpret_cast<const __m256i *>(buffer));
__m256i input_hi =
_mm256_loadu_si256(reinterpret_cast<const __m256i *>(buffer + 32));
uint64_t bs_bits =
cmp_mask_against_input_mini(input_lo, input_hi, _mm256_set1_epi8('\\'));
uint64_t start_edges = bs_bits & ~(bs_bits << 1);
uint64_t even_start_mask = even_bits ^ prev_iter_ends_odd_backslash;
uint64_t even_starts = start_edges & even_start_mask;
uint64_t odd_starts = start_edges & ~even_start_mask;
uint64_t even_carries = bs_bits + even_starts;
uint64_t odd_carries;
// bool iter_ends_odd_backslash =
add_overflow(bs_bits, odd_starts, &odd_carries);
odd_carries |= prev_iter_ends_odd_backslash;
// prev_iter_ends_odd_backslash = iter_ends_odd_backslash ? 0x1ULL : 0x0ULL;
// // we never use it
uint64_t even_carry_ends = even_carries & ~bs_bits;
uint64_t odd_carry_ends = odd_carries & ~bs_bits;
uint64_t even_start_odd_end = even_carry_ends & odd_bits;
uint64_t odd_start_even_end = odd_carry_ends & even_bits;
uint64_t odd_ends = even_start_odd_end | odd_start_even_end;
uint64_t quote_bits =
cmp_mask_against_input_mini(input_lo, input_hi, _mm256_set1_epi8('"'));
quote_bits = quote_bits & ~odd_ends;
uint64_t quote_mask = _mm_cvtsi128_si64(_mm_clmulepi64_si128(
_mm_set_epi64x(0ULL, quote_bits), _mm_set1_epi8(0xFF), 0));
quote_mask ^= prev_iter_inside_quote;
// prev_iter_inside_quote = (uint64_t)((int64_t)quote_mask >> 63);// we
// don't need this anymore
__m256i mask_20 = _mm256_set1_epi8(0x20); // c==32
__m256i mask_70 =
_mm256_set1_epi8(0x70); // adding 0x70 does not check low 4-bits
// but moves any value >= 16 above 128
__m256i lut_cntrl = _mm256_setr_epi8(
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0x00,
0x00, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0xFF, 0xFF, 0x00, 0x00, 0xFF, 0x00, 0x00);
__m256i tmp_ws_lo = _mm256_or_si256(
_mm256_cmpeq_epi8(mask_20, input_lo),
_mm256_shuffle_epi8(lut_cntrl, _mm256_adds_epu8(mask_70, input_lo)));
__m256i tmp_ws_hi = _mm256_or_si256(
_mm256_cmpeq_epi8(mask_20, input_hi),
_mm256_shuffle_epi8(lut_cntrl, _mm256_adds_epu8(mask_70, input_hi)));
uint64_t ws_res_0 = static_cast<uint32_t>(_mm256_movemask_epi8(tmp_ws_lo));
uint64_t ws_res_1 = _mm256_movemask_epi8(tmp_ws_hi);
uint64_t whitespace = (ws_res_0 | (ws_res_1 << 32));
whitespace &= ~quote_mask;
if (len - idx < 64) {
whitespace |= UINT64_C(0xFFFFFFFFFFFFFFFF) << (len - idx);
}
int mask1 = non_whitespace & 0xFFFF;
int mask2 = (non_whitespace >> 16) & 0xFFFF;
int mask3 = (non_whitespace >> 32) & 0xFFFF;
int mask4 = (non_whitespace >> 48) & 0xFFFF;
x1 = skinnycleanm128(x1, mask1);
x2 = skinnycleanm128(x2, mask2);
x3 = skinnycleanm128(x3, mask3);
x4 = skinnycleanm128(x4, mask4);
int pop1 = hamming(non_whitespace & 0xFFFF);
int pop2 = hamming(non_whitespace & UINT64_C(0xFFFFFFFF));
int pop3 = hamming(non_whitespace) & UINT64_C(0xFFFFFFFFFFFF));
