]> git.proxmox.com Git - mirror_qemu.git/blame - tcg/tcg-op-gvec.c
projects / mirror_qemu.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
tcg: Add tcg_gen_vec_add{sub}8_i32
[mirror_qemu.git] / tcg / tcg-op-gvec.c
CommitLineData
db432672
RH		 1/*
2 * Generic vector operation expansion
3 *
4 * Copyright (c) 2018 Linaro
5 *
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
fb0343d5 9 * version 2.1 of the License, or (at your option) any later version.
db432672
RH		 10 *
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.
15 *
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/>.
18 */
19
20#include "qemu/osdep.h"
dcb32f1d
PMD		 21#include "tcg/tcg.h"
22#include "tcg/tcg-op.h"
23#include "tcg/tcg-op-gvec.h"
db725815 24#include "qemu/main-loop.h"
dcb32f1d 25#include "tcg/tcg-gvec-desc.h"
db432672
RH		 26
27#define MAX_UNROLL 4
28
53229a77
RH		 29#ifdef CONFIG_DEBUG_TCG
30static const TCGOpcode vecop_list_empty[1] = { 0 };
31#else
32#define vecop_list_empty NULL
33#endif
34
35
db432672
RH		 36/* Verify vector size and alignment rules. OFS should be the OR of all
37 of the operand offsets so that we can check them all at once. */
38static void check_size_align(uint32_t oprsz, uint32_t maxsz, uint32_t ofs)
39{
e2e7168a
RH		 40 uint32_t max_align;
41
42 switch (oprsz) {
43 case 8:
44 case 16:
45 case 32:
46 tcg_debug_assert(oprsz <= maxsz);
47 break;
48 default:
49 tcg_debug_assert(oprsz == maxsz);
50 break;
51 }
52 tcg_debug_assert(maxsz <= (8 << SIMD_MAXSZ_BITS));
53
54 max_align = maxsz >= 16 ? 15 : 7;
db432672
RH		 55 tcg_debug_assert((maxsz & max_align) == 0);
56 tcg_debug_assert((ofs & max_align) == 0);
57}
58
59/* Verify vector overlap rules for two operands. */
60static void check_overlap_2(uint32_t d, uint32_t a, uint32_t s)
61{
62 tcg_debug_assert(d == a || d + s <= a || a + s <= d);
63}
64
65/* Verify vector overlap rules for three operands. */
66static void check_overlap_3(uint32_t d, uint32_t a, uint32_t b, uint32_t s)
67{
68 check_overlap_2(d, a, s);
69 check_overlap_2(d, b, s);
70 check_overlap_2(a, b, s);
71}
72
73/* Verify vector overlap rules for four operands. */
74static void check_overlap_4(uint32_t d, uint32_t a, uint32_t b,
75 uint32_t c, uint32_t s)
76{
77 check_overlap_2(d, a, s);
78 check_overlap_2(d, b, s);
79 check_overlap_2(d, c, s);
80 check_overlap_2(a, b, s);
81 check_overlap_2(a, c, s);
82 check_overlap_2(b, c, s);
83}
84
85/* Create a descriptor from components. */
86uint32_t simd_desc(uint32_t oprsz, uint32_t maxsz, int32_t data)
87{
88 uint32_t desc = 0;
89
e2e7168a
RH		 90 check_size_align(oprsz, maxsz, 0);
91 tcg_debug_assert(data == sextract32(data, 0, SIMD_DATA_BITS));
db432672
RH		 92
93 oprsz = (oprsz / 8) - 1;
94 maxsz = (maxsz / 8) - 1;
e2e7168a
RH		 95
96 /*
97 * We have just asserted in check_size_align that either
98 * oprsz is {8,16,32} or matches maxsz. Encode the final
99 * case with '2', as that would otherwise map to 24.
100 */
101 if (oprsz == maxsz) {
102 oprsz = 2;
103 }
104
db432672
RH		 105 desc = deposit32(desc, SIMD_OPRSZ_SHIFT, SIMD_OPRSZ_BITS, oprsz);
106 desc = deposit32(desc, SIMD_MAXSZ_SHIFT, SIMD_MAXSZ_BITS, maxsz);
107 desc = deposit32(desc, SIMD_DATA_SHIFT, SIMD_DATA_BITS, data);
108
109 return desc;
110}
111
112/* Generate a call to a gvec-style helper with two vector operands. */
113void tcg_gen_gvec_2_ool(uint32_t dofs, uint32_t aofs,
114 uint32_t oprsz, uint32_t maxsz, int32_t data,
115 gen_helper_gvec_2 *fn)
116{
117 TCGv_ptr a0, a1;
88d4005b 118 TCGv_i32 desc = tcg_constant_i32(simd_desc(oprsz, maxsz, data));
db432672
RH		 119
120 a0 = tcg_temp_new_ptr();
121 a1 = tcg_temp_new_ptr();
122
123 tcg_gen_addi_ptr(a0, cpu_env, dofs);
124 tcg_gen_addi_ptr(a1, cpu_env, aofs);
125
126 fn(a0, a1, desc);
127
128 tcg_temp_free_ptr(a0);
129 tcg_temp_free_ptr(a1);
db432672
RH		 130}
131
22fc3527
RH		 132/* Generate a call to a gvec-style helper with two vector operands
133 and one scalar operand. */
134void tcg_gen_gvec_2i_ool(uint32_t dofs, uint32_t aofs, TCGv_i64 c,
135 uint32_t oprsz, uint32_t maxsz, int32_t data,
136 gen_helper_gvec_2i *fn)
137{
138 TCGv_ptr a0, a1;
88d4005b 139 TCGv_i32 desc = tcg_constant_i32(simd_desc(oprsz, maxsz, data));
22fc3527
RH		 140
141 a0 = tcg_temp_new_ptr();
142 a1 = tcg_temp_new_ptr();
143
144 tcg_gen_addi_ptr(a0, cpu_env, dofs);
145 tcg_gen_addi_ptr(a1, cpu_env, aofs);
146
147 fn(a0, a1, c, desc);
148
149 tcg_temp_free_ptr(a0);
150 tcg_temp_free_ptr(a1);
22fc3527
RH		 151}
152
db432672
RH		 153/* Generate a call to a gvec-style helper with three vector operands. */
154void tcg_gen_gvec_3_ool(uint32_t dofs, uint32_t aofs, uint32_t bofs,
155 uint32_t oprsz, uint32_t maxsz, int32_t data,
156 gen_helper_gvec_3 *fn)
157{
158 TCGv_ptr a0, a1, a2;
88d4005b 159 TCGv_i32 desc = tcg_constant_i32(simd_desc(oprsz, maxsz, data));
db432672
RH		 160
161 a0 = tcg_temp_new_ptr();
162 a1 = tcg_temp_new_ptr();
163 a2 = tcg_temp_new_ptr();
164
165 tcg_gen_addi_ptr(a0, cpu_env, dofs);
166 tcg_gen_addi_ptr(a1, cpu_env, aofs);
167 tcg_gen_addi_ptr(a2, cpu_env, bofs);
168
169 fn(a0, a1, a2, desc);
170
171 tcg_temp_free_ptr(a0);
172 tcg_temp_free_ptr(a1);
173 tcg_temp_free_ptr(a2);
db432672
RH		 174}
175
176/* Generate a call to a gvec-style helper with four vector operands. */
177void tcg_gen_gvec_4_ool(uint32_t dofs, uint32_t aofs, uint32_t bofs,
178 uint32_t cofs, uint32_t oprsz, uint32_t maxsz,
179 int32_t data, gen_helper_gvec_4 *fn)
180{
181 TCGv_ptr a0, a1, a2, a3;
88d4005b 182 TCGv_i32 desc = tcg_constant_i32(simd_desc(oprsz, maxsz, data));
db432672
RH		 183
184 a0 = tcg_temp_new_ptr();
185 a1 = tcg_temp_new_ptr();
186 a2 = tcg_temp_new_ptr();
187 a3 = tcg_temp_new_ptr();
188
189 tcg_gen_addi_ptr(a0, cpu_env, dofs);
190 tcg_gen_addi_ptr(a1, cpu_env, aofs);
191 tcg_gen_addi_ptr(a2, cpu_env, bofs);
192 tcg_gen_addi_ptr(a3, cpu_env, cofs);
193
194 fn(a0, a1, a2, a3, desc);
195
196 tcg_temp_free_ptr(a0);
197 tcg_temp_free_ptr(a1);
198 tcg_temp_free_ptr(a2);
199 tcg_temp_free_ptr(a3);
db432672
RH		 200}
201
202/* Generate a call to a gvec-style helper with five vector operands. */
203void tcg_gen_gvec_5_ool(uint32_t dofs, uint32_t aofs, uint32_t bofs,
204 uint32_t cofs, uint32_t xofs, uint32_t oprsz,
205 uint32_t maxsz, int32_t data, gen_helper_gvec_5 *fn)
206{
207 TCGv_ptr a0, a1, a2, a3, a4;
88d4005b 208 TCGv_i32 desc = tcg_constant_i32(simd_desc(oprsz, maxsz, data));
db432672
RH		 209
210 a0 = tcg_temp_new_ptr();
211 a1 = tcg_temp_new_ptr();
212 a2 = tcg_temp_new_ptr();
213 a3 = tcg_temp_new_ptr();
214 a4 = tcg_temp_new_ptr();
215
216 tcg_gen_addi_ptr(a0, cpu_env, dofs);
217 tcg_gen_addi_ptr(a1, cpu_env, aofs);
218 tcg_gen_addi_ptr(a2, cpu_env, bofs);
219 tcg_gen_addi_ptr(a3, cpu_env, cofs);
220 tcg_gen_addi_ptr(a4, cpu_env, xofs);
221
222 fn(a0, a1, a2, a3, a4, desc);
223
224 tcg_temp_free_ptr(a0);
225 tcg_temp_free_ptr(a1);
226 tcg_temp_free_ptr(a2);
227 tcg_temp_free_ptr(a3);
228 tcg_temp_free_ptr(a4);
db432672
RH		 229}
230
231/* Generate a call to a gvec-style helper with three vector operands
232 and an extra pointer operand. */
233void tcg_gen_gvec_2_ptr(uint32_t dofs, uint32_t aofs,
234 TCGv_ptr ptr, uint32_t oprsz, uint32_t maxsz,
235 int32_t data, gen_helper_gvec_2_ptr *fn)
236{
237 TCGv_ptr a0, a1;
88d4005b 238 TCGv_i32 desc = tcg_constant_i32(simd_desc(oprsz, maxsz, data));
db432672
RH		 239
240 a0 = tcg_temp_new_ptr();
241 a1 = tcg_temp_new_ptr();
242
243 tcg_gen_addi_ptr(a0, cpu_env, dofs);
244 tcg_gen_addi_ptr(a1, cpu_env, aofs);
245
246 fn(a0, a1, ptr, desc);
247
248 tcg_temp_free_ptr(a0);
249 tcg_temp_free_ptr(a1);
db432672
RH		 250}
251
252/* Generate a call to a gvec-style helper with three vector operands
253 and an extra pointer operand. */
254void tcg_gen_gvec_3_ptr(uint32_t dofs, uint32_t aofs, uint32_t bofs,
255 TCGv_ptr ptr, uint32_t oprsz, uint32_t maxsz,
256 int32_t data, gen_helper_gvec_3_ptr *fn)
257{
258 TCGv_ptr a0, a1, a2;
88d4005b 259 TCGv_i32 desc = tcg_constant_i32(simd_desc(oprsz, maxsz, data));
db432672
RH		 260
261 a0 = tcg_temp_new_ptr();
262 a1 = tcg_temp_new_ptr();
263 a2 = tcg_temp_new_ptr();
264
265 tcg_gen_addi_ptr(a0, cpu_env, dofs);
266 tcg_gen_addi_ptr(a1, cpu_env, aofs);
267 tcg_gen_addi_ptr(a2, cpu_env, bofs);
268
269 fn(a0, a1, a2, ptr, desc);
270
271 tcg_temp_free_ptr(a0);
272 tcg_temp_free_ptr(a1);
273 tcg_temp_free_ptr(a2);
db432672
RH		 274}
275
276/* Generate a call to a gvec-style helper with four vector operands
277 and an extra pointer operand. */
278void tcg_gen_gvec_4_ptr(uint32_t dofs, uint32_t aofs, uint32_t bofs,
279 uint32_t cofs, TCGv_ptr ptr, uint32_t oprsz,
280 uint32_t maxsz, int32_t data,
281 gen_helper_gvec_4_ptr *fn)
282{
283 TCGv_ptr a0, a1, a2, a3;
88d4005b 284 TCGv_i32 desc = tcg_constant_i32(simd_desc(oprsz, maxsz, data));
db432672
RH		 285
286 a0 = tcg_temp_new_ptr();
287 a1 = tcg_temp_new_ptr();
288 a2 = tcg_temp_new_ptr();
289 a3 = tcg_temp_new_ptr();
290
291 tcg_gen_addi_ptr(a0, cpu_env, dofs);
292 tcg_gen_addi_ptr(a1, cpu_env, aofs);
293 tcg_gen_addi_ptr(a2, cpu_env, bofs);
294 tcg_gen_addi_ptr(a3, cpu_env, cofs);
295
296 fn(a0, a1, a2, a3, ptr, desc);
297
298 tcg_temp_free_ptr(a0);
299 tcg_temp_free_ptr(a1);
300 tcg_temp_free_ptr(a2);
301 tcg_temp_free_ptr(a3);
db432672
RH		 302}
303
24459716
RH		 304/* Generate a call to a gvec-style helper with five vector operands
305 and an extra pointer operand. */
306void tcg_gen_gvec_5_ptr(uint32_t dofs, uint32_t aofs, uint32_t bofs,
307 uint32_t cofs, uint32_t eofs, TCGv_ptr ptr,
308 uint32_t oprsz, uint32_t maxsz, int32_t data,
309 gen_helper_gvec_5_ptr *fn)
310{
311 TCGv_ptr a0, a1, a2, a3, a4;
88d4005b 312 TCGv_i32 desc = tcg_constant_i32(simd_desc(oprsz, maxsz, data));
24459716
RH		 313
314 a0 = tcg_temp_new_ptr();
315 a1 = tcg_temp_new_ptr();
316 a2 = tcg_temp_new_ptr();
317 a3 = tcg_temp_new_ptr();
318 a4 = tcg_temp_new_ptr();
319
320 tcg_gen_addi_ptr(a0, cpu_env, dofs);
321 tcg_gen_addi_ptr(a1, cpu_env, aofs);
322 tcg_gen_addi_ptr(a2, cpu_env, bofs);
323 tcg_gen_addi_ptr(a3, cpu_env, cofs);
324 tcg_gen_addi_ptr(a4, cpu_env, eofs);
325
326 fn(a0, a1, a2, a3, a4, ptr, desc);
327
328 tcg_temp_free_ptr(a0);
329 tcg_temp_free_ptr(a1);
330 tcg_temp_free_ptr(a2);
331 tcg_temp_free_ptr(a3);
332 tcg_temp_free_ptr(a4);
24459716
RH		 333}
334
db432672
RH		 335/* Return true if we want to implement something of OPRSZ bytes
336 in units of LNSZ. This limits the expansion of inline code. */
337static inline bool check_size_impl(uint32_t oprsz, uint32_t lnsz)
338{
f47db80c
RH		 339 uint32_t q, r;
340
341 if (oprsz < lnsz) {
342 return false;
343 }
344
345 q = oprsz / lnsz;
346 r = oprsz % lnsz;
347 tcg_debug_assert((r & 7) == 0);
348
349 if (lnsz < 16) {
350 /* For sizes below 16, accept no remainder. */
351 if (r != 0) {
352 return false;
353 }
354 } else {
355 /*
356 * Recall that ARM SVE allows vector sizes that are not a
357 * power of 2, but always a multiple of 16. The intent is
358 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
359 * In addition, expand_clr needs to handle a multiple of 8.
360 * Thus we can handle the tail with one more operation per
361 * diminishing power of 2.
362 */
363 q += ctpop32(r);
499748d7 364 }
f47db80c
RH		 365
366 return q <= MAX_UNROLL;
db432672
RH		 367}
368
369static void expand_clr(uint32_t dofs, uint32_t maxsz);
370
371/* Duplicate C as per VECE. */
372uint64_t (dup_const)(unsigned vece, uint64_t c)
373{
374 switch (vece) {
375 case MO_8:
376 return 0x0101010101010101ull * (uint8_t)c;
377 case MO_16:
378 return 0x0001000100010001ull * (uint16_t)c;
379 case MO_32:
380 return 0x0000000100000001ull * (uint32_t)c;
381 case MO_64:
382 return c;
383 default:
384 g_assert_not_reached();
385 }
386}
387
388/* Duplicate IN into OUT as per VECE. */
614dd4f3 389void tcg_gen_dup_i32(unsigned vece, TCGv_i32 out, TCGv_i32 in)
db432672
RH		 390{
391 switch (vece) {
392 case MO_8:
393 tcg_gen_ext8u_i32(out, in);
394 tcg_gen_muli_i32(out, out, 0x01010101);
395 break;
396 case MO_16:
397 tcg_gen_deposit_i32(out, in, in, 16, 16);
398 break;
399 case MO_32:
400 tcg_gen_mov_i32(out, in);
401 break;
402 default:
403 g_assert_not_reached();
404 }
405}
406
614dd4f3 407void tcg_gen_dup_i64(unsigned vece, TCGv_i64 out, TCGv_i64 in)
db432672
RH		 408{
409 switch (vece) {
410 case MO_8:
411 tcg_gen_ext8u_i64(out, in);
412 tcg_gen_muli_i64(out, out, 0x0101010101010101ull);
413 break;
414 case MO_16:
415 tcg_gen_ext16u_i64(out, in);
416 tcg_gen_muli_i64(out, out, 0x0001000100010001ull);
417 break;
418 case MO_32:
419 tcg_gen_deposit_i64(out, in, in, 32, 32);
420 break;
421 case MO_64:
422 tcg_gen_mov_i64(out, in);
423 break;
424 default:
425 g_assert_not_reached();
426 }
427}
428
adb196cb
RH		 429/* Select a supported vector type for implementing an operation on SIZE
430 * bytes. If OP is 0, assume that the real operation to be performed is
431 * required by all backends. Otherwise, make sure than OP can be performed
432 * on elements of size VECE in the selected type. Do not select V64 if
433 * PREFER_I64 is true. Return 0 if no vector type is selected.
434 */
53229a77
RH		 435static TCGType choose_vector_type(const TCGOpcode *list, unsigned vece,
436 uint32_t size, bool prefer_i64)
adb196cb 437{
f47db80c
RH		 438 /*
439 * Recall that ARM SVE allows vector sizes that are not a
440 * power of 2, but always a multiple of 16. The intent is
441 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
442 * It is hard to imagine a case in which v256 is supported
443 * but v128 is not, but check anyway.
444 * In addition, expand_clr needs to handle a multiple of 8.
445 */
446 if (TCG_TARGET_HAS_v256 &&
447 check_size_impl(size, 32) &&
448 tcg_can_emit_vecop_list(list, TCG_TYPE_V256, vece) &&
449 (!(size & 16) ||
450 (TCG_TARGET_HAS_v128 &&
451 tcg_can_emit_vecop_list(list, TCG_TYPE_V128, vece))) &&
452 (!(size & 8) ||
453 (TCG_TARGET_HAS_v64 &&
454 tcg_can_emit_vecop_list(list, TCG_TYPE_V64, vece)))) {
455 return TCG_TYPE_V256;
adb196cb 456 }
f47db80c
RH		 457 if (TCG_TARGET_HAS_v128 &&
458 check_size_impl(size, 16) &&
459 tcg_can_emit_vecop_list(list, TCG_TYPE_V128, vece) &&
460 (!(size & 8) ||
461 (TCG_TARGET_HAS_v64 &&
462 tcg_can_emit_vecop_list(list, TCG_TYPE_V64, vece)))) {
adb196cb
RH		 463 return TCG_TYPE_V128;
464 }
465 if (TCG_TARGET_HAS_v64 && !prefer_i64 && check_size_impl(size, 8)
53229a77 466 && tcg_can_emit_vecop_list(list, TCG_TYPE_V64, vece)) {
adb196cb
RH		 467 return TCG_TYPE_V64;
468 }
469 return 0;
470}
471
37ee55a0
RH		 472static void do_dup_store(TCGType type, uint32_t dofs, uint32_t oprsz,
473 uint32_t maxsz, TCGv_vec t_vec)
474{
475 uint32_t i = 0;
476
f47db80c
RH		 477 tcg_debug_assert(oprsz >= 8);
478
479 /*
480 * This may be expand_clr for the tail of an operation, e.g.
481 * oprsz == 8 && maxsz == 64. The first 8 bytes of this store
482 * are misaligned wrt the maximum vector size, so do that first.
