graph-lock: Fix GRAPH_RDLOCK_GUARD*() to be reader lock
[qemu/armbru.git] / tcg / tcg-op-vec.c
blobaeeb2435cb12d1b2532f6d66eef12520cbd13646
1 /*
2 * Tiny Code Generator for QEMU
4 * Copyright (c) 2018 Linaro, Inc.
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
21 #include "tcg/tcg.h"
22 #include "tcg/tcg-temp-internal.h"
23 #include "tcg/tcg-op.h"
24 #include "tcg/tcg-mo.h"
25 #include "tcg-internal.h"
28 /* Reduce the number of ifdefs below. This assumes that all uses of
29 TCGV_HIGH and TCGV_LOW are properly protected by a conditional that
30 the compiler can eliminate. */
31 #if TCG_TARGET_REG_BITS == 64
32 extern TCGv_i32 TCGV_LOW_link_error(TCGv_i64);
33 extern TCGv_i32 TCGV_HIGH_link_error(TCGv_i64);
34 #define TCGV_LOW TCGV_LOW_link_error
35 #define TCGV_HIGH TCGV_HIGH_link_error
36 #endif
39 * Vector optional opcode tracking.
40 * Except for the basic logical operations (and, or, xor), and
41 * data movement (mov, ld, st, dupi), many vector opcodes are
42 * optional and may not be supported on the host. Thank Intel
43 * for the irregularity in their instruction set.
45 * The gvec expanders allow custom vector operations to be composed,
46 * generally via the .fniv callback in the GVecGen* structures. At
47 * the same time, in deciding whether to use this hook we need to
48 * know if the host supports the required operations. This is
49 * presented as an array of opcodes, terminated by 0. Each opcode
50 * is assumed to be expanded with the given VECE.
52 * For debugging, we want to validate this array. Therefore, when
53 * tcg_ctx->vec_opt_opc is non-NULL, the tcg_gen_*_vec expanders
54 * will validate that their opcode is present in the list.
56 #ifdef CONFIG_DEBUG_TCG
57 void tcg_assert_listed_vecop(TCGOpcode op)
59 const TCGOpcode *p = tcg_ctx->vecop_list;
60 if (p) {
61 for (; *p; ++p) {
62 if (*p == op) {
63 return;
66 g_assert_not_reached();
69 #endif
71 bool tcg_can_emit_vecop_list(const TCGOpcode *list,
72 TCGType type, unsigned vece)
74 if (list == NULL) {
75 return true;
78 for (; *list; ++list) {
79 TCGOpcode opc = *list;
81 #ifdef CONFIG_DEBUG_TCG
82 switch (opc) {
83 case INDEX_op_and_vec:
84 case INDEX_op_or_vec:
85 case INDEX_op_xor_vec:
86 case INDEX_op_mov_vec:
87 case INDEX_op_dup_vec:
88 case INDEX_op_dup2_vec:
89 case INDEX_op_ld_vec:
90 case INDEX_op_st_vec:
91 case INDEX_op_bitsel_vec:
92 /* These opcodes are mandatory and should not be listed. */
93 g_assert_not_reached();
94 case INDEX_op_not_vec:
95 /* These opcodes have generic expansions using the above. */
96 g_assert_not_reached();
97 default:
98 break;
100 #endif
102 if (tcg_can_emit_vec_op(opc, type, vece)) {
103 continue;
107 * The opcode list is created by front ends based on what they
108 * actually invoke. We must mirror the logic in the routines
109 * below for generic expansions using other opcodes.
