]>
Commit | Line | Data |
---|---|---|
78e138bc PM |
1 | /* |
2 | * ARM translation: AArch32 VFP instructions | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * Copyright (c) 2005-2007 CodeSourcery | |
6 | * Copyright (c) 2007 OpenedHand, Ltd. | |
7 | * Copyright (c) 2019 Linaro, Ltd. | |
8 | * | |
9 | * This library is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU Lesser General Public | |
11 | * License as published by the Free Software Foundation; either | |
12 | * version 2 of the License, or (at your option) any later version. | |
13 | * | |
14 | * This library is distributed in the hope that it will be useful, | |
15 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * Lesser General Public License for more details. | |
18 | * | |
19 | * You should have received a copy of the GNU Lesser General Public | |
20 | * License along with this library; if not, see <http://www.gnu.org/licenses/>. | |
21 | */ | |
22 | ||
23 | /* | |
24 | * This file is intended to be included from translate.c; it uses | |
25 | * some macros and definitions provided by that file. | |
26 | * It might be possible to convert it to a standalone .c file eventually. | |
27 | */ | |
28 | ||
29 | /* Include the generated VFP decoder */ | |
30 | #include "decode-vfp.inc.c" | |
31 | #include "decode-vfp-uncond.inc.c" | |
06db8196 | 32 | |
d6a092d4 PM |
33 | /* |
34 | * The imm8 encodes the sign bit, enough bits to represent an exponent in | |
35 | * the range 01....1xx to 10....0xx, and the most significant 4 bits of | |
36 | * the mantissa; see VFPExpandImm() in the v8 ARM ARM. | |
37 | */ | |
38 | uint64_t vfp_expand_imm(int size, uint8_t imm8) | |
39 | { | |
40 | uint64_t imm; | |
41 | ||
42 | switch (size) { | |
43 | case MO_64: | |
44 | imm = (extract32(imm8, 7, 1) ? 0x8000 : 0) | | |
45 | (extract32(imm8, 6, 1) ? 0x3fc0 : 0x4000) | | |
46 | extract32(imm8, 0, 6); | |
47 | imm <<= 48; | |
48 | break; | |
49 | case MO_32: | |
50 | imm = (extract32(imm8, 7, 1) ? 0x8000 : 0) | | |
51 | (extract32(imm8, 6, 1) ? 0x3e00 : 0x4000) | | |
52 | (extract32(imm8, 0, 6) << 3); | |
53 | imm <<= 16; | |
54 | break; | |
55 | case MO_16: | |
56 | imm = (extract32(imm8, 7, 1) ? 0x8000 : 0) | | |
57 | (extract32(imm8, 6, 1) ? 0x3000 : 0x4000) | | |
58 | (extract32(imm8, 0, 6) << 6); | |
59 | break; | |
60 | default: | |
61 | g_assert_not_reached(); | |
62 | } | |
63 | return imm; | |
64 | } | |
65 | ||
b623d803 PM |
66 | /* |
67 | * Return the offset of a 16-bit half of the specified VFP single-precision | |
68 | * register. If top is true, returns the top 16 bits; otherwise the bottom | |
69 | * 16 bits. | |
70 | */ | |
71 | static inline long vfp_f16_offset(unsigned reg, bool top) | |
72 | { | |
73 | long offs = vfp_reg_offset(false, reg); | |
74 | #ifdef HOST_WORDS_BIGENDIAN | |
75 | if (!top) { | |
76 | offs += 2; | |
77 | } | |
78 | #else | |
79 | if (top) { | |
80 | offs += 2; | |
81 | } | |
82 | #endif | |
83 | return offs; | |
84 | } | |
85 | ||
06db8196 PM |
86 | /* |
87 | * Check that VFP access is enabled. If it is, do the necessary | |
88 | * M-profile lazy-FP handling and then return true. | |
89 | * If not, emit code to generate an appropriate exception and | |
90 | * return false. | |
91 | * The ignore_vfp_enabled argument specifies that we should ignore | |
92 | * whether VFP is enabled via FPEXC[EN]: this should be true for FMXR/FMRX | |
93 | * accesses to FPSID, FPEXC, MVFR0, MVFR1, MVFR2, and false for all other insns. | |
94 | */ | |
95 | static bool full_vfp_access_check(DisasContext *s, bool ignore_vfp_enabled) | |
96 | { | |
97 | if (s->fp_excp_el) { | |
98 | if (arm_dc_feature(s, ARM_FEATURE_M)) { | |
99 | gen_exception_insn(s, 4, EXCP_NOCP, syn_uncategorized(), | |
100 | s->fp_excp_el); | |
101 | } else { | |
102 | gen_exception_insn(s, 4, EXCP_UDEF, | |
103 | syn_fp_access_trap(1, 0xe, false), | |
104 | s->fp_excp_el); | |
105 | } | |
106 | return false; | |
107 | } | |
108 | ||
109 | if (!s->vfp_enabled && !ignore_vfp_enabled) { | |
110 | assert(!arm_dc_feature(s, ARM_FEATURE_M)); | |
111 | gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(), | |
112 | default_exception_el(s)); | |
113 | return false; | |
114 | } | |
115 | ||
116 | if (arm_dc_feature(s, ARM_FEATURE_M)) { | |
117 | /* Handle M-profile lazy FP state mechanics */ | |
118 | ||
119 | /* Trigger lazy-state preservation if necessary */ | |
120 | if (s->v7m_lspact) { | |
121 | /* | |
122 | * Lazy state saving affects external memory and also the NVIC, | |
123 | * so we must mark it as an IO operation for icount. | |
124 | */ | |
125 | if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) { | |
126 | gen_io_start(); | |
127 | } | |
128 | gen_helper_v7m_preserve_fp_state(cpu_env); | |
129 | if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) { | |
130 | gen_io_end(); | |
131 | } | |
132 | /* | |
133 | * If the preserve_fp_state helper doesn't throw an exception | |
134 | * then it will clear LSPACT; we don't need to repeat this for | |
135 | * any further FP insns in this TB. | |
136 | */ | |
137 | s->v7m_lspact = false; | |
138 | } | |
139 | ||
140 | /* Update ownership of FP context: set FPCCR.S to match current state */ | |
141 | if (s->v8m_fpccr_s_wrong) { | |
142 | TCGv_i32 tmp; | |
143 | ||
144 | tmp = load_cpu_field(v7m.fpccr[M_REG_S]); | |
145 | if (s->v8m_secure) { | |
146 | tcg_gen_ori_i32(tmp, tmp, R_V7M_FPCCR_S_MASK); | |
147 | } else { | |
148 | tcg_gen_andi_i32(tmp, tmp, ~R_V7M_FPCCR_S_MASK); | |
149 | } | |
150 | store_cpu_field(tmp, v7m.fpccr[M_REG_S]); | |
151 | /* Don't need to do this for any further FP insns in this TB */ | |
152 | s->v8m_fpccr_s_wrong = false; | |
153 | } | |
154 | ||