483 */
484 if (dofs & 8) {
485 tcg_gen_stl_vec(t_vec, cpu_env, dofs + i, TCG_TYPE_V64);
486 i += 8;
487 }
488
37ee55a0
RH		 489 switch (type) {
490 case TCG_TYPE_V256:
491 /*
492 * Recall that ARM SVE allows vector sizes that are not a
493 * power of 2, but always a multiple of 16. The intent is
494 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
495 */
496 for (; i + 32 <= oprsz; i += 32) {
497 tcg_gen_stl_vec(t_vec, cpu_env, dofs + i, TCG_TYPE_V256);
498 }
499 /* fallthru */
500 case TCG_TYPE_V128:
501 for (; i + 16 <= oprsz; i += 16) {
502 tcg_gen_stl_vec(t_vec, cpu_env, dofs + i, TCG_TYPE_V128);
503 }
504 break;
505 case TCG_TYPE_V64:
506 for (; i < oprsz; i += 8) {
507 tcg_gen_stl_vec(t_vec, cpu_env, dofs + i, TCG_TYPE_V64);
508 }
509 break;
510 default:
511 g_assert_not_reached();
512 }
513
514 if (oprsz < maxsz) {
515 expand_clr(dofs + oprsz, maxsz - oprsz);
516 }
517}
518
db432672
RH		 519/* Set OPRSZ bytes at DOFS to replications of IN_32, IN_64 or IN_C.
520 * Only one of IN_32 or IN_64 may be set;
521 * IN_C is used if IN_32 and IN_64 are unset.
522 */
523static void do_dup(unsigned vece, uint32_t dofs, uint32_t oprsz,
524 uint32_t maxsz, TCGv_i32 in_32, TCGv_i64 in_64,
525 uint64_t in_c)
526{
527 TCGType type;
528 TCGv_i64 t_64;
529 TCGv_i32 t_32, t_desc;
530 TCGv_ptr t_ptr;
531 uint32_t i;
532
533 assert(vece <= (in_32 ? MO_32 : MO_64));
534 assert(in_32 == NULL || in_64 == NULL);
535
536 /* If we're storing 0, expand oprsz to maxsz. */
537 if (in_32 == NULL && in_64 == NULL) {
538 in_c = dup_const(vece, in_c);
539 if (in_c == 0) {
540 oprsz = maxsz;
6d3ef048
RH		 541 vece = MO_8;
542 } else if (in_c == dup_const(MO_8, in_c)) {
543 vece = MO_8;
db432672
RH		 544 }
545 }
546
adb196cb
RH		 547 /* Implement inline with a vector type, if possible.
548 * Prefer integer when 64-bit host and no variable dup.
549 */
53229a77 550 type = choose_vector_type(NULL, vece, oprsz,
adb196cb
RH		 551 (TCG_TARGET_REG_BITS == 64 && in_32 == NULL
552 && (in_64 == NULL || vece == MO_64)));
db432672
RH		 553 if (type != 0) {
554 TCGv_vec t_vec = tcg_temp_new_vec(type);
555
556 if (in_32) {
557 tcg_gen_dup_i32_vec(vece, t_vec, in_32);
558 } else if (in_64) {
559 tcg_gen_dup_i64_vec(vece, t_vec, in_64);
560 } else {
37ee55a0 561 tcg_gen_dupi_vec(vece, t_vec, in_c);
db432672 562 }
37ee55a0 563 do_dup_store(type, dofs, oprsz, maxsz, t_vec);
db432672 564 tcg_temp_free_vec(t_vec);
37ee55a0 565 return;
db432672
RH		 566 }
567
568 /* Otherwise, inline with an integer type, unless "large". */
569 if (check_size_impl(oprsz, TCG_TARGET_REG_BITS / 8)) {
570 t_64 = NULL;
571 t_32 = NULL;
572
573 if (in_32) {
574 /* We are given a 32-bit variable input. For a 64-bit host,
575 use a 64-bit operation unless the 32-bit operation would
576 be simple enough. */
577 if (TCG_TARGET_REG_BITS == 64
578 && (vece != MO_32 || !check_size_impl(oprsz, 4))) {
579 t_64 = tcg_temp_new_i64();
580 tcg_gen_extu_i32_i64(t_64, in_32);
614dd4f3 581 tcg_gen_dup_i64(vece, t_64, t_64);
db432672
RH		 582 } else {
583 t_32 = tcg_temp_new_i32();
614dd4f3 584 tcg_gen_dup_i32(vece, t_32, in_32);
db432672
RH		 585 }
586 } else if (in_64) {
587 /* We are given a 64-bit variable input. */
588 t_64 = tcg_temp_new_i64();
614dd4f3 589 tcg_gen_dup_i64(vece, t_64, in_64);
db432672
RH		 590 } else {
591 /* We are given a constant input. */
592 /* For 64-bit hosts, use 64-bit constants for "simple" constants
593 or when we'd need too many 32-bit stores, or when a 64-bit
594 constant is really required. */
595 if (vece == MO_64
596 || (TCG_TARGET_REG_BITS == 64
597 && (in_c == 0 || in_c == -1
598 || !check_size_impl(oprsz, 4)))) {
88d4005b 599 t_64 = tcg_constant_i64(in_c);
db432672 600 } else {
88d4005b 601 t_32 = tcg_constant_i32(in_c);
db432672
RH		 602 }
603 }
604
605 /* Implement inline if we picked an implementation size above. */
606 if (t_32) {
607 for (i = 0; i < oprsz; i += 4) {
608 tcg_gen_st_i32(t_32, cpu_env, dofs + i);
609 }
610 tcg_temp_free_i32(t_32);
611 goto done;
612 }
613 if (t_64) {
614 for (i = 0; i < oprsz; i += 8) {
615 tcg_gen_st_i64(t_64, cpu_env, dofs + i);
616 }
617 tcg_temp_free_i64(t_64);
618 goto done;
adb196cb 619 }
db432672
RH		 620 }
621
622 /* Otherwise implement out of line. */
623 t_ptr = tcg_temp_new_ptr();
624 tcg_gen_addi_ptr(t_ptr, cpu_env, dofs);
6d3ef048
RH		 625
626 /*
627 * This may be expand_clr for the tail of an operation, e.g.
628 * oprsz == 8 && maxsz == 64. The size of the clear is misaligned
629 * wrt simd_desc and will assert. Simply pass all replicated byte
630 * stores through to memset.
631 */
632 if (oprsz == maxsz && vece == MO_8) {
633 TCGv_ptr t_size = tcg_const_ptr(oprsz);
634 TCGv_i32 t_val;
635
636 if (in_32) {
637 t_val = in_32;
638 } else if (in_64) {
639 t_val = tcg_temp_new_i32();
640 tcg_gen_extrl_i64_i32(t_val, in_64);
641 } else {
88d4005b 642 t_val = tcg_constant_i32(in_c);
6d3ef048
RH		 643 }
644 gen_helper_memset(t_ptr, t_ptr, t_val, t_size);
645
88d4005b 646 if (in_64) {
6d3ef048
RH		 647 tcg_temp_free_i32(t_val);
648 }
649 tcg_temp_free_ptr(t_size);
650 tcg_temp_free_ptr(t_ptr);
651 return;
652 }
653
88d4005b 654 t_desc = tcg_constant_i32(simd_desc(oprsz, maxsz, 0));
db432672
RH		 655
656 if (vece == MO_64) {
657 if (in_64) {
658 gen_helper_gvec_dup64(t_ptr, t_desc, in_64);
659 } else {
88d4005b 660 t_64 = tcg_constant_i64(in_c);
db432672 661 gen_helper_gvec_dup64(t_ptr, t_desc, t_64);
db432672
RH		 662 }
663 } else {
664 typedef void dup_fn(TCGv_ptr, TCGv_i32, TCGv_i32);
665 static dup_fn * const fns[3] = {
666 gen_helper_gvec_dup8,
667 gen_helper_gvec_dup16,
668 gen_helper_gvec_dup32
669 };
670
671 if (in_32) {
672 fns[vece](t_ptr, t_desc, in_32);
88d4005b 673 } else if (in_64) {
db432672 674 t_32 = tcg_temp_new_i32();
88d4005b
RH		 675 tcg_gen_extrl_i64_i32(t_32, in_64);
676 fns[vece](t_ptr, t_desc, t_32);
677 tcg_temp_free_i32(t_32);
678 } else {
679 if (vece == MO_8) {
680 in_c &= 0xff;
db432672 681 } else if (vece == MO_16) {
88d4005b 682 in_c &= 0xffff;
db432672 683 }
88d4005b 684 t_32 = tcg_constant_i32(in_c);
db432672 685 fns[vece](t_ptr, t_desc, t_32);
db432672
RH		 686 }
687 }
688
689 tcg_temp_free_ptr(t_ptr);
db432672
RH		 690 return;
691
692 done:
693 if (oprsz < maxsz) {
694 expand_clr(dofs + oprsz, maxsz - oprsz);
695 }
696}
697
698/* Likewise, but with zero. */
699static void expand_clr(uint32_t dofs, uint32_t maxsz)
700{
701 do_dup(MO_8, dofs, maxsz, maxsz, NULL, NULL, 0);
702}
703
704/* Expand OPSZ bytes worth of two-operand operations using i32 elements. */
705static void expand_2_i32(uint32_t dofs, uint32_t aofs, uint32_t oprsz,
ac09ae62 706 bool load_dest, void (*fni)(TCGv_i32, TCGv_i32))
db432672
RH		 707{
708 TCGv_i32 t0 = tcg_temp_new_i32();
ac09ae62 709 TCGv_i32 t1 = tcg_temp_new_i32();
db432672
RH		 710 uint32_t i;
711
712 for (i = 0; i < oprsz; i += 4) {
713 tcg_gen_ld_i32(t0, cpu_env, aofs + i);
ac09ae62
RH		 714 if (load_dest) {
715 tcg_gen_ld_i32(t1, cpu_env, dofs + i);
716 }
717 fni(t1, t0);
718 tcg_gen_st_i32(t1, cpu_env, dofs + i);
db432672
RH		 719 }
720 tcg_temp_free_i32(t0);
ac09ae62 721 tcg_temp_free_i32(t1);
db432672
RH		 722}
723
d0ec9796
RH		 724static void expand_2i_i32(uint32_t dofs, uint32_t aofs, uint32_t oprsz,
725 int32_t c, bool load_dest,
726 void (*fni)(TCGv_i32, TCGv_i32, int32_t))
727{
728 TCGv_i32 t0 = tcg_temp_new_i32();
729 TCGv_i32 t1 = tcg_temp_new_i32();
730 uint32_t i;
731
732 for (i = 0; i < oprsz; i += 4) {
733 tcg_gen_ld_i32(t0, cpu_env, aofs + i);
734 if (load_dest) {
735 tcg_gen_ld_i32(t1, cpu_env, dofs + i);
736 }
737 fni(t1, t0, c);
738 tcg_gen_st_i32(t1, cpu_env, dofs + i);
739 }
740 tcg_temp_free_i32(t0);
741 tcg_temp_free_i32(t1);
742}
743
22fc3527
RH		 744static void expand_2s_i32(uint32_t dofs, uint32_t aofs, uint32_t oprsz,
745 TCGv_i32 c, bool scalar_first,
746 void (*fni)(TCGv_i32, TCGv_i32, TCGv_i32))
747{
748 TCGv_i32 t0 = tcg_temp_new_i32();
749 TCGv_i32 t1 = tcg_temp_new_i32();
750 uint32_t i;
751
752 for (i = 0; i < oprsz; i += 4) {
753 tcg_gen_ld_i32(t0, cpu_env, aofs + i);
754 if (scalar_first) {
755 fni(t1, c, t0);
756 } else {
757 fni(t1, t0, c);
758 }
759 tcg_gen_st_i32(t1, cpu_env, dofs + i);
760 }
761 tcg_temp_free_i32(t0);
762 tcg_temp_free_i32(t1);
763}
764
db432672
RH		 765/* Expand OPSZ bytes worth of three-operand operations using i32 elements. */
766static void expand_3_i32(uint32_t dofs, uint32_t aofs,
767 uint32_t bofs, uint32_t oprsz, bool load_dest,
768 void (*fni)(TCGv_i32, TCGv_i32, TCGv_i32))
769{
770 TCGv_i32 t0 = tcg_temp_new_i32();
771 TCGv_i32 t1 = tcg_temp_new_i32();
772 TCGv_i32 t2 = tcg_temp_new_i32();
773 uint32_t i;
774
775 for (i = 0; i < oprsz; i += 4) {
776 tcg_gen_ld_i32(t0, cpu_env, aofs + i);
777 tcg_gen_ld_i32(t1, cpu_env, bofs + i);
778 if (load_dest) {
779 tcg_gen_ld_i32(t2, cpu_env, dofs + i);
780 }
781 fni(t2, t0, t1);
782 tcg_gen_st_i32(t2, cpu_env, dofs + i);
783 }
784 tcg_temp_free_i32(t2);
785 tcg_temp_free_i32(t1);
786 tcg_temp_free_i32(t0);
787}
788
e1227bb6
DH		 789static void expand_3i_i32(uint32_t dofs, uint32_t aofs, uint32_t bofs,
790 uint32_t oprsz, int32_t c, bool load_dest,
791 void (*fni)(TCGv_i32, TCGv_i32, TCGv_i32, int32_t))
792{
793 TCGv_i32 t0 = tcg_temp_new_i32();
794 TCGv_i32 t1 = tcg_temp_new_i32();
795 TCGv_i32 t2 = tcg_temp_new_i32();
796 uint32_t i;
797
798 for (i = 0; i < oprsz; i += 4) {
799 tcg_gen_ld_i32(t0, cpu_env, aofs + i);
800 tcg_gen_ld_i32(t1, cpu_env, bofs + i);
801 if (load_dest) {
802 tcg_gen_ld_i32(t2, cpu_env, dofs + i);
803 }
804 fni(t2, t0, t1, c);
805 tcg_gen_st_i32(t2, cpu_env, dofs + i);
806 }
807 tcg_temp_free_i32(t0);
808 tcg_temp_free_i32(t1);
809 tcg_temp_free_i32(t2);
810}
811
db432672
RH		 812/* Expand OPSZ bytes worth of three-operand operations using i32 elements. */
813static void expand_4_i32(uint32_t dofs, uint32_t aofs, uint32_t bofs,
5d6acdd4 814 uint32_t cofs, uint32_t oprsz, bool write_aofs,
db432672
RH		 815 void (*fni)(TCGv_i32, TCGv_i32, TCGv_i32, TCGv_i32))
816{
817 TCGv_i32 t0 = tcg_temp_new_i32();
818 TCGv_i32 t1 = tcg_temp_new_i32();
819 TCGv_i32 t2 = tcg_temp_new_i32();
820 TCGv_i32 t3 = tcg_temp_new_i32();
821 uint32_t i;
822
823 for (i = 0; i < oprsz; i += 4) {
824 tcg_gen_ld_i32(t1, cpu_env, aofs + i);
825 tcg_gen_ld_i32(t2, cpu_env, bofs + i);
826 tcg_gen_ld_i32(t3, cpu_env, cofs + i);
827 fni(t0, t1, t2, t3);
828 tcg_gen_st_i32(t0, cpu_env, dofs + i);
5d6acdd4
RH		 829 if (write_aofs) {
830 tcg_gen_st_i32(t1, cpu_env, aofs + i);
831 }
db432672
RH		 832 }
833 tcg_temp_free_i32(t3);
834 tcg_temp_free_i32(t2);
835 tcg_temp_free_i32(t1);
836 tcg_temp_free_i32(t0);
837}
838
839/* Expand OPSZ bytes worth of two-operand operations using i64 elements. */
840static void expand_2_i64(uint32_t dofs, uint32_t aofs, uint32_t oprsz,
ac09ae62 841 bool load_dest, void (*fni)(TCGv_i64, TCGv_i64))
db432672
RH		 842{
843 TCGv_i64 t0 = tcg_temp_new_i64();
ac09ae62 844 TCGv_i64 t1 = tcg_temp_new_i64();
db432672
RH		 845 uint32_t i;
846
847 for (i = 0; i < oprsz; i += 8) {
848 tcg_gen_ld_i64(t0, cpu_env, aofs + i);
ac09ae62
RH		 849 if (load_dest) {
850 tcg_gen_ld_i64(t1, cpu_env, dofs + i);
851 }
852 fni(t1, t0);
853 tcg_gen_st_i64(t1, cpu_env, dofs + i);
db432672
RH		 854 }
855 tcg_temp_free_i64(t0);
ac09ae62 856 tcg_temp_free_i64(t1);
db432672
RH		 857}
858
d0ec9796
RH		 859static void expand_2i_i64(uint32_t dofs, uint32_t aofs, uint32_t oprsz,
860 int64_t c, bool load_dest,
861 void (*fni)(TCGv_i64, TCGv_i64, int64_t))
862{
863 TCGv_i64 t0 = tcg_temp_new_i64();
864 TCGv_i64 t1 = tcg_temp_new_i64();
865 uint32_t i;
866
867 for (i = 0; i < oprsz; i += 8) {
868 tcg_gen_ld_i64(t0, cpu_env, aofs + i);
869 if (load_dest) {
870 tcg_gen_ld_i64(t1, cpu_env, dofs + i);
871 }
872 fni(t1, t0, c);
873 tcg_gen_st_i64(t1, cpu_env, dofs + i);
874 }
875 tcg_temp_free_i64(t0);
876 tcg_temp_free_i64(t1);
877}
878
22fc3527
RH		 879static void expand_2s_i64(uint32_t dofs, uint32_t aofs, uint32_t oprsz,
880 TCGv_i64 c, bool scalar_first,
881 void (*fni)(TCGv_i64, TCGv_i64, TCGv_i64))
882{
883 TCGv_i64 t0 = tcg_temp_new_i64();
884 TCGv_i64 t1 = tcg_temp_new_i64();
885 uint32_t i;
886
887 for (i = 0; i < oprsz; i += 8) {
888 tcg_gen_ld_i64(t0, cpu_env, aofs + i);
889 if (scalar_first) {
890 fni(t1, c, t0);
891 } else {
892 fni(t1, t0, c);
893 }
894 tcg_gen_st_i64(t1, cpu_env, dofs + i);
895 }
896 tcg_temp_free_i64(t0);
897 tcg_temp_free_i64(t1);
898}
899
db432672
RH		 900/* Expand OPSZ bytes worth of three-operand operations using i64 elements. */
901static void expand_3_i64(uint32_t dofs, uint32_t aofs,
902 uint32_t bofs, uint32_t oprsz, bool load_dest,
903 void (*fni)(TCGv_i64, TCGv_i64, TCGv_i64))
904{
905 TCGv_i64 t0 = tcg_temp_new_i64();
906 TCGv_i64 t1 = tcg_temp_new_i64();
907 TCGv_i64 t2 = tcg_temp_new_i64();
908 uint32_t i;
909
910 for (i = 0; i < oprsz; i += 8) {
911 tcg_gen_ld_i64(t0, cpu_env, aofs + i);
912 tcg_gen_ld_i64(t1, cpu_env, bofs + i);
913 if (load_dest) {
914 tcg_gen_ld_i64(t2, cpu_env, dofs + i);
915 }
916 fni(t2, t0, t1);
917 tcg_gen_st_i64(t2, cpu_env, dofs + i);
918 }
919 tcg_temp_free_i64(t2);
920 tcg_temp_free_i64(t1);
921 tcg_temp_free_i64(t0);
922}
923
e1227bb6
DH		 924static void expand_3i_i64(uint32_t dofs, uint32_t aofs, uint32_t bofs,
925 uint32_t oprsz, int64_t c, bool load_dest,
926 void (*fni)(TCGv_i64, TCGv_i64, TCGv_i64, int64_t))
927{
928 TCGv_i64 t0 = tcg_temp_new_i64();
929 TCGv_i64 t1 = tcg_temp_new_i64();
930 TCGv_i64 t2 = tcg_temp_new_i64();
931 uint32_t i;
932
933 for (i = 0; i < oprsz; i += 8) {
934 tcg_gen_ld_i64(t0, cpu_env, aofs + i);
935 tcg_gen_ld_i64(t1, cpu_env, bofs + i);
936 if (load_dest) {
937 tcg_gen_ld_i64(t2, cpu_env, dofs + i);
938 }
939 fni(t2, t0, t1, c);
940 tcg_gen_st_i64(t2, cpu_env, dofs + i);
941 }
942 tcg_temp_free_i64(t0);
943 tcg_temp_free_i64(t1);
944 tcg_temp_free_i64(t2);
945}
946
db432672
RH		 947/* Expand OPSZ bytes worth of three-operand operations using i64 elements. */
948static void expand_4_i64(uint32_t dofs, uint32_t aofs, uint32_t bofs,
5d6acdd4 949 uint32_t cofs, uint32_t oprsz, bool write_aofs,
db432672
RH		 950 void (*fni)(TCGv_i64, TCGv_i64, TCGv_i64, TCGv_i64))
951{
952 TCGv_i64 t0 = tcg_temp_new_i64();
953 TCGv_i64 t1 = tcg_temp_new_i64();
954 TCGv_i64 t2 = tcg_temp_new_i64();
955 TCGv_i64 t3 = tcg_temp_new_i64();
956 uint32_t i;
957
958 for (i = 0; i < oprsz; i += 8) {
959 tcg_gen_ld_i64(t1, cpu_env, aofs + i);
960 tcg_gen_ld_i64(t2, cpu_env, bofs + i);
961 tcg_gen_ld_i64(t3, cpu_env, cofs + i);
962 fni(t0, t1, t2, t3);
963 tcg_gen_st_i64(t0, cpu_env, dofs + i);
5d6acdd4
RH		 964 if (write_aofs) {
965 tcg_gen_st_i64(t1, cpu_env, aofs + i);
966 }
db432672
RH		 967 }
968 tcg_temp_free_i64(t3);
969 tcg_temp_free_i64(t2);
970 tcg_temp_free_i64(t1);
971 tcg_temp_free_i64(t0);
972}
973
974/* Expand OPSZ bytes worth of two-operand operations using host vectors. */
975static void expand_2_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
976 uint32_t oprsz, uint32_t tysz, TCGType type,
ac09ae62 977 bool load_dest,
db432672
RH		 978 void (*fni)(unsigned, TCGv_vec, TCGv_vec))
979{
980 TCGv_vec t0 = tcg_temp_new_vec(type);
ac09ae62 981 TCGv_vec t1 = tcg_temp_new_vec(type);
db432672
RH		 982 uint32_t i;
983
984 for (i = 0; i < oprsz; i += tysz) {
985 tcg_gen_ld_vec(t0, cpu_env, aofs + i);
ac09ae62
RH		 986 if (load_dest) {
987 tcg_gen_ld_vec(t1, cpu_env, dofs + i);
988 }
989 fni(vece, t1, t0);
990 tcg_gen_st_vec(t1, cpu_env, dofs + i);
db432672
RH		 991 }
992 tcg_temp_free_vec(t0);
ac09ae62 993 tcg_temp_free_vec(t1);
db432672
RH		 994}
995
d0ec9796
RH		 996/* Expand OPSZ bytes worth of two-vector operands and an immediate operand
997 using host vectors. */
998static void expand_2i_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
999 uint32_t oprsz, uint32_t tysz, TCGType type,
1000 int64_t c, bool load_dest,
1001 void (*fni)(unsigned, TCGv_vec, TCGv_vec, int64_t))
1002{
1003 TCGv_vec t0 = tcg_temp_new_vec(type);
1004 TCGv_vec t1 = tcg_temp_new_vec(type);
1005 uint32_t i;
1006
1007 for (i = 0; i < oprsz; i += tysz) {
1008 tcg_gen_ld_vec(t0, cpu_env, aofs + i);
1009 if (load_dest) {
1010 tcg_gen_ld_vec(t1, cpu_env, dofs + i);
1011 }
1012 fni(vece, t1, t0, c);
1013 tcg_gen_st_vec(t1, cpu_env, dofs + i);
1014 }
1015 tcg_temp_free_vec(t0);
1016 tcg_temp_free_vec(t1);
1017}
1018
22fc3527
RH		 1019static void expand_2s_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
1020 uint32_t oprsz, uint32_t tysz, TCGType type,
1021 TCGv_vec c, bool scalar_first,
1022 void (*fni)(unsigned, TCGv_vec, TCGv_vec, TCGv_vec))
1023{
1024 TCGv_vec t0 = tcg_temp_new_vec(type);
1025 TCGv_vec t1 = tcg_temp_new_vec(type);
1026 uint32_t i;
1027
1028 for (i = 0; i < oprsz; i += tysz) {
1029 tcg_gen_ld_vec(t0, cpu_env, aofs + i);
1030 if (scalar_first) {
1031 fni(vece, t1, c, t0);
1032 } else {
1033 fni(vece, t1, t0, c);
1034 }
1035 tcg_gen_st_vec(t1, cpu_env, dofs + i);
1036 }
1037 tcg_temp_free_vec(t0);
1038 tcg_temp_free_vec(t1);
1039}
1040
db432672
RH		 1041/* Expand OPSZ bytes worth of three-operand operations using host vectors. */
1042static void expand_3_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
1043 uint32_t bofs, uint32_t oprsz,
1044 uint32_t tysz, TCGType type, bool load_dest,
1045 void (*fni)(unsigned, TCGv_vec, TCGv_vec, TCGv_vec))
1046{
1047 TCGv_vec t0 = tcg_temp_new_vec(type);
1048 TCGv_vec t1 = tcg_temp_new_vec(type);
1049 TCGv_vec t2 = tcg_temp_new_vec(type);
1050 uint32_t i;
1051
1052 for (i = 0; i < oprsz; i += tysz) {
1053 tcg_gen_ld_vec(t0, cpu_env, aofs + i);
1054 tcg_gen_ld_vec(t1, cpu_env, bofs + i);
1055 if (load_dest) {
1056 tcg_gen_ld_vec(t2, cpu_env, dofs + i);
1057 }
1058 fni(vece, t2, t0, t1);
1059 tcg_gen_st_vec(t2, cpu_env, dofs + i);
1060 }
1061 tcg_temp_free_vec(t2);
1062 tcg_temp_free_vec(t1);
1063 tcg_temp_free_vec(t0);
1064}
1065
e1227bb6
DH		 1066/*
1067 * Expand OPSZ bytes worth of three-vector operands and an immediate operand
1068 * using host vectors.