111 switch (opc) {
112 case INDEX_op_neg_vec:
113 if (tcg_can_emit_vec_op(INDEX_op_sub_vec, type, vece)) {
114 continue;
116 break;
117 case INDEX_op_abs_vec:
118 if (tcg_can_emit_vec_op(INDEX_op_sub_vec, type, vece)
119 && (tcg_can_emit_vec_op(INDEX_op_smax_vec, type, vece) > 0
120 || tcg_can_emit_vec_op(INDEX_op_sari_vec, type, vece) > 0
121 || tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece))) {
122 continue;
124 break;
125 case INDEX_op_usadd_vec:
126 if (tcg_can_emit_vec_op(INDEX_op_umin_vec, type, vece) ||
127 tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece)) {
128 continue;
130 break;
131 case INDEX_op_ussub_vec:
132 if (tcg_can_emit_vec_op(INDEX_op_umax_vec, type, vece) ||
133 tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece)) {
134 continue;
136 break;
137 case INDEX_op_cmpsel_vec:
138 case INDEX_op_smin_vec:
139 case INDEX_op_smax_vec:
140 case INDEX_op_umin_vec:
141 case INDEX_op_umax_vec:
142 if (tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece)) {
143 continue;
145 break;
146 default:
147 break;
149 return false;
151 return true;
154 void vec_gen_2(TCGOpcode opc, TCGType type, unsigned vece, TCGArg r, TCGArg a)
156 TCGOp *op = tcg_emit_op(opc, 2);
157 TCGOP_VECL(op) = type - TCG_TYPE_V64;
158 TCGOP_VECE(op) = vece;
159 op->args[0] = r;
160 op->args[1] = a;
163 void vec_gen_3(TCGOpcode opc, TCGType type, unsigned vece,
164 TCGArg r, TCGArg a, TCGArg b)
166 TCGOp *op = tcg_emit_op(opc, 3);
167 TCGOP_VECL(op) = type - TCG_TYPE_V64;
168 TCGOP_VECE(op) = vece;
169 op->args[0] = r;
170 op->args[1] = a;
171 op->args[2] = b;
174 void vec_gen_4(TCGOpcode opc, TCGType type, unsigned vece,
175 TCGArg r, TCGArg a, TCGArg b, TCGArg c)
177 TCGOp *op = tcg_emit_op(opc, 4);
178 TCGOP_VECL(op) = type - TCG_TYPE_V64;
179 TCGOP_VECE(op) = vece;
180 op->args[0] = r;
181 op->args[1] = a;
182 op->args[2] = b;
183 op->args[3] = c;
186 static void vec_gen_6(TCGOpcode opc, TCGType type, unsigned vece, TCGArg r,
187 TCGArg a, TCGArg b, TCGArg c, TCGArg d, TCGArg e)
189 TCGOp *op = tcg_emit_op(opc, 6);
190 TCGOP_VECL(op) = type - TCG_TYPE_V64;
191 TCGOP_VECE(op) = vece;
192 op->args[0] = r;
193 op->args[1] = a;
194 op->args[2] = b;
195 op->args[3] = c;
196 op->args[4] = d;
197 op->args[5] = e;
200 static void vec_gen_op2(TCGOpcode opc, unsigned vece, TCGv_vec r, TCGv_vec a)
202 TCGTemp *rt = tcgv_vec_temp(r);
203 TCGTemp *at = tcgv_vec_temp(a);
204 TCGType type = rt->base_type;
206 /* Must enough inputs for the output. */
207 tcg_debug_assert(at->base_type >= type);
208 vec_gen_2(opc, type, vece, temp_arg(rt), temp_arg(at));
211 static void vec_gen_op3(TCGOpcode opc, unsigned vece,
212 TCGv_vec r, TCGv_vec a, TCGv_vec b)
214 TCGTemp *rt = tcgv_vec_temp(r);
215 TCGTemp *at = tcgv_vec_temp(a);
216 TCGTemp *bt = tcgv_vec_temp(b);
217 TCGType type = rt->base_type;
219 /* Must enough inputs for the output. */
220 tcg_debug_assert(at->base_type >= type);
221 tcg_debug_assert(bt->base_type >= type);
222 vec_gen_3(opc, type, vece, temp_arg(rt), temp_arg(at), temp_arg(bt));
225 void tcg_gen_mov_vec(TCGv_vec r, TCGv_vec a)
227 if (r != a) {
228 vec_gen_op2(INDEX_op_mov_vec, 0, r, a);