155 | if (s->v7m_new_fp_ctxt_needed) { | |
156 | /* | |
157 | * Create new FP context by updating CONTROL.FPCA, CONTROL.SFPA | |
158 | * and the FPSCR. | |
159 | */ | |
160 | TCGv_i32 control, fpscr; | |
161 | uint32_t bits = R_V7M_CONTROL_FPCA_MASK; | |
162 | ||
163 | fpscr = load_cpu_field(v7m.fpdscr[s->v8m_secure]); | |
164 | gen_helper_vfp_set_fpscr(cpu_env, fpscr); | |
165 | tcg_temp_free_i32(fpscr); | |
166 | /* | |
167 | * We don't need to arrange to end the TB, because the only | |
168 | * parts of FPSCR which we cache in the TB flags are the VECLEN | |
169 | * and VECSTRIDE, and those don't exist for M-profile. | |
170 | */ | |
171 | ||
172 | if (s->v8m_secure) { | |
173 | bits |= R_V7M_CONTROL_SFPA_MASK; | |
174 | } | |
175 | control = load_cpu_field(v7m.control[M_REG_S]); | |
176 | tcg_gen_ori_i32(control, control, bits); | |
177 | store_cpu_field(control, v7m.control[M_REG_S]); | |
178 | /* Don't need to do this for any further FP insns in this TB */ | |
179 | s->v7m_new_fp_ctxt_needed = false; | |
180 | } | |
181 | } | |
182 | ||
183 | return true; | |
184 | } | |
b3ff4b87 PM |
185 | |
186 | /* | |
187 | * The most usual kind of VFP access check, for everything except | |
188 | * FMXR/FMRX to the always-available special registers. | |
189 | */ | |
190 | static bool vfp_access_check(DisasContext *s) | |
191 | { | |
192 | return full_vfp_access_check(s, false); | |
193 | } | |
f7bbb8f3 PM |
194 | |
195 | static bool trans_VSEL(DisasContext *s, arg_VSEL *a) | |
196 | { | |
197 | uint32_t rd, rn, rm; | |
198 | bool dp = a->dp; | |
199 | ||
200 | if (!dc_isar_feature(aa32_vsel, s)) { | |
201 | return false; | |
202 | } | |
203 | ||
204 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
205 | if (dp && !dc_isar_feature(aa32_fp_d32, s) && | |
206 | ((a->vm | a->vn | a->vd) & 0x10)) { | |
207 | return false; | |
208 | } | |
209 | rd = a->vd; | |
210 | rn = a->vn; | |
211 | rm = a->vm; | |
212 | ||
213 | if (!vfp_access_check(s)) { | |
214 | return true; | |
215 | } | |
216 | ||
217 | if (dp) { | |
218 | TCGv_i64 frn, frm, dest; | |
219 | TCGv_i64 tmp, zero, zf, nf, vf; | |
220 | ||
221 | zero = tcg_const_i64(0); | |
222 | ||
223 | frn = tcg_temp_new_i64(); | |
224 | frm = tcg_temp_new_i64(); | |
225 | dest = tcg_temp_new_i64(); | |
226 | ||
227 | zf = tcg_temp_new_i64(); | |
228 | nf = tcg_temp_new_i64(); | |
229 | vf = tcg_temp_new_i64(); | |
230 | ||
231 | tcg_gen_extu_i32_i64(zf, cpu_ZF); | |
232 | tcg_gen_ext_i32_i64(nf, cpu_NF); | |
233 | tcg_gen_ext_i32_i64(vf, cpu_VF); | |
234 | ||
160f3b64 PM |
235 | neon_load_reg64(frn, rn); |
236 | neon_load_reg64(frm, rm); | |
f7bbb8f3 PM |
237 | switch (a->cc) { |
238 | case 0: /* eq: Z */ | |
239 | tcg_gen_movcond_i64(TCG_COND_EQ, dest, zf, zero, | |
240 | frn, frm); | |
241 | break; | |
242 | case 1: /* vs: V */ | |
243 | tcg_gen_movcond_i64(TCG_COND_LT, dest, vf, zero, | |
244 | frn, frm); | |
245 | break; | |
246 | case 2: /* ge: N == V -> N ^ V == 0 */ | |
247 | tmp = tcg_temp_new_i64(); | |
248 | tcg_gen_xor_i64(tmp, vf, nf); | |
249 | tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero, | |
250 | frn, frm); | |
251 | tcg_temp_free_i64(tmp); | |
252 | break; | |
253 | case 3: /* gt: !Z && N == V */ | |
254 | tcg_gen_movcond_i64(TCG_COND_NE, dest, zf, zero, | |
255 | frn, frm); | |
256 | tmp = tcg_temp_new_i64(); | |
257 | tcg_gen_xor_i64(tmp, vf, nf); | |
258 | tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero, | |
259 | dest, frm); | |
260 | tcg_temp_free_i64(tmp); | |
261 | break; | |
262 | } | |
160f3b64 | 263 | neon_store_reg64(dest, rd); |
f7bbb8f3 PM |
264 | tcg_temp_free_i64(frn); |
265 | tcg_temp_free_i64(frm); | |
266 | tcg_temp_free_i64(dest); | |
267 | ||
268 | tcg_temp_free_i64(zf); | |
269 | tcg_temp_free_i64(nf); | |
270 | tcg_temp_free_i64(vf); | |
271 | ||
272 | tcg_temp_free_i64(zero); | |
273 | } else { | |
274 | TCGv_i32 frn, frm, dest; | |
275 | TCGv_i32 tmp, zero; | |
276 | ||
277 | zero = tcg_const_i32(0); | |
278 | ||
279 | frn = tcg_temp_new_i32(); | |
280 | frm = tcg_temp_new_i32(); | |
281 | dest = tcg_temp_new_i32(); | |
160f3b64 PM |
282 | neon_load_reg32(frn, rn); |
283 | neon_load_reg32(frm, rm); | |
f7bbb8f3 PM |
284 | switch (a->cc) { |
285 | case 0: /* eq: Z */ | |
286 | tcg_gen_movcond_i32(TCG_COND_EQ, dest, cpu_ZF, zero, | |
287 | frn, frm); | |
288 | break; | |
289 | case 1: /* vs: V */ | |
290 | tcg_gen_movcond_i32(TCG_COND_LT, dest, cpu_VF, zero, | |
291 | frn, frm); | |
292 | break; | |
293 | case 2: /* ge: N == V -> N ^ V == 0 */ | |
294 | tmp = tcg_temp_new_i32(); | |
295 | tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF); | |
296 | tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero, | |
297 | frn, frm); | |
298 | tcg_temp_free_i32(tmp); | |
299 | break; | |
300 | case 3: /* gt: !Z && N == V */ | |
301 | tcg_gen_movcond_i32(TCG_COND_NE, dest, cpu_ZF, zero, | |
302 | frn, frm); | |
303 | tmp = tcg_temp_new_i32(); | |
304 | tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF); | |
305 | tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero, | |
306 | dest, frm); | |
307 | tcg_temp_free_i32(tmp); | |
308 | break; | |
309 | } | |
160f3b64 | 310 | neon_store_reg32(dest, rd); |
f7bbb8f3 PM |
311 | tcg_temp_free_i32(frn); |
312 | tcg_temp_free_i32(frm); | |
313 | tcg_temp_free_i32(dest); | |
314 | ||
315 | tcg_temp_free_i32(zero); | |
316 | } | |
317 | ||
318 | return true; | |
319 | } | |
320 | ||
321 | static bool trans_VMINMAXNM(DisasContext *s, arg_VMINMAXNM *a) | |
322 | { | |
323 | uint32_t rd, rn, rm; | |
324 | bool dp = a->dp; | |
325 | bool vmin = a->op; | |
326 | TCGv_ptr fpst; | |
327 | ||
328 | if (!dc_isar_feature(aa32_vminmaxnm, s)) { | |
329 | return false; | |