1069 */
1070static void expand_3i_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
1071 uint32_t bofs, uint32_t oprsz, uint32_t tysz,
1072 TCGType type, int64_t c, bool load_dest,
1073 void (*fni)(unsigned, TCGv_vec, TCGv_vec, TCGv_vec,
1074 int64_t))
1075{
1076 TCGv_vec t0 = tcg_temp_new_vec(type);
1077 TCGv_vec t1 = tcg_temp_new_vec(type);
1078 TCGv_vec t2 = tcg_temp_new_vec(type);
1079 uint32_t i;
1080
1081 for (i = 0; i < oprsz; i += tysz) {
1082 tcg_gen_ld_vec(t0, cpu_env, aofs + i);
1083 tcg_gen_ld_vec(t1, cpu_env, bofs + i);
1084 if (load_dest) {
1085 tcg_gen_ld_vec(t2, cpu_env, dofs + i);
1086 }
1087 fni(vece, t2, t0, t1, c);
1088 tcg_gen_st_vec(t2, cpu_env, dofs + i);
1089 }
1090 tcg_temp_free_vec(t0);
1091 tcg_temp_free_vec(t1);
1092 tcg_temp_free_vec(t2);
1093}
1094
db432672
RH		 1095/* Expand OPSZ bytes worth of four-operand operations using host vectors. */
1096static void expand_4_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
1097 uint32_t bofs, uint32_t cofs, uint32_t oprsz,
5d6acdd4 1098 uint32_t tysz, TCGType type, bool write_aofs,
db432672
RH		 1099 void (*fni)(unsigned, TCGv_vec, TCGv_vec,
1100 TCGv_vec, TCGv_vec))
1101{
1102 TCGv_vec t0 = tcg_temp_new_vec(type);
1103 TCGv_vec t1 = tcg_temp_new_vec(type);
1104 TCGv_vec t2 = tcg_temp_new_vec(type);
1105 TCGv_vec t3 = tcg_temp_new_vec(type);
1106 uint32_t i;
1107
1108 for (i = 0; i < oprsz; i += tysz) {
1109 tcg_gen_ld_vec(t1, cpu_env, aofs + i);
1110 tcg_gen_ld_vec(t2, cpu_env, bofs + i);
1111 tcg_gen_ld_vec(t3, cpu_env, cofs + i);
1112 fni(vece, t0, t1, t2, t3);
1113 tcg_gen_st_vec(t0, cpu_env, dofs + i);
5d6acdd4
RH		 1114 if (write_aofs) {
1115 tcg_gen_st_vec(t1, cpu_env, aofs + i);
1116 }
db432672
RH		 1117 }
1118 tcg_temp_free_vec(t3);
1119 tcg_temp_free_vec(t2);
1120 tcg_temp_free_vec(t1);
1121 tcg_temp_free_vec(t0);
1122}
1123
1124/* Expand a vector two-operand operation. */
1125void tcg_gen_gvec_2(uint32_t dofs, uint32_t aofs,
1126 uint32_t oprsz, uint32_t maxsz, const GVecGen2 *g)
1127{
53229a77
RH		 1128 const TCGOpcode *this_list = g->opt_opc ? : vecop_list_empty;
1129 const TCGOpcode *hold_list = tcg_swap_vecop_list(this_list);
adb196cb
RH		 1130 TCGType type;
1131 uint32_t some;
1132
db432672
RH		 1133 check_size_align(oprsz, maxsz, dofs | aofs);
1134 check_overlap_2(dofs, aofs, maxsz);
1135
adb196cb
RH		 1136 type = 0;
1137 if (g->fniv) {
53229a77 1138 type = choose_vector_type(g->opt_opc, g->vece, oprsz, g->prefer_i64);
adb196cb
RH		 1139 }
1140 switch (type) {
1141 case TCG_TYPE_V256:
1142 /* Recall that ARM SVE allows vector sizes that are not a
1143 * power of 2, but always a multiple of 16. The intent is
1144 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
1145 */
1146 some = QEMU_ALIGN_DOWN(oprsz, 32);
ac09ae62
RH		 1147 expand_2_vec(g->vece, dofs, aofs, some, 32, TCG_TYPE_V256,
1148 g->load_dest, g->fniv);
db432672 1149 if (some == oprsz) {
adb196cb 1150 break;
db432672
RH		 1151 }
1152 dofs += some;
1153 aofs += some;
1154 oprsz -= some;
1155 maxsz -= some;
adb196cb
RH		 1156 /* fallthru */
1157 case TCG_TYPE_V128:
ac09ae62
RH		 1158 expand_2_vec(g->vece, dofs, aofs, oprsz, 16, TCG_TYPE_V128,
1159 g->load_dest, g->fniv);
adb196cb
RH		 1160 break;
1161 case TCG_TYPE_V64:
ac09ae62
RH		 1162 expand_2_vec(g->vece, dofs, aofs, oprsz, 8, TCG_TYPE_V64,
1163 g->load_dest, g->fniv);
adb196cb
RH		 1164 break;
1165
1166 case 0:
1167 if (g->fni8 && check_size_impl(oprsz, 8)) {
ac09ae62 1168 expand_2_i64(dofs, aofs, oprsz, g->load_dest, g->fni8);
adb196cb 1169 } else if (g->fni4 && check_size_impl(oprsz, 4)) {
ac09ae62 1170 expand_2_i32(dofs, aofs, oprsz, g->load_dest, g->fni4);
adb196cb
RH		 1171 } else {
1172 assert(g->fno != NULL);
1173 tcg_gen_gvec_2_ool(dofs, aofs, oprsz, maxsz, g->data, g->fno);
53229a77 1174 oprsz = maxsz;
adb196cb
RH		 1175 }
1176 break;
1177
1178 default:
1179 g_assert_not_reached();
db432672 1180 }
53229a77 1181 tcg_swap_vecop_list(hold_list);
db432672 1182
db432672
RH		 1183 if (oprsz < maxsz) {
1184 expand_clr(dofs + oprsz, maxsz - oprsz);
1185 }
1186}
1187
22fc3527 1188/* Expand a vector operation with two vectors and an immediate. */
d0ec9796
RH		 1189void tcg_gen_gvec_2i(uint32_t dofs, uint32_t aofs, uint32_t oprsz,
1190 uint32_t maxsz, int64_t c, const GVecGen2i *g)
1191{
53229a77
RH		 1192 const TCGOpcode *this_list = g->opt_opc ? : vecop_list_empty;
1193 const TCGOpcode *hold_list = tcg_swap_vecop_list(this_list);
adb196cb
RH		 1194 TCGType type;
1195 uint32_t some;
1196
d0ec9796
RH		 1197 check_size_align(oprsz, maxsz, dofs | aofs);
1198 check_overlap_2(dofs, aofs, maxsz);
1199
adb196cb
RH		 1200 type = 0;
1201 if (g->fniv) {
53229a77 1202 type = choose_vector_type(g->opt_opc, g->vece, oprsz, g->prefer_i64);
adb196cb
RH		 1203 }
1204 switch (type) {
1205 case TCG_TYPE_V256:
1206 /* Recall that ARM SVE allows vector sizes that are not a
1207 * power of 2, but always a multiple of 16. The intent is
1208 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
1209 */
1210 some = QEMU_ALIGN_DOWN(oprsz, 32);
d0ec9796
RH		 1211 expand_2i_vec(g->vece, dofs, aofs, some, 32, TCG_TYPE_V256,
1212 c, g->load_dest, g->fniv);
1213 if (some == oprsz) {
adb196cb 1214 break;
d0ec9796
RH		 1215 }
1216 dofs += some;
1217 aofs += some;
1218 oprsz -= some;
1219 maxsz -= some;
adb196cb
RH		 1220 /* fallthru */
1221 case TCG_TYPE_V128:
d0ec9796
RH		 1222 expand_2i_vec(g->vece, dofs, aofs, oprsz, 16, TCG_TYPE_V128,
1223 c, g->load_dest, g->fniv);
adb196cb
RH		 1224 break;
1225 case TCG_TYPE_V64:
d0ec9796
RH		 1226 expand_2i_vec(g->vece, dofs, aofs, oprsz, 8, TCG_TYPE_V64,
1227 c, g->load_dest, g->fniv);
adb196cb
RH		 1228 break;
1229
1230 case 0:
1231 if (g->fni8 && check_size_impl(oprsz, 8)) {
1232 expand_2i_i64(dofs, aofs, oprsz, c, g->load_dest, g->fni8);
1233 } else if (g->fni4 && check_size_impl(oprsz, 4)) {
1234 expand_2i_i32(dofs, aofs, oprsz, c, g->load_dest, g->fni4);
22fc3527 1235 } else {
adb196cb
RH		 1236 if (g->fno) {
1237 tcg_gen_gvec_2_ool(dofs, aofs, oprsz, maxsz, c, g->fno);
1238 } else {
88d4005b 1239 TCGv_i64 tcg_c = tcg_constant_i64(c);
adb196cb
RH		 1240 tcg_gen_gvec_2i_ool(dofs, aofs, tcg_c, oprsz,
1241 maxsz, c, g->fnoi);
adb196cb 1242 }
53229a77 1243 oprsz = maxsz;
22fc3527 1244 }
adb196cb
RH		 1245 break;
1246
1247 default:
1248 g_assert_not_reached();
d0ec9796 1249 }
53229a77 1250 tcg_swap_vecop_list(hold_list);
d0ec9796 1251
d0ec9796
RH		 1252 if (oprsz < maxsz) {
1253 expand_clr(dofs + oprsz, maxsz - oprsz);
1254 }
1255}
1256
22fc3527
RH		 1257/* Expand a vector operation with two vectors and a scalar. */
1258void tcg_gen_gvec_2s(uint32_t dofs, uint32_t aofs, uint32_t oprsz,
1259 uint32_t maxsz, TCGv_i64 c, const GVecGen2s *g)
1260{
1261 TCGType type;
1262
1263 check_size_align(oprsz, maxsz, dofs | aofs);
1264 check_overlap_2(dofs, aofs, maxsz);
1265
1266 type = 0;
1267 if (g->fniv) {
53229a77 1268 type = choose_vector_type(g->opt_opc, g->vece, oprsz, g->prefer_i64);
22fc3527
RH		 1269 }
1270 if (type != 0) {
53229a77
RH		 1271 const TCGOpcode *this_list = g->opt_opc ? : vecop_list_empty;
1272 const TCGOpcode *hold_list = tcg_swap_vecop_list(this_list);
22fc3527 1273 TCGv_vec t_vec = tcg_temp_new_vec(type);
adb196cb 1274 uint32_t some;
22fc3527
RH		 1275
1276 tcg_gen_dup_i64_vec(g->vece, t_vec, c);
1277
22fc3527
RH		 1278 switch (type) {
1279 case TCG_TYPE_V256:
adb196cb
RH		 1280 /* Recall that ARM SVE allows vector sizes that are not a
1281 * power of 2, but always a multiple of 16. The intent is
1282 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
1283 */
1284 some = QEMU_ALIGN_DOWN(oprsz, 32);
1285 expand_2s_vec(g->vece, dofs, aofs, some, 32, TCG_TYPE_V256,
1286 t_vec, g->scalar_first, g->fniv);
1287 if (some == oprsz) {
1288 break;
22fc3527 1289 }
adb196cb
RH		 1290 dofs += some;
1291 aofs += some;
1292 oprsz -= some;
1293 maxsz -= some;
22fc3527
RH		 1294 /* fallthru */
1295
1296 case TCG_TYPE_V128:
1297 expand_2s_vec(g->vece, dofs, aofs, oprsz, 16, TCG_TYPE_V128,
1298 t_vec, g->scalar_first, g->fniv);
1299 break;
1300
1301 case TCG_TYPE_V64:
1302 expand_2s_vec(g->vece, dofs, aofs, oprsz, 8, TCG_TYPE_V64,
1303 t_vec, g->scalar_first, g->fniv);
1304 break;
1305
1306 default:
1307 g_assert_not_reached();
1308 }
1309 tcg_temp_free_vec(t_vec);
53229a77 1310 tcg_swap_vecop_list(hold_list);
22fc3527
RH		 1311 } else if (g->fni8 && check_size_impl(oprsz, 8)) {
1312 TCGv_i64 t64 = tcg_temp_new_i64();
1313
614dd4f3 1314 tcg_gen_dup_i64(g->vece, t64, c);
22fc3527
RH		 1315 expand_2s_i64(dofs, aofs, oprsz, t64, g->scalar_first, g->fni8);
1316 tcg_temp_free_i64(t64);
1317 } else if (g->fni4 && check_size_impl(oprsz, 4)) {
1318 TCGv_i32 t32 = tcg_temp_new_i32();
1319
1320 tcg_gen_extrl_i64_i32(t32, c);
614dd4f3 1321 tcg_gen_dup_i32(g->vece, t32, t32);
22fc3527
RH		 1322 expand_2s_i32(dofs, aofs, oprsz, t32, g->scalar_first, g->fni4);
1323 tcg_temp_free_i32(t32);
1324 } else {
1325 tcg_gen_gvec_2i_ool(dofs, aofs, c, oprsz, maxsz, 0, g->fno);
1326 return;
1327 }
1328
1329 if (oprsz < maxsz) {
1330 expand_clr(dofs + oprsz, maxsz - oprsz);
1331 }
1332}
1333
db432672
RH		 1334/* Expand a vector three-operand operation. */
1335void tcg_gen_gvec_3(uint32_t dofs, uint32_t aofs, uint32_t bofs,
1336 uint32_t oprsz, uint32_t maxsz, const GVecGen3 *g)
1337{
53229a77
RH		 1338 const TCGOpcode *this_list = g->opt_opc ? : vecop_list_empty;
1339 const TCGOpcode *hold_list = tcg_swap_vecop_list(this_list);
adb196cb
RH		 1340 TCGType type;
1341 uint32_t some;
1342
db432672
RH		 1343 check_size_align(oprsz, maxsz, dofs | aofs | bofs);
1344 check_overlap_3(dofs, aofs, bofs, maxsz);
1345
adb196cb
RH		 1346 type = 0;
1347 if (g->fniv) {
53229a77 1348 type = choose_vector_type(g->opt_opc, g->vece, oprsz, g->prefer_i64);
adb196cb
RH		 1349 }
1350 switch (type) {
1351 case TCG_TYPE_V256:
1352 /* Recall that ARM SVE allows vector sizes that are not a
1353 * power of 2, but always a multiple of 16. The intent is
1354 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
1355 */
1356 some = QEMU_ALIGN_DOWN(oprsz, 32);
db432672
RH		 1357 expand_3_vec(g->vece, dofs, aofs, bofs, some, 32, TCG_TYPE_V256,
1358 g->load_dest, g->fniv);
1359 if (some == oprsz) {
adb196cb 1360 break;
db432672
RH		 1361 }
1362 dofs += some;
1363 aofs += some;
1364 bofs += some;
1365 oprsz -= some;
1366 maxsz -= some;
adb196cb
RH		 1367 /* fallthru */
1368 case TCG_TYPE_V128:
db432672
RH		 1369 expand_3_vec(g->vece, dofs, aofs, bofs, oprsz, 16, TCG_TYPE_V128,
1370 g->load_dest, g->fniv);
adb196cb
RH		 1371 break;
1372 case TCG_TYPE_V64:
db432672
RH		 1373 expand_3_vec(g->vece, dofs, aofs, bofs, oprsz, 8, TCG_TYPE_V64,
1374 g->load_dest, g->fniv);
adb196cb
RH		 1375 break;
1376
1377 case 0:
1378 if (g->fni8 && check_size_impl(oprsz, 8)) {
1379 expand_3_i64(dofs, aofs, bofs, oprsz, g->load_dest, g->fni8);
1380 } else if (g->fni4 && check_size_impl(oprsz, 4)) {
1381 expand_3_i32(dofs, aofs, bofs, oprsz, g->load_dest, g->fni4);
1382 } else {
1383 assert(g->fno != NULL);
1384 tcg_gen_gvec_3_ool(dofs, aofs, bofs, oprsz,
1385 maxsz, g->data, g->fno);
53229a77 1386 oprsz = maxsz;
adb196cb
RH		 1387 }
1388 break;
1389
1390 default:
1391 g_assert_not_reached();
db432672 1392 }
53229a77 1393 tcg_swap_vecop_list(hold_list);
db432672 1394
db432672
RH		 1395 if (oprsz < maxsz) {
1396 expand_clr(dofs + oprsz, maxsz - oprsz);
1397 }
1398}
1399
e1227bb6
DH		 1400/* Expand a vector operation with three vectors and an immediate. */
1401void tcg_gen_gvec_3i(uint32_t dofs, uint32_t aofs, uint32_t bofs,
1402 uint32_t oprsz, uint32_t maxsz, int64_t c,
1403 const GVecGen3i *g)
1404{
53229a77
RH		 1405 const TCGOpcode *this_list = g->opt_opc ? : vecop_list_empty;
1406 const TCGOpcode *hold_list = tcg_swap_vecop_list(this_list);
e1227bb6
DH		 1407 TCGType type;
1408 uint32_t some;
1409
1410 check_size_align(oprsz, maxsz, dofs | aofs | bofs);
1411 check_overlap_3(dofs, aofs, bofs, maxsz);
1412
1413 type = 0;
1414 if (g->fniv) {
53229a77 1415 type = choose_vector_type(g->opt_opc, g->vece, oprsz, g->prefer_i64);
e1227bb6
DH		 1416 }
1417 switch (type) {
1418 case TCG_TYPE_V256:
1419 /*
1420 * Recall that ARM SVE allows vector sizes that are not a
1421 * power of 2, but always a multiple of 16. The intent is
1422 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
1423 */
1424 some = QEMU_ALIGN_DOWN(oprsz, 32);
1425 expand_3i_vec(g->vece, dofs, aofs, bofs, some, 32, TCG_TYPE_V256,
1426 c, g->load_dest, g->fniv);
1427 if (some == oprsz) {
1428 break;
1429 }
1430 dofs += some;
1431 aofs += some;
1432 bofs += some;
1433 oprsz -= some;
1434 maxsz -= some;
1435 /* fallthru */
1436 case TCG_TYPE_V128:
1437 expand_3i_vec(g->vece, dofs, aofs, bofs, oprsz, 16, TCG_TYPE_V128,
1438 c, g->load_dest, g->fniv);
1439 break;
1440 case TCG_TYPE_V64:
1441 expand_3i_vec(g->vece, dofs, aofs, bofs, oprsz, 8, TCG_TYPE_V64,
1442 c, g->load_dest, g->fniv);
1443 break;
1444
1445 case 0:
1446 if (g->fni8 && check_size_impl(oprsz, 8)) {
1447 expand_3i_i64(dofs, aofs, bofs, oprsz, c, g->load_dest, g->fni8);
1448 } else if (g->fni4 && check_size_impl(oprsz, 4)) {
1449 expand_3i_i32(dofs, aofs, bofs, oprsz, c, g->load_dest, g->fni4);
1450 } else {
1451 assert(g->fno != NULL);
1452 tcg_gen_gvec_3_ool(dofs, aofs, bofs, oprsz, maxsz, c, g->fno);
53229a77 1453 oprsz = maxsz;
e1227bb6
DH		 1454 }
1455 break;
1456
1457 default:
1458 g_assert_not_reached();
1459 }
53229a77 1460 tcg_swap_vecop_list(hold_list);
e1227bb6
DH		 1461
1462 if (oprsz < maxsz) {
1463 expand_clr(dofs + oprsz, maxsz - oprsz);
1464 }
1465}
1466
db432672
RH		 1467/* Expand a vector four-operand operation. */
1468void tcg_gen_gvec_4(uint32_t dofs, uint32_t aofs, uint32_t bofs, uint32_t cofs,
1469 uint32_t oprsz, uint32_t maxsz, const GVecGen4 *g)
1470{
53229a77
RH		 1471 const TCGOpcode *this_list = g->opt_opc ? : vecop_list_empty;
1472 const TCGOpcode *hold_list = tcg_swap_vecop_list(this_list);