int pop4 = hamming(non_whitespace);
_mm_storeu_si128(reinterpret_cast<__m128i *>(out), x1);
_mm_storeu_si128(reinterpret_cast<__m128i *>(out + pop1), x2);
_mm_storeu_si128(reinterpret_cast<__m128i *>(out + pop2), x3);
_mm_storeu_si128(reinterpret_cast<__m128i *>(out + pop3), x4);
out += pop4;
}
*out = '\0'; // NULL termination
return out - initout;
}
size_t oldjson_minify(const uint8_t *buf, size_t len, uint8_t *out) {
// Useful constant masks
const uint64_t even_bits = 0x5555555555555555ULL;
const uint64_t odd_bits = ~even_bits;
uint8_t *initout(out);
uint64_t prev_iter_ends_odd_backslash =
0ULL; // either 0 or 1, but a 64-bit value
uint64_t prev_iter_inside_quote = 0ULL; // either all zeros or all ones
size_t idx = 0;
if (len >= 64) {
size_t avx_len = len - 63;
for (; idx < avx_len; idx += 64) {
__m256i input_lo =
_mm256_loadu_si256(reinterpret_cast<const __m256i *>(buf + idx + 0));
__m256i input_hi =
_mm256_loadu_si256(reinterpret_cast<const __m256i *>(buf + idx + 32));
uint64_t bs_bits = cmp_mask_against_input_mini(input_lo, input_hi,
_mm256_set1_epi8('\\'));
uint64_t start_edges = bs_bits & ~(bs_bits << 1);
uint64_t even_start_mask = even_bits ^ prev_iter_ends_odd_backslash;
uint64_t even_starts = start_edges & even_start_mask;
uint64_t odd_starts = start_edges & ~even_start_mask;
uint64_t even_carries = bs_bits + even_starts;
uint64_t odd_carries;
bool iter_ends_odd_backslash =
add_overflow(bs_bits, odd_starts, &odd_carries);
odd_carries |= prev_iter_ends_odd_backslash;
prev_iter_ends_odd_backslash = iter_ends_odd_backslash ? 0x1ULL : 0x0ULL;
uint64_t even_carry_ends = even_carries & ~bs_bits;
uint64_t odd_carry_ends = odd_carries & ~bs_bits;
uint64_t even_start_odd_end = even_carry_ends & odd_bits;
uint64_t odd_start_even_end = odd_carry_ends & even_bits;
uint64_t odd_ends = even_start_odd_end | odd_start_even_end;
uint64_t quote_bits = cmp_mask_against_input_mini(input_lo, input_hi,
_mm256_set1_epi8('"'));
quote_bits = quote_bits & ~odd_ends;
uint64_t quote_mask = _mm_cvtsi128_si64(_mm_clmulepi64_si128(
_mm_set_epi64x(0ULL, quote_bits), _mm_set1_epi8(0xFF), 0));
quote_mask ^= prev_iter_inside_quote;
prev_iter_inside_quote = static_cast<uint64_t>(
static_cast<int64_t>(quote_mask) >>
63); // might be undefined behavior, should be fully defined in C++20,
// ok according to John Regher from Utah University
const __m256i low_nibble_mask = _mm256_setr_epi8(
// 0 9 a b c d
16, 0, 0, 0, 0, 0, 0, 0, 0, 8, 12, 1, 2, 9, 0, 0, 16, 0, 0, 0, 0, 0,
0, 0, 0, 8, 12, 1, 2, 9, 0, 0);
const __m256i high_nibble_mask = _mm256_setr_epi8(
// 0 2 3 5 7
8, 0, 18, 4, 0, 1, 0, 1, 0, 0, 0, 3, 2, 1, 0, 0, 8, 0, 18, 4, 0, 1, 0,
1, 0, 0, 0, 3, 2, 1, 0, 0);
__m256i whitespace_shufti_mask = _mm256_set1_epi8(0x18);
__m256i v_lo = _mm256_and_si256(
_mm256_shuffle_epi8(low_nibble_mask, input_lo),
