aboutsummaryrefslogtreecommitdiff
path: root/tcg/tcg-op-vec.c
blob: 84af210bc0b9f931e4f874bccdb43fb55af0181a (plain)
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
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
/*
 * Tiny Code Generator for QEMU
 *
 * Copyright (c) 2018 Linaro, Inc.
 *
 * This library is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * This library 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
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
 */

#include "qemu/osdep.h"
#include "tcg/tcg.h"
#include "tcg/tcg-temp-internal.h"
#include "tcg/tcg-op-common.h"
#include "tcg/tcg-mo.h"
#include "tcg-internal.h"

/*
 * Vector optional opcode tracking.
 * Except for the basic logical operations (and, or, xor), and
 * data movement (mov, ld, st, dupi), many vector opcodes are
 * optional and may not be supported on the host.  Thank Intel
 * for the irregularity in their instruction set.
 *
 * The gvec expanders allow custom vector operations to be composed,
 * generally via the .fniv callback in the GVecGen* structures.  At
 * the same time, in deciding whether to use this hook we need to
 * know if the host supports the required operations.  This is
 * presented as an array of opcodes, terminated by 0.  Each opcode
 * is assumed to be expanded with the given VECE.
 *
 * For debugging, we want to validate this array.  Therefore, when
 * tcg_ctx->vec_opt_opc is non-NULL, the tcg_gen_*_vec expanders
 * will validate that their opcode is present in the list.
 */
static void tcg_assert_listed_vecop(TCGOpcode op)
{
#ifdef CONFIG_DEBUG_TCG
    const TCGOpcode *p = tcg_ctx->vecop_list;
    if (p) {
        for (; *p; ++p) {
            if (*p == op) {
                return;
            }
        }
        g_assert_not_reached();
    }
#endif
}

bool tcg_can_emit_vecop_list(const TCGOpcode *list,
                             TCGType type, unsigned vece)
{
    if (list == NULL) {
        return true;
    }

    for (; *list; ++list) {
        TCGOpcode opc = *list;

#ifdef CONFIG_DEBUG_TCG
        switch (opc) {
        case INDEX_op_and_vec:
        case INDEX_op_or_vec:
        case INDEX_op_xor_vec:
        case INDEX_op_mov_vec:
        case INDEX_op_dup_vec:
        case INDEX_op_dup2_vec:
        case INDEX_op_ld_vec:
        case INDEX_op_st_vec:
        case INDEX_op_bitsel_vec:
            /* These opcodes are mandatory and should not be listed.  */
            g_assert_not_reached();
        case INDEX_op_not_vec:
            /* These opcodes have generic expansions using the above.  */
            g_assert_not_reached();
        default:
            break;
        }
#endif

        if (tcg_can_emit_vec_op(opc, type, vece)) {
            continue;
        }

        /*
         * The opcode list is created by front ends based on what they
         * actually invoke.  We must mirror the logic in the routines
         * below for generic expansions using other opcodes.
         */
        switch (opc) {
        case INDEX_op_neg_vec:
            if (tcg_can_emit_vec_op(INDEX_op_sub_vec, type, vece)) {
                continue;
            }
            break;
        case INDEX_op_abs_vec:
            if (tcg_can_emit_vec_op(INDEX_op_sub_vec, type, vece)
                && (tcg_can_emit_vec_op(INDEX_op_smax_vec, type, vece) > 0
                    || tcg_can_emit_vec_op(INDEX_op_sari_vec, type, vece) > 0
                    || tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece))) {
                continue;
            }
            break;
        case INDEX_op_usadd_vec:
            if (tcg_can_emit_vec_op(INDEX_op_umin_vec, type, vece) ||
                tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece)) {
                continue;
            }
            break;
        case INDEX_op_ussub_vec:
            if (tcg_can_emit_vec_op(INDEX_op_umax_vec, type, vece) ||
                tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece)) {
                continue;
            }
            break;
        case INDEX_op_cmpsel_vec:
        case INDEX_op_smin_vec:
        case INDEX_op_smax_vec:
        case INDEX_op_umin_vec:
        case INDEX_op_umax_vec:
            if (tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece)) {
                continue;
            }
            break;
        default:
            break;
        }
        return false;
    }
    return true;
}