232 void tcg_gen_dupi_vec(unsigned vece, TCGv_vec r, uint64_t a)
234 TCGTemp *rt = tcgv_vec_temp(r);
235 tcg_gen_mov_vec(r, tcg_constant_vec(rt->base_type, vece, a));
238 void tcg_gen_dup_i64_vec(unsigned vece, TCGv_vec r, TCGv_i64 a)
240 TCGArg ri = tcgv_vec_arg(r);
241 TCGTemp *rt = arg_temp(ri);
242 TCGType type = rt->base_type;
244 if (TCG_TARGET_REG_BITS == 64) {
245 TCGArg ai = tcgv_i64_arg(a);
246 vec_gen_2(INDEX_op_dup_vec, type, vece, ri, ai);
247 } else if (vece == MO_64) {
248 TCGArg al = tcgv_i32_arg(TCGV_LOW(a));
249 TCGArg ah = tcgv_i32_arg(TCGV_HIGH(a));
250 vec_gen_3(INDEX_op_dup2_vec, type, MO_64, ri, al, ah);
251 } else {
252 TCGArg ai = tcgv_i32_arg(TCGV_LOW(a));
253 vec_gen_2(INDEX_op_dup_vec, type, vece, ri, ai);
257 void tcg_gen_dup_i32_vec(unsigned vece, TCGv_vec r, TCGv_i32 a)
259 TCGArg ri = tcgv_vec_arg(r);
260 TCGArg ai = tcgv_i32_arg(a);
261 TCGTemp *rt = arg_temp(ri);
262 TCGType type = rt->base_type;
264 vec_gen_2(INDEX_op_dup_vec, type, vece, ri, ai);
267 void tcg_gen_dup_mem_vec(unsigned vece, TCGv_vec r, TCGv_ptr b,
268 tcg_target_long ofs)
270 TCGArg ri = tcgv_vec_arg(r);
271 TCGArg bi = tcgv_ptr_arg(b);
272 TCGTemp *rt = arg_temp(ri);
273 TCGType type = rt->base_type;
275 vec_gen_3(INDEX_op_dupm_vec, type, vece, ri, bi, ofs);
278 static void vec_gen_ldst(TCGOpcode opc, TCGv_vec r, TCGv_ptr b, TCGArg o)
280 TCGArg ri = tcgv_vec_arg(r);
281 TCGArg bi = tcgv_ptr_arg(b);
282 TCGTemp *rt = arg_temp(ri);
283 TCGType type = rt->base_type;
285 vec_gen_3(opc, type, 0, ri, bi, o);
288 void tcg_gen_ld_vec(TCGv_vec r, TCGv_ptr b, TCGArg o)
290 vec_gen_ldst(INDEX_op_ld_vec, r, b, o);
293 void tcg_gen_st_vec(TCGv_vec r, TCGv_ptr b, TCGArg o)
295 vec_gen_ldst(INDEX_op_st_vec, r, b, o);
298 void tcg_gen_stl_vec(TCGv_vec r, TCGv_ptr b, TCGArg o, TCGType low_type)
300 TCGArg ri = tcgv_vec_arg(r);
301 TCGArg bi = tcgv_ptr_arg(b);
302 TCGTemp *rt = arg_temp(ri);
303 TCGType type = rt->base_type;
305 tcg_debug_assert(low_type >= TCG_TYPE_V64);
306 tcg_debug_assert(low_type <= type);
307 vec_gen_3(INDEX_op_st_vec, low_type, 0, ri, bi, o);
310 void tcg_gen_and_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
312 vec_gen_op3(INDEX_op_and_vec, 0, r, a, b);
315 void tcg_gen_or_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
317 vec_gen_op3(INDEX_op_or_vec, 0, r, a, b);
320 void tcg_gen_xor_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
322 vec_gen_op3(INDEX_op_xor_vec, 0, r, a, b);
325 void tcg_gen_andc_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
327 if (TCG_TARGET_HAS_andc_vec) {
328 vec_gen_op3(INDEX_op_andc_vec, 0, r, a, b);
329 } else {
330 TCGv_vec t = tcg_temp_new_vec_matching(r);
331 tcg_gen_not_vec(0, t, b);
332 tcg_gen_and_vec(0, r, a, t);
333 tcg_temp_free_vec(t);
337 void tcg_gen_orc_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
339 if (TCG_TARGET_HAS_orc_vec) {
340 vec_gen_op3(INDEX_op_orc_vec, 0, r, a, b);
341 } else {
342 TCGv_vec t = tcg_temp_new_vec_matching(r);
343 tcg_gen_not_vec(0, t, b);
344 tcg_gen_or_vec(0, r, a, t);
345 tcg_temp_free_vec(t);
349 void tcg_gen_nand_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