_mm256_shuffle_epi8(high_nibble_mask,
_mm256_and_si256(_mm256_srli_epi32(input_lo, 4),
_mm256_set1_epi8(0x7f))));
__m256i v_hi = _mm256_and_si256(
_mm256_shuffle_epi8(low_nibble_mask, input_hi),
_mm256_shuffle_epi8(high_nibble_mask,
_mm256_and_si256(_mm256_srli_epi32(input_hi, 4),
_mm256_set1_epi8(0x7f))));
__m256i tmp_ws_lo = _mm256_cmpeq_epi8(
_mm256_and_si256(v_lo, whitespace_shufti_mask), _mm256_set1_epi8(0));
__m256i tmp_ws_hi = _mm256_cmpeq_epi8(
_mm256_and_si256(v_hi, whitespace_shufti_mask), _mm256_set1_epi8(0));
uint64_t ws_res_0 =
static_cast<uint32_t>(_mm256_movemask_epi8(tmp_ws_lo));
uint64_t ws_res_1 = _mm256_movemask_epi8(tmp_ws_hi);
uint64_t whitespace = ~(ws_res_0 | (ws_res_1 << 32));
whitespace &= ~quote_mask;
int mask1 = whitespace & 0xFFFF;
int mask2 = (whitespace >> 16) & 0xFFFF;
int mask3 = (whitespace >> 32) & 0xFFFF;
int mask4 = (whitespace >> 48) & 0xFFFF;
int pop1 = hamming((~whitespace) & 0xFFFF);
int pop2 = hamming((~whitespace) & UINT64_C(0xFFFFFFFF));
int pop3 = hamming((~whitespace) & UINT64_C(0xFFFFFFFFFFFF));
int pop4 = hamming((~whitespace));
__m256i vmask1 = _mm256_loadu2_m128i(
reinterpret_cast<const __m128i *>(mask128_epi8) + (mask2 & 0x7FFF),
reinterpret_cast<const __m128i *>(mask128_epi8) + (mask1 & 0x7FFF));
__m256i vmask2 = _mm256_loadu2_m128i(
reinterpret_cast<const __m128i *>(mask128_epi8) + (mask4 & 0x7FFF),
reinterpret_cast<const __m128i *>(mask128_epi8) + (mask3 & 0x7FFF));
__m256i result1 = _mm256_shuffle_epi8(input_lo, vmask1);
__m256i result2 = _mm256_shuffle_epi8(input_hi, vmask2);
_mm256_storeu2_m128i(reinterpret_cast<__m128i *>(out + pop1),
reinterpret_cast<__m128i *>(out), result1);
_mm256_storeu2_m128i(reinterpret_cast<__m128i *>(out + pop3),
reinterpret_cast<__m128i *>(out + pop2), result2);
out += pop4;
}
}
// we finish off the job... copying and pasting the code is not ideal here,
// but it gets the job done.
if (idx < len) {
uint8_t buffer[64];
memset(buffer, 0, 64);
memcpy(buffer, buf + idx, len - idx);
__m256i input_lo =
_mm256_loadu_si256(reinterpret_cast<const __m256i *>(buffer));
__m256i input_hi =
_mm256_loadu_si256(reinterpret_cast<const __m256i *>(buffer + 32));
uint64_t bs_bits =
cmp_mask_against_input_mini(input_lo, input_hi, _mm256_set1_epi8('\\'));
uint64_t start_edges = bs_bits & ~(bs_bits << 1);
uint64_t even_start_mask = even_bits ^ prev_iter_ends_odd_backslash;
uint64_t even_starts = start_edges & even_start_mask;
uint64_t odd_starts = start_edges & ~even_start_mask;
uint64_t even_carries = bs_bits + even_starts;
uint64_t odd_carries;
// bool iter_ends_odd_backslash =
add_overflow(bs_bits, odd_starts, &odd_carries);
odd_carries |= prev_iter_ends_odd_backslash;
// prev_iter_ends_odd_backslash = iter_ends_odd_backslash ? 0x1ULL : 0x0ULL;
// // we never use it