adb196cb
RH		 1473 TCGType type;
1474 uint32_t some;
1475
db432672
RH		 1476 check_size_align(oprsz, maxsz, dofs | aofs | bofs | cofs);
1477 check_overlap_4(dofs, aofs, bofs, cofs, maxsz);
1478
adb196cb
RH		 1479 type = 0;
1480 if (g->fniv) {
53229a77 1481 type = choose_vector_type(g->opt_opc, g->vece, oprsz, g->prefer_i64);
adb196cb
RH		 1482 }
1483 switch (type) {
1484 case TCG_TYPE_V256:
1485 /* Recall that ARM SVE allows vector sizes that are not a
1486 * power of 2, but always a multiple of 16. The intent is
1487 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
1488 */
1489 some = QEMU_ALIGN_DOWN(oprsz, 32);
db432672 1490 expand_4_vec(g->vece, dofs, aofs, bofs, cofs, some,
5d6acdd4 1491 32, TCG_TYPE_V256, g->write_aofs, g->fniv);
db432672 1492 if (some == oprsz) {
adb196cb 1493 break;
db432672
RH		 1494 }
1495 dofs += some;
1496 aofs += some;
1497 bofs += some;
1498 cofs += some;
1499 oprsz -= some;
1500 maxsz -= some;
adb196cb
RH		 1501 /* fallthru */
1502 case TCG_TYPE_V128:
db432672 1503 expand_4_vec(g->vece, dofs, aofs, bofs, cofs, oprsz,
5d6acdd4 1504 16, TCG_TYPE_V128, g->write_aofs, g->fniv);
adb196cb
RH		 1505 break;
1506 case TCG_TYPE_V64:
db432672 1507 expand_4_vec(g->vece, dofs, aofs, bofs, cofs, oprsz,
5d6acdd4 1508 8, TCG_TYPE_V64, g->write_aofs, g->fniv);
adb196cb
RH		 1509 break;
1510
1511 case 0:
1512 if (g->fni8 && check_size_impl(oprsz, 8)) {
5d6acdd4
RH		 1513 expand_4_i64(dofs, aofs, bofs, cofs, oprsz,
1514 g->write_aofs, g->fni8);
adb196cb 1515 } else if (g->fni4 && check_size_impl(oprsz, 4)) {
5d6acdd4
RH		 1516 expand_4_i32(dofs, aofs, bofs, cofs, oprsz,
1517 g->write_aofs, g->fni4);
adb196cb
RH		 1518 } else {
1519 assert(g->fno != NULL);
1520 tcg_gen_gvec_4_ool(dofs, aofs, bofs, cofs,
1521 oprsz, maxsz, g->data, g->fno);
53229a77 1522 oprsz = maxsz;
adb196cb
RH		 1523 }
1524 break;
1525
1526 default:
1527 g_assert_not_reached();
db432672 1528 }
53229a77 1529 tcg_swap_vecop_list(hold_list);
db432672 1530
db432672
RH		 1531 if (oprsz < maxsz) {
1532 expand_clr(dofs + oprsz, maxsz - oprsz);
1533 }
1534}
1535
1536/*
1537 * Expand specific vector operations.
1538 */
1539
1540static void vec_mov2(unsigned vece, TCGv_vec a, TCGv_vec b)
1541{
1542 tcg_gen_mov_vec(a, b);
1543}
1544
1545void tcg_gen_gvec_mov(unsigned vece, uint32_t dofs, uint32_t aofs,
1546 uint32_t oprsz, uint32_t maxsz)
1547{
1548 static const GVecGen2 g = {
1549 .fni8 = tcg_gen_mov_i64,
1550 .fniv = vec_mov2,
1551 .fno = gen_helper_gvec_mov,
1552 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
1553 };
1554 if (dofs != aofs) {
1555 tcg_gen_gvec_2(dofs, aofs, oprsz, maxsz, &g);
1556 } else {
1557 check_size_align(oprsz, maxsz, dofs);
1558 if (oprsz < maxsz) {
1559 expand_clr(dofs + oprsz, maxsz - oprsz);
1560 }
1561 }
1562}
1563
1564void tcg_gen_gvec_dup_i32(unsigned vece, uint32_t dofs, uint32_t oprsz,
1565 uint32_t maxsz, TCGv_i32 in)
1566{
1567 check_size_align(oprsz, maxsz, dofs);
1568 tcg_debug_assert(vece <= MO_32);
1569 do_dup(vece, dofs, oprsz, maxsz, in, NULL, 0);
1570}
1571
1572void tcg_gen_gvec_dup_i64(unsigned vece, uint32_t dofs, uint32_t oprsz,
1573 uint32_t maxsz, TCGv_i64 in)
1574{
1575 check_size_align(oprsz, maxsz, dofs);
1576 tcg_debug_assert(vece <= MO_64);
1577 do_dup(vece, dofs, oprsz, maxsz, NULL, in, 0);
1578}
1579
1580void tcg_gen_gvec_dup_mem(unsigned vece, uint32_t dofs, uint32_t aofs,
1581 uint32_t oprsz, uint32_t maxsz)
1582{
532ba368 1583 check_size_align(oprsz, maxsz, dofs);
37ee55a0 1584 if (vece <= MO_64) {
532ba368 1585 TCGType type = choose_vector_type(NULL, vece, oprsz, 0);
37ee55a0
RH		 1586 if (type != 0) {
1587 TCGv_vec t_vec = tcg_temp_new_vec(type);
1588 tcg_gen_dup_mem_vec(vece, t_vec, cpu_env, aofs);
1589 do_dup_store(type, dofs, oprsz, maxsz, t_vec);
1590 tcg_temp_free_vec(t_vec);
532ba368
RH		 1591 } else if (vece <= MO_32) {
1592 TCGv_i32 in = tcg_temp_new_i32();
1593 switch (vece) {
1594 case MO_8:
1595 tcg_gen_ld8u_i32(in, cpu_env, aofs);
1596 break;
1597 case MO_16:
1598 tcg_gen_ld16u_i32(in, cpu_env, aofs);
1599 break;
1600 default:
1601 tcg_gen_ld_i32(in, cpu_env, aofs);
1602 break;
1603 }
1604 do_dup(vece, dofs, oprsz, maxsz, in, NULL, 0);
1605 tcg_temp_free_i32(in);
1606 } else {
1607 TCGv_i64 in = tcg_temp_new_i64();
1608 tcg_gen_ld_i64(in, cpu_env, aofs);
1609 do_dup(vece, dofs, oprsz, maxsz, NULL, in, 0);
1610 tcg_temp_free_i64(in);
db432672 1611 }
fe4b0b5b 1612 } else if (vece == 4) {
db432672 1613 /* 128-bit duplicate. */
db432672
RH		 1614 int i;
1615
db432672
RH		 1616 tcg_debug_assert(oprsz >= 16);
1617 if (TCG_TARGET_HAS_v128) {
1618 TCGv_vec in = tcg_temp_new_vec(TCG_TYPE_V128);
1619
1620 tcg_gen_ld_vec(in, cpu_env, aofs);
6a176461 1621 for (i = (aofs == dofs) * 16; i < oprsz; i += 16) {
db432672
RH		 1622 tcg_gen_st_vec(in, cpu_env, dofs + i);
1623 }
1624 tcg_temp_free_vec(in);
1625 } else {
1626 TCGv_i64 in0 = tcg_temp_new_i64();
1627 TCGv_i64 in1 = tcg_temp_new_i64();
1628
1629 tcg_gen_ld_i64(in0, cpu_env, aofs);
1630 tcg_gen_ld_i64(in1, cpu_env, aofs + 8);
6a176461 1631 for (i = (aofs == dofs) * 16; i < oprsz; i += 16) {
db432672
RH		 1632 tcg_gen_st_i64(in0, cpu_env, dofs + i);
1633 tcg_gen_st_i64(in1, cpu_env, dofs + i + 8);
1634 }
1635 tcg_temp_free_i64(in0);
1636 tcg_temp_free_i64(in1);
1637 }
532ba368
RH		 1638 if (oprsz < maxsz) {
1639 expand_clr(dofs + oprsz, maxsz - oprsz);
1640 }
fe4b0b5b
RH		 1641 } else if (vece == 5) {
1642 /* 256-bit duplicate. */
1643 int i;
1644
1645 tcg_debug_assert(oprsz >= 32);
1646 tcg_debug_assert(oprsz % 32 == 0);
1647 if (TCG_TARGET_HAS_v256) {
1648 TCGv_vec in = tcg_temp_new_vec(TCG_TYPE_V256);
1649
1650 tcg_gen_ld_vec(in, cpu_env, aofs);
1651 for (i = (aofs == dofs) * 32; i < oprsz; i += 32) {
1652 tcg_gen_st_vec(in, cpu_env, dofs + i);
1653 }
1654 tcg_temp_free_vec(in);
1655 } else if (TCG_TARGET_HAS_v128) {
1656 TCGv_vec in0 = tcg_temp_new_vec(TCG_TYPE_V128);
1657 TCGv_vec in1 = tcg_temp_new_vec(TCG_TYPE_V128);
1658
1659 tcg_gen_ld_vec(in0, cpu_env, aofs);
1660 tcg_gen_ld_vec(in1, cpu_env, aofs + 16);
1661 for (i = (aofs == dofs) * 32; i < oprsz; i += 32) {
1662 tcg_gen_st_vec(in0, cpu_env, dofs + i);
1663 tcg_gen_st_vec(in1, cpu_env, dofs + i + 16);
1664 }
1665 tcg_temp_free_vec(in0);
1666 tcg_temp_free_vec(in1);
1667 } else {
1668 TCGv_i64 in[4];
1669 int j;
1670
1671 for (j = 0; j < 4; ++j) {
1672 in[j] = tcg_temp_new_i64();
1673 tcg_gen_ld_i64(in[j], cpu_env, aofs + j * 8);
1674 }
1675 for (i = (aofs == dofs) * 32; i < oprsz; i += 32) {
1676 for (j = 0; j < 4; ++j) {
1677 tcg_gen_st_i64(in[j], cpu_env, dofs + i + j * 8);
1678 }
1679 }
1680 for (j = 0; j < 4; ++j) {
1681 tcg_temp_free_i64(in[j]);
1682 }
1683 }
1684 if (oprsz < maxsz) {
1685 expand_clr(dofs + oprsz, maxsz - oprsz);
1686 }
1687 } else {
1688 g_assert_not_reached();
db432672
RH		 1689 }
1690}
1691
44c94677
RH		 1692void tcg_gen_gvec_dup_imm(unsigned vece, uint32_t dofs, uint32_t oprsz,
1693 uint32_t maxsz, uint64_t x)
1694{
1695 check_size_align(oprsz, maxsz, dofs);
1696 do_dup(vece, dofs, oprsz, maxsz, NULL, NULL, x);
1697}
1698
db432672
RH		 1699void tcg_gen_gvec_not(unsigned vece, uint32_t dofs, uint32_t aofs,
1700 uint32_t oprsz, uint32_t maxsz)
1701{
1702 static const GVecGen2 g = {
1703 .fni8 = tcg_gen_not_i64,
1704 .fniv = tcg_gen_not_vec,
1705 .fno = gen_helper_gvec_not,
1706 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
1707 };
1708 tcg_gen_gvec_2(dofs, aofs, oprsz, maxsz, &g);
1709}
1710
1711/* Perform a vector addition using normal addition and a mask. The mask
1712 should be the sign bit of each lane. This 6-operation form is more
1713 efficient than separate additions when there are 4 or more lanes in
1714 the 64-bit operation. */
1715static void gen_addv_mask(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b, TCGv_i64 m)
1716{
1717 TCGv_i64 t1 = tcg_temp_new_i64();
1718 TCGv_i64 t2 = tcg_temp_new_i64();
1719 TCGv_i64 t3 = tcg_temp_new_i64();
1720
1721 tcg_gen_andc_i64(t1, a, m);
1722 tcg_gen_andc_i64(t2, b, m);
1723 tcg_gen_xor_i64(t3, a, b);
1724 tcg_gen_add_i64(d, t1, t2);
1725 tcg_gen_and_i64(t3, t3, m);
1726 tcg_gen_xor_i64(d, d, t3);
1727
1728 tcg_temp_free_i64(t1);
1729 tcg_temp_free_i64(t2);
1730 tcg_temp_free_i64(t3);
1731}
1732
1733void tcg_gen_vec_add8_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
1734{
88d4005b 1735 TCGv_i64 m = tcg_constant_i64(dup_const(MO_8, 0x80));
db432672 1736 gen_addv_mask(d, a, b, m);
db432672
RH		 1737}
1738
448e7aa2
LZ		 1739void tcg_gen_vec_add8_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
1740{
1741 TCGv_i32 m = tcg_constant_i32((int32_t)dup_const(MO_8, 0x80));
1742 TCGv_i32 t1 = tcg_temp_new_i32();
1743 TCGv_i32 t2 = tcg_temp_new_i32();
1744 TCGv_i32 t3 = tcg_temp_new_i32();
1745
1746 tcg_gen_andc_i32(t1, a, m);
1747 tcg_gen_andc_i32(t2, b, m);
1748 tcg_gen_xor_i32(t3, a, b);
1749 tcg_gen_add_i32(d, t1, t2);
1750 tcg_gen_and_i32(t3, t3, m);
1751 tcg_gen_xor_i32(d, d, t3);
1752
1753 tcg_temp_free_i32(t1);
1754 tcg_temp_free_i32(t2);
1755 tcg_temp_free_i32(t3);
1756}
1757
db432672
RH		 1758void tcg_gen_vec_add16_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
1759{
88d4005b 1760 TCGv_i64 m = tcg_constant_i64(dup_const(MO_16, 0x8000));
db432672 1761 gen_addv_mask(d, a, b, m);
db432672
RH		 1762}
1763
3d066e5d
LZ		 1764void tcg_gen_vec_add16_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
1765{
1766 TCGv_i32 t1 = tcg_temp_new_i32();
1767 TCGv_i32 t2 = tcg_temp_new_i32();
1768
1769 tcg_gen_andi_i32(t1, a, ~0xffff);
1770 tcg_gen_add_i32(t2, a, b);
1771 tcg_gen_add_i32(t1, t1, b);
1772 tcg_gen_deposit_i32(d, t1, t2, 0, 16);
1773
1774 tcg_temp_free_i32(t1);
1775 tcg_temp_free_i32(t2);
1776}
1777
db432672
RH		 1778void tcg_gen_vec_add32_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
1779{
1780 TCGv_i64 t1 = tcg_temp_new_i64();
1781 TCGv_i64 t2 = tcg_temp_new_i64();
1782
1783 tcg_gen_andi_i64(t1, a, ~0xffffffffull);
1784 tcg_gen_add_i64(t2, a, b);
1785 tcg_gen_add_i64(t1, t1, b);
1786 tcg_gen_deposit_i64(d, t1, t2, 0, 32);
1787
1788 tcg_temp_free_i64(t1);
1789 tcg_temp_free_i64(t2);
1790}
1791
53229a77
RH		 1792static const TCGOpcode vecop_list_add[] = { INDEX_op_add_vec, 0 };
1793
db432672
RH		 1794void tcg_gen_gvec_add(unsigned vece, uint32_t dofs, uint32_t aofs,
1795 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
1796{
1797 static const GVecGen3 g[4] = {
1798 { .fni8 = tcg_gen_vec_add8_i64,
1799 .fniv = tcg_gen_add_vec,
1800 .fno = gen_helper_gvec_add8,
53229a77 1801 .opt_opc = vecop_list_add,
db432672
RH		 1802 .vece = MO_8 },
1803 { .fni8 = tcg_gen_vec_add16_i64,
1804 .fniv = tcg_gen_add_vec,
1805 .fno = gen_helper_gvec_add16,
53229a77 1806 .opt_opc = vecop_list_add,
db432672
RH		 1807 .vece = MO_16 },
1808 { .fni4 = tcg_gen_add_i32,
1809 .fniv = tcg_gen_add_vec,
1810 .fno = gen_helper_gvec_add32,
53229a77 1811 .opt_opc = vecop_list_add,
db432672
RH		 1812 .vece = MO_32 },
1813 { .fni8 = tcg_gen_add_i64,
1814 .fniv = tcg_gen_add_vec,
1815 .fno = gen_helper_gvec_add64,
53229a77 1816 .opt_opc = vecop_list_add,
db432672
RH		 1817 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
1818 .vece = MO_64 },
1819 };
1820
1821 tcg_debug_assert(vece <= MO_64);
1822 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
1823}
1824
22fc3527
RH		 1825void tcg_gen_gvec_adds(unsigned vece, uint32_t dofs, uint32_t aofs,
1826 TCGv_i64 c, uint32_t oprsz, uint32_t maxsz)
1827{
1828 static const GVecGen2s g[4] = {
1829 { .fni8 = tcg_gen_vec_add8_i64,
1830 .fniv = tcg_gen_add_vec,
1831 .fno = gen_helper_gvec_adds8,
53229a77 1832 .opt_opc = vecop_list_add,
22fc3527
RH		 1833 .vece = MO_8 },
1834 { .fni8 = tcg_gen_vec_add16_i64,
1835 .fniv = tcg_gen_add_vec,
1836 .fno = gen_helper_gvec_adds16,
53229a77 1837 .opt_opc = vecop_list_add,
22fc3527
RH		 1838 .vece = MO_16 },
1839 { .fni4 = tcg_gen_add_i32,
1840 .fniv = tcg_gen_add_vec,
1841 .fno = gen_helper_gvec_adds32,
53229a77 1842 .opt_opc = vecop_list_add,
22fc3527
RH		 1843 .vece = MO_32 },
1844 { .fni8 = tcg_gen_add_i64,
1845 .fniv = tcg_gen_add_vec,
1846 .fno = gen_helper_gvec_adds64,
53229a77 1847 .opt_opc = vecop_list_add,
22fc3527
RH		 1848 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
1849 .vece = MO_64 },
1850 };
1851
1852 tcg_debug_assert(vece <= MO_64);
1853 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, c, &g[vece]);
1854}
1855
1856void tcg_gen_gvec_addi(unsigned vece, uint32_t dofs, uint32_t aofs,
1857 int64_t c, uint32_t oprsz, uint32_t maxsz)
1858{
88d4005b 1859 TCGv_i64 tmp = tcg_constant_i64(c);
22fc3527 1860 tcg_gen_gvec_adds(vece, dofs, aofs, tmp, oprsz, maxsz);
22fc3527
RH		 1861}
1862
53229a77
RH		 1863static const TCGOpcode vecop_list_sub[] = { INDEX_op_sub_vec, 0 };
1864
22fc3527
RH		 1865void tcg_gen_gvec_subs(unsigned vece, uint32_t dofs, uint32_t aofs,
1866 TCGv_i64 c, uint32_t oprsz, uint32_t maxsz)
1867{
1868 static const GVecGen2s g[4] = {
1869 { .fni8 = tcg_gen_vec_sub8_i64,
1870 .fniv = tcg_gen_sub_vec,
1871 .fno = gen_helper_gvec_subs8,
53229a77 1872 .opt_opc = vecop_list_sub,
22fc3527
RH		 1873 .vece = MO_8 },
1874 { .fni8 = tcg_gen_vec_sub16_i64,
1875 .fniv = tcg_gen_sub_vec,
1876 .fno = gen_helper_gvec_subs16,
53229a77 1877 .opt_opc = vecop_list_sub,
22fc3527
RH		 1878 .vece = MO_16 },
1879 { .fni4 = tcg_gen_sub_i32,
1880 .fniv = tcg_gen_sub_vec,
1881 .fno = gen_helper_gvec_subs32,
53229a77 1882 .opt_opc = vecop_list_sub,
22fc3527
RH		 1883 .vece = MO_32 },
1884 { .fni8 = tcg_gen_sub_i64,
1885 .fniv = tcg_gen_sub_vec,
1886 .fno = gen_helper_gvec_subs64,
53229a77 1887 .opt_opc = vecop_list_sub,
22fc3527
RH		 1888 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
1889 .vece = MO_64 },
1890 };
1891
1892 tcg_debug_assert(vece <= MO_64);
1893 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, c, &g[vece]);
1894}
1895
db432672
RH		 1896/* Perform a vector subtraction using normal subtraction and a mask.
