forked from DataDog/zstd
-
Notifications
You must be signed in to change notification settings - Fork 1
/
huf_decompress_amd64.S
595 lines (505 loc) · 14.7 KB
/
huf_decompress_amd64.S
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
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
/*
* Copyright (c) Meta Platforms, Inc. and affiliates.
* All rights reserved.
*
* This source code is licensed under both the BSD-style license (found in the
* LICENSE file in the root directory of this source tree) and the GPLv2 (found
* in the COPYING file in the root directory of this source tree).
* You may select, at your option, one of the above-listed licenses.
*/
#include "portability_macros.h"
#if defined(__ELF__) && defined(__GNUC__)
/* Stack marking
* ref: https://wiki.gentoo.org/wiki/Hardened/GNU_stack_quickstart
*/
.section .note.GNU-stack,"",%progbits
#if defined(__aarch64__)
/* Mark that this assembly supports BTI & PAC, because it is empty for aarch64.
* See: https://github.com/facebook/zstd/issues/3841
* See: https://gcc.godbolt.org/z/sqr5T4ffK
* See: https://lore.kernel.org/linux-arm-kernel/[email protected]/
* See: https://reviews.llvm.org/D62609
*/
.pushsection .note.gnu.property, "a"
.p2align 3
.long 4 /* size of the name - "GNU\0" */
.long 0x10 /* size of descriptor */
.long 0x5 /* NT_GNU_PROPERTY_TYPE_0 */
.asciz "GNU"
.long 0xc0000000 /* pr_type - GNU_PROPERTY_AARCH64_FEATURE_1_AND */
.long 4 /* pr_datasz - 4 bytes */
.long 3 /* pr_data - GNU_PROPERTY_AARCH64_FEATURE_1_BTI | GNU_PROPERTY_AARCH64_FEATURE_1_PAC */
.p2align 3 /* pr_padding - bring everything to 8 byte alignment */
.popsection
#endif
#endif
#if ZSTD_ENABLE_ASM_X86_64_BMI2
/* Calling convention:
*
* %rdi contains the first argument: HUF_DecompressAsmArgs*.
* %rbp isn't maintained (no frame pointer).
* %rsp contains the stack pointer that grows down.
* No red-zone is assumed, only addresses >= %rsp are used.
* All register contents are preserved.
*
* TODO: Support Windows calling convention.
*/
ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X1_usingDTable_internal_fast_asm_loop)
ZSTD_HIDE_ASM_FUNCTION(HUF_decompress4X2_usingDTable_internal_fast_asm_loop)
ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X2_usingDTable_internal_fast_asm_loop)
ZSTD_HIDE_ASM_FUNCTION(_HUF_decompress4X1_usingDTable_internal_fast_asm_loop)
.global HUF_decompress4X1_usingDTable_internal_fast_asm_loop
.global HUF_decompress4X2_usingDTable_internal_fast_asm_loop
.global _HUF_decompress4X1_usingDTable_internal_fast_asm_loop
.global _HUF_decompress4X2_usingDTable_internal_fast_asm_loop
.text
/* Sets up register mappings for clarity.
* op[], bits[], dtable & ip[0] each get their own register.
* ip[1,2,3] & olimit alias var[].
* %rax is a scratch register.
*/
#define op0 rsi
#define op1 rbx
#define op2 rcx
#define op3 rdi
#define ip0 r8
#define ip1 r9
#define ip2 r10
#define ip3 r11
#define bits0 rbp
#define bits1 rdx
#define bits2 r12
#define bits3 r13
#define dtable r14
#define olimit r15
/* var[] aliases ip[1,2,3] & olimit
* ip[1,2,3] are saved every iteration.
* olimit is only used in compute_olimit.
*/
#define var0 r15
#define var1 r9
#define var2 r10
#define var3 r11
/* 32-bit var registers */
#define vard0 r15d
#define vard1 r9d
#define vard2 r10d
#define vard3 r11d
/* Calls X(N) for each stream 0, 1, 2, 3. */
#define FOR_EACH_STREAM(X) \
X(0); \
X(1); \
X(2); \
X(3)
/* Calls X(N, idx) for each stream 0, 1, 2, 3. */
#define FOR_EACH_STREAM_WITH_INDEX(X, idx) \
X(0, idx); \
X(1, idx); \
X(2, idx); \
X(3, idx)
/* Define both _HUF_* & HUF_* symbols because MacOS
* C symbols are prefixed with '_' & Linux symbols aren't.