351 if (TCG_TARGET_HAS_nand_vec) {
352 vec_gen_op3(INDEX_op_nand_vec, 0, r, a, b);
353 } else {
354 tcg_gen_and_vec(0, r, a, b);
355 tcg_gen_not_vec(0, r, r);
359 void tcg_gen_nor_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
361 if (TCG_TARGET_HAS_nor_vec) {
362 vec_gen_op3(INDEX_op_nor_vec, 0, r, a, b);
363 } else {
364 tcg_gen_or_vec(0, r, a, b);
365 tcg_gen_not_vec(0, r, r);
369 void tcg_gen_eqv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
371 if (TCG_TARGET_HAS_eqv_vec) {
372 vec_gen_op3(INDEX_op_eqv_vec, 0, r, a, b);
373 } else {
374 tcg_gen_xor_vec(0, r, a, b);
375 tcg_gen_not_vec(0, r, r);
379 static bool do_op2(unsigned vece, TCGv_vec r, TCGv_vec a, TCGOpcode opc)
381 TCGTemp *rt = tcgv_vec_temp(r);
382 TCGTemp *at = tcgv_vec_temp(a);
383 TCGArg ri = temp_arg(rt);
384 TCGArg ai = temp_arg(at);
385 TCGType type = rt->base_type;
386 int can;
388 tcg_debug_assert(at->base_type >= type);
389 tcg_assert_listed_vecop(opc);
390 can = tcg_can_emit_vec_op(opc, type, vece);
391 if (can > 0) {
392 vec_gen_2(opc, type, vece, ri, ai);
393 } else if (can < 0) {
394 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
395 tcg_expand_vec_op(opc, type, vece, ri, ai);
396 tcg_swap_vecop_list(hold_list);
397 } else {
398 return false;
400 return true;
403 void tcg_gen_not_vec(unsigned vece, TCGv_vec r, TCGv_vec a)
405 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
407 if (!TCG_TARGET_HAS_not_vec || !do_op2(vece, r, a, INDEX_op_not_vec)) {
408 tcg_gen_xor_vec(0, r, a, tcg_constant_vec_matching(r, 0, -1));
410 tcg_swap_vecop_list(hold_list);
413 void tcg_gen_neg_vec(unsigned vece, TCGv_vec r, TCGv_vec a)
415 const TCGOpcode *hold_list;
417 tcg_assert_listed_vecop(INDEX_op_neg_vec);
418 hold_list = tcg_swap_vecop_list(NULL);
420 if (!TCG_TARGET_HAS_neg_vec || !do_op2(vece, r, a, INDEX_op_neg_vec)) {
421 tcg_gen_sub_vec(vece, r, tcg_constant_vec_matching(r, vece, 0), a);
423 tcg_swap_vecop_list(hold_list);
426 void tcg_gen_abs_vec(unsigned vece, TCGv_vec r, TCGv_vec a)
428 const TCGOpcode *hold_list;
430 tcg_assert_listed_vecop(INDEX_op_abs_vec);
431 hold_list = tcg_swap_vecop_list(NULL);
433 if (!do_op2(vece, r, a, INDEX_op_abs_vec)) {
434 TCGType type = tcgv_vec_temp(r)->base_type;
435 TCGv_vec t = tcg_temp_new_vec(type);
437 tcg_debug_assert(tcg_can_emit_vec_op(INDEX_op_sub_vec, type, vece));
438 if (tcg_can_emit_vec_op(INDEX_op_smax_vec, type, vece) > 0) {
439 tcg_gen_neg_vec(vece, t, a);
440 tcg_gen_smax_vec(vece, r, a, t);
441 } else {
442 if (tcg_can_emit_vec_op(INDEX_op_sari_vec, type, vece) > 0) {
443 tcg_gen_sari_vec(vece, t, a, (8 << vece) - 1);
444 } else {
445 tcg_gen_cmp_vec(TCG_COND_LT, vece, t, a,
446 tcg_constant_vec(type, vece, 0));
448 tcg_gen_xor_vec(vece, r, a, t);
449 tcg_gen_sub_vec(vece, r, r, t);
452 tcg_temp_free_vec(t);
454 tcg_swap_vecop_list(hold_list);
457 static void do_shifti(TCGOpcode opc, unsigned vece,
458 TCGv_vec r, TCGv_vec a, int64_t i)
460 TCGTemp *rt = tcgv_vec_temp(r);
461 TCGTemp *at = tcgv_vec_temp(a);
462 TCGArg ri = temp_arg(rt);
463 TCGArg ai = temp_arg(at);
464 TCGType type = rt->base_type;
465 int can;