1897 Compare gen_addv_mask above. */
1898static void gen_subv_mask(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b, TCGv_i64 m)
1899{
1900 TCGv_i64 t1 = tcg_temp_new_i64();
1901 TCGv_i64 t2 = tcg_temp_new_i64();
1902 TCGv_i64 t3 = tcg_temp_new_i64();
1903
1904 tcg_gen_or_i64(t1, a, m);
1905 tcg_gen_andc_i64(t2, b, m);
1906 tcg_gen_eqv_i64(t3, a, b);
1907 tcg_gen_sub_i64(d, t1, t2);
1908 tcg_gen_and_i64(t3, t3, m);
1909 tcg_gen_xor_i64(d, d, t3);
1910
1911 tcg_temp_free_i64(t1);
1912 tcg_temp_free_i64(t2);
1913 tcg_temp_free_i64(t3);
1914}
1915
1916void tcg_gen_vec_sub8_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
1917{
88d4005b 1918 TCGv_i64 m = tcg_constant_i64(dup_const(MO_8, 0x80));
db432672 1919 gen_subv_mask(d, a, b, m);
db432672
RH		 1920}
1921
448e7aa2
LZ		 1922void tcg_gen_vec_sub8_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
1923{
1924 TCGv_i32 m = tcg_constant_i32((int32_t)dup_const(MO_8, 0x80));
1925 TCGv_i32 t1 = tcg_temp_new_i32();
1926 TCGv_i32 t2 = tcg_temp_new_i32();
1927 TCGv_i32 t3 = tcg_temp_new_i32();
1928
1929 tcg_gen_or_i32(t1, a, m);
1930 tcg_gen_andc_i32(t2, b, m);
1931 tcg_gen_eqv_i32(t3, a, b);
1932 tcg_gen_sub_i32(d, t1, t2);
1933 tcg_gen_and_i32(t3, t3, m);
1934 tcg_gen_xor_i32(d, d, t3);
1935
1936 tcg_temp_free_i32(t1);
1937 tcg_temp_free_i32(t2);
1938 tcg_temp_free_i32(t3);
1939}
1940
db432672
RH		 1941void tcg_gen_vec_sub16_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
1942{
88d4005b 1943 TCGv_i64 m = tcg_constant_i64(dup_const(MO_16, 0x8000));
db432672 1944 gen_subv_mask(d, a, b, m);
db432672
RH		 1945}
1946
3d066e5d
LZ		 1947void tcg_gen_vec_sub16_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
1948{
1949 TCGv_i32 t1 = tcg_temp_new_i32();
1950 TCGv_i32 t2 = tcg_temp_new_i32();
1951
1952 tcg_gen_andi_i32(t1, b, ~0xffff);
1953 tcg_gen_sub_i32(t2, a, b);
1954 tcg_gen_sub_i32(t1, a, t1);
1955 tcg_gen_deposit_i32(d, t1, t2, 0, 16);
1956
1957 tcg_temp_free_i32(t1);
1958 tcg_temp_free_i32(t2);
1959}
1960
db432672
RH		 1961void tcg_gen_vec_sub32_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
1962{
1963 TCGv_i64 t1 = tcg_temp_new_i64();
1964 TCGv_i64 t2 = tcg_temp_new_i64();
1965
1966 tcg_gen_andi_i64(t1, b, ~0xffffffffull);
1967 tcg_gen_sub_i64(t2, a, b);
1968 tcg_gen_sub_i64(t1, a, t1);
1969 tcg_gen_deposit_i64(d, t1, t2, 0, 32);
1970
1971 tcg_temp_free_i64(t1);
1972 tcg_temp_free_i64(t2);
1973}
1974
1975void tcg_gen_gvec_sub(unsigned vece, uint32_t dofs, uint32_t aofs,
1976 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
1977{
1978 static const GVecGen3 g[4] = {
1979 { .fni8 = tcg_gen_vec_sub8_i64,
1980 .fniv = tcg_gen_sub_vec,
1981 .fno = gen_helper_gvec_sub8,
53229a77 1982 .opt_opc = vecop_list_sub,
db432672
RH		 1983 .vece = MO_8 },
1984 { .fni8 = tcg_gen_vec_sub16_i64,
1985 .fniv = tcg_gen_sub_vec,
1986 .fno = gen_helper_gvec_sub16,
53229a77 1987 .opt_opc = vecop_list_sub,
db432672
RH		 1988 .vece = MO_16 },
1989 { .fni4 = tcg_gen_sub_i32,
1990 .fniv = tcg_gen_sub_vec,
1991 .fno = gen_helper_gvec_sub32,
53229a77 1992 .opt_opc = vecop_list_sub,
db432672
RH		 1993 .vece = MO_32 },
1994 { .fni8 = tcg_gen_sub_i64,
1995 .fniv = tcg_gen_sub_vec,
1996 .fno = gen_helper_gvec_sub64,
53229a77 1997 .opt_opc = vecop_list_sub,
db432672
RH		 1998 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
1999 .vece = MO_64 },
2000 };
2001
2002 tcg_debug_assert(vece <= MO_64);
2003 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
2004}
2005
53229a77
RH		 2006static const TCGOpcode vecop_list_mul[] = { INDEX_op_mul_vec, 0 };
2007
3774030a
RH		 2008void tcg_gen_gvec_mul(unsigned vece, uint32_t dofs, uint32_t aofs,
2009 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2010{
2011 static const GVecGen3 g[4] = {
2012 { .fniv = tcg_gen_mul_vec,
2013 .fno = gen_helper_gvec_mul8,
53229a77 2014 .opt_opc = vecop_list_mul,
3774030a
RH		 2015 .vece = MO_8 },
2016 { .fniv = tcg_gen_mul_vec,
2017 .fno = gen_helper_gvec_mul16,
53229a77 2018 .opt_opc = vecop_list_mul,
3774030a
RH		 2019 .vece = MO_16 },
2020 { .fni4 = tcg_gen_mul_i32,
2021 .fniv = tcg_gen_mul_vec,
2022 .fno = gen_helper_gvec_mul32,
53229a77 2023 .opt_opc = vecop_list_mul,
3774030a
RH		 2024 .vece = MO_32 },
2025 { .fni8 = tcg_gen_mul_i64,
2026 .fniv = tcg_gen_mul_vec,
2027 .fno = gen_helper_gvec_mul64,
53229a77 2028 .opt_opc = vecop_list_mul,
3774030a
RH		 2029 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2030 .vece = MO_64 },
2031 };
2032
2033 tcg_debug_assert(vece <= MO_64);
2034 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
2035}
2036
22fc3527
RH		 2037void tcg_gen_gvec_muls(unsigned vece, uint32_t dofs, uint32_t aofs,
2038 TCGv_i64 c, uint32_t oprsz, uint32_t maxsz)
2039{
2040 static const GVecGen2s g[4] = {
2041 { .fniv = tcg_gen_mul_vec,
2042 .fno = gen_helper_gvec_muls8,
53229a77 2043 .opt_opc = vecop_list_mul,
22fc3527
RH		 2044 .vece = MO_8 },
2045 { .fniv = tcg_gen_mul_vec,
2046 .fno = gen_helper_gvec_muls16,
53229a77 2047 .opt_opc = vecop_list_mul,
22fc3527
RH		 2048 .vece = MO_16 },
2049 { .fni4 = tcg_gen_mul_i32,
2050 .fniv = tcg_gen_mul_vec,
2051 .fno = gen_helper_gvec_muls32,
53229a77 2052 .opt_opc = vecop_list_mul,
22fc3527
RH		 2053 .vece = MO_32 },
2054 { .fni8 = tcg_gen_mul_i64,
2055 .fniv = tcg_gen_mul_vec,
2056 .fno = gen_helper_gvec_muls64,
53229a77 2057 .opt_opc = vecop_list_mul,
22fc3527
RH		 2058 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2059 .vece = MO_64 },
2060 };
2061
2062 tcg_debug_assert(vece <= MO_64);
2063 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, c, &g[vece]);
2064}
2065
2066void tcg_gen_gvec_muli(unsigned vece, uint32_t dofs, uint32_t aofs,
2067 int64_t c, uint32_t oprsz, uint32_t maxsz)
2068{
88d4005b 2069 TCGv_i64 tmp = tcg_constant_i64(c);
22fc3527 2070 tcg_gen_gvec_muls(vece, dofs, aofs, tmp, oprsz, maxsz);
22fc3527
RH		 2071}
2072
f49b12c6
RH		 2073void tcg_gen_gvec_ssadd(unsigned vece, uint32_t dofs, uint32_t aofs,
2074 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2075{
53229a77 2076 static const TCGOpcode vecop_list[] = { INDEX_op_ssadd_vec, 0 };
f49b12c6 2077 static const GVecGen3 g[4] = {
8afaf050
RH		 2078 { .fniv = tcg_gen_ssadd_vec,
2079 .fno = gen_helper_gvec_ssadd8,
53229a77 2080 .opt_opc = vecop_list,
8afaf050
RH		 2081 .vece = MO_8 },
2082 { .fniv = tcg_gen_ssadd_vec,
2083 .fno = gen_helper_gvec_ssadd16,
53229a77 2084 .opt_opc = vecop_list,
8afaf050
RH		 2085 .vece = MO_16 },
2086 { .fniv = tcg_gen_ssadd_vec,
2087 .fno = gen_helper_gvec_ssadd32,
53229a77 2088 .opt_opc = vecop_list,
8afaf050
RH		 2089 .vece = MO_32 },
2090 { .fniv = tcg_gen_ssadd_vec,
2091 .fno = gen_helper_gvec_ssadd64,
53229a77 2092 .opt_opc = vecop_list,
8afaf050 2093 .vece = MO_64 },
f49b12c6
RH		 2094 };
2095 tcg_debug_assert(vece <= MO_64);
2096 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
2097}
2098
2099void tcg_gen_gvec_sssub(unsigned vece, uint32_t dofs, uint32_t aofs,
2100 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2101{
53229a77 2102 static const TCGOpcode vecop_list[] = { INDEX_op_sssub_vec, 0 };
f49b12c6 2103 static const GVecGen3 g[4] = {
8afaf050
RH		 2104 { .fniv = tcg_gen_sssub_vec,
2105 .fno = gen_helper_gvec_sssub8,
53229a77 2106 .opt_opc = vecop_list,
8afaf050
RH		 2107 .vece = MO_8 },
2108 { .fniv = tcg_gen_sssub_vec,
2109 .fno = gen_helper_gvec_sssub16,
53229a77 2110 .opt_opc = vecop_list,
8afaf050
RH		 2111 .vece = MO_16 },
2112 { .fniv = tcg_gen_sssub_vec,
2113 .fno = gen_helper_gvec_sssub32,
53229a77 2114 .opt_opc = vecop_list,
8afaf050
RH		 2115 .vece = MO_32 },
2116 { .fniv = tcg_gen_sssub_vec,
2117 .fno = gen_helper_gvec_sssub64,
53229a77 2118 .opt_opc = vecop_list,
8afaf050 2119 .vece = MO_64 },
f49b12c6
RH		 2120 };
2121 tcg_debug_assert(vece <= MO_64);
2122 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
2123}
2124
8afaf050 2125static void tcg_gen_usadd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
f49b12c6 2126{
88d4005b 2127 TCGv_i32 max = tcg_constant_i32(-1);
f49b12c6
RH		 2128 tcg_gen_add_i32(d, a, b);
2129 tcg_gen_movcond_i32(TCG_COND_LTU, d, d, a, max, d);
f49b12c6
RH		 2130}
2131
8afaf050 2132static void tcg_gen_usadd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
f49b12c6 2133{
88d4005b 2134 TCGv_i64 max = tcg_constant_i64(-1);
f49b12c6
RH		 2135 tcg_gen_add_i64(d, a, b);
2136 tcg_gen_movcond_i64(TCG_COND_LTU, d, d, a, max, d);
f49b12c6
RH		 2137}
2138
2139void tcg_gen_gvec_usadd(unsigned vece, uint32_t dofs, uint32_t aofs,
2140 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2141{
53229a77 2142 static const TCGOpcode vecop_list[] = { INDEX_op_usadd_vec, 0 };
f49b12c6 2143 static const GVecGen3 g[4] = {
8afaf050
RH		 2144 { .fniv = tcg_gen_usadd_vec,
2145 .fno = gen_helper_gvec_usadd8,
53229a77 2146 .opt_opc = vecop_list,
8afaf050
RH		 2147 .vece = MO_8 },
2148 { .fniv = tcg_gen_usadd_vec,
2149 .fno = gen_helper_gvec_usadd16,
53229a77 2150 .opt_opc = vecop_list,
8afaf050
RH		 2151 .vece = MO_16 },
2152 { .fni4 = tcg_gen_usadd_i32,
2153 .fniv = tcg_gen_usadd_vec,
f49b12c6 2154 .fno = gen_helper_gvec_usadd32,
53229a77 2155 .opt_opc = vecop_list,
f49b12c6 2156 .vece = MO_32 },
8afaf050
RH		 2157 { .fni8 = tcg_gen_usadd_i64,
2158 .fniv = tcg_gen_usadd_vec,
f49b12c6 2159 .fno = gen_helper_gvec_usadd64,
53229a77 2160 .opt_opc = vecop_list,
f49b12c6
RH		 2161 .vece = MO_64 }
2162 };
2163 tcg_debug_assert(vece <= MO_64);
2164 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
2165}
2166
8afaf050 2167static void tcg_gen_ussub_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
f49b12c6 2168{
88d4005b 2169 TCGv_i32 min = tcg_constant_i32(0);
f49b12c6
RH		 2170 tcg_gen_sub_i32(d, a, b);
2171 tcg_gen_movcond_i32(TCG_COND_LTU, d, a, b, min, d);
f49b12c6
RH		 2172}
2173
8afaf050 2174static void tcg_gen_ussub_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
f49b12c6 2175{
88d4005b 2176 TCGv_i64 min = tcg_constant_i64(0);
f49b12c6
RH		 2177 tcg_gen_sub_i64(d, a, b);
2178 tcg_gen_movcond_i64(TCG_COND_LTU, d, a, b, min, d);
f49b12c6
RH		 2179}
2180
2181void tcg_gen_gvec_ussub(unsigned vece, uint32_t dofs, uint32_t aofs,
2182 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2183{
53229a77 2184 static const TCGOpcode vecop_list[] = { INDEX_op_ussub_vec, 0 };
f49b12c6 2185 static const GVecGen3 g[4] = {
8afaf050
RH		 2186 { .fniv = tcg_gen_ussub_vec,
2187 .fno = gen_helper_gvec_ussub8,
53229a77 2188 .opt_opc = vecop_list,
8afaf050
RH		 2189 .vece = MO_8 },
2190 { .fniv = tcg_gen_ussub_vec,
2191 .fno = gen_helper_gvec_ussub16,
53229a77 2192 .opt_opc = vecop_list,
8afaf050
RH		 2193 .vece = MO_16 },
2194 { .fni4 = tcg_gen_ussub_i32,
2195 .fniv = tcg_gen_ussub_vec,
f49b12c6 2196 .fno = gen_helper_gvec_ussub32,
53229a77 2197 .opt_opc = vecop_list,
f49b12c6 2198 .vece = MO_32 },
8afaf050
RH		 2199 { .fni8 = tcg_gen_ussub_i64,
2200 .fniv = tcg_gen_ussub_vec,
f49b12c6 2201 .fno = gen_helper_gvec_ussub64,
53229a77 2202 .opt_opc = vecop_list,
f49b12c6
RH		 2203 .vece = MO_64 }
2204 };
2205 tcg_debug_assert(vece <= MO_64);
2206 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
dd0a0fcd
RH		 2207}
2208
2209void tcg_gen_gvec_smin(unsigned vece, uint32_t dofs, uint32_t aofs,
2210 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2211{
53229a77 2212 static const TCGOpcode vecop_list[] = { INDEX_op_smin_vec, 0 };
dd0a0fcd
RH		 2213 static const GVecGen3 g[4] = {
2214 { .fniv = tcg_gen_smin_vec,
2215 .fno = gen_helper_gvec_smin8,
53229a77 2216 .opt_opc = vecop_list,
dd0a0fcd
RH		 2217 .vece = MO_8 },
2218 { .fniv = tcg_gen_smin_vec,
2219 .fno = gen_helper_gvec_smin16,
53229a77 2220 .opt_opc = vecop_list,
dd0a0fcd
RH		 2221 .vece = MO_16 },
2222 { .fni4 = tcg_gen_smin_i32,
2223 .fniv = tcg_gen_smin_vec,
2224 .fno = gen_helper_gvec_smin32,
53229a77 2225 .opt_opc = vecop_list,
dd0a0fcd
RH		 2226 .vece = MO_32 },
2227 { .fni8 = tcg_gen_smin_i64,
2228 .fniv = tcg_gen_smin_vec,
2229 .fno = gen_helper_gvec_smin64,
53229a77 2230 .opt_opc = vecop_list,
dd0a0fcd
RH		 2231 .vece = MO_64 }
2232 };
2233 tcg_debug_assert(vece <= MO_64);
2234 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
2235}
2236
2237void tcg_gen_gvec_umin(unsigned vece, uint32_t dofs, uint32_t aofs,
2238 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2239{
53229a77 2240 static const TCGOpcode vecop_list[] = { INDEX_op_umin_vec, 0 };
dd0a0fcd
RH		 2241 static const GVecGen3 g[4] = {
2242 { .fniv = tcg_gen_umin_vec,
2243 .fno = gen_helper_gvec_umin8,
53229a77 2244 .opt_opc = vecop_list,
dd0a0fcd
RH		 2245 .vece = MO_8 },
2246 { .fniv = tcg_gen_umin_vec,
2247 .fno = gen_helper_gvec_umin16,
53229a77 2248 .opt_opc = vecop_list,
dd0a0fcd
RH		 2249 .vece = MO_16 },
2250 { .fni4 = tcg_gen_umin_i32,
2251 .fniv = tcg_gen_umin_vec,
2252 .fno = gen_helper_gvec_umin32,
53229a77 2253 .opt_opc = vecop_list,
dd0a0fcd
RH		 2254 .vece = MO_32 },
2255 { .fni8 = tcg_gen_umin_i64,
2256 .fniv = tcg_gen_umin_vec,
2257 .fno = gen_helper_gvec_umin64,
53229a77 2258 .opt_opc = vecop_list,
dd0a0fcd
RH		 2259 .vece = MO_64 }
2260 };
2261 tcg_debug_assert(vece <= MO_64);
2262 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
2263}
2264
2265void tcg_gen_gvec_smax(unsigned vece, uint32_t dofs, uint32_t aofs,
2266 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2267{
53229a77 2268 static const TCGOpcode vecop_list[] = { INDEX_op_smax_vec, 0 };
dd0a0fcd
RH		 2269 static const GVecGen3 g[4] = {
2270 { .fniv = tcg_gen_smax_vec,
2271 .fno = gen_helper_gvec_smax8,
53229a77 2272 .opt_opc = vecop_list,
dd0a0fcd
RH		 2273 .vece = MO_8 },
2274 { .fniv = tcg_gen_smax_vec,
2275 .fno = gen_helper_gvec_smax16,
53229a77 2276 .opt_opc = vecop_list,
dd0a0fcd
RH		 2277 .vece = MO_16 },
2278 { .fni4 = tcg_gen_smax_i32,
2279 .fniv = tcg_gen_smax_vec,
2280 .fno = gen_helper_gvec_smax32,
53229a77 2281 .opt_opc = vecop_list,
dd0a0fcd
RH		 2282 .vece = MO_32 },
2283 { .fni8 = tcg_gen_smax_i64,
2284 .fniv = tcg_gen_smax_vec,
2285 .fno = gen_helper_gvec_smax64,
53229a77 2286 .opt_opc = vecop_list,
dd0a0fcd
RH		 2287 .vece = MO_64 }
2288 };
2289 tcg_debug_assert(vece <= MO_64);
2290 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
2291}
2292
2293void tcg_gen_gvec_umax(unsigned vece, uint32_t dofs, uint32_t aofs,
2294 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2295{
53229a77 2296 static const TCGOpcode vecop_list[] = { INDEX_op_umax_vec, 0 };
dd0a0fcd
RH		 2297 static const GVecGen3 g[4] = {
2298 { .fniv = tcg_gen_umax_vec,
2299 .fno = gen_helper_gvec_umax8,
53229a77 2300 .opt_opc = vecop_list,
dd0a0fcd
RH		 2301 .vece = MO_8 },
2302 { .fniv = tcg_gen_umax_vec,
2303 .fno = gen_helper_gvec_umax16,
53229a77 2304 .opt_opc = vecop_list,
dd0a0fcd
RH		 2305 .vece = MO_16 },
2306 { .fni4 = tcg_gen_umax_i32,
2307 .fniv = tcg_gen_umax_vec,
2308 .fno = gen_helper_gvec_umax32,
53229a77 2309 .opt_opc = vecop_list,
dd0a0fcd
RH		 2310 .vece = MO_32 },
2311 { .fni8 = tcg_gen_umax_i64,
2312 .fniv = tcg_gen_umax_vec,
2313 .fno = gen_helper_gvec_umax64,
53229a77 2314 .opt_opc = vecop_list,
dd0a0fcd
RH		 2315 .vece = MO_64 }
2316 };
2317 tcg_debug_assert(vece <= MO_64);
2318 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
f49b12c6
RH		 2319}
2320
db432672
RH		 2321/* Perform a vector negation using normal negation and a mask.