*/
_HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
HUF_decompress4X1_usingDTable_internal_fast_asm_loop:
ZSTD_CET_ENDBRANCH
/* Save all registers - even if they are callee saved for simplicity. */
push %rax
push %rbx
push %rcx
push %rdx
push %rbp
push %rsi
push %rdi
push %r8
push %r9
push %r10
push %r11
push %r12
push %r13
push %r14
push %r15
/* Read HUF_DecompressAsmArgs* args from %rax */
movq %rdi, %rax
movq 0(%rax), %ip0
movq 8(%rax), %ip1
movq 16(%rax), %ip2
movq 24(%rax), %ip3
movq 32(%rax), %op0
movq 40(%rax), %op1
movq 48(%rax), %op2
movq 56(%rax), %op3
movq 64(%rax), %bits0
movq 72(%rax), %bits1
movq 80(%rax), %bits2
movq 88(%rax), %bits3
movq 96(%rax), %dtable
push %rax /* argument */
push 104(%rax) /* ilowest */
push 112(%rax) /* oend */
push %olimit /* olimit space */
subq $24, %rsp
.L_4X1_compute_olimit:
/* Computes how many iterations we can do safely
* %r15, %rax may be clobbered
* rbx, rdx must be saved
* op3 & ip0 mustn't be clobbered
*/
movq %rbx, 0(%rsp)
movq %rdx, 8(%rsp)
movq 32(%rsp), %rax /* rax = oend */
subq %op3, %rax /* rax = oend - op3 */
/* r15 = (oend - op3) / 5 */
movabsq $-3689348814741910323, %rdx
mulq %rdx
movq %rdx, %r15
shrq $2, %r15
movq %ip0, %rax /* rax = ip0 */
movq 40(%rsp), %rdx /* rdx = ilowest */
subq %rdx, %rax /* rax = ip0 - ilowest */
movq %rax, %rbx /* rbx = ip0 - ilowest */
/* rdx = (ip0 - ilowest) / 7 */
movabsq $2635249153387078803, %rdx
mulq %rdx
subq %rdx, %rbx
shrq %rbx
addq %rbx, %rdx
shrq $2, %rdx
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
/* r15 = r15 * 5 */
leaq (%r15, %r15, 4), %r15
/* olimit = op3 + r15 */
addq %op3, %olimit
movq 8(%rsp), %rdx
movq 0(%rsp), %rbx
/* If (op3 + 20 > olimit) */
movq %op3, %rax /* rax = op3 */
cmpq %rax, %olimit /* op3 == olimit */
je .L_4X1_exit
/* If (ip1 < ip0) go to exit */
cmpq %ip0, %ip1
jb .L_4X1_exit
/* If (ip2 < ip1) go to exit */
cmpq %ip1, %ip2
jb .L_4X1_exit
/* If (ip3 < ip2) go to exit */
cmpq %ip2, %ip3
jb .L_4X1_exit
/* Reads top 11 bits from bits[n]
* Loads dt[bits[n]] into var[n]
*/
#define GET_NEXT_DELT(n) \
movq $53, %var##n; \
shrxq %var##n, %bits##n, %var##n; \
movzwl (%dtable,%var##n,2),%vard##n
/* var[n] must contain the DTable entry computed with GET_NEXT_DELT
* Moves var[n] to %rax
* bits[n] <<= var[n] & 63
* op[n][idx] = %rax >> 8
* %ah is a way to access bits [8, 16) of %rax
*/
#define DECODE_FROM_DELT(n, idx) \
movq %var##n, %rax; \
shlxq %var##n, %bits##n, %bits##n; \
movb %ah, idx(%op##n)
/* Assumes GET_NEXT_DELT has been called.