467 tcg_debug_assert(at->base_type == type);
468 tcg_debug_assert(i >= 0 && i < (8 << vece));
469 tcg_assert_listed_vecop(opc);
471 if (i == 0) {
472 tcg_gen_mov_vec(r, a);
473 return;
476 can = tcg_can_emit_vec_op(opc, type, vece);
477 if (can > 0) {
478 vec_gen_3(opc, type, vece, ri, ai, i);
479 } else {
480 /* We leave the choice of expansion via scalar or vector shift
481 to the target. Often, but not always, dupi can feed a vector
482 shift easier than a scalar. */
483 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
484 tcg_debug_assert(can < 0);
485 tcg_expand_vec_op(opc, type, vece, ri, ai, i);
486 tcg_swap_vecop_list(hold_list);
490 void tcg_gen_shli_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
492 do_shifti(INDEX_op_shli_vec, vece, r, a, i);
495 void tcg_gen_shri_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
497 do_shifti(INDEX_op_shri_vec, vece, r, a, i);
500 void tcg_gen_sari_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
502 do_shifti(INDEX_op_sari_vec, vece, r, a, i);
505 void tcg_gen_rotli_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
507 do_shifti(INDEX_op_rotli_vec, vece, r, a, i);
510 void tcg_gen_rotri_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i)
512 int bits = 8 << vece;
513 tcg_debug_assert(i >= 0 && i < bits);
514 do_shifti(INDEX_op_rotli_vec, vece, r, a, -i & (bits - 1));
517 void tcg_gen_cmp_vec(TCGCond cond, unsigned vece,
518 TCGv_vec r, TCGv_vec a, TCGv_vec b)
520 TCGTemp *rt = tcgv_vec_temp(r);
521 TCGTemp *at = tcgv_vec_temp(a);
522 TCGTemp *bt = tcgv_vec_temp(b);
523 TCGArg ri = temp_arg(rt);
524 TCGArg ai = temp_arg(at);
525 TCGArg bi = temp_arg(bt);
526 TCGType type = rt->base_type;
527 int can;
529 tcg_debug_assert(at->base_type >= type);
530 tcg_debug_assert(bt->base_type >= type);
531 tcg_assert_listed_vecop(INDEX_op_cmp_vec);
532 can = tcg_can_emit_vec_op(INDEX_op_cmp_vec, type, vece);
533 if (can > 0) {
534 vec_gen_4(INDEX_op_cmp_vec, type, vece, ri, ai, bi, cond);
535 } else {
536 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
537 tcg_debug_assert(can < 0);
538 tcg_expand_vec_op(INDEX_op_cmp_vec, type, vece, ri, ai, bi, cond);
539 tcg_swap_vecop_list(hold_list);
543 static bool do_op3(unsigned vece, TCGv_vec r, TCGv_vec a,
544 TCGv_vec b, TCGOpcode opc)
546 TCGTemp *rt = tcgv_vec_temp(r);
547 TCGTemp *at = tcgv_vec_temp(a);
548 TCGTemp *bt = tcgv_vec_temp(b);
549 TCGArg ri = temp_arg(rt);
550 TCGArg ai = temp_arg(at);
551 TCGArg bi = temp_arg(bt);
552 TCGType type = rt->base_type;
553 int can;
555 tcg_debug_assert(at->base_type >= type);
556 tcg_debug_assert(bt->base_type >= type);
557 tcg_assert_listed_vecop(opc);
558 can = tcg_can_emit_vec_op(opc, type, vece);
559 if (can > 0) {
560 vec_gen_3(opc, type, vece, ri, ai, bi);
561 } else if (can < 0) {
562 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
563 tcg_expand_vec_op(opc, type, vece, ri, ai, bi);
564 tcg_swap_vecop_list(hold_list);
565 } else {
566 return false;
568 return true;
571 static void do_op3_nofail(unsigned vece, TCGv_vec r, TCGv_vec a,
572 TCGv_vec b, TCGOpcode opc)
574 bool ok = do_op3(vece, r, a, b, opc);
575 tcg_debug_assert(ok);