void vec_gen_2(TCGOpcode opc, TCGType type, unsigned vece, TCGArg r, TCGArg a)
{
    TCGOp *op = tcg_emit_op(opc, 2);
    TCGOP_VECL(op) = type - TCG_TYPE_V64;
    TCGOP_VECE(op) = vece;
    op->args[0] = r;
    op->args[1] = a;
}

void vec_gen_3(TCGOpcode opc, TCGType type, unsigned vece,
               TCGArg r, TCGArg a, TCGArg b)
{
    TCGOp *op = tcg_emit_op(opc, 3);
    TCGOP_VECL(op) = type - TCG_TYPE_V64;
    TCGOP_VECE(op) = vece;
    op->args[0] = r;
    op->args[1] = a;
    op->args[2] = b;
}

void vec_gen_4(TCGOpcode opc, TCGType type, unsigned vece,
               TCGArg r, TCGArg a, TCGArg b, TCGArg c)
{
    TCGOp *op = tcg_emit_op(opc, 4);
    TCGOP_VECL(op) = type - TCG_TYPE_V64;
    TCGOP_VECE(op) = vece;
    op->args[0] = r;
    op->args[1] = a;
    op->args[2] = b;
    op->args[3] = c;
}

static void vec_gen_6(TCGOpcode opc, TCGType type, unsigned vece, TCGArg r,
                      TCGArg a, TCGArg b, TCGArg c, TCGArg d, TCGArg e)
{
    TCGOp *op = tcg_emit_op(opc, 6);
    TCGOP_VECL(op) = type - TCG_TYPE_V64;
    TCGOP_VECE(op) = vece;
    op->args[0] = r;
    op->args[1] = a;
    op->args[2] = b;
    op->args[3] = c;
    op->args[4] = d;
    op->args[5] = e;
}

static void vec_gen_op2(TCGOpcode opc, unsigned vece, TCGv_vec r, TCGv_vec a)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    TCGTemp *at = tcgv_vec_temp(a);
    TCGType type = rt->base_type;

    /* Must enough inputs for the output.  */
    tcg_debug_assert(at->base_type >= type);
    vec_gen_2(opc, type, vece, temp_arg(rt), temp_arg(at));
}

static void vec_gen_op3(TCGOpcode opc, unsigned vece,
                        TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    TCGTemp *at = tcgv_vec_temp(a);
    TCGTemp *bt = tcgv_vec_temp(b);
    TCGType type = rt->base_type;

    /* Must enough inputs for the output.  */
    tcg_debug_assert(at->base_type >= type);
    tcg_debug_assert(bt->base_type >= type);
    vec_gen_3(opc, type, vece, temp_arg(rt), temp_arg(at), temp_arg(bt));
}

void tcg_gen_mov_vec(TCGv_vec r, TCGv_vec a)
{
    if (r != a) {
        vec_gen_op2(INDEX_op_mov_vec, 0, r, a);
    }
}

void tcg_gen_dupi_vec(unsigned vece, TCGv_vec r, uint64_t a)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    tcg_gen_mov_vec(r, tcg_constant_vec(rt->base_type, vece, a));
}

void tcg_gen_dup_i64_vec(unsigned vece, TCGv_vec r, TCGv_i64 a)
{
    TCGArg ri = tcgv_vec_arg(r);
    TCGTemp *rt = arg_temp(ri);
    TCGType type = rt->base_type;