uint64_t even_carry_ends = even_carries & ~bs_bits;
uint64_t odd_carry_ends = odd_carries & ~bs_bits;
uint64_t even_start_odd_end = even_carry_ends & odd_bits;
uint64_t odd_start_even_end = odd_carry_ends & even_bits;
uint64_t odd_ends = even_start_odd_end | odd_start_even_end;
uint64_t quote_bits =
cmp_mask_against_input_mini(input_lo, input_hi, _mm256_set1_epi8('"'));
quote_bits = quote_bits & ~odd_ends;
uint64_t quote_mask = _mm_cvtsi128_si64(_mm_clmulepi64_si128(
_mm_set_epi64x(0ULL, quote_bits), _mm_set1_epi8(0xFF), 0));
quote_mask ^= prev_iter_inside_quote;
// prev_iter_inside_quote = (uint64_t)((int64_t)quote_mask >> 63);// we
// don't need this anymore
__m256i mask_20 = _mm256_set1_epi8(0x20); // c==32
__m256i mask_70 =
_mm256_set1_epi8(0x70); // adding 0x70 does not check low 4-bits
// but moves any value >= 16 above 128
__m256i lut_cntrl = _mm256_setr_epi8(
0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xFF, 0x00,
0x00, 0xFF, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
0x00, 0xFF, 0xFF, 0x00, 0x00, 0xFF, 0x00, 0x00);
__m256i tmp_ws_lo = _mm256_or_si256(
_mm256_cmpeq_epi8(mask_20, input_lo),
_mm256_shuffle_epi8(lut_cntrl, _mm256_adds_epu8(mask_70, input_lo)));
__m256i tmp_ws_hi = _mm256_or_si256(
_mm256_cmpeq_epi8(mask_20, input_hi),
_mm256_shuffle_epi8(lut_cntrl, _mm256_adds_epu8(mask_70, input_hi)));
uint64_t ws_res_0 = static_cast<uint32_t>(_mm256_movemask_epi8(tmp_ws_lo));
uint64_t ws_res_1 = _mm256_movemask_epi8(tmp_ws_hi);
uint64_t whitespace = (ws_res_0 | (ws_res_1 << 32));
whitespace &= ~quote_mask;
if (len - idx < 64) {
whitespace |= UINT64_C(0xFFFFFFFFFFFFFFFF) << (len - idx);
}
int mask1 = whitespace & 0xFFFF;
int mask2 = (whitespace >> 16) & 0xFFFF;
int mask3 = (whitespace >> 32) & 0xFFFF;
int mask4 = (whitespace >> 48) & 0xFFFF;
int pop1 = hamming((~whitespace) & 0xFFFF);
int pop2 = hamming((~whitespace) & UINT64_C(0xFFFFFFFF));
int pop3 = hamming((~whitespace) & UINT64_C(0xFFFFFFFFFFFF));
int pop4 = hamming((~whitespace));
__m256i vmask1 = _mm256_loadu2_m128i(
reinterpret_cast<const __m128i *>(mask128_epi8) + (mask2 & 0x7FFF),
reinterpret_cast<const __m128i *>(mask128_epi8) + (mask1 & 0x7FFF));
__m256i vmask2 = _mm256_loadu2_m128i(
reinterpret_cast<const __m128i *>(mask128_epi8) + (mask4 & 0x7FFF),
reinterpret_cast<const __m128i *>(mask128_epi8) + (mask3 & 0x7FFF));
__m256i result1 = _mm256_shuffle_epi8(input_lo, vmask1);
__m256i result2 = _mm256_shuffle_epi8(input_hi, vmask2);
_mm256_storeu2_m128i(reinterpret_cast<__m128i *>(buffer + pop1),
reinterpret_cast<__m128i *>(buffer), result1);
_mm256_storeu2_m128i(reinterpret_cast<__m128i *>(buffer + pop3),
reinterpret_cast<__m128i *>(buffer + pop2), result2);
memcpy(out, buffer, pop4);
out += pop4;
}
*out = '\0'; // NULL termination
return out - initout;
}
} // namespace simdjson
#endif