* Calls DECODE_FROM_DELT then GET_NEXT_DELT
*/
#define DECODE_AND_GET_NEXT(n, idx) \
DECODE_FROM_DELT(n, idx); \
GET_NEXT_DELT(n) \
/* // ctz & nbBytes is stored in bits[n]
* // nbBits is stored in %rax
* ctz = CTZ[bits[n]]
* nbBits = ctz & 7
* nbBytes = ctz >> 3
* op[n] += 5
* ip[n] -= nbBytes
* // Note: x86-64 is little-endian ==> no bswap
* bits[n] = MEM_readST(ip[n]) | 1
* bits[n] <<= nbBits
*/
#define RELOAD_BITS(n) \
bsfq %bits##n, %bits##n; \
movq %bits##n, %rax; \
andq $7, %rax; \
shrq $3, %bits##n; \
leaq 5(%op##n), %op##n; \
subq %bits##n, %ip##n; \
movq (%ip##n), %bits##n; \
orq $1, %bits##n; \
shlx %rax, %bits##n, %bits##n
/* Store clobbered variables on the stack */
movq %olimit, 24(%rsp)
movq %ip1, 0(%rsp)
movq %ip2, 8(%rsp)
movq %ip3, 16(%rsp)
/* Call GET_NEXT_DELT for each stream */
FOR_EACH_STREAM(GET_NEXT_DELT)
.p2align 6
.L_4X1_loop_body:
/* Decode 5 symbols in each of the 4 streams (20 total)
* Must have called GET_NEXT_DELT for each stream
*/
FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 0)
FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 1)
FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 2)
FOR_EACH_STREAM_WITH_INDEX(DECODE_AND_GET_NEXT, 3)
FOR_EACH_STREAM_WITH_INDEX(DECODE_FROM_DELT, 4)
/* Load ip[1,2,3] from stack (var[] aliases them)
* ip[] is needed for RELOAD_BITS
* Each will be stored back to the stack after RELOAD
*/
movq 0(%rsp), %ip1
movq 8(%rsp), %ip2
movq 16(%rsp), %ip3
/* Reload each stream & fetch the next table entry
* to prepare for the next iteration
*/
RELOAD_BITS(0)
GET_NEXT_DELT(0)
RELOAD_BITS(1)
movq %ip1, 0(%rsp)
GET_NEXT_DELT(1)
RELOAD_BITS(2)
movq %ip2, 8(%rsp)
GET_NEXT_DELT(2)
RELOAD_BITS(3)
movq %ip3, 16(%rsp)
GET_NEXT_DELT(3)
/* If op3 < olimit: continue the loop */
cmp %op3, 24(%rsp)
ja .L_4X1_loop_body
/* Reload ip[1,2,3] from stack */
movq 0(%rsp), %ip1
movq 8(%rsp), %ip2
movq 16(%rsp), %ip3
/* Re-compute olimit */
jmp .L_4X1_compute_olimit
#undef GET_NEXT_DELT
#undef DECODE_FROM_DELT
#undef DECODE
#undef RELOAD_BITS
.L_4X1_exit:
addq $24, %rsp
/* Restore stack (oend & olimit) */
pop %rax /* olimit */
pop %rax /* oend */
pop %rax /* ilowest */
pop %rax /* arg */
/* Save ip / op / bits */
movq %ip0, 0(%rax)
movq %ip1, 8(%rax)
movq %ip2, 16(%rax)
movq %ip3, 24(%rax)
movq %op0, 32(%rax)
movq %op1, 40(%rax)
movq %op2, 48(%rax)
movq %op3, 56(%rax)
movq %bits0, 64(%rax)
movq %bits1, 72(%rax)
movq %bits2, 80(%rax)
movq %bits3, 88(%rax)
/* Restore registers */
pop %r15
pop %r14
pop %r13
pop %r12
pop %r11
pop %r10
pop %r9
pop %r8
pop %rdi
pop %rsi
pop %rbp
pop %rdx
pop %rcx
pop %rbx
pop %rax
ret
_HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
HUF_decompress4X2_usingDTable_internal_fast_asm_loop:
ZSTD_CET_ENDBRANCH
/* Save all registers - even if they are callee saved for simplicity. */
push %rax
push %rbx
push %rcx
push %rdx
push %rbp
push %rsi
push %rdi
push %r8
push %r9
push %r10
push %r11
push %r12
push %r13
push %r14
push %r15
movq %rdi, %rax
movq 0(%rax), %ip0
movq 8(%rax), %ip1
movq 16(%rax), %ip2
movq 24(%rax), %ip3
movq 32(%rax), %op0
movq 40(%rax), %op1
movq 48(%rax), %op2
movq 56(%rax), %op3
movq 64(%rax), %bits0
movq 72(%rax), %bits1
movq 80(%rax), %bits2
movq 88(%rax), %bits3
movq 96(%rax), %dtable
push %rax /* argument */
push %rax /* olimit */
push 104(%rax) /* ilowest */
movq 112(%rax), %rax
push %rax /* oend3 */
movq %op3, %rax
push %rax /* oend2 */
movq %op2, %rax
push %rax /* oend1 */
movq %op1, %rax