    if (TCG_TARGET_REG_BITS == 64) {
        TCGArg ai = tcgv_i64_arg(a);
        vec_gen_2(INDEX_op_dup_vec, type, vece, ri, ai);
    } else if (vece == MO_64) {
        TCGArg al = tcgv_i32_arg(TCGV_LOW(a));
        TCGArg ah = tcgv_i32_arg(TCGV_HIGH(a));
        vec_gen_3(INDEX_op_dup2_vec, type, MO_64, ri, al, ah);
    } else {
        TCGArg ai = tcgv_i32_arg(TCGV_LOW(a));
        vec_gen_2(INDEX_op_dup_vec, type, vece, ri, ai);
    }
}

void tcg_gen_dup_i32_vec(unsigned vece, TCGv_vec r, TCGv_i32 a)
{
    TCGArg ri = tcgv_vec_arg(r);
    TCGArg ai = tcgv_i32_arg(a);
    TCGTemp *rt = arg_temp(ri);
    TCGType type = rt->base_type;

    vec_gen_2(INDEX_op_dup_vec, type, vece, ri, ai);
}

void tcg_gen_dup_mem_vec(unsigned vece, TCGv_vec r, TCGv_ptr b,
                         tcg_target_long ofs)
{
    TCGArg ri = tcgv_vec_arg(r);
    TCGArg bi = tcgv_ptr_arg(b);
    TCGTemp *rt = arg_temp(ri);
    TCGType type = rt->base_type;

    vec_gen_3(INDEX_op_dupm_vec, type, vece, ri, bi, ofs);
}

static void vec_gen_ldst(TCGOpcode opc, TCGv_vec r, TCGv_ptr b, TCGArg o)
{
    TCGArg ri = tcgv_vec_arg(r);
    TCGArg bi = tcgv_ptr_arg(b);
    TCGTemp *rt = arg_temp(ri);
    TCGType type = rt->base_type;

    vec_gen_3(opc, type, 0, ri, bi, o);
}

void tcg_gen_ld_vec(TCGv_vec r, TCGv_ptr b, TCGArg o)
{
    vec_gen_ldst(INDEX_op_ld_vec, r, b, o);
}

void tcg_gen_st_vec(TCGv_vec r, TCGv_ptr b, TCGArg o)
{
    vec_gen_ldst(INDEX_op_st_vec, r, b, o);
}

void tcg_gen_stl_vec(TCGv_vec r, TCGv_ptr b, TCGArg o, TCGType low_type)
{
    TCGArg ri = tcgv_vec_arg(r);
    TCGArg bi = tcgv_ptr_arg(b);
    TCGTemp *rt = arg_temp(ri);
    TCGType type = rt->base_type;

    tcg_debug_assert(low_type >= TCG_TYPE_V64);
    tcg_debug_assert(low_type <= type);
    vec_gen_3(INDEX_op_st_vec, low_type, 0, ri, bi, o);
}

void tcg_gen_and_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    vec_gen_op3(INDEX_op_and_vec, 0, r, a, b);
}

void tcg_gen_or_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    vec_gen_op3(INDEX_op_or_vec, 0, r, a, b);
}

void tcg_gen_xor_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    vec_gen_op3(INDEX_op_xor_vec, 0, r, a, b);
}

void tcg_gen_andc_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    if (TCG_TARGET_HAS_andc_vec) {
        vec_gen_op3(INDEX_op_andc_vec, 0, r, a, b);
    } else {
        TCGv_vec t = tcg_temp_new_vec_matching(r);
        tcg_gen_not_vec(0, t, b);
        tcg_gen_and_vec(0, r, a, t);
        tcg_temp_free_vec(t);
    }
}

void tcg_gen_orc_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    if (TCG_TARGET_HAS_orc_vec) {
        vec_gen_op3(INDEX_op_orc_vec, 0, r, a, b);
    } else {
        TCGv_vec t = tcg_temp_new_vec_matching(r);
        tcg_gen_not_vec(0, t, b);
        tcg_gen_or_vec(0, r, a, t);
        tcg_temp_free_vec(t);
    }
}