578 void tcg_gen_add_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
580 do_op3_nofail(vece, r, a, b, INDEX_op_add_vec);
583 void tcg_gen_sub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
585 do_op3_nofail(vece, r, a, b, INDEX_op_sub_vec);
588 void tcg_gen_mul_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
590 do_op3_nofail(vece, r, a, b, INDEX_op_mul_vec);
593 void tcg_gen_ssadd_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
595 do_op3_nofail(vece, r, a, b, INDEX_op_ssadd_vec);
598 void tcg_gen_usadd_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
600 if (!do_op3(vece, r, a, b, INDEX_op_usadd_vec)) {
601 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
602 TCGv_vec t = tcg_temp_new_vec_matching(r);
604 /* usadd(a, b) = min(a, ~b) + b */
605 tcg_gen_not_vec(vece, t, b);
606 tcg_gen_umin_vec(vece, t, t, a);
607 tcg_gen_add_vec(vece, r, t, b);
609 tcg_temp_free_vec(t);
610 tcg_swap_vecop_list(hold_list);
614 void tcg_gen_sssub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
616 do_op3_nofail(vece, r, a, b, INDEX_op_sssub_vec);
619 void tcg_gen_ussub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
621 if (!do_op3(vece, r, a, b, INDEX_op_ussub_vec)) {
622 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
623 TCGv_vec t = tcg_temp_new_vec_matching(r);
625 /* ussub(a, b) = max(a, b) - b */
626 tcg_gen_umax_vec(vece, t, a, b);
627 tcg_gen_sub_vec(vece, r, t, b);
629 tcg_temp_free_vec(t);
630 tcg_swap_vecop_list(hold_list);
634 static void do_minmax(unsigned vece, TCGv_vec r, TCGv_vec a,
635 TCGv_vec b, TCGOpcode opc, TCGCond cond)
637 if (!do_op3(vece, r, a, b, opc)) {
638 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
639 tcg_gen_cmpsel_vec(cond, vece, r, a, b, a, b);
640 tcg_swap_vecop_list(hold_list);
644 void tcg_gen_smin_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
646 do_minmax(vece, r, a, b, INDEX_op_smin_vec, TCG_COND_LT);
649 void tcg_gen_umin_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
651 do_minmax(vece, r, a, b, INDEX_op_umin_vec, TCG_COND_LTU);
654 void tcg_gen_smax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
656 do_minmax(vece, r, a, b, INDEX_op_smax_vec, TCG_COND_GT);
659 void tcg_gen_umax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
661 do_minmax(vece, r, a, b, INDEX_op_umax_vec, TCG_COND_GTU);
664 void tcg_gen_shlv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
666 do_op3_nofail(vece, r, a, b, INDEX_op_shlv_vec);
669 void tcg_gen_shrv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
671 do_op3_nofail(vece, r, a, b, INDEX_op_shrv_vec);
674 void tcg_gen_sarv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
676 do_op3_nofail(vece, r, a, b, INDEX_op_sarv_vec);
679 void tcg_gen_rotlv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
681 do_op3_nofail(vece, r, a, b, INDEX_op_rotlv_vec);
684 void tcg_gen_rotrv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
686 do_op3_nofail(vece, r, a, b, INDEX_op_rotrv_vec);
689 static void do_shifts(unsigned vece, TCGv_vec r, TCGv_vec a,
690 TCGv_i32 s, TCGOpcode opc)
692 TCGTemp *rt = tcgv_vec_temp(r);