push %rax /* oend0 */
/* Scratch space */
subq $8, %rsp
.L_4X2_compute_olimit:
/* Computes how many iterations we can do safely
* %r15, %rax may be clobbered
* rdx must be saved
* op[1,2,3,4] & ip0 mustn't be clobbered
*/
movq %rdx, 0(%rsp)
/* We can consume up to 7 input bytes each iteration. */
movq %ip0, %rax /* rax = ip0 */
movq 40(%rsp), %rdx /* rdx = ilowest */
subq %rdx, %rax /* rax = ip0 - ilowest */
movq %rax, %r15 /* r15 = ip0 - ilowest */
/* rdx = rax / 7 */
movabsq $2635249153387078803, %rdx
mulq %rdx
subq %rdx, %r15
shrq %r15
addq %r15, %rdx
shrq $2, %rdx
/* r15 = (ip0 - ilowest) / 7 */
movq %rdx, %r15
/* r15 = min(r15, min(oend0 - op0, oend1 - op1, oend2 - op2, oend3 - op3) / 10) */
movq 8(%rsp), %rax /* rax = oend0 */
subq %op0, %rax /* rax = oend0 - op0 */
movq 16(%rsp), %rdx /* rdx = oend1 */
subq %op1, %rdx /* rdx = oend1 - op1 */
cmpq %rax, %rdx
cmova %rax, %rdx /* rdx = min(%rdx, %rax) */
movq 24(%rsp), %rax /* rax = oend2 */
subq %op2, %rax /* rax = oend2 - op2 */
cmpq %rax, %rdx
cmova %rax, %rdx /* rdx = min(%rdx, %rax) */
movq 32(%rsp), %rax /* rax = oend3 */
subq %op3, %rax /* rax = oend3 - op3 */
cmpq %rax, %rdx
cmova %rax, %rdx /* rdx = min(%rdx, %rax) */
movabsq $-3689348814741910323, %rax
mulq %rdx
shrq $3, %rdx /* rdx = rdx / 10 */
/* r15 = min(%rdx, %r15) */
cmpq %rdx, %r15
cmova %rdx, %r15
/* olimit = op3 + 5 * r15 */
movq %r15, %rax
leaq (%op3, %rax, 4), %olimit
addq %rax, %olimit
movq 0(%rsp), %rdx
/* If (op3 + 10 > olimit) */
movq %op3, %rax /* rax = op3 */
cmpq %rax, %olimit /* op3 == olimit */
je .L_4X2_exit
/* If (ip1 < ip0) go to exit */
cmpq %ip0, %ip1
jb .L_4X2_exit
/* If (ip2 < ip1) go to exit */
cmpq %ip1, %ip2
jb .L_4X2_exit
/* If (ip3 < ip2) go to exit */
cmpq %ip2, %ip3
jb .L_4X2_exit
#define DECODE(n, idx) \
movq %bits##n, %rax; \
shrq $53, %rax; \
movzwl 0(%dtable,%rax,4),%r8d; \
movzbl 2(%dtable,%rax,4),%r15d; \
movzbl 3(%dtable,%rax,4),%eax; \
movw %r8w, (%op##n); \
shlxq %r15, %bits##n, %bits##n; \
addq %rax, %op##n
#define RELOAD_BITS(n) \
bsfq %bits##n, %bits##n; \
movq %bits##n, %rax; \
shrq $3, %bits##n; \
andq $7, %rax; \
subq %bits##n, %ip##n; \
movq (%ip##n), %bits##n; \
orq $1, %bits##n; \
shlxq %rax, %bits##n, %bits##n
movq %olimit, 48(%rsp)
.p2align 6
.L_4X2_loop_body:
/* We clobber r8, so store it on the stack */
movq %r8, 0(%rsp)
/* Decode 5 symbols from each of the 4 streams (20 symbols total). */
FOR_EACH_STREAM_WITH_INDEX(DECODE, 0)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 1)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 2)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 3)
FOR_EACH_STREAM_WITH_INDEX(DECODE, 4)
/* Reload r8 */
movq 0(%rsp), %r8
FOR_EACH_STREAM(RELOAD_BITS)
cmp %op3, 48(%rsp)
ja .L_4X2_loop_body
jmp .L_4X2_compute_olimit
#undef DECODE
#undef RELOAD_BITS
.L_4X2_exit:
addq $8, %rsp
/* Restore stack (oend & olimit) */
pop %rax /* oend0 */
pop %rax /* oend1 */
pop %rax /* oend2 */
pop %rax /* oend3 */
pop %rax /* ilowest */
pop %rax /* olimit */
pop %rax /* arg */
/* Save ip / op / bits */
movq %ip0, 0(%rax)
movq %ip1, 8(%rax)
movq %ip2, 16(%rax)
movq %ip3, 24(%rax)
movq %op0, 32(%rax)
movq %op1, 40(%rax)
movq %op2, 48(%rax)
movq %op3, 56(%rax)
movq %bits0, 64(%rax)
movq %bits1, 72(%rax)
movq %bits2, 80(%rax)
movq %bits3, 88(%rax)
/* Restore registers */
pop %r15
pop %r14
pop %r13
pop %r12
pop %r11
pop %r10
pop %r9
pop %r8
pop %rdi
pop %rsi
pop %rbp
pop %rdx
pop %rcx
pop %rbx
pop %rax
ret
#endif