2322 Compare gen_subv_mask above. */
2323static void gen_negv_mask(TCGv_i64 d, TCGv_i64 b, TCGv_i64 m)
2324{
2325 TCGv_i64 t2 = tcg_temp_new_i64();
2326 TCGv_i64 t3 = tcg_temp_new_i64();
2327
2328 tcg_gen_andc_i64(t3, m, b);
2329 tcg_gen_andc_i64(t2, b, m);
2330 tcg_gen_sub_i64(d, m, t2);
2331 tcg_gen_xor_i64(d, d, t3);
2332
2333 tcg_temp_free_i64(t2);
2334 tcg_temp_free_i64(t3);
2335}
2336
2337void tcg_gen_vec_neg8_i64(TCGv_i64 d, TCGv_i64 b)
2338{
88d4005b 2339 TCGv_i64 m = tcg_constant_i64(dup_const(MO_8, 0x80));
db432672 2340 gen_negv_mask(d, b, m);
db432672
RH		 2341}
2342
2343void tcg_gen_vec_neg16_i64(TCGv_i64 d, TCGv_i64 b)
2344{
88d4005b 2345 TCGv_i64 m = tcg_constant_i64(dup_const(MO_16, 0x8000));
db432672 2346 gen_negv_mask(d, b, m);
db432672
RH		 2347}
2348
2349void tcg_gen_vec_neg32_i64(TCGv_i64 d, TCGv_i64 b)
2350{
2351 TCGv_i64 t1 = tcg_temp_new_i64();
2352 TCGv_i64 t2 = tcg_temp_new_i64();
2353
2354 tcg_gen_andi_i64(t1, b, ~0xffffffffull);
2355 tcg_gen_neg_i64(t2, b);
2356 tcg_gen_neg_i64(t1, t1);
2357 tcg_gen_deposit_i64(d, t1, t2, 0, 32);
2358
2359 tcg_temp_free_i64(t1);
2360 tcg_temp_free_i64(t2);
2361}
2362
2363void tcg_gen_gvec_neg(unsigned vece, uint32_t dofs, uint32_t aofs,
2364 uint32_t oprsz, uint32_t maxsz)
2365{
53229a77 2366 static const TCGOpcode vecop_list[] = { INDEX_op_neg_vec, 0 };
db432672
RH		 2367 static const GVecGen2 g[4] = {
2368 { .fni8 = tcg_gen_vec_neg8_i64,
2369 .fniv = tcg_gen_neg_vec,
2370 .fno = gen_helper_gvec_neg8,
53229a77 2371 .opt_opc = vecop_list,
db432672
RH		 2372 .vece = MO_8 },
2373 { .fni8 = tcg_gen_vec_neg16_i64,
2374 .fniv = tcg_gen_neg_vec,
2375 .fno = gen_helper_gvec_neg16,
53229a77 2376 .opt_opc = vecop_list,
db432672
RH		 2377 .vece = MO_16 },
2378 { .fni4 = tcg_gen_neg_i32,
2379 .fniv = tcg_gen_neg_vec,
2380 .fno = gen_helper_gvec_neg32,
53229a77 2381 .opt_opc = vecop_list,
db432672
RH		 2382 .vece = MO_32 },
2383 { .fni8 = tcg_gen_neg_i64,
2384 .fniv = tcg_gen_neg_vec,
2385 .fno = gen_helper_gvec_neg64,
53229a77 2386 .opt_opc = vecop_list,
db432672
RH		 2387 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2388 .vece = MO_64 },
2389 };
2390
2391 tcg_debug_assert(vece <= MO_64);
2392 tcg_gen_gvec_2(dofs, aofs, oprsz, maxsz, &g[vece]);
2393}
2394
bcefc902
RH		 2395static void gen_absv_mask(TCGv_i64 d, TCGv_i64 b, unsigned vece)
2396{
2397 TCGv_i64 t = tcg_temp_new_i64();
2398 int nbit = 8 << vece;
2399
2400 /* Create -1 for each negative element. */
2401 tcg_gen_shri_i64(t, b, nbit - 1);
2402 tcg_gen_andi_i64(t, t, dup_const(vece, 1));
2403 tcg_gen_muli_i64(t, t, (1 << nbit) - 1);
2404
2405 /*
e7e8f33f 2406 * Invert (via xor -1) and add one.
bcefc902
RH		 2407 * Because of the ordering the msb is cleared,
2408 * so we never have carry into the next element.
2409 */
2410 tcg_gen_xor_i64(d, b, t);
e7e8f33f
SL		 2411 tcg_gen_andi_i64(t, t, dup_const(vece, 1));
2412 tcg_gen_add_i64(d, d, t);
bcefc902
RH		 2413
2414 tcg_temp_free_i64(t);
2415}
2416
2417static void tcg_gen_vec_abs8_i64(TCGv_i64 d, TCGv_i64 b)
2418{
2419 gen_absv_mask(d, b, MO_8);
2420}
2421
2422static void tcg_gen_vec_abs16_i64(TCGv_i64 d, TCGv_i64 b)
2423{
2424 gen_absv_mask(d, b, MO_16);
2425}
2426
2427void tcg_gen_gvec_abs(unsigned vece, uint32_t dofs, uint32_t aofs,
2428 uint32_t oprsz, uint32_t maxsz)
2429{
2430 static const TCGOpcode vecop_list[] = { INDEX_op_abs_vec, 0 };
2431 static const GVecGen2 g[4] = {
2432 { .fni8 = tcg_gen_vec_abs8_i64,
2433 .fniv = tcg_gen_abs_vec,
2434 .fno = gen_helper_gvec_abs8,
2435 .opt_opc = vecop_list,
2436 .vece = MO_8 },
2437 { .fni8 = tcg_gen_vec_abs16_i64,
2438 .fniv = tcg_gen_abs_vec,
2439 .fno = gen_helper_gvec_abs16,
2440 .opt_opc = vecop_list,
2441 .vece = MO_16 },
2442 { .fni4 = tcg_gen_abs_i32,
2443 .fniv = tcg_gen_abs_vec,
2444 .fno = gen_helper_gvec_abs32,
2445 .opt_opc = vecop_list,
2446 .vece = MO_32 },
2447 { .fni8 = tcg_gen_abs_i64,
2448 .fniv = tcg_gen_abs_vec,
2449 .fno = gen_helper_gvec_abs64,
2450 .opt_opc = vecop_list,
2451 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2452 .vece = MO_64 },
2453 };
2454
2455 tcg_debug_assert(vece <= MO_64);
2456 tcg_gen_gvec_2(dofs, aofs, oprsz, maxsz, &g[vece]);
2457}
2458
db432672
RH		 2459void tcg_gen_gvec_and(unsigned vece, uint32_t dofs, uint32_t aofs,
2460 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2461{
2462 static const GVecGen3 g = {
2463 .fni8 = tcg_gen_and_i64,
2464 .fniv = tcg_gen_and_vec,
2465 .fno = gen_helper_gvec_and,
db432672
RH		 2466 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2467 };
9a9eda78
RH		 2468
2469 if (aofs == bofs) {
2470 tcg_gen_gvec_mov(vece, dofs, aofs, oprsz, maxsz);
2471 } else {
2472 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
2473 }
db432672
RH		 2474}
2475
2476void tcg_gen_gvec_or(unsigned vece, uint32_t dofs, uint32_t aofs,
2477 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2478{
2479 static const GVecGen3 g = {
2480 .fni8 = tcg_gen_or_i64,
2481 .fniv = tcg_gen_or_vec,
2482 .fno = gen_helper_gvec_or,
db432672
RH		 2483 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2484 };
9a9eda78
RH		 2485
2486 if (aofs == bofs) {
2487 tcg_gen_gvec_mov(vece, dofs, aofs, oprsz, maxsz);
2488 } else {
2489 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
2490 }
db432672
RH		 2491}
2492
2493void tcg_gen_gvec_xor(unsigned vece, uint32_t dofs, uint32_t aofs,
2494 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2495{
2496 static const GVecGen3 g = {
2497 .fni8 = tcg_gen_xor_i64,
2498 .fniv = tcg_gen_xor_vec,
2499 .fno = gen_helper_gvec_xor,
db432672
RH		 2500 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2501 };
9a9eda78
RH		 2502
2503 if (aofs == bofs) {
03ddb6f3 2504 tcg_gen_gvec_dup_imm(MO_64, dofs, oprsz, maxsz, 0);
9a9eda78
RH		 2505 } else {
2506 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
2507 }
db432672
RH		 2508}
2509
2510void tcg_gen_gvec_andc(unsigned vece, uint32_t dofs, uint32_t aofs,
2511 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2512{
2513 static const GVecGen3 g = {
2514 .fni8 = tcg_gen_andc_i64,
2515 .fniv = tcg_gen_andc_vec,
2516 .fno = gen_helper_gvec_andc,
db432672
RH		 2517 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2518 };
9a9eda78
RH		 2519
2520 if (aofs == bofs) {
03ddb6f3 2521 tcg_gen_gvec_dup_imm(MO_64, dofs, oprsz, maxsz, 0);
9a9eda78
RH		 2522 } else {
2523 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
2524 }
db432672
RH		 2525}
2526
2527void tcg_gen_gvec_orc(unsigned vece, uint32_t dofs, uint32_t aofs,
2528 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2529{
2530 static const GVecGen3 g = {
2531 .fni8 = tcg_gen_orc_i64,
2532 .fniv = tcg_gen_orc_vec,
2533 .fno = gen_helper_gvec_orc,
db432672
RH		 2534 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2535 };
9a9eda78
RH		 2536
2537 if (aofs == bofs) {
03ddb6f3 2538 tcg_gen_gvec_dup_imm(MO_64, dofs, oprsz, maxsz, -1);
9a9eda78 2539 } else {
f550805d
RH		 2540 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
2541 }
2542}
2543
2544void tcg_gen_gvec_nand(unsigned vece, uint32_t dofs, uint32_t aofs,
2545 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2546{
2547 static const GVecGen3 g = {
2548 .fni8 = tcg_gen_nand_i64,
2549 .fniv = tcg_gen_nand_vec,
2550 .fno = gen_helper_gvec_nand,
2551 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2552 };
2553
2554 if (aofs == bofs) {
2555 tcg_gen_gvec_not(vece, dofs, aofs, oprsz, maxsz);
2556 } else {
2557 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
2558 }
2559}
2560
2561void tcg_gen_gvec_nor(unsigned vece, uint32_t dofs, uint32_t aofs,
2562 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2563{
2564 static const GVecGen3 g = {
2565 .fni8 = tcg_gen_nor_i64,
2566 .fniv = tcg_gen_nor_vec,
2567 .fno = gen_helper_gvec_nor,
2568 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2569 };
2570
2571 if (aofs == bofs) {
2572 tcg_gen_gvec_not(vece, dofs, aofs, oprsz, maxsz);
2573 } else {
2574 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
2575 }
2576}
2577
2578void tcg_gen_gvec_eqv(unsigned vece, uint32_t dofs, uint32_t aofs,
2579 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
2580{
2581 static const GVecGen3 g = {
2582 .fni8 = tcg_gen_eqv_i64,
2583 .fniv = tcg_gen_eqv_vec,
2584 .fno = gen_helper_gvec_eqv,
2585 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2586 };
2587
2588 if (aofs == bofs) {
03ddb6f3 2589 tcg_gen_gvec_dup_imm(MO_64, dofs, oprsz, maxsz, -1);
f550805d 2590 } else {
9a9eda78
RH		 2591 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
2592 }
db432672 2593}
d0ec9796 2594
22fc3527
RH		 2595static const GVecGen2s gop_ands = {
2596 .fni8 = tcg_gen_and_i64,
2597 .fniv = tcg_gen_and_vec,
2598 .fno = gen_helper_gvec_ands,
22fc3527
RH		 2599 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2600 .vece = MO_64
2601};
2602
2603void tcg_gen_gvec_ands(unsigned vece, uint32_t dofs, uint32_t aofs,
2604 TCGv_i64 c, uint32_t oprsz, uint32_t maxsz)
2605{
2606 TCGv_i64 tmp = tcg_temp_new_i64();
614dd4f3 2607 tcg_gen_dup_i64(vece, tmp, c);
22fc3527
RH		 2608 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, tmp, &gop_ands);
2609 tcg_temp_free_i64(tmp);
2610}
2611
2612void tcg_gen_gvec_andi(unsigned vece, uint32_t dofs, uint32_t aofs,
2613 int64_t c, uint32_t oprsz, uint32_t maxsz)
2614{
88d4005b 2615 TCGv_i64 tmp = tcg_constant_i64(dup_const(vece, c));
22fc3527 2616 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, tmp, &gop_ands);
22fc3527
RH		 2617}
2618
2619static const GVecGen2s gop_xors = {
2620 .fni8 = tcg_gen_xor_i64,
2621 .fniv = tcg_gen_xor_vec,
2622 .fno = gen_helper_gvec_xors,
22fc3527
RH		 2623 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2624 .vece = MO_64
2625};
2626
2627void tcg_gen_gvec_xors(unsigned vece, uint32_t dofs, uint32_t aofs,
2628 TCGv_i64 c, uint32_t oprsz, uint32_t maxsz)
2629{
2630 TCGv_i64 tmp = tcg_temp_new_i64();
614dd4f3 2631 tcg_gen_dup_i64(vece, tmp, c);
22fc3527
RH		 2632 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, tmp, &gop_xors);
2633 tcg_temp_free_i64(tmp);
2634}
2635
2636void tcg_gen_gvec_xori(unsigned vece, uint32_t dofs, uint32_t aofs,
2637 int64_t c, uint32_t oprsz, uint32_t maxsz)
2638{
88d4005b 2639 TCGv_i64 tmp = tcg_constant_i64(dup_const(vece, c));
22fc3527 2640 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, tmp, &gop_xors);
22fc3527
RH		 2641}
2642
2643static const GVecGen2s gop_ors = {
2644 .fni8 = tcg_gen_or_i64,
2645 .fniv = tcg_gen_or_vec,
2646 .fno = gen_helper_gvec_ors,
22fc3527
RH		 2647 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2648 .vece = MO_64
2649};
2650
2651void tcg_gen_gvec_ors(unsigned vece, uint32_t dofs, uint32_t aofs,
2652 TCGv_i64 c, uint32_t oprsz, uint32_t maxsz)
2653{
2654 TCGv_i64 tmp = tcg_temp_new_i64();
614dd4f3 2655 tcg_gen_dup_i64(vece, tmp, c);
22fc3527
RH		 2656 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, tmp, &gop_ors);
2657 tcg_temp_free_i64(tmp);
2658}
2659
2660void tcg_gen_gvec_ori(unsigned vece, uint32_t dofs, uint32_t aofs,
2661 int64_t c, uint32_t oprsz, uint32_t maxsz)
2662{
88d4005b 2663 TCGv_i64 tmp = tcg_constant_i64(dup_const(vece, c));
22fc3527 2664 tcg_gen_gvec_2s(dofs, aofs, oprsz, maxsz, tmp, &gop_ors);
22fc3527
RH		 2665}
2666
d0ec9796
RH		 2667void tcg_gen_vec_shl8i_i64(TCGv_i64 d, TCGv_i64 a, int64_t c)
2668{
2669 uint64_t mask = dup_const(MO_8, 0xff << c);
2670 tcg_gen_shli_i64(d, a, c);
2671 tcg_gen_andi_i64(d, d, mask);
2672}
2673
2674void tcg_gen_vec_shl16i_i64(TCGv_i64 d, TCGv_i64 a, int64_t c)
2675{
2676 uint64_t mask = dup_const(MO_16, 0xffff << c);
2677 tcg_gen_shli_i64(d, a, c);
2678 tcg_gen_andi_i64(d, d, mask);
2679}
2680
2681void tcg_gen_gvec_shli(unsigned vece, uint32_t dofs, uint32_t aofs,
2682 int64_t shift, uint32_t oprsz, uint32_t maxsz)
2683{
53229a77 2684 static const TCGOpcode vecop_list[] = { INDEX_op_shli_vec, 0 };
d0ec9796
RH		 2685 static const GVecGen2i g[4] = {
2686 { .fni8 = tcg_gen_vec_shl8i_i64,
2687 .fniv = tcg_gen_shli_vec,
2688 .fno = gen_helper_gvec_shl8i,
53229a77 2689 .opt_opc = vecop_list,
d0ec9796
RH		 2690 .vece = MO_8 },
2691 { .fni8 = tcg_gen_vec_shl16i_i64,
2692 .fniv = tcg_gen_shli_vec,
2693 .fno = gen_helper_gvec_shl16i,
53229a77 2694 .opt_opc = vecop_list,
d0ec9796
RH		 2695 .vece = MO_16 },
2696 { .fni4 = tcg_gen_shli_i32,
2697 .fniv = tcg_gen_shli_vec,
2698 .fno = gen_helper_gvec_shl32i,
53229a77 2699 .opt_opc = vecop_list,
d0ec9796
RH		 2700 .vece = MO_32 },
2701 { .fni8 = tcg_gen_shli_i64,
2702 .fniv = tcg_gen_shli_vec,
2703 .fno = gen_helper_gvec_shl64i,
53229a77 2704 .opt_opc = vecop_list,
d0ec9796
RH		 2705 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2706 .vece = MO_64 },
2707 };
2708
2709 tcg_debug_assert(vece <= MO_64);
2710 tcg_debug_assert(shift >= 0 && shift < (8 << vece));
2711 if (shift == 0) {
2712 tcg_gen_gvec_mov(vece, dofs, aofs, oprsz, maxsz);
2713 } else {
2714 tcg_gen_gvec_2i(dofs, aofs, oprsz, maxsz, shift, &g[vece]);
2715 }
2716}
2717
2718void tcg_gen_vec_shr8i_i64(TCGv_i64 d, TCGv_i64 a, int64_t c)
2719{
2720 uint64_t mask = dup_const(MO_8, 0xff >> c);
2721 tcg_gen_shri_i64(d, a, c);
2722 tcg_gen_andi_i64(d, d, mask);
2723}
2724
2725void tcg_gen_vec_shr16i_i64(TCGv_i64 d, TCGv_i64 a, int64_t c)
2726{
2727 uint64_t mask = dup_const(MO_16, 0xffff >> c);
2728 tcg_gen_shri_i64(d, a, c);
2729 tcg_gen_andi_i64(d, d, mask);
2730}
2731