void tcg_gen_nand_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    if (TCG_TARGET_HAS_nand_vec) {
        vec_gen_op3(INDEX_op_nand_vec, 0, r, a, b);
    } else {
        tcg_gen_and_vec(0, r, a, b);
        tcg_gen_not_vec(0, r, r);
    }
}

void tcg_gen_nor_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    if (TCG_TARGET_HAS_nor_vec) {
        vec_gen_op3(INDEX_op_nor_vec, 0, r, a, b);
    } else {
        tcg_gen_or_vec(0, r, a, b);
        tcg_gen_not_vec(0, r, r);
    }
}

void tcg_gen_eqv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    if (TCG_TARGET_HAS_eqv_vec) {
        vec_gen_op3(INDEX_op_eqv_vec, 0, r, a, b);
    } else {
        tcg_gen_xor_vec(0, r, a, b);
        tcg_gen_not_vec(0, r, r);
    }
}

static bool do_op2(unsigned vece, TCGv_vec r, TCGv_vec a, TCGOpcode opc)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    TCGTemp *at = tcgv_vec_temp(a);
    TCGArg ri = temp_arg(rt);
    TCGArg ai = temp_arg(at);
    TCGType type = rt->base_type;
    int can;

    tcg_debug_assert(at->base_type >= type);
    tcg_assert_listed_vecop(opc);
    can = tcg_can_emit_vec_op(opc, type, vece);
    if (can > 0) {
        vec_gen_2(opc, type, vece, ri, ai);
    } else if (can < 0) {
        const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
        tcg_expand_vec_op(opc, type, vece, ri, ai);
        tcg_swap_vecop_list(hold_list);
    } else {
        return false;
    }
    return true;
}

void tcg_gen_not_vec(unsigned vece, TCGv_vec r, TCGv_vec a)
{
    if (TCG_TARGET_HAS_not_vec) {
        vec_gen_op2(INDEX_op_not_vec, 0, r, a);
    } else {
        tcg_gen_xor_vec(0, r, a, tcg_constant_vec_matching(r, 0, -1));
    }
}

void tcg_gen_neg_vec(unsigned vece, TCGv_vec r, TCGv_vec a)
{
    const TCGOpcode *hold_list;

    tcg_assert_listed_vecop(INDEX_op_neg_vec);
    hold_list = tcg_swap_vecop_list(NULL);

    if (!TCG_TARGET_HAS_neg_vec || !do_op2(vece, r, a, INDEX_op_neg_vec)) {
        tcg_gen_sub_vec(vece, r, tcg_constant_vec_matching(r, vece, 0), a);
    }
    tcg_swap_vecop_list(hold_list);
}

void tcg_gen_abs_vec(unsigned vece, TCGv_vec r, TCGv_vec a)
{
    const TCGOpcode *hold_list;

    tcg_assert_listed_vecop(INDEX_op_abs_vec);
    hold_list = tcg_swap_vecop_list(NULL);

    if (!do_op2(vece, r, a, INDEX_op_abs_vec)) {
        TCGType type = tcgv_vec_temp(r)->base_type;
        TCGv_vec t = tcg_temp_new_vec(type);

        tcg_debug_assert(tcg_can_emit_vec_op(INDEX_op_sub_vec, type, vece));
        if (tcg_can_emit_vec_op(INDEX_op_smax_vec, type, vece) > 0) {
            tcg_gen_neg_vec(vece, t, a);
            tcg_gen_smax_vec(vece, r, a, t);
        } else {
            if (tcg_can_emit_vec_op(INDEX_op_sari_vec, type, vece) > 0) {
                tcg_gen_sari_vec(vece, t, a, (8 << vece) - 1);
            } else {
                tcg_gen_cmp_vec(TCG_COND_LT, vece, t, a,
                                tcg_constant_vec(type, vece, 0));
            }
            tcg_gen_xor_vec(vece, r, a, t);
            tcg_gen_sub_vec(vece, r, r, t);
        }