693 TCGTemp *at = tcgv_vec_temp(a);
694 TCGTemp *st = tcgv_i32_temp(s);
695 TCGArg ri = temp_arg(rt);
696 TCGArg ai = temp_arg(at);
697 TCGArg si = temp_arg(st);
698 TCGType type = rt->base_type;
699 int can;
701 tcg_debug_assert(at->base_type >= type);
702 tcg_assert_listed_vecop(opc);
703 can = tcg_can_emit_vec_op(opc, type, vece);
704 if (can > 0) {
705 vec_gen_3(opc, type, vece, ri, ai, si);
706 } else if (can < 0) {
707 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
708 tcg_expand_vec_op(opc, type, vece, ri, ai, si);
709 tcg_swap_vecop_list(hold_list);
710 } else {
711 g_assert_not_reached();
715 void tcg_gen_shls_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_i32 b)
717 do_shifts(vece, r, a, b, INDEX_op_shls_vec);
720 void tcg_gen_shrs_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_i32 b)
722 do_shifts(vece, r, a, b, INDEX_op_shrs_vec);
725 void tcg_gen_sars_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_i32 b)
727 do_shifts(vece, r, a, b, INDEX_op_sars_vec);
730 void tcg_gen_rotls_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_i32 s)
732 do_shifts(vece, r, a, s, INDEX_op_rotls_vec);
735 void tcg_gen_bitsel_vec(unsigned vece, TCGv_vec r, TCGv_vec a,
736 TCGv_vec b, TCGv_vec c)
738 TCGTemp *rt = tcgv_vec_temp(r);
739 TCGTemp *at = tcgv_vec_temp(a);
740 TCGTemp *bt = tcgv_vec_temp(b);
741 TCGTemp *ct = tcgv_vec_temp(c);
742 TCGType type = rt->base_type;
744 tcg_debug_assert(at->base_type >= type);
745 tcg_debug_assert(bt->base_type >= type);
746 tcg_debug_assert(ct->base_type >= type);
748 if (TCG_TARGET_HAS_bitsel_vec) {
749 vec_gen_4(INDEX_op_bitsel_vec, type, MO_8,
750 temp_arg(rt), temp_arg(at), temp_arg(bt), temp_arg(ct));
751 } else {
752 TCGv_vec t = tcg_temp_new_vec(type);
753 tcg_gen_and_vec(MO_8, t, a, b);
754 tcg_gen_andc_vec(MO_8, r, c, a);
755 tcg_gen_or_vec(MO_8, r, r, t);
756 tcg_temp_free_vec(t);
760 void tcg_gen_cmpsel_vec(TCGCond cond, unsigned vece, TCGv_vec r,
761 TCGv_vec a, TCGv_vec b, TCGv_vec c, TCGv_vec d)
763 TCGTemp *rt = tcgv_vec_temp(r);
764 TCGTemp *at = tcgv_vec_temp(a);
765 TCGTemp *bt = tcgv_vec_temp(b);
766 TCGTemp *ct = tcgv_vec_temp(c);
767 TCGTemp *dt = tcgv_vec_temp(d);
768 TCGArg ri = temp_arg(rt);
769 TCGArg ai = temp_arg(at);
770 TCGArg bi = temp_arg(bt);
771 TCGArg ci = temp_arg(ct);
772 TCGArg di = temp_arg(dt);
773 TCGType type = rt->base_type;
774 const TCGOpcode *hold_list;
775 int can;
777 tcg_debug_assert(at->base_type >= type);
778 tcg_debug_assert(bt->base_type >= type);
779 tcg_debug_assert(ct->base_type >= type);
780 tcg_debug_assert(dt->base_type >= type);
782 tcg_assert_listed_vecop(INDEX_op_cmpsel_vec);
783 hold_list = tcg_swap_vecop_list(NULL);
784 can = tcg_can_emit_vec_op(INDEX_op_cmpsel_vec, type, vece);
786 if (can > 0) {
787 vec_gen_6(INDEX_op_cmpsel_vec, type, vece, ri, ai, bi, ci, di, cond);
788 } else if (can < 0) {
789 tcg_expand_vec_op(INDEX_op_cmpsel_vec, type, vece,
790 ri, ai, bi, ci, di, cond);
791 } else {
792 TCGv_vec t = tcg_temp_new_vec(type);
793 tcg_gen_cmp_vec(cond, vece, t, a, b);
794 tcg_gen_bitsel_vec(vece, r, t, c, d);
795 tcg_temp_free_vec(t);
797 tcg_swap_vecop_list(hold_list);