330 | } | |
331 | ||
332 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
333 | if (dp && !dc_isar_feature(aa32_fp_d32, s) && | |
334 | ((a->vm | a->vn | a->vd) & 0x10)) { | |
335 | return false; | |
336 | } | |
337 | rd = a->vd; | |
338 | rn = a->vn; | |
339 | rm = a->vm; | |
340 | ||
341 | if (!vfp_access_check(s)) { | |
342 | return true; | |
343 | } | |
344 | ||
345 | fpst = get_fpstatus_ptr(0); | |
346 | ||
347 | if (dp) { | |
348 | TCGv_i64 frn, frm, dest; | |
349 | ||
350 | frn = tcg_temp_new_i64(); | |
351 | frm = tcg_temp_new_i64(); | |
352 | dest = tcg_temp_new_i64(); | |
353 | ||
160f3b64 PM |
354 | neon_load_reg64(frn, rn); |
355 | neon_load_reg64(frm, rm); | |
f7bbb8f3 PM |
356 | if (vmin) { |
357 | gen_helper_vfp_minnumd(dest, frn, frm, fpst); | |
358 | } else { | |
359 | gen_helper_vfp_maxnumd(dest, frn, frm, fpst); | |
360 | } | |
160f3b64 | 361 | neon_store_reg64(dest, rd); |
f7bbb8f3 PM |
362 | tcg_temp_free_i64(frn); |
363 | tcg_temp_free_i64(frm); | |
364 | tcg_temp_free_i64(dest); | |
365 | } else { | |
366 | TCGv_i32 frn, frm, dest; | |
367 | ||
368 | frn = tcg_temp_new_i32(); | |
369 | frm = tcg_temp_new_i32(); | |
370 | dest = tcg_temp_new_i32(); | |
371 | ||
160f3b64 PM |
372 | neon_load_reg32(frn, rn); |
373 | neon_load_reg32(frm, rm); | |
f7bbb8f3 PM |
374 | if (vmin) { |
375 | gen_helper_vfp_minnums(dest, frn, frm, fpst); | |
376 | } else { | |
377 | gen_helper_vfp_maxnums(dest, frn, frm, fpst); | |
378 | } | |
160f3b64 | 379 | neon_store_reg32(dest, rd); |
f7bbb8f3 PM |
380 | tcg_temp_free_i32(frn); |
381 | tcg_temp_free_i32(frm); | |
382 | tcg_temp_free_i32(dest); | |
383 | } | |
384 | ||
385 | tcg_temp_free_ptr(fpst); | |
386 | return true; | |
387 | } | |
388 | ||
389 | /* | |
390 | * Table for converting the most common AArch32 encoding of | |
391 | * rounding mode to arm_fprounding order (which matches the | |
392 | * common AArch64 order); see ARM ARM pseudocode FPDecodeRM(). | |
393 | */ | |
394 | static const uint8_t fp_decode_rm[] = { | |
395 | FPROUNDING_TIEAWAY, | |
396 | FPROUNDING_TIEEVEN, | |
397 | FPROUNDING_POSINF, | |
398 | FPROUNDING_NEGINF, | |
399 | }; | |
400 | ||
401 | static bool trans_VRINT(DisasContext *s, arg_VRINT *a) | |
402 | { | |
403 | uint32_t rd, rm; | |
404 | bool dp = a->dp; | |
405 | TCGv_ptr fpst; | |
406 | TCGv_i32 tcg_rmode; | |
407 | int rounding = fp_decode_rm[a->rm]; | |
408 | ||
409 | if (!dc_isar_feature(aa32_vrint, s)) { | |
410 | return false; | |
411 | } | |
412 | ||
413 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
414 | if (dp && !dc_isar_feature(aa32_fp_d32, s) && | |
415 | ((a->vm | a->vd) & 0x10)) { | |
416 | return false; | |
417 | } | |
418 | rd = a->vd; | |
419 | rm = a->vm; | |
420 | ||
421 | if (!vfp_access_check(s)) { | |
422 | return true; | |
423 | } | |
424 | ||
425 | fpst = get_fpstatus_ptr(0); | |
426 | ||
427 | tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding)); | |
428 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
429 | ||
430 | if (dp) { | |
431 | TCGv_i64 tcg_op; | |
432 | TCGv_i64 tcg_res; | |
433 | tcg_op = tcg_temp_new_i64(); | |
434 | tcg_res = tcg_temp_new_i64(); | |
160f3b64 | 435 | neon_load_reg64(tcg_op, rm); |
f7bbb8f3 | 436 | gen_helper_rintd(tcg_res, tcg_op, fpst); |
160f3b64 | 437 | neon_store_reg64(tcg_res, rd); |
f7bbb8f3 PM |
438 | tcg_temp_free_i64(tcg_op); |
439 | tcg_temp_free_i64(tcg_res); | |
440 | } else { | |
441 | TCGv_i32 tcg_op; | |
442 | TCGv_i32 tcg_res; | |
443 | tcg_op = tcg_temp_new_i32(); | |
444 | tcg_res = tcg_temp_new_i32(); | |
160f3b64 | 445 | neon_load_reg32(tcg_op, rm); |
f7bbb8f3 | 446 | gen_helper_rints(tcg_res, tcg_op, fpst); |
160f3b64 | 447 | neon_store_reg32(tcg_res, rd); |
f7bbb8f3 PM |
448 | tcg_temp_free_i32(tcg_op); |
449 | tcg_temp_free_i32(tcg_res); | |
450 | } | |
451 | ||
452 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
453 | tcg_temp_free_i32(tcg_rmode); | |
454 | ||
455 | tcg_temp_free_ptr(fpst); | |
456 | return true; | |
457 | } | |
458 | ||
459 | static bool trans_VCVT(DisasContext *s, arg_VCVT *a) | |
460 | { | |
461 | uint32_t rd, rm; | |
462 | bool dp = a->dp; | |
463 | TCGv_ptr fpst; | |
464 | TCGv_i32 tcg_rmode, tcg_shift; | |
465 | int rounding = fp_decode_rm[a->rm]; | |
466 | bool is_signed = a->op; | |
467 | ||
468 | if (!dc_isar_feature(aa32_vcvt_dr, s)) { | |
469 | return false; | |
470 | } | |
471 | ||
472 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
473 | if (dp && !dc_isar_feature(aa32_fp_d32, s) && (a->vm & 0x10)) { | |
474 | return false; | |
475 | } | |
476 | rd = a->vd; | |
477 | rm = a->vm; | |
478 | ||
479 | if (!vfp_access_check(s)) { | |
480 | return true; | |
481 | } | |
482 | ||
483 | fpst = get_fpstatus_ptr(0); | |
484 | ||
485 | tcg_shift = tcg_const_i32(0); | |
486 | ||
487 | tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding)); | |
488 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
489 | ||
490 | if (dp) { | |
491 | TCGv_i64 tcg_double, tcg_res; | |
492 | TCGv_i32 tcg_tmp; | |
493 | tcg_double = tcg_temp_new_i64(); | |
494 | tcg_res = tcg_temp_new_i64(); | |
495 | tcg_tmp = tcg_temp_new_i32(); | |
160f3b64 | 496 | neon_load_reg64(tcg_double, rm); |
f7bbb8f3 PM |
497 | if (is_signed) { |
498 | gen_helper_vfp_tosld(tcg_res, tcg_double, tcg_shift, fpst); | |
499 | } else { | |
500 | gen_helper_vfp_tould(tcg_res, tcg_double, tcg_shift, fpst); | |
501 | } | |
502 | tcg_gen_extrl_i64_i32(tcg_tmp, tcg_res); | |
160f3b64 | 503 | neon_store_reg32(tcg_tmp, rd); |
f7bbb8f3 PM |
504 | tcg_temp_free_i32(tcg_tmp); |
505 | tcg_temp_free_i64(tcg_res); | |
506 | tcg_temp_free_i64(tcg_double); | |
507 | } else { | |
508 | TCGv_i32 tcg_single, tcg_res; | |