        tcg_temp_free_vec(t);
    }
    tcg_swap_vecop_list(hold_list);
}

static void do_shifti(TCGOpcode opc, unsigned vece,
                      TCGv_vec r, TCGv_vec a, int64_t i)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    TCGTemp *at = tcgv_vec_temp(a);
    TCGArg ri = temp_arg(rt);
    TCGArg ai = temp_arg(at);
    TCGType type = rt->base_type;
    int can;

    tcg_debug_assert(at->base_type == type);
    tcg_debug_assert(i >= 0 && i < (8 << vece));
    tcg_assert_listed_vecop(opc);

    if (i == 0) {
        tcg_gen_mov_vec(r, a);
        return;
    }

    can = tcg_can_emit_vec_op(opc, type, vece);
    if (can > 0) {
        vec_gen_3(opc, type, vece, ri, ai, i);
    } else {
        /* We leave the choice of expansion via scalar or vector shift
           to the target.  Often, but not always, dupi can feed a vector
           shift easier than a scalar.  */
        const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
        tcg_debug_assert(can < 0);
        tcg_expand_vec_op(opc, type, vece, ri, ai, i);
        tcg_swap_vecop_list(hold_list);
    }
}

void tcg_gen_shli_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
{
    do_shifti(INDEX_op_shli_vec, vece, r, a, i);
}

void tcg_gen_shri_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
{
    do_shifti(INDEX_op_shri_vec, vece, r, a, i);
}

void tcg_gen_sari_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
{
    do_shifti(INDEX_op_sari_vec, vece, r, a, i);
}

void tcg_gen_rotli_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
{
    do_shifti(INDEX_op_rotli_vec, vece, r, a, i);
}

void tcg_gen_rotri_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
{
    int bits = 8 << vece;
    tcg_debug_assert(i >= 0 && i < bits);
    do_shifti(INDEX_op_rotli_vec, vece, r, a, -i & (bits - 1));
}

void tcg_gen_cmp_vec(TCGCond cond, unsigned vece,
                     TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    TCGTemp *at = tcgv_vec_temp(a);
    TCGTemp *bt = tcgv_vec_temp(b);
    TCGTemp *tt = NULL;
    TCGArg ri = temp_arg(rt);
    TCGArg ai = temp_arg(at);
    TCGArg bi = temp_arg(bt);
    TCGArg ti;
    TCGType type = rt->base_type;
    int can;

    tcg_debug_assert(at->base_type >= type);
    tcg_debug_assert(bt->base_type >= type);
    tcg_assert_listed_vecop(INDEX_op_cmp_vec);
    can = tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece);

    if (!TCG_TARGET_HAS_tst_vec && is_tst_cond(cond)) {
        tt = tcg_temp_new_internal(type, TEMP_EBB);
        ti = temp_arg(tt);
        vec_gen_3(INDEX_op_and_vec, type, 0, ti, ai, bi);
        at = tt;
        ai = ti;
        bt = tcg_constant_internal(type, 0);
        bi = temp_arg(bt);
        cond = tcg_tst_eqne_cond(cond);
    }

    if (can > 0) {
        vec_gen_4(INDEX_op_cmp_vec, type, vece, ri, ai, bi, cond);
    } else {
        const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
        tcg_debug_assert(can < 0);
        tcg_expand_vec_op(INDEX_op_cmp_vec, type, vece, ri, ai, bi, cond);
        tcg_swap_vecop_list(hold_list);
    }

    if (tt) {
        tcg_temp_free_internal(tt);
    }
}

static bool do_op3(unsigned vece, TCGv_vec r, TCGv_vec a,
                   TCGv_vec b, TCGOpcode opc)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    TCGTemp *at = tcgv_vec_temp(a);
    TCGTemp *bt = tcgv_vec_temp(b);
    TCGArg ri = temp_arg(rt);
    TCGArg ai = temp_arg(at);
    TCGArg bi = temp_arg(bt);
    TCGType type = rt->base_type;
    int can;