2732void tcg_gen_gvec_shri(unsigned vece, uint32_t dofs, uint32_t aofs,
2733 int64_t shift, uint32_t oprsz, uint32_t maxsz)
2734{
53229a77 2735 static const TCGOpcode vecop_list[] = { INDEX_op_shri_vec, 0 };
d0ec9796
RH		 2736 static const GVecGen2i g[4] = {
2737 { .fni8 = tcg_gen_vec_shr8i_i64,
2738 .fniv = tcg_gen_shri_vec,
2739 .fno = gen_helper_gvec_shr8i,
53229a77 2740 .opt_opc = vecop_list,
d0ec9796
RH		 2741 .vece = MO_8 },
2742 { .fni8 = tcg_gen_vec_shr16i_i64,
2743 .fniv = tcg_gen_shri_vec,
2744 .fno = gen_helper_gvec_shr16i,
53229a77 2745 .opt_opc = vecop_list,
d0ec9796
RH		 2746 .vece = MO_16 },
2747 { .fni4 = tcg_gen_shri_i32,
2748 .fniv = tcg_gen_shri_vec,
2749 .fno = gen_helper_gvec_shr32i,
53229a77 2750 .opt_opc = vecop_list,
d0ec9796
RH		 2751 .vece = MO_32 },
2752 { .fni8 = tcg_gen_shri_i64,
2753 .fniv = tcg_gen_shri_vec,
2754 .fno = gen_helper_gvec_shr64i,
53229a77 2755 .opt_opc = vecop_list,
d0ec9796
RH		 2756 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2757 .vece = MO_64 },
2758 };
2759
2760 tcg_debug_assert(vece <= MO_64);
2761 tcg_debug_assert(shift >= 0 && shift < (8 << vece));
2762 if (shift == 0) {
2763 tcg_gen_gvec_mov(vece, dofs, aofs, oprsz, maxsz);
2764 } else {
2765 tcg_gen_gvec_2i(dofs, aofs, oprsz, maxsz, shift, &g[vece]);
2766 }
2767}
2768
2769void tcg_gen_vec_sar8i_i64(TCGv_i64 d, TCGv_i64 a, int64_t c)
2770{
2771 uint64_t s_mask = dup_const(MO_8, 0x80 >> c);
2772 uint64_t c_mask = dup_const(MO_8, 0xff >> c);
2773 TCGv_i64 s = tcg_temp_new_i64();
2774
2775 tcg_gen_shri_i64(d, a, c);
2776 tcg_gen_andi_i64(s, d, s_mask); /* isolate (shifted) sign bit */
2777 tcg_gen_muli_i64(s, s, (2 << c) - 2); /* replicate isolated signs */
2778 tcg_gen_andi_i64(d, d, c_mask); /* clear out bits above sign */
2779 tcg_gen_or_i64(d, d, s); /* include sign extension */
2780 tcg_temp_free_i64(s);
2781}
2782
2783void tcg_gen_vec_sar16i_i64(TCGv_i64 d, TCGv_i64 a, int64_t c)
2784{
2785 uint64_t s_mask = dup_const(MO_16, 0x8000 >> c);
2786 uint64_t c_mask = dup_const(MO_16, 0xffff >> c);
2787 TCGv_i64 s = tcg_temp_new_i64();
2788
2789 tcg_gen_shri_i64(d, a, c);
2790 tcg_gen_andi_i64(s, d, s_mask); /* isolate (shifted) sign bit */
2791 tcg_gen_andi_i64(d, d, c_mask); /* clear out bits above sign */
2792 tcg_gen_muli_i64(s, s, (2 << c) - 2); /* replicate isolated signs */
2793 tcg_gen_or_i64(d, d, s); /* include sign extension */
2794 tcg_temp_free_i64(s);
2795}
2796
2797void tcg_gen_gvec_sari(unsigned vece, uint32_t dofs, uint32_t aofs,
2798 int64_t shift, uint32_t oprsz, uint32_t maxsz)
2799{
53229a77 2800 static const TCGOpcode vecop_list[] = { INDEX_op_sari_vec, 0 };
d0ec9796
RH		 2801 static const GVecGen2i g[4] = {
2802 { .fni8 = tcg_gen_vec_sar8i_i64,
2803 .fniv = tcg_gen_sari_vec,
2804 .fno = gen_helper_gvec_sar8i,
53229a77 2805 .opt_opc = vecop_list,
d0ec9796
RH		 2806 .vece = MO_8 },
2807 { .fni8 = tcg_gen_vec_sar16i_i64,
2808 .fniv = tcg_gen_sari_vec,
2809 .fno = gen_helper_gvec_sar16i,
53229a77 2810 .opt_opc = vecop_list,
d0ec9796
RH		 2811 .vece = MO_16 },
2812 { .fni4 = tcg_gen_sari_i32,
2813 .fniv = tcg_gen_sari_vec,
2814 .fno = gen_helper_gvec_sar32i,
53229a77 2815 .opt_opc = vecop_list,
d0ec9796
RH		 2816 .vece = MO_32 },
2817 { .fni8 = tcg_gen_sari_i64,
2818 .fniv = tcg_gen_sari_vec,
2819 .fno = gen_helper_gvec_sar64i,
53229a77 2820 .opt_opc = vecop_list,
d0ec9796
RH		 2821 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2822 .vece = MO_64 },
2823 };
2824
2825 tcg_debug_assert(vece <= MO_64);
2826 tcg_debug_assert(shift >= 0 && shift < (8 << vece));
2827 if (shift == 0) {
2828 tcg_gen_gvec_mov(vece, dofs, aofs, oprsz, maxsz);
2829 } else {
2830 tcg_gen_gvec_2i(dofs, aofs, oprsz, maxsz, shift, &g[vece]);
2831 }
2832}
212be173 2833
b0f7e744
RH		 2834void tcg_gen_vec_rotl8i_i64(TCGv_i64 d, TCGv_i64 a, int64_t c)
2835{
2836 uint64_t mask = dup_const(MO_8, 0xff << c);
2837
2838 tcg_gen_shli_i64(d, a, c);
2839 tcg_gen_shri_i64(a, a, 8 - c);
2840 tcg_gen_andi_i64(d, d, mask);
2841 tcg_gen_andi_i64(a, a, ~mask);
2842 tcg_gen_or_i64(d, d, a);
2843}
2844
2845void tcg_gen_vec_rotl16i_i64(TCGv_i64 d, TCGv_i64 a, int64_t c)
2846{
2847 uint64_t mask = dup_const(MO_16, 0xffff << c);
2848
2849 tcg_gen_shli_i64(d, a, c);
2850 tcg_gen_shri_i64(a, a, 16 - c);
2851 tcg_gen_andi_i64(d, d, mask);
2852 tcg_gen_andi_i64(a, a, ~mask);
2853 tcg_gen_or_i64(d, d, a);
2854}
2855
2856void tcg_gen_gvec_rotli(unsigned vece, uint32_t dofs, uint32_t aofs,
2857 int64_t shift, uint32_t oprsz, uint32_t maxsz)
2858{
2859 static const TCGOpcode vecop_list[] = { INDEX_op_rotli_vec, 0 };
2860 static const GVecGen2i g[4] = {
2861 { .fni8 = tcg_gen_vec_rotl8i_i64,
2862 .fniv = tcg_gen_rotli_vec,
2863 .fno = gen_helper_gvec_rotl8i,
2864 .opt_opc = vecop_list,
2865 .vece = MO_8 },
2866 { .fni8 = tcg_gen_vec_rotl16i_i64,
2867 .fniv = tcg_gen_rotli_vec,
2868 .fno = gen_helper_gvec_rotl16i,
2869 .opt_opc = vecop_list,
2870 .vece = MO_16 },
2871 { .fni4 = tcg_gen_rotli_i32,
2872 .fniv = tcg_gen_rotli_vec,
2873 .fno = gen_helper_gvec_rotl32i,
2874 .opt_opc = vecop_list,
2875 .vece = MO_32 },
2876 { .fni8 = tcg_gen_rotli_i64,
2877 .fniv = tcg_gen_rotli_vec,
2878 .fno = gen_helper_gvec_rotl64i,
2879 .opt_opc = vecop_list,
2880 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
2881 .vece = MO_64 },
2882 };
2883
2884 tcg_debug_assert(vece <= MO_64);
2885 tcg_debug_assert(shift >= 0 && shift < (8 << vece));
2886 if (shift == 0) {
2887 tcg_gen_gvec_mov(vece, dofs, aofs, oprsz, maxsz);
2888 } else {
2889 tcg_gen_gvec_2i(dofs, aofs, oprsz, maxsz, shift, &g[vece]);
2890 }
2891}
2892
2893void tcg_gen_gvec_rotri(unsigned vece, uint32_t dofs, uint32_t aofs,
2894 int64_t shift, uint32_t oprsz, uint32_t maxsz)
2895{
2896 tcg_debug_assert(vece <= MO_64);
2897 tcg_debug_assert(shift >= 0 && shift < (8 << vece));
2898 tcg_gen_gvec_rotli(vece, dofs, aofs, -shift & ((8 << vece) - 1),
2899 oprsz, maxsz);
2900}
2901
b4578cd9
RH		 2902/*
2903 * Specialized generation vector shifts by a non-constant scalar.
2904 */
2905
2906typedef struct {
2907 void (*fni4)(TCGv_i32, TCGv_i32, TCGv_i32);
2908 void (*fni8)(TCGv_i64, TCGv_i64, TCGv_i64);
2909 void (*fniv_s)(unsigned, TCGv_vec, TCGv_vec, TCGv_i32);
2910 void (*fniv_v)(unsigned, TCGv_vec, TCGv_vec, TCGv_vec);
2911 gen_helper_gvec_2 *fno[4];
2912 TCGOpcode s_list[2];
2913 TCGOpcode v_list[2];
2914} GVecGen2sh;
2915
2916static void expand_2sh_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
2917 uint32_t oprsz, uint32_t tysz, TCGType type,
2918 TCGv_i32 shift,
2919 void (*fni)(unsigned, TCGv_vec, TCGv_vec, TCGv_i32))
2920{
2921 TCGv_vec t0 = tcg_temp_new_vec(type);
2922 uint32_t i;
2923
2924 for (i = 0; i < oprsz; i += tysz) {
2925 tcg_gen_ld_vec(t0, cpu_env, aofs + i);
2926 fni(vece, t0, t0, shift);
2927 tcg_gen_st_vec(t0, cpu_env, dofs + i);
2928 }
2929 tcg_temp_free_vec(t0);
2930}
2931
2932static void
2933do_gvec_shifts(unsigned vece, uint32_t dofs, uint32_t aofs, TCGv_i32 shift,
2934 uint32_t oprsz, uint32_t maxsz, const GVecGen2sh *g)
2935{
2936 TCGType type;
2937 uint32_t some;
2938
2939 check_size_align(oprsz, maxsz, dofs | aofs);
2940 check_overlap_2(dofs, aofs, maxsz);
2941
2942 /* If the backend has a scalar expansion, great. */
2943 type = choose_vector_type(g->s_list, vece, oprsz, vece == MO_64);
2944 if (type) {
2945 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
2946 switch (type) {
2947 case TCG_TYPE_V256:
2948 some = QEMU_ALIGN_DOWN(oprsz, 32);
2949 expand_2sh_vec(vece, dofs, aofs, some, 32,
2950 TCG_TYPE_V256, shift, g->fniv_s);
2951 if (some == oprsz) {
2952 break;
2953 }
2954 dofs += some;
2955 aofs += some;
2956 oprsz -= some;
2957 maxsz -= some;
2958 /* fallthru */
2959 case TCG_TYPE_V128:
2960 expand_2sh_vec(vece, dofs, aofs, oprsz, 16,
2961 TCG_TYPE_V128, shift, g->fniv_s);
2962 break;
2963 case TCG_TYPE_V64:
2964 expand_2sh_vec(vece, dofs, aofs, oprsz, 8,
2965 TCG_TYPE_V64, shift, g->fniv_s);
2966 break;
2967 default:
2968 g_assert_not_reached();
2969 }
2970 tcg_swap_vecop_list(hold_list);
2971 goto clear_tail;
2972 }
2973
2974 /* If the backend supports variable vector shifts, also cool. */
2975 type = choose_vector_type(g->v_list, vece, oprsz, vece == MO_64);
2976 if (type) {
2977 const TCGOpcode *hold_list = tcg_swap_vecop_list(NULL);
2978 TCGv_vec v_shift = tcg_temp_new_vec(type);
2979
2980 if (vece == MO_64) {
2981 TCGv_i64 sh64 = tcg_temp_new_i64();
2982 tcg_gen_extu_i32_i64(sh64, shift);
2983 tcg_gen_dup_i64_vec(MO_64, v_shift, sh64);
2984 tcg_temp_free_i64(sh64);
2985 } else {
2986 tcg_gen_dup_i32_vec(vece, v_shift, shift);
2987 }
2988
2989 switch (type) {
2990 case TCG_TYPE_V256:
2991 some = QEMU_ALIGN_DOWN(oprsz, 32);
2992 expand_2s_vec(vece, dofs, aofs, some, 32, TCG_TYPE_V256,
2993 v_shift, false, g->fniv_v);
2994 if (some == oprsz) {
2995 break;
2996 }
2997 dofs += some;
2998 aofs += some;
2999 oprsz -= some;
3000 maxsz -= some;
3001 /* fallthru */
3002 case TCG_TYPE_V128:
3003 expand_2s_vec(vece, dofs, aofs, oprsz, 16, TCG_TYPE_V128,
3004 v_shift, false, g->fniv_v);
3005 break;
3006 case TCG_TYPE_V64:
3007 expand_2s_vec(vece, dofs, aofs, oprsz, 8, TCG_TYPE_V64,
3008 v_shift, false, g->fniv_v);
3009 break;
3010 default:
3011 g_assert_not_reached();
3012 }
3013 tcg_temp_free_vec(v_shift);
3014 tcg_swap_vecop_list(hold_list);
3015 goto clear_tail;
3016 }
3017
3018 /* Otherwise fall back to integral... */
3019 if (vece == MO_32 && check_size_impl(oprsz, 4)) {
3020 expand_2s_i32(dofs, aofs, oprsz, shift, false, g->fni4);
3021 } else if (vece == MO_64 && check_size_impl(oprsz, 8)) {
3022 TCGv_i64 sh64 = tcg_temp_new_i64();
3023 tcg_gen_extu_i32_i64(sh64, shift);
3024 expand_2s_i64(dofs, aofs, oprsz, sh64, false, g->fni8);
3025 tcg_temp_free_i64(sh64);
3026 } else {
3027 TCGv_ptr a0 = tcg_temp_new_ptr();
3028 TCGv_ptr a1 = tcg_temp_new_ptr();
3029 TCGv_i32 desc = tcg_temp_new_i32();
3030
3031 tcg_gen_shli_i32(desc, shift, SIMD_DATA_SHIFT);
3032 tcg_gen_ori_i32(desc, desc, simd_desc(oprsz, maxsz, 0));
3033 tcg_gen_addi_ptr(a0, cpu_env, dofs);
3034 tcg_gen_addi_ptr(a1, cpu_env, aofs);
3035
3036 g->fno[vece](a0, a1, desc);
3037
3038 tcg_temp_free_ptr(a0);
3039 tcg_temp_free_ptr(a1);
3040 tcg_temp_free_i32(desc);
3041 return;
3042 }
3043
3044 clear_tail:
3045 if (oprsz < maxsz) {
3046 expand_clr(dofs + oprsz, maxsz - oprsz);
3047 }
3048}
3049
3050void tcg_gen_gvec_shls(unsigned vece, uint32_t dofs, uint32_t aofs,
3051 TCGv_i32 shift, uint32_t oprsz, uint32_t maxsz)
3052{
3053 static const GVecGen2sh g = {
3054 .fni4 = tcg_gen_shl_i32,
3055 .fni8 = tcg_gen_shl_i64,
3056 .fniv_s = tcg_gen_shls_vec,
3057 .fniv_v = tcg_gen_shlv_vec,
3058 .fno = {
3059 gen_helper_gvec_shl8i,
3060 gen_helper_gvec_shl16i,
3061 gen_helper_gvec_shl32i,
3062 gen_helper_gvec_shl64i,
3063 },
3064 .s_list = { INDEX_op_shls_vec, 0 },
3065 .v_list = { INDEX_op_shlv_vec, 0 },
3066 };
3067
3068 tcg_debug_assert(vece <= MO_64);
3069 do_gvec_shifts(vece, dofs, aofs, shift, oprsz, maxsz, &g);
3070}
3071
3072void tcg_gen_gvec_shrs(unsigned vece, uint32_t dofs, uint32_t aofs,
3073 TCGv_i32 shift, uint32_t oprsz, uint32_t maxsz)
3074{
3075 static const GVecGen2sh g = {
3076 .fni4 = tcg_gen_shr_i32,
3077 .fni8 = tcg_gen_shr_i64,
3078 .fniv_s = tcg_gen_shrs_vec,
3079 .fniv_v = tcg_gen_shrv_vec,
3080 .fno = {
3081 gen_helper_gvec_shr8i,
3082 gen_helper_gvec_shr16i,
3083 gen_helper_gvec_shr32i,
3084 gen_helper_gvec_shr64i,
3085 },
3086 .s_list = { INDEX_op_shrs_vec, 0 },
3087 .v_list = { INDEX_op_shrv_vec, 0 },
3088 };
3089
3090 tcg_debug_assert(vece <= MO_64);
3091 do_gvec_shifts(vece, dofs, aofs, shift, oprsz, maxsz, &g);
3092}
3093
3094void tcg_gen_gvec_sars(unsigned vece, uint32_t dofs, uint32_t aofs,
3095 TCGv_i32 shift, uint32_t oprsz, uint32_t maxsz)
3096{
3097 static const GVecGen2sh g = {
3098 .fni4 = tcg_gen_sar_i32,
3099 .fni8 = tcg_gen_sar_i64,
3100 .fniv_s = tcg_gen_sars_vec,
3101 .fniv_v = tcg_gen_sarv_vec,
3102 .fno = {
3103 gen_helper_gvec_sar8i,
3104 gen_helper_gvec_sar16i,
3105 gen_helper_gvec_sar32i,
3106 gen_helper_gvec_sar64i,
3107 },
3108 .s_list = { INDEX_op_sars_vec, 0 },
3109 .v_list = { INDEX_op_sarv_vec, 0 },
3110 };
3111
3112 tcg_debug_assert(vece <= MO_64);
3113 do_gvec_shifts(vece, dofs, aofs, shift, oprsz, maxsz, &g);
3114}
3115
23850a74
RH		 3116void tcg_gen_gvec_rotls(unsigned vece, uint32_t dofs, uint32_t aofs,
3117 TCGv_i32 shift, uint32_t oprsz, uint32_t maxsz)
3118{
3119 static const GVecGen2sh g = {
3120 .fni4 = tcg_gen_rotl_i32,
3121 .fni8 = tcg_gen_rotl_i64,
3122 .fniv_s = tcg_gen_rotls_vec,
3123 .fniv_v = tcg_gen_rotlv_vec,
3124 .fno = {
3125 gen_helper_gvec_rotl8i,
3126 gen_helper_gvec_rotl16i,
3127 gen_helper_gvec_rotl32i,
3128 gen_helper_gvec_rotl64i,
3129 },
3130 .s_list = { INDEX_op_rotls_vec, 0 },
3131 .v_list = { INDEX_op_rotlv_vec, 0 },
3132 };
3133
3134 tcg_debug_assert(vece <= MO_64);
3135 do_gvec_shifts(vece, dofs, aofs, shift, oprsz, maxsz, &g);
3136}
3137
5ee5c14c
RH		 3138/*
3139 * Expand D = A << (B % element bits)
3140 *
3141 * Unlike scalar shifts, where it is easy for the target front end
3142 * to include the modulo as part of the expansion. If the target
3143 * naturally includes the modulo as part of the operation, great!
3144 * If the target has some other behaviour from out-of-range shifts,
3145 * then it could not use this function anyway, and would need to
3146 * do it's own expansion with custom functions.