509 | tcg_single = tcg_temp_new_i32(); | |
510 | tcg_res = tcg_temp_new_i32(); | |
160f3b64 | 511 | neon_load_reg32(tcg_single, rm); |
f7bbb8f3 PM |
512 | if (is_signed) { |
513 | gen_helper_vfp_tosls(tcg_res, tcg_single, tcg_shift, fpst); | |
514 | } else { | |
515 | gen_helper_vfp_touls(tcg_res, tcg_single, tcg_shift, fpst); | |
516 | } | |
160f3b64 | 517 | neon_store_reg32(tcg_res, rd); |
f7bbb8f3 PM |
518 | tcg_temp_free_i32(tcg_res); |
519 | tcg_temp_free_i32(tcg_single); | |
520 | } | |
521 | ||
522 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
523 | tcg_temp_free_i32(tcg_rmode); | |
524 | ||
525 | tcg_temp_free_i32(tcg_shift); | |
526 | ||
527 | tcg_temp_free_ptr(fpst); | |
528 | ||
529 | return true; | |
530 | } | |
9851ed92 PM |
531 | |
532 | static bool trans_VMOV_to_gp(DisasContext *s, arg_VMOV_to_gp *a) | |
533 | { | |
534 | /* VMOV scalar to general purpose register */ | |
535 | TCGv_i32 tmp; | |
536 | int pass; | |
537 | uint32_t offset; | |
538 | ||
539 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
540 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vn & 0x10)) { | |
541 | return false; | |
542 | } | |
543 | ||
544 | offset = a->index << a->size; | |
545 | pass = extract32(offset, 2, 1); | |
546 | offset = extract32(offset, 0, 2) * 8; | |
547 | ||
548 | if (a->size != 2 && !arm_dc_feature(s, ARM_FEATURE_NEON)) { | |
549 | return false; | |
550 | } | |
551 | ||
552 | if (!vfp_access_check(s)) { | |
553 | return true; | |
554 | } | |
555 | ||
556 | tmp = neon_load_reg(a->vn, pass); | |
557 | switch (a->size) { | |
558 | case 0: | |
559 | if (offset) { | |
560 | tcg_gen_shri_i32(tmp, tmp, offset); | |
561 | } | |
562 | if (a->u) { | |
563 | gen_uxtb(tmp); | |
564 | } else { | |
565 | gen_sxtb(tmp); | |
566 | } | |
567 | break; | |
568 | case 1: | |
569 | if (a->u) { | |
570 | if (offset) { | |
571 | tcg_gen_shri_i32(tmp, tmp, 16); | |
572 | } else { | |
573 | gen_uxth(tmp); | |
574 | } | |
575 | } else { | |
576 | if (offset) { | |
577 | tcg_gen_sari_i32(tmp, tmp, 16); | |
578 | } else { | |
579 | gen_sxth(tmp); | |
580 | } | |
581 | } | |
582 | break; | |
583 | case 2: | |
584 | break; | |
585 | } | |
586 | store_reg(s, a->rt, tmp); | |
587 | ||
588 | return true; | |
589 | } | |
590 | ||
591 | static bool trans_VMOV_from_gp(DisasContext *s, arg_VMOV_from_gp *a) | |
592 | { | |
593 | /* VMOV general purpose register to scalar */ | |
594 | TCGv_i32 tmp, tmp2; | |
595 | int pass; | |
596 | uint32_t offset; | |
597 | ||
598 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
599 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vn & 0x10)) { | |
600 | return false; | |
601 | } | |
602 | ||
603 | offset = a->index << a->size; | |
604 | pass = extract32(offset, 2, 1); | |
605 | offset = extract32(offset, 0, 2) * 8; | |
606 | ||
607 | if (a->size != 2 && !arm_dc_feature(s, ARM_FEATURE_NEON)) { | |
608 | return false; | |
609 | } | |
610 | ||
611 | if (!vfp_access_check(s)) { | |
612 | return true; | |
613 | } | |
614 | ||
615 | tmp = load_reg(s, a->rt); | |
616 | switch (a->size) { | |
617 | case 0: | |
618 | tmp2 = neon_load_reg(a->vn, pass); | |
619 | tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 8); | |
620 | tcg_temp_free_i32(tmp2); | |
621 | break; | |
622 | case 1: | |
623 | tmp2 = neon_load_reg(a->vn, pass); | |
624 | tcg_gen_deposit_i32(tmp, tmp2, tmp, offset, 16); | |
625 | tcg_temp_free_i32(tmp2); | |
626 | break; | |
627 | case 2: | |
628 | break; | |
629 | } | |
630 | neon_store_reg(a->vn, pass, tmp); | |
631 | ||
632 | return true; | |
633 | } | |
634 | ||
635 | static bool trans_VDUP(DisasContext *s, arg_VDUP *a) | |
636 | { | |
637 | /* VDUP (general purpose register) */ | |
638 | TCGv_i32 tmp; | |
639 | int size, vec_size; | |
640 | ||
641 | if (!arm_dc_feature(s, ARM_FEATURE_NEON)) { | |
642 | return false; | |
643 | } | |
644 | ||
645 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
646 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vn & 0x10)) { | |
647 | return false; | |
648 | } | |
649 | ||
650 | if (a->b && a->e) { | |
651 | return false; | |
652 | } | |
653 | ||
654 | if (a->q && (a->vn & 1)) { | |
655 | return false; | |
656 | } | |
657 | ||
658 | vec_size = a->q ? 16 : 8; | |
659 | if (a->b) { | |
660 | size = 0; | |
661 | } else if (a->e) { | |
662 | size = 1; | |
663 | } else { | |
664 | size = 2; | |
665 | } | |
666 | ||
667 | if (!vfp_access_check(s)) { | |
668 | return true; | |
669 | } | |
670 | ||
671 | tmp = load_reg(s, a->rt); | |
672 | tcg_gen_gvec_dup_i32(size, neon_reg_offset(a->vn, 0), | |
673 | vec_size, vec_size, tmp); | |
674 | tcg_temp_free_i32(tmp); | |
675 | ||
676 | return true; | |
677 | } | |
a9ab5001 PM |
678 | |
679 | static bool trans_VMSR_VMRS(DisasContext *s, arg_VMSR_VMRS *a) | |
680 | { | |
681 | TCGv_i32 tmp; | |
682 | bool ignore_vfp_enabled = false; | |
683 | ||
684 | if (arm_dc_feature(s, ARM_FEATURE_M)) { | |
685 | /* | |
686 | * The only M-profile VFP vmrs/vmsr sysreg is FPSCR. | |
687 | * Writes to R15 are UNPREDICTABLE; we choose to undef. | |
688 | */ | |
689 | if (a->rt == 15 || a->reg != ARM_VFP_FPSCR) { | |
690 | return false; | |
691 | } | |
692 | } | |
693 | ||
694 | switch (a->reg) { | |
695 | case ARM_VFP_FPSID: | |
696 | /* | |
697 | * VFPv2 allows access to FPSID from userspace; VFPv3 restricts | |
698 | * all ID registers to privileged access only. | |
699 | */ | |
700 | if (IS_USER(s) && arm_dc_feature(s, ARM_FEATURE_VFP3)) { | |
701 | return false; | |
702 | } | |
703 | ignore_vfp_enabled = true; | |
704 | break; | |
705 | case ARM_VFP_MVFR0: | |
706 | case ARM_VFP_MVFR1: | |
707 | if (IS_USER(s) || !arm_dc_feature(s, ARM_FEATURE_MVFR)) { | |
708 | return false; | |
709 | } | |