    tcg_debug_assert(at->base_type >= type);
    tcg_debug_assert(bt->base_type >= type);
    tcg_assert_listed_vecop(opc);
    can = tcg_can_emit_vec_op(opc, type, vece);
    if (can > 0) {
        vec_gen_3(opc, type, vece, ri, ai, bi);
    } else if (can < 0) {
        const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
        tcg_expand_vec_op(opc, type, vece, ri, ai, bi);
        tcg_swap_vecop_list(hold_list);
    } else {
        return false;
    }
    return true;
}

static void do_op3_nofail(unsigned vece, TCGv_vec r, TCGv_vec a,
                          TCGv_vec b, TCGOpcode opc)
{
    bool ok = do_op3(vece, r, a, b, opc);
    tcg_debug_assert(ok);
}

void tcg_gen_add_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_add_vec);
}

void tcg_gen_sub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_sub_vec);
}

void tcg_gen_mul_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_mul_vec);
}

void tcg_gen_ssadd_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_ssadd_vec);
}

void tcg_gen_usadd_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    if (!do_op3(vece, r, a, b, INDEX_op_usadd_vec)) {
        const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
        TCGv_vec t = tcg_temp_new_vec_matching(r);

        /* usadd(a, b) = min(a, ~b) + b */
        tcg_gen_not_vec(vece, t, b);
        tcg_gen_umin_vec(vece, t, t, a);
        tcg_gen_add_vec(vece, r, t, b);

        tcg_temp_free_vec(t);
        tcg_swap_vecop_list(hold_list);
    }
}

void tcg_gen_sssub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_sssub_vec);
}

void tcg_gen_ussub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    if (!do_op3(vece, r, a, b, INDEX_op_ussub_vec)) {
        const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
        TCGv_vec t = tcg_temp_new_vec_matching(r);

        /* ussub(a, b) = max(a, b) - b */
        tcg_gen_umax_vec(vece, t, a, b);
        tcg_gen_sub_vec(vece, r, t, b);

        tcg_temp_free_vec(t);
        tcg_swap_vecop_list(hold_list);
    }
}

static void do_minmax(unsigned vece, TCGv_vec r, TCGv_vec a,
                      TCGv_vec b, TCGOpcode opc, TCGCond cond)
{
    if (!do_op3(vece, r, a, b, opc)) {
        const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
        tcg_gen_cmpsel_vec(cond, vece, r, a, b, a, b);
        tcg_swap_vecop_list(hold_list);
    }
}

void tcg_gen_smin_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_minmax(vece, r, a, b, INDEX_op_smin_vec, TCG_COND_LT);
}

void tcg_gen_umin_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_minmax(vece, r, a, b, INDEX_op_umin_vec, TCG_COND_LTU);
}

void tcg_gen_smax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_minmax(vece, r, a, b, INDEX_op_smax_vec, TCG_COND_GT);
}

void tcg_gen_umax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_minmax(vece, r, a, b, INDEX_op_umax_vec, TCG_COND_GTU);
}

void tcg_gen_shlv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_shlv_vec);
}

void tcg_gen_shrv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_shrv_vec);
}

void tcg_gen_sarv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_sarv_vec);
}

void tcg_gen_rotlv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_rotlv_vec);
}

void tcg_gen_rotrv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
    do_op3_nofail(vece, r, a, b, INDEX_op_rotrv_vec);
}

static void do_shifts(unsigned vece, TCGv_vec r, TCGv_vec a,
                      TCGv_i32 s, TCGOpcode opc)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    TCGTemp *at = tcgv_vec_temp(a);
    TCGTemp *st = tcgv_i32_temp(s);
    TCGArg ri = temp_arg(rt);
    TCGArg ai = temp_arg(at);
    TCGArg si = temp_arg(st);
    TCGType type = rt->base_type;
    int can;