3147 */
3148static void tcg_gen_shlv_mod_vec(unsigned vece, TCGv_vec d,
3149 TCGv_vec a, TCGv_vec b)
3150{
3151 TCGv_vec t = tcg_temp_new_vec_matching(d);
88d4005b 3152 TCGv_vec m = tcg_constant_vec_matching(d, vece, (8 << vece) - 1);
5ee5c14c 3153
88d4005b 3154 tcg_gen_and_vec(vece, t, b, m);
5ee5c14c
RH		 3155 tcg_gen_shlv_vec(vece, d, a, t);
3156 tcg_temp_free_vec(t);
3157}
3158
3159static void tcg_gen_shl_mod_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3160{
3161 TCGv_i32 t = tcg_temp_new_i32();
3162
3163 tcg_gen_andi_i32(t, b, 31);
3164 tcg_gen_shl_i32(d, a, t);
3165 tcg_temp_free_i32(t);
3166}
3167
3168static void tcg_gen_shl_mod_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
3169{
3170 TCGv_i64 t = tcg_temp_new_i64();
3171
3172 tcg_gen_andi_i64(t, b, 63);
3173 tcg_gen_shl_i64(d, a, t);
3174 tcg_temp_free_i64(t);
3175}
3176
3177void tcg_gen_gvec_shlv(unsigned vece, uint32_t dofs, uint32_t aofs,
3178 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
3179{
3180 static const TCGOpcode vecop_list[] = { INDEX_op_shlv_vec, 0 };
3181 static const GVecGen3 g[4] = {
3182 { .fniv = tcg_gen_shlv_mod_vec,
3183 .fno = gen_helper_gvec_shl8v,
3184 .opt_opc = vecop_list,
3185 .vece = MO_8 },
3186 { .fniv = tcg_gen_shlv_mod_vec,
3187 .fno = gen_helper_gvec_shl16v,
3188 .opt_opc = vecop_list,
3189 .vece = MO_16 },
3190 { .fni4 = tcg_gen_shl_mod_i32,
3191 .fniv = tcg_gen_shlv_mod_vec,
3192 .fno = gen_helper_gvec_shl32v,
3193 .opt_opc = vecop_list,
3194 .vece = MO_32 },
3195 { .fni8 = tcg_gen_shl_mod_i64,
3196 .fniv = tcg_gen_shlv_mod_vec,
3197 .fno = gen_helper_gvec_shl64v,
3198 .opt_opc = vecop_list,
3199 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3200 .vece = MO_64 },
3201 };
3202
3203 tcg_debug_assert(vece <= MO_64);
3204 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
3205}
3206
3207/*
3208 * Similarly for logical right shifts.
3209 */
3210
3211static void tcg_gen_shrv_mod_vec(unsigned vece, TCGv_vec d,
3212 TCGv_vec a, TCGv_vec b)
3213{
3214 TCGv_vec t = tcg_temp_new_vec_matching(d);
88d4005b 3215 TCGv_vec m = tcg_constant_vec_matching(d, vece, (8 << vece) - 1);
5ee5c14c 3216
88d4005b 3217 tcg_gen_and_vec(vece, t, b, m);
5ee5c14c
RH		 3218 tcg_gen_shrv_vec(vece, d, a, t);
3219 tcg_temp_free_vec(t);
3220}
3221
3222static void tcg_gen_shr_mod_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3223{
3224 TCGv_i32 t = tcg_temp_new_i32();
3225
3226 tcg_gen_andi_i32(t, b, 31);
3227 tcg_gen_shr_i32(d, a, t);
3228 tcg_temp_free_i32(t);
3229}
3230
3231static void tcg_gen_shr_mod_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
3232{
3233 TCGv_i64 t = tcg_temp_new_i64();
3234
3235 tcg_gen_andi_i64(t, b, 63);
3236 tcg_gen_shr_i64(d, a, t);
3237 tcg_temp_free_i64(t);
3238}
3239
3240void tcg_gen_gvec_shrv(unsigned vece, uint32_t dofs, uint32_t aofs,
3241 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
3242{
3243 static const TCGOpcode vecop_list[] = { INDEX_op_shrv_vec, 0 };
3244 static const GVecGen3 g[4] = {
3245 { .fniv = tcg_gen_shrv_mod_vec,
3246 .fno = gen_helper_gvec_shr8v,
3247 .opt_opc = vecop_list,
3248 .vece = MO_8 },
3249 { .fniv = tcg_gen_shrv_mod_vec,
3250 .fno = gen_helper_gvec_shr16v,
3251 .opt_opc = vecop_list,
3252 .vece = MO_16 },
3253 { .fni4 = tcg_gen_shr_mod_i32,
3254 .fniv = tcg_gen_shrv_mod_vec,
3255 .fno = gen_helper_gvec_shr32v,
3256 .opt_opc = vecop_list,
3257 .vece = MO_32 },
3258 { .fni8 = tcg_gen_shr_mod_i64,
3259 .fniv = tcg_gen_shrv_mod_vec,
3260 .fno = gen_helper_gvec_shr64v,
3261 .opt_opc = vecop_list,
3262 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3263 .vece = MO_64 },
3264 };
3265
3266 tcg_debug_assert(vece <= MO_64);
3267 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
3268}
3269
3270/*
3271 * Similarly for arithmetic right shifts.
3272 */
3273
3274static void tcg_gen_sarv_mod_vec(unsigned vece, TCGv_vec d,
3275 TCGv_vec a, TCGv_vec b)
3276{
3277 TCGv_vec t = tcg_temp_new_vec_matching(d);
88d4005b 3278 TCGv_vec m = tcg_constant_vec_matching(d, vece, (8 << vece) - 1);
5ee5c14c 3279
88d4005b 3280 tcg_gen_and_vec(vece, t, b, m);
5ee5c14c
RH		 3281 tcg_gen_sarv_vec(vece, d, a, t);
3282 tcg_temp_free_vec(t);
3283}
3284
3285static void tcg_gen_sar_mod_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3286{
3287 TCGv_i32 t = tcg_temp_new_i32();
3288
3289 tcg_gen_andi_i32(t, b, 31);
3290 tcg_gen_sar_i32(d, a, t);
3291 tcg_temp_free_i32(t);
3292}
3293
3294static void tcg_gen_sar_mod_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
3295{
3296 TCGv_i64 t = tcg_temp_new_i64();
3297
3298 tcg_gen_andi_i64(t, b, 63);
3299 tcg_gen_sar_i64(d, a, t);
3300 tcg_temp_free_i64(t);
3301}
3302
3303void tcg_gen_gvec_sarv(unsigned vece, uint32_t dofs, uint32_t aofs,
3304 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
3305{
3306 static const TCGOpcode vecop_list[] = { INDEX_op_sarv_vec, 0 };
3307 static const GVecGen3 g[4] = {
3308 { .fniv = tcg_gen_sarv_mod_vec,
3309 .fno = gen_helper_gvec_sar8v,
3310 .opt_opc = vecop_list,
3311 .vece = MO_8 },
3312 { .fniv = tcg_gen_sarv_mod_vec,
3313 .fno = gen_helper_gvec_sar16v,
3314 .opt_opc = vecop_list,
3315 .vece = MO_16 },
3316 { .fni4 = tcg_gen_sar_mod_i32,
3317 .fniv = tcg_gen_sarv_mod_vec,
3318 .fno = gen_helper_gvec_sar32v,
3319 .opt_opc = vecop_list,
3320 .vece = MO_32 },
3321 { .fni8 = tcg_gen_sar_mod_i64,
3322 .fniv = tcg_gen_sarv_mod_vec,
3323 .fno = gen_helper_gvec_sar64v,
3324 .opt_opc = vecop_list,
3325 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3326 .vece = MO_64 },
3327 };
3328
3329 tcg_debug_assert(vece <= MO_64);
3330 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
3331}
3332
5d0ceda9
RH		 3333/*
3334 * Similarly for rotates.
3335 */
3336
3337static void tcg_gen_rotlv_mod_vec(unsigned vece, TCGv_vec d,
3338 TCGv_vec a, TCGv_vec b)
3339{
3340 TCGv_vec t = tcg_temp_new_vec_matching(d);
88d4005b 3341 TCGv_vec m = tcg_constant_vec_matching(d, vece, (8 << vece) - 1);
5d0ceda9 3342
88d4005b 3343 tcg_gen_and_vec(vece, t, b, m);
5d0ceda9
RH		 3344 tcg_gen_rotlv_vec(vece, d, a, t);
3345 tcg_temp_free_vec(t);
3346}
3347
3348static void tcg_gen_rotl_mod_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3349{
3350 TCGv_i32 t = tcg_temp_new_i32();
3351
3352 tcg_gen_andi_i32(t, b, 31);
3353 tcg_gen_rotl_i32(d, a, t);
3354 tcg_temp_free_i32(t);
3355}
3356
3357static void tcg_gen_rotl_mod_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
3358{
3359 TCGv_i64 t = tcg_temp_new_i64();
3360
3361 tcg_gen_andi_i64(t, b, 63);
3362 tcg_gen_rotl_i64(d, a, t);
3363 tcg_temp_free_i64(t);
3364}
3365
3366void tcg_gen_gvec_rotlv(unsigned vece, uint32_t dofs, uint32_t aofs,
3367 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
3368{
3369 static const TCGOpcode vecop_list[] = { INDEX_op_rotlv_vec, 0 };
3370 static const GVecGen3 g[4] = {
3371 { .fniv = tcg_gen_rotlv_mod_vec,
3372 .fno = gen_helper_gvec_rotl8v,
3373 .opt_opc = vecop_list,
3374 .vece = MO_8 },
3375 { .fniv = tcg_gen_rotlv_mod_vec,
3376 .fno = gen_helper_gvec_rotl16v,
3377 .opt_opc = vecop_list,
3378 .vece = MO_16 },
3379 { .fni4 = tcg_gen_rotl_mod_i32,
3380 .fniv = tcg_gen_rotlv_mod_vec,
3381 .fno = gen_helper_gvec_rotl32v,
3382 .opt_opc = vecop_list,
3383 .vece = MO_32 },
3384 { .fni8 = tcg_gen_rotl_mod_i64,
3385 .fniv = tcg_gen_rotlv_mod_vec,
3386 .fno = gen_helper_gvec_rotl64v,
3387 .opt_opc = vecop_list,
3388 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3389 .vece = MO_64 },
3390 };
3391
3392 tcg_debug_assert(vece <= MO_64);
3393 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
3394}
3395
3396static void tcg_gen_rotrv_mod_vec(unsigned vece, TCGv_vec d,
3397 TCGv_vec a, TCGv_vec b)
3398{
3399 TCGv_vec t = tcg_temp_new_vec_matching(d);
88d4005b 3400 TCGv_vec m = tcg_constant_vec_matching(d, vece, (8 << vece) - 1);
5d0ceda9 3401
88d4005b 3402 tcg_gen_and_vec(vece, t, b, m);
5d0ceda9
RH		 3403 tcg_gen_rotrv_vec(vece, d, a, t);
3404 tcg_temp_free_vec(t);
3405}
3406
3407static void tcg_gen_rotr_mod_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
3408{
3409 TCGv_i32 t = tcg_temp_new_i32();
3410
3411 tcg_gen_andi_i32(t, b, 31);
3412 tcg_gen_rotr_i32(d, a, t);
3413 tcg_temp_free_i32(t);
3414}
3415
3416static void tcg_gen_rotr_mod_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
3417{
3418 TCGv_i64 t = tcg_temp_new_i64();
3419
3420 tcg_gen_andi_i64(t, b, 63);
3421 tcg_gen_rotr_i64(d, a, t);
3422 tcg_temp_free_i64(t);
3423}
3424
3425void tcg_gen_gvec_rotrv(unsigned vece, uint32_t dofs, uint32_t aofs,
3426 uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
3427{
3428 static const TCGOpcode vecop_list[] = { INDEX_op_rotrv_vec, 0 };
3429 static const GVecGen3 g[4] = {
3430 { .fniv = tcg_gen_rotrv_mod_vec,
3431 .fno = gen_helper_gvec_rotr8v,
3432 .opt_opc = vecop_list,
3433 .vece = MO_8 },
3434 { .fniv = tcg_gen_rotrv_mod_vec,
3435 .fno = gen_helper_gvec_rotr16v,
3436 .opt_opc = vecop_list,
3437 .vece = MO_16 },
3438 { .fni4 = tcg_gen_rotr_mod_i32,
3439 .fniv = tcg_gen_rotrv_mod_vec,
3440 .fno = gen_helper_gvec_rotr32v,
3441 .opt_opc = vecop_list,
3442 .vece = MO_32 },
3443 { .fni8 = tcg_gen_rotr_mod_i64,
3444 .fniv = tcg_gen_rotrv_mod_vec,
3445 .fno = gen_helper_gvec_rotr64v,
3446 .opt_opc = vecop_list,
3447 .prefer_i64 = TCG_TARGET_REG_BITS == 64,
3448 .vece = MO_64 },
3449 };
3450
3451 tcg_debug_assert(vece <= MO_64);
3452 tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
3453}
3454
212be173
RH		 3455/* Expand OPSZ bytes worth of three-operand operations using i32 elements. */
3456static void expand_cmp_i32(uint32_t dofs, uint32_t aofs, uint32_t bofs,
3457 uint32_t oprsz, TCGCond cond)
3458{
3459 TCGv_i32 t0 = tcg_temp_new_i32();
3460 TCGv_i32 t1 = tcg_temp_new_i32();
3461 uint32_t i;
3462
3463 for (i = 0; i < oprsz; i += 4) {
3464 tcg_gen_ld_i32(t0, cpu_env, aofs + i);
3465 tcg_gen_ld_i32(t1, cpu_env, bofs + i);
3466 tcg_gen_setcond_i32(cond, t0, t0, t1);
3467 tcg_gen_neg_i32(t0, t0);
3468 tcg_gen_st_i32(t0, cpu_env, dofs + i);
3469 }
3470 tcg_temp_free_i32(t1);
3471 tcg_temp_free_i32(t0);
3472}
3473
3474static void expand_cmp_i64(uint32_t dofs, uint32_t aofs, uint32_t bofs,
3475 uint32_t oprsz, TCGCond cond)
3476{
3477 TCGv_i64 t0 = tcg_temp_new_i64();
3478 TCGv_i64 t1 = tcg_temp_new_i64();
3479 uint32_t i;
3480
3481 for (i = 0; i < oprsz; i += 8) {
3482 tcg_gen_ld_i64(t0, cpu_env, aofs + i);
3483 tcg_gen_ld_i64(t1, cpu_env, bofs + i);
3484 tcg_gen_setcond_i64(cond, t0, t0, t1);
3485 tcg_gen_neg_i64(t0, t0);
3486 tcg_gen_st_i64(t0, cpu_env, dofs + i);
3487 }
3488 tcg_temp_free_i64(t1);
3489 tcg_temp_free_i64(t0);
3490}
3491
3492static void expand_cmp_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
3493 uint32_t bofs, uint32_t oprsz, uint32_t tysz,
3494 TCGType type, TCGCond cond)
3495{
3496 TCGv_vec t0 = tcg_temp_new_vec(type);
3497 TCGv_vec t1 = tcg_temp_new_vec(type);
3498 uint32_t i;
3499
3500 for (i = 0; i < oprsz; i += tysz) {
3501 tcg_gen_ld_vec(t0, cpu_env, aofs + i);
3502 tcg_gen_ld_vec(t1, cpu_env, bofs + i);
3503 tcg_gen_cmp_vec(cond, vece, t0, t0, t1);
3504 tcg_gen_st_vec(t0, cpu_env, dofs + i);
3505 }
3506 tcg_temp_free_vec(t1);
3507 tcg_temp_free_vec(t0);
3508}
3509
3510void tcg_gen_gvec_cmp(TCGCond cond, unsigned vece, uint32_t dofs,
3511 uint32_t aofs, uint32_t bofs,
3512 uint32_t oprsz, uint32_t maxsz)
3513{
53229a77 3514 static const TCGOpcode cmp_list[] = { INDEX_op_cmp_vec, 0 };
212be173
RH		 3515 static gen_helper_gvec_3 * const eq_fn[4] = {
3516 gen_helper_gvec_eq8, gen_helper_gvec_eq16,
3517 gen_helper_gvec_eq32, gen_helper_gvec_eq64
3518 };
3519 static gen_helper_gvec_3 * const ne_fn[4] = {
3520 gen_helper_gvec_ne8, gen_helper_gvec_ne16,
3521 gen_helper_gvec_ne32, gen_helper_gvec_ne64
3522 };
3523 static gen_helper_gvec_3 * const lt_fn[4] = {
3524 gen_helper_gvec_lt8, gen_helper_gvec_lt16,
3525 gen_helper_gvec_lt32, gen_helper_gvec_lt64
3526 };
3527 static gen_helper_gvec_3 * const le_fn[4] = {
3528 gen_helper_gvec_le8, gen_helper_gvec_le16,
3529 gen_helper_gvec_le32, gen_helper_gvec_le64
3530 };
3531 static gen_helper_gvec_3 * const ltu_fn[4] = {
3532 gen_helper_gvec_ltu8, gen_helper_gvec_ltu16,
3533 gen_helper_gvec_ltu32, gen_helper_gvec_ltu64
3534 };
3535 static gen_helper_gvec_3 * const leu_fn[4] = {
3536 gen_helper_gvec_leu8, gen_helper_gvec_leu16,
3537 gen_helper_gvec_leu32, gen_helper_gvec_leu64
3538 };
3539 static gen_helper_gvec_3 * const * const fns[16] = {
3540 [TCG_COND_EQ] = eq_fn,
3541 [TCG_COND_NE] = ne_fn,
3542 [TCG_COND_LT] = lt_fn,
3543 [TCG_COND_LE] = le_fn,
3544 [TCG_COND_LTU] = ltu_fn,
3545 [TCG_COND_LEU] = leu_fn,
3546 };
53229a77
RH		 3547
3548 const TCGOpcode *hold_list;
adb196cb
RH		 3549 TCGType type;
3550 uint32_t some;
212be173
RH		 3551
3552 check_size_align(oprsz, maxsz, dofs | aofs | bofs);
3553 check_overlap_3(dofs, aofs, bofs, maxsz);
3554
3555 if (cond == TCG_COND_NEVER || cond == TCG_COND_ALWAYS) {
3556 do_dup(MO_8, dofs, oprsz, maxsz,
3557 NULL, NULL, -(cond == TCG_COND_ALWAYS));
3558 return;
3559 }
3560
53229a77
RH		 3561 /*
3562 * Implement inline with a vector type, if possible.
adb196cb
RH		 3563 * Prefer integer when 64-bit host and 64-bit comparison.
3564 */
53229a77
RH		 3565 hold_list = tcg_swap_vecop_list(cmp_list);
3566 type = choose_vector_type(cmp_list, vece, oprsz,
adb196cb
RH		 3567 TCG_TARGET_REG_BITS == 64 && vece == MO_64);
3568 switch (type) {
3569 case TCG_TYPE_V256:
3570 /* Recall that ARM SVE allows vector sizes that are not a
3571 * power of 2, but always a multiple of 16. The intent is
3572 * that e.g. size == 80 would be expanded with 2x32 + 1x16.
3573 */
3574 some = QEMU_ALIGN_DOWN(oprsz, 32);
212be173
RH		 3575 expand_cmp_vec(vece, dofs, aofs, bofs, some, 32, TCG_TYPE_V256, cond);
3576 if (some == oprsz) {
adb196cb 3577 break;
212be173
RH		 3578 }
3579 dofs += some;
3580 aofs += some;
3581 bofs += some;
3582 oprsz -= some;
3583 maxsz -= some;
adb196cb
RH		 3584 /* fallthru */
3585 case TCG_TYPE_V128:
212be173 3586 expand_cmp_vec(vece, dofs, aofs, bofs, oprsz, 16, TCG_TYPE_V128, cond);
adb196cb
RH		 3587 break;
3588 case TCG_TYPE_V64:
212be173 3589 expand_cmp_vec(vece, dofs, aofs, bofs, oprsz, 8, TCG_TYPE_V64, cond);
adb196cb
RH		 3590 break;
3591
3592 case 0:
3593 if (vece == MO_64 && check_size_impl(oprsz, 8)) {
3594 expand_cmp_i64(dofs, aofs, bofs, oprsz, cond);
3595 } else if (vece == MO_32 && check_size_impl(oprsz, 4)) {
3596 expand_cmp_i32(dofs, aofs, bofs, oprsz, cond);
3597 } else {
3598 gen_helper_gvec_3 * const *fn = fns[cond];
3599
3600 if (fn == NULL) {
3601 uint32_t tmp;
3602 tmp = aofs, aofs = bofs, bofs = tmp;
3603 cond = tcg_swap_cond(cond);
3604 fn = fns[cond];
3605 assert(fn != NULL);
3606 }
3607 tcg_gen_gvec_3_ool(dofs, aofs, bofs, oprsz, maxsz, 0, fn[vece]);
53229a77 3608 oprsz = maxsz;
212be173 3609 }
adb196cb
RH		 3610 break;
3611
3612 default:
3613 g_assert_not_reached();
212be173 3614 }
53229a77 3615 tcg_swap_vecop_list(hold_list);
212be173 3616
212be173
RH		 3617 if (oprsz < maxsz) {
3618 expand_clr(dofs + oprsz, maxsz - oprsz);
3619 }
3620}
38dc1294
RH		 3621
3622static void tcg_gen_bitsel_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b, TCGv_i64 c)
3623{
3624 TCGv_i64 t = tcg_temp_new_i64();
3625
3626 tcg_gen_and_i64(t, b, a);
3627 tcg_gen_andc_i64(d, c, a);
3628 tcg_gen_or_i64(d, d, t);
3629 tcg_temp_free_i64(t);
3630}
3631
3632void tcg_gen_gvec_bitsel(unsigned vece, uint32_t dofs, uint32_t aofs,
3633 uint32_t bofs, uint32_t cofs,
3634 uint32_t oprsz, uint32_t maxsz)
3635{
3636 static const GVecGen4 g = {
3637 .fni8 = tcg_gen_bitsel_i64,
3638 .fniv = tcg_gen_bitsel_vec,
3639 .fno = gen_helper_gvec_bitsel,
3640 };
3641
3642 tcg_gen_gvec_4(dofs, aofs, bofs, cofs, oprsz, maxsz, &g);
3643}
QEMU mirror