710 | ignore_vfp_enabled = true; | |
711 | break; | |
712 | case ARM_VFP_MVFR2: | |
713 | if (IS_USER(s) || !arm_dc_feature(s, ARM_FEATURE_V8)) { | |
714 | return false; | |
715 | } | |
716 | ignore_vfp_enabled = true; | |
717 | break; | |
718 | case ARM_VFP_FPSCR: | |
719 | break; | |
720 | case ARM_VFP_FPEXC: | |
721 | if (IS_USER(s)) { | |
722 | return false; | |
723 | } | |
724 | ignore_vfp_enabled = true; | |
725 | break; | |
726 | case ARM_VFP_FPINST: | |
727 | case ARM_VFP_FPINST2: | |
728 | /* Not present in VFPv3 */ | |
729 | if (IS_USER(s) || arm_dc_feature(s, ARM_FEATURE_VFP3)) { | |
730 | return false; | |
731 | } | |
732 | break; | |
733 | default: | |
734 | return false; | |
735 | } | |
736 | ||
737 | if (!full_vfp_access_check(s, ignore_vfp_enabled)) { | |
738 | return true; | |
739 | } | |
740 | ||
741 | if (a->l) { | |
742 | /* VMRS, move VFP special register to gp register */ | |
743 | switch (a->reg) { | |
744 | case ARM_VFP_FPSID: | |
745 | case ARM_VFP_FPEXC: | |
746 | case ARM_VFP_FPINST: | |
747 | case ARM_VFP_FPINST2: | |
748 | case ARM_VFP_MVFR0: | |
749 | case ARM_VFP_MVFR1: | |
750 | case ARM_VFP_MVFR2: | |
751 | tmp = load_cpu_field(vfp.xregs[a->reg]); | |
752 | break; | |
753 | case ARM_VFP_FPSCR: | |
754 | if (a->rt == 15) { | |
755 | tmp = load_cpu_field(vfp.xregs[ARM_VFP_FPSCR]); | |
756 | tcg_gen_andi_i32(tmp, tmp, 0xf0000000); | |
757 | } else { | |
758 | tmp = tcg_temp_new_i32(); | |
759 | gen_helper_vfp_get_fpscr(tmp, cpu_env); | |
760 | } | |
761 | break; | |
762 | default: | |
763 | g_assert_not_reached(); | |
764 | } | |
765 | ||
766 | if (a->rt == 15) { | |
767 | /* Set the 4 flag bits in the CPSR. */ | |
768 | gen_set_nzcv(tmp); | |
769 | tcg_temp_free_i32(tmp); | |
770 | } else { | |
771 | store_reg(s, a->rt, tmp); | |
772 | } | |
773 | } else { | |
774 | /* VMSR, move gp register to VFP special register */ | |
775 | switch (a->reg) { | |
776 | case ARM_VFP_FPSID: | |
777 | case ARM_VFP_MVFR0: | |
778 | case ARM_VFP_MVFR1: | |
779 | case ARM_VFP_MVFR2: | |
780 | /* Writes are ignored. */ | |
781 | break; | |
782 | case ARM_VFP_FPSCR: | |
783 | tmp = load_reg(s, a->rt); | |
784 | gen_helper_vfp_set_fpscr(cpu_env, tmp); | |
785 | tcg_temp_free_i32(tmp); | |
786 | gen_lookup_tb(s); | |
787 | break; | |
788 | case ARM_VFP_FPEXC: | |
789 | /* | |
790 | * TODO: VFP subarchitecture support. | |
791 | * For now, keep the EN bit only | |
792 | */ | |
793 | tmp = load_reg(s, a->rt); | |
794 | tcg_gen_andi_i32(tmp, tmp, 1 << 30); | |
795 | store_cpu_field(tmp, vfp.xregs[a->reg]); | |
796 | gen_lookup_tb(s); | |
797 | break; | |
798 | case ARM_VFP_FPINST: | |
799 | case ARM_VFP_FPINST2: | |
800 | tmp = load_reg(s, a->rt); | |
801 | store_cpu_field(tmp, vfp.xregs[a->reg]); | |
802 | break; | |
803 | default: | |
804 | g_assert_not_reached(); | |
805 | } | |
806 | } | |
807 | ||
808 | return true; | |
809 | } | |
810 | ||
811 | static bool trans_VMOV_single(DisasContext *s, arg_VMOV_single *a) | |
812 | { | |
813 | TCGv_i32 tmp; | |
814 | ||
815 | if (!vfp_access_check(s)) { | |
816 | return true; | |
817 | } | |
818 | ||
819 | if (a->l) { | |
820 | /* VFP to general purpose register */ | |
821 | tmp = tcg_temp_new_i32(); | |
822 | neon_load_reg32(tmp, a->vn); | |
823 | if (a->rt == 15) { | |
824 | /* Set the 4 flag bits in the CPSR. */ | |
825 | gen_set_nzcv(tmp); | |
826 | tcg_temp_free_i32(tmp); | |
827 | } else { | |
828 | store_reg(s, a->rt, tmp); | |
829 | } | |
830 | } else { | |
831 | /* general purpose register to VFP */ | |
832 | tmp = load_reg(s, a->rt); | |
833 | neon_store_reg32(tmp, a->vn); | |
834 | tcg_temp_free_i32(tmp); | |
835 | } | |
836 | ||
837 | return true; | |
838 | } | |
81f68110 PM |
839 | |
840 | static bool trans_VMOV_64_sp(DisasContext *s, arg_VMOV_64_sp *a) | |
841 | { | |
842 | TCGv_i32 tmp; | |
843 | ||
844 | /* | |
845 | * VMOV between two general-purpose registers and two single precision | |
846 | * floating point registers | |
847 | */ | |
848 | if (!vfp_access_check(s)) { | |
849 | return true; | |
850 | } | |
851 | ||
852 | if (a->op) { | |
853 | /* fpreg to gpreg */ | |
854 | tmp = tcg_temp_new_i32(); | |
855 | neon_load_reg32(tmp, a->vm); | |
856 | store_reg(s, a->rt, tmp); | |
857 | tmp = tcg_temp_new_i32(); | |
858 | neon_load_reg32(tmp, a->vm + 1); | |
859 | store_reg(s, a->rt2, tmp); | |
860 | } else { | |
861 | /* gpreg to fpreg */ | |
862 | tmp = load_reg(s, a->rt); | |
863 | neon_store_reg32(tmp, a->vm); | |
864 | tmp = load_reg(s, a->rt2); | |
865 | neon_store_reg32(tmp, a->vm + 1); | |
866 | } | |
867 | ||
868 | return true; | |
869 | } | |
870 | ||
871 | static bool trans_VMOV_64_dp(DisasContext *s, arg_VMOV_64_sp *a) | |
872 | { | |
873 | TCGv_i32 tmp; | |
874 | ||
875 | /* | |
876 | * VMOV between two general-purpose registers and one double precision | |
877 | * floating point register | |
878 | */ | |
879 | ||
880 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
881 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vm & 0x10)) { | |
882 | return false; | |
883 | } | |
884 | ||
885 | if (!vfp_access_check(s)) { | |
886 | return true; | |
887 | } | |
888 | ||
889 | if (a->op) { | |
890 | /* fpreg to gpreg */ | |
891 | tmp = tcg_temp_new_i32(); | |
892 | neon_load_reg32(tmp, a->vm * 2); | |
893 | store_reg(s, a->rt, tmp); | |
894 | tmp = tcg_temp_new_i32(); | |
895 | neon_load_reg32(tmp, a->vm * 2 + 1); | |
896 | store_reg(s, a->rt2, tmp); | |
897 | } else { | |
898 | /* gpreg to fpreg */ | |
899 | tmp = load_reg(s, a->rt); | |
900 | neon_store_reg32(tmp, a->vm * 2); | |
901 | tcg_temp_free_i32(tmp); | |
902 | tmp = load_reg(s, a->rt2); | |
903 | neon_store_reg32(tmp, a->vm * 2 + 1); | |
904 | tcg_temp_free_i32(tmp); | |