    tcg_debug_assert(at->base_type >= type);
    tcg_assert_listed_vecop(opc);
    can = tcg_can_emit_vec_op(opc, type, vece);
    if (can > 0) {
        vec_gen_3(opc, type, vece, ri, ai, si);
    } else if (can < 0) {
        const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
        tcg_expand_vec_op(opc, type, vece, ri, ai, si);
        tcg_swap_vecop_list(hold_list);
    } else {
        g_assert_not_reached();
    }
}

void tcg_gen_shls_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_i32 b)
{
    do_shifts(vece, r, a, b, INDEX_op_shls_vec);
}

void tcg_gen_shrs_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_i32 b)
{
    do_shifts(vece, r, a, b, INDEX_op_shrs_vec);
}

void tcg_gen_sars_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_i32 b)
{
    do_shifts(vece, r, a, b, INDEX_op_sars_vec);
}

void tcg_gen_rotls_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_i32 s)
{
    do_shifts(vece, r, a, s, INDEX_op_rotls_vec);
}

void tcg_gen_bitsel_vec(unsigned vece, TCGv_vec r, TCGv_vec a,
                        TCGv_vec b, TCGv_vec c)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    TCGTemp *at = tcgv_vec_temp(a);
    TCGTemp *bt = tcgv_vec_temp(b);
    TCGTemp *ct = tcgv_vec_temp(c);
    TCGType type = rt->base_type;

    tcg_debug_assert(at->base_type >= type);
    tcg_debug_assert(bt->base_type >= type);
    tcg_debug_assert(ct->base_type >= type);

    if (TCG_TARGET_HAS_bitsel_vec) {
        vec_gen_4(INDEX_op_bitsel_vec, type, MO_8,
                  temp_arg(rt), temp_arg(at), temp_arg(bt), temp_arg(ct));
    } else {
        TCGv_vec t = tcg_temp_new_vec(type);
        tcg_gen_and_vec(MO_8, t, a, b);
        tcg_gen_andc_vec(MO_8, r, c, a);
        tcg_gen_or_vec(MO_8, r, r, t);
        tcg_temp_free_vec(t);
    }
}

void tcg_gen_cmpsel_vec(TCGCond cond, unsigned vece, TCGv_vec r,
                        TCGv_vec a, TCGv_vec b, TCGv_vec c, TCGv_vec d)
{
    TCGTemp *rt = tcgv_vec_temp(r);
    TCGTemp *at = tcgv_vec_temp(a);
    TCGTemp *bt = tcgv_vec_temp(b);
    TCGTemp *ct = tcgv_vec_temp(c);
    TCGTemp *dt = tcgv_vec_temp(d);
    TCGArg ri = temp_arg(rt);
    TCGArg ai = temp_arg(at);
    TCGArg bi = temp_arg(bt);
    TCGArg ci = temp_arg(ct);
    TCGArg di = temp_arg(dt);
    TCGType type = rt->base_type;
    const TCGOpcode *hold_list;
    int can;

    tcg_debug_assert(at->base_type >= type);
    tcg_debug_assert(bt->base_type >= type);
    tcg_debug_assert(ct->base_type >= type);
    tcg_debug_assert(dt->base_type >= type);

    tcg_assert_listed_vecop(INDEX_op_cmpsel_vec);
    hold_list = tcg_swap_vecop_list(NULL);
    can = tcg_can_emit_vec_op(INDEX_op_cmpsel_vec, type, vece);

    if (can > 0) {
        vec_gen_6(INDEX_op_cmpsel_vec, type, vece, ri, ai, bi, ci, di, cond);
    } else if (can < 0) {
        tcg_expand_vec_op(INDEX_op_cmpsel_vec, type, vece,
                          ri, ai, bi, ci, di, cond);
    } else {
        TCGv_vec t = tcg_temp_new_vec(type);
        tcg_gen_cmp_vec(cond, vece, t, a, b);
        tcg_gen_bitsel_vec(vece, r, t, c, d);
        tcg_temp_free_vec(t);
    }
    tcg_swap_vecop_list(hold_list);
}