905 | } | |
906 | ||
907 | return true; | |
908 | } | |
79b02a3b PM |
909 | |
910 | static bool trans_VLDR_VSTR_sp(DisasContext *s, arg_VLDR_VSTR_sp *a) | |
911 | { | |
912 | uint32_t offset; | |
3993d040 | 913 | TCGv_i32 addr, tmp; |
79b02a3b PM |
914 | |
915 | if (!vfp_access_check(s)) { | |
916 | return true; | |
917 | } | |
918 | ||
919 | offset = a->imm << 2; | |
920 | if (!a->u) { | |
921 | offset = -offset; | |
922 | } | |
923 | ||
924 | if (s->thumb && a->rn == 15) { | |
925 | /* This is actually UNPREDICTABLE */ | |
926 | addr = tcg_temp_new_i32(); | |
927 | tcg_gen_movi_i32(addr, s->pc & ~2); | |
928 | } else { | |
929 | addr = load_reg(s, a->rn); | |
930 | } | |
931 | tcg_gen_addi_i32(addr, addr, offset); | |
3993d040 | 932 | tmp = tcg_temp_new_i32(); |
79b02a3b | 933 | if (a->l) { |
3993d040 PM |
934 | gen_aa32_ld32u(s, tmp, addr, get_mem_index(s)); |
935 | neon_store_reg32(tmp, a->vd); | |
79b02a3b | 936 | } else { |
3993d040 PM |
937 | neon_load_reg32(tmp, a->vd); |
938 | gen_aa32_st32(s, tmp, addr, get_mem_index(s)); | |
79b02a3b | 939 | } |
3993d040 | 940 | tcg_temp_free_i32(tmp); |
79b02a3b PM |
941 | tcg_temp_free_i32(addr); |
942 | ||
943 | return true; | |
944 | } | |
945 | ||
946 | static bool trans_VLDR_VSTR_dp(DisasContext *s, arg_VLDR_VSTR_sp *a) | |
947 | { | |
948 | uint32_t offset; | |
949 | TCGv_i32 addr; | |
3993d040 | 950 | TCGv_i64 tmp; |
79b02a3b PM |
951 | |
952 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
953 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vd & 0x10)) { | |
954 | return false; | |
955 | } | |
956 | ||
957 | if (!vfp_access_check(s)) { | |
958 | return true; | |
959 | } | |
960 | ||
961 | offset = a->imm << 2; | |
962 | if (!a->u) { | |
963 | offset = -offset; | |
964 | } | |
965 | ||
966 | if (s->thumb && a->rn == 15) { | |
967 | /* This is actually UNPREDICTABLE */ | |
968 | addr = tcg_temp_new_i32(); | |
969 | tcg_gen_movi_i32(addr, s->pc & ~2); | |
970 | } else { | |
971 | addr = load_reg(s, a->rn); | |
972 | } | |
973 | tcg_gen_addi_i32(addr, addr, offset); | |
3993d040 | 974 | tmp = tcg_temp_new_i64(); |
79b02a3b | 975 | if (a->l) { |
3993d040 PM |
976 | gen_aa32_ld64(s, tmp, addr, get_mem_index(s)); |
977 | neon_store_reg64(tmp, a->vd); | |
79b02a3b | 978 | } else { |
3993d040 PM |
979 | neon_load_reg64(tmp, a->vd); |
980 | gen_aa32_st64(s, tmp, addr, get_mem_index(s)); | |
79b02a3b | 981 | } |
3993d040 | 982 | tcg_temp_free_i64(tmp); |
79b02a3b PM |
983 | tcg_temp_free_i32(addr); |
984 | ||
985 | return true; | |
986 | } | |
fa288de2 PM |
987 | |
988 | static bool trans_VLDM_VSTM_sp(DisasContext *s, arg_VLDM_VSTM_sp *a) | |
989 | { | |
990 | uint32_t offset; | |
3993d040 | 991 | TCGv_i32 addr, tmp; |
fa288de2 PM |
992 | int i, n; |
993 | ||
994 | n = a->imm; | |
995 | ||
996 | if (n == 0 || (a->vd + n) > 32) { | |
997 | /* | |
998 | * UNPREDICTABLE cases for bad immediates: we choose to | |
999 | * UNDEF to avoid generating huge numbers of TCG ops | |
1000 | */ | |
1001 | return false; | |
1002 | } | |
1003 | if (a->rn == 15 && a->w) { | |
1004 | /* writeback to PC is UNPREDICTABLE, we choose to UNDEF */ | |
1005 | return false; | |
1006 | } | |
1007 | ||
1008 | if (!vfp_access_check(s)) { | |
1009 | return true; | |
1010 | } | |
1011 | ||
1012 | if (s->thumb && a->rn == 15) { | |
1013 | /* This is actually UNPREDICTABLE */ | |
1014 | addr = tcg_temp_new_i32(); | |
1015 | tcg_gen_movi_i32(addr, s->pc & ~2); | |
1016 | } else { | |
1017 | addr = load_reg(s, a->rn); | |
1018 | } | |
1019 | if (a->p) { | |
1020 | /* pre-decrement */ | |
1021 | tcg_gen_addi_i32(addr, addr, -(a->imm << 2)); | |
1022 | } | |
1023 | ||
1024 | if (s->v8m_stackcheck && a->rn == 13 && a->w) { | |
1025 | /* | |
1026 | * Here 'addr' is the lowest address we will store to, | |
1027 | * and is either the old SP (if post-increment) or | |
1028 | * the new SP (if pre-decrement). For post-increment | |
1029 | * where the old value is below the limit and the new | |
1030 | * value is above, it is UNKNOWN whether the limit check | |
1031 | * triggers; we choose to trigger. | |
1032 | */ | |
1033 | gen_helper_v8m_stackcheck(cpu_env, addr); | |
1034 | } | |
1035 | ||
1036 | offset = 4; | |
3993d040 | 1037 | tmp = tcg_temp_new_i32(); |
fa288de2 PM |
1038 | for (i = 0; i < n; i++) { |
1039 | if (a->l) { | |
1040 | /* load */ | |
3993d040 PM |
1041 | gen_aa32_ld32u(s, tmp, addr, get_mem_index(s)); |
1042 | neon_store_reg32(tmp, a->vd + i); | |
fa288de2 PM |
1043 | } else { |
1044 | /* store */ | |
3993d040 PM |
1045 | neon_load_reg32(tmp, a->vd + i); |
1046 | gen_aa32_st32(s, tmp, addr, get_mem_index(s)); | |
fa288de2 PM |
1047 | } |
1048 | tcg_gen_addi_i32(addr, addr, offset); | |
1049 | } | |
3993d040 | 1050 | tcg_temp_free_i32(tmp); |
fa288de2 PM |
1051 | if (a->w) { |
1052 | /* writeback */ | |
1053 | if (a->p) { | |
1054 | offset = -offset * n; | |
1055 | tcg_gen_addi_i32(addr, addr, offset); | |
1056 | } | |
1057 | store_reg(s, a->rn, addr); | |
1058 | } else { | |
1059 | tcg_temp_free_i32(addr); | |
1060 | } | |
1061 | ||
1062 | return true; | |
1063 | } | |
1064 | ||
1065 | static bool trans_VLDM_VSTM_dp(DisasContext *s, arg_VLDM_VSTM_dp *a) | |
1066 | { | |
1067 | uint32_t offset; | |
1068 | TCGv_i32 addr; | |
3993d040 | 1069 | TCGv_i64 tmp; |
fa288de2 PM |
1070 | int i, n; |
1071 | ||
1072 | n = a->imm >> 1; | |
1073 | ||
1074 | if (n == 0 || (a->vd + n) > 32 || n > 16) { | |
1075 | /* | |
1076 | * UNPREDICTABLE cases for bad immediates: we choose to | |
1077 | * UNDEF to avoid generating huge numbers of TCG ops | |
1078 | */ | |
1079 | return false; | |
1080 | } | |
1081 | if (a->rn == 15 && a->w) { | |
1082 | /* writeback to PC is UNPREDICTABLE, we choose to UNDEF */ | |
1083 | return false; | |
1084 | } | |
1085 | ||
1086 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
1087 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vd + n) > 16) { | |
1088 | return false; | |
1089 | } | |
1090 | ||
1091 | if (!vfp_access_check(s)) { | |
1092 | return true; | |
1093 | } | |
1094 | ||
1095 | if (s->thumb && a->rn == 15) { | |
1096 | /* This is actually UNPREDICTABLE */ | |
1097 | addr = tcg_temp_new_i32(); | |
1098 | tcg_gen_movi_i32(addr, s->pc & ~2); | |
1099 | } else { | |
1100 | addr = load_reg(s, a->rn); | |
1101 | } | |
1102 | if (a->p) { | |
1103 | /* pre-decrement */ | |
1104 | tcg_gen_addi_i32(addr, addr, -(a->imm << 2)); | |
1105 | } | |
1106 | ||
1107 | if (s->v8m_stackcheck && a->rn == 13 && a->w) { | |
1108 | /* | |
1109 | * Here 'addr' is the lowest address we will store to, | |
1110 | * and is either the old SP (if post-increment) or | |
1111 | * the new SP (if pre-decrement). For post-increment | |
1112 | * where the old value is below the limit and the new | |
1113 | * value is above, it is UNKNOWN whether the limit check | |
1114 | * triggers; we choose to trigger. | |
1115 | */ | |
1116 | gen_helper_v8m_stackcheck(cpu_env, addr); | |
1117 | } | |
1118 | ||
1119 | offset = 8; | |
3993d040 | 1120 | tmp = tcg_temp_new_i64(); |
fa288de2 PM |
1121 | for (i = 0; i < n; i++) { |
1122 | if (a->l) { | |
1123 | /* load */ | |
3993d040 PM |
1124 | gen_aa32_ld64(s, tmp, addr, get_mem_index(s)); |
1125 | neon_store_reg64(tmp, a->vd + i); | |
fa288de2 PM |
1126 | } else { |
1127 | /* store */ | |
3993d040 PM |
1128 | neon_load_reg64(tmp, a->vd + i); |
1129 | gen_aa32_st64(s, tmp, addr, get_mem_index(s)); | |
fa288de2 PM |
1130 | } |
1131 | tcg_gen_addi_i32(addr, addr, offset); | |
1132 | } | |
3993d040 | 1133 | tcg_temp_free_i64(tmp); |
fa288de2 PM |
1134 | if (a->w) { |
1135 | /* writeback */ | |
1136 | if (a->p) { | |
1137 | offset = -offset * n; | |
1138 | } else if (a->imm & 1) { | |
1139 | offset = 4; | |
1140 | } else { | |
1141 | offset = 0; | |
1142 | } | |
1143 | ||
1144 | if (offset != 0) { | |
1145 | tcg_gen_addi_i32(addr, addr, offset); | |
1146 | } | |
1147 | store_reg(s, a->rn, addr); | |
1148 | } else { | |
1149 | tcg_temp_free_i32(addr); | |
1150 | } | |
1151 | ||
1152 | return true; | |
1153 | } | |
266bd25c PM |
1154 | |
1155 | /* | |
1156 | * Types for callbacks for do_vfp_3op_sp() and do_vfp_3op_dp(). | |
1157 | * The callback should emit code to write a value to vd. If | |
1158 | * do_vfp_3op_{sp,dp}() was passed reads_vd then the TCGv vd | |
1159 | * will contain the old value of the relevant VFP register; | |
1160 | * otherwise it must be written to only. | |
1161 | */ | |
1162 | typedef void VFPGen3OpSPFn(TCGv_i32 vd, | |
1163 | TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst); | |
1164 | typedef void VFPGen3OpDPFn(TCGv_i64 vd, | |
1165 | TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst); | |
1166 | ||
90287e22 PM |
1167 | /* |
1168 | * Types for callbacks for do_vfp_2op_sp() and do_vfp_2op_dp(). | |
1169 | * The callback should emit code to write a value to vd (which | |
1170 | * should be written to only). | |
1171 | */ | |
1172 | typedef void VFPGen2OpSPFn(TCGv_i32 vd, TCGv_i32 vm); | |
1173 | typedef void VFPGen2OpDPFn(TCGv_i64 vd, TCGv_i64 vm); | |
1174 | ||
18cf951a PM |
1175 | /* |
1176 | * Return true if the specified S reg is in a scalar bank | |
1177 | * (ie if it is s0..s7) | |
1178 | */ | |
1179 | static inline bool vfp_sreg_is_scalar(int reg) | |
1180 | { | |
1181 | return (reg & 0x18) == 0; | |
1182 | } | |
1183 | ||
1184 | /* | |
1185 | * Return true if the specified D reg is in a scalar bank | |
1186 | * (ie if it is d0..d3 or d16..d19) | |
1187 | */ | |
1188 | static inline bool vfp_dreg_is_scalar(int reg) | |
1189 | { | |
1190 | return (reg & 0xc) == 0; | |
1191 | } | |
1192 | ||
1193 | /* | |
1194 | * Advance the S reg number forwards by delta within its bank | |
1195 | * (ie increment the low 3 bits but leave the rest the same) | |
1196 | */ | |
1197 | static inline int vfp_advance_sreg(int reg, int delta) | |
1198 | { | |
1199 | return ((reg + delta) & 0x7) | (reg & ~0x7); | |
1200 | } | |
1201 | ||
1202 | /* | |
1203 | * Advance the D reg number forwards by delta within its bank | |
1204 | * (ie increment the low 2 bits but leave the rest the same) | |
1205 | */ | |
1206 | static inline int vfp_advance_dreg(int reg, int delta) | |
1207 | { | |
1208 | return ((reg + delta) & 0x3) | (reg & ~0x3); | |
1209 | } | |
1210 | ||
266bd25c PM |
1211 | /* |
1212 | * Perform a 3-operand VFP data processing instruction. fn is the | |
1213 | * callback to do the actual operation; this function deals with the | |
1214 | * code to handle looping around for VFP vector processing. | |
1215 | */ | |
1216 | static bool do_vfp_3op_sp(DisasContext *s, VFPGen3OpSPFn *fn, | |
1217 | int vd, int vn, int vm, bool reads_vd) | |
1218 | { | |
1219 | uint32_t delta_m = 0; | |
1220 | uint32_t delta_d = 0; | |
266bd25c PM |
1221 | int veclen = s->vec_len; |
1222 | TCGv_i32 f0, f1, fd; | |
1223 | TCGv_ptr fpst; | |
1224 | ||
1225 | if (!dc_isar_feature(aa32_fpshvec, s) && | |
1226 | (veclen != 0 || s->vec_stride != 0)) { | |
1227 | return false; | |
1228 | } | |
1229 | ||
1230 | if (!vfp_access_check(s)) { | |
1231 | return true; | |
1232 | } | |
1233 | ||
1234 | if (veclen > 0) { | |
266bd25c | 1235 | /* Figure out what type of vector operation this is. */ |
18cf951a | 1236 | if (vfp_sreg_is_scalar(vd)) { |
266bd25c PM |
1237 | /* scalar */ |
1238 | veclen = 0; | |
1239 | } else { | |
1240 | delta_d = s->vec_stride + 1; | |
1241 | ||
18cf951a | 1242 | if (vfp_sreg_is_scalar(vm)) { |
266bd25c PM |
1243 | /* mixed scalar/vector */ |
1244 | delta_m = 0; | |
1245 | } else { | |
1246 | /* vector */ | |
1247 | delta_m = delta_d; | |
1248 | } | |
1249 | } | |
1250 | } | |
1251 | ||
1252 | f0 = tcg_temp_new_i32(); | |
1253 | f1 = tcg_temp_new_i32(); | |
1254 | fd = tcg_temp_new_i32(); | |
1255 | fpst = get_fpstatus_ptr(0); | |
1256 | ||
1257 | neon_load_reg32(f0, vn); | |
1258 | neon_load_reg32(f1, vm); | |
1259 | ||
1260 | for (;;) { | |
1261 | if (reads_vd) { | |
1262 | neon_load_reg32(fd, vd); | |
1263 | } | |
1264 | fn(fd, f0, f1, fpst); | |
1265 | neon_store_reg32(fd, vd); | |
1266 | ||
1267 | if (veclen == 0) { | |
1268 | break; | |
1269 | } | |
1270 | ||
1271 | /* Set up the operands for the next iteration */ | |
1272 | veclen--; | |
18cf951a PM |
1273 | vd = vfp_advance_sreg(vd, delta_d); |
1274 | vn = vfp_advance_sreg(vn, delta_d); | |
266bd25c PM |
1275 | neon_load_reg32(f0, vn); |
1276 | if (delta_m) { | |
18cf951a | 1277 | vm = vfp_advance_sreg(vm, delta_m); |
266bd25c PM |
1278 | neon_load_reg32(f1, vm); |
1279 | } | |
1280 | } | |
1281 | ||
1282 | tcg_temp_free_i32(f0); | |
1283 | tcg_temp_free_i32(f1); | |
1284 | tcg_temp_free_i32(fd); | |
1285 | tcg_temp_free_ptr(fpst); | |
1286 | ||
1287 | return true; | |
1288 | } | |
1289 | ||
1290 | static bool do_vfp_3op_dp(DisasContext *s, VFPGen3OpDPFn *fn, | |
1291 | int vd, int vn, int vm, bool reads_vd) | |
1292 | { | |
1293 | uint32_t delta_m = 0; | |
1294 | uint32_t delta_d = 0; | |
266bd25c PM |
1295 | int veclen = s->vec_len; |
1296 | TCGv_i64 f0, f1, fd; | |
1297 | TCGv_ptr fpst; | |
1298 | ||
1299 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
1300 | if (!dc_isar_feature(aa32_fp_d32, s) && ((vd | vn | vm) & 0x10)) { | |
1301 | return false; | |
1302 | } | |
1303 | ||
1304 | if (!dc_isar_feature(aa32_fpshvec, s) && | |
1305 | (veclen != 0 || s->vec_stride != 0)) { | |
1306 | return false; | |
1307 | } | |
1308 | ||
1309 | if (!vfp_access_check(s)) { | |
1310 | return true; | |
1311 | } | |
1312 | ||
1313 | if (veclen > 0) { | |
266bd25c | 1314 | /* Figure out what type of vector operation this is. */ |
18cf951a | 1315 | if (vfp_dreg_is_scalar(vd)) { |
266bd25c PM |
1316 | /* scalar */ |
1317 | veclen = 0; | |
1318 | } else { | |
1319 | delta_d = (s->vec_stride >> 1) + 1; | |
1320 | ||
18cf951a | 1321 | if (vfp_dreg_is_scalar(vm)) { |
266bd25c PM |
1322 | /* mixed scalar/vector */ |
1323 | delta_m = 0; | |
1324 | } else { | |
1325 | /* vector */ | |
1326 | delta_m = delta_d; | |
1327 | } | |
1328 | } | |
1329 | } | |
1330 | ||
1331 | f0 = tcg_temp_new_i64(); | |
1332 | f1 = tcg_temp_new_i64(); | |
1333 | fd = tcg_temp_new_i64(); | |
1334 | fpst = get_fpstatus_ptr(0); | |
1335 | ||
1336 | neon_load_reg64(f0, vn); | |
1337 | neon_load_reg64(f1, vm); | |
1338 | ||
1339 | for (;;) { | |
1340 | if (reads_vd) { | |
1341 | neon_load_reg64(fd, vd); | |
1342 | } | |
1343 | fn(fd, f0, f1, fpst); | |
1344 | neon_store_reg64(fd, vd); | |
1345 | ||
1346 | if (veclen == 0) { | |
1347 | break; | |
1348 | } | |
1349 | /* Set up the operands for the next iteration */ | |
1350 | veclen--; | |
18cf951a PM |
1351 | vd = vfp_advance_dreg(vd, delta_d); |
1352 | vn = vfp_advance_dreg(vn, delta_d); | |
266bd25c PM |
1353 | neon_load_reg64(f0, vn); |
1354 | if (delta_m) { | |
18cf951a | 1355 | vm = vfp_advance_dreg(vm, delta_m); |
266bd25c PM |
1356 | neon_load_reg64(f1, vm); |
1357 | } | |
1358 | } | |
1359 | ||
1360 | tcg_temp_free_i64(f0); | |
1361 | tcg_temp_free_i64(f1); | |
1362 | tcg_temp_free_i64(fd); | |
1363 | tcg_temp_free_ptr(fpst); | |
1364 | ||
1365 | return true; | |
1366 | } | |
1367 | ||
90287e22 PM |
1368 | static bool do_vfp_2op_sp(DisasContext *s, VFPGen2OpSPFn *fn, int vd, int vm) |
1369 | { | |
1370 | uint32_t delta_m = 0; | |
1371 | uint32_t delta_d = 0; | |
90287e22 PM |
1372 | int veclen = s->vec_len; |
1373 | TCGv_i32 f0, fd; | |
1374 | ||
1375 | if (!dc_isar_feature(aa32_fpshvec, s) && | |
1376 | (veclen != 0 || s->vec_stride != 0)) { | |
1377 | return false; | |
1378 | } | |
1379 | ||
1380 | if (!vfp_access_check(s)) { | |
1381 | return true; | |
1382 | } | |
1383 | ||
1384 | if (veclen > 0) { | |
90287e22 | 1385 | /* Figure out what type of vector operation this is. */ |
18cf951a | 1386 | if (vfp_sreg_is_scalar(vd)) { |
90287e22 PM |
1387 | /* scalar */ |
1388 | veclen = 0; | |
1389 | } else { | |
1390 | delta_d = s->vec_stride + 1; | |
1391 | ||
18cf951a | 1392 | if (vfp_sreg_is_scalar(vm)) { |
90287e22 PM |
1393 | /* mixed scalar/vector */ |
1394 | delta_m = 0; | |
1395 | } else { | |
1396 | /* vector */ | |
1397 | delta_m = delta_d; | |
1398 | } | |
1399 | } | |
1400 | } | |
1401 | ||
1402 | f0 = tcg_temp_new_i32(); | |
1403 | fd = tcg_temp_new_i32(); | |
1404 | ||
1405 | neon_load_reg32(f0, vm); | |
1406 | ||
1407 | for (;;) { | |
1408 | fn(fd, f0); | |
1409 | neon_store_reg32(fd, vd); | |
1410 | ||
1411 | if (veclen == 0) { | |
1412 | break; | |
1413 | } | |
1414 | ||
1415 | if (delta_m == 0) { | |
1416 | /* single source one-many */ | |
1417 | while (veclen--) { | |
18cf951a | 1418 | vd = vfp_advance_sreg(vd, delta_d); |
90287e22 PM |
1419 | neon_store_reg32(fd, vd); |
1420 | } | |
1421 | break; | |
1422 | } | |
1423 | ||
1424 | /* Set up the operands for the next iteration */ | |
1425 | veclen--; | |
18cf951a PM |
1426 | vd = vfp_advance_sreg(vd, delta_d); |
1427 | vm = vfp_advance_sreg(vm, delta_m); | |
90287e22 PM |
1428 | neon_load_reg32(f0, vm); |
1429 | } | |
1430 | ||
1431 | tcg_temp_free_i32(f0); | |
1432 | tcg_temp_free_i32(fd); | |
1433 | ||
1434 | return true; | |
1435 | } | |
1436 | ||
1437 | static bool do_vfp_2op_dp(DisasContext *s, VFPGen2OpDPFn *fn, int vd, int vm) | |
1438 | { | |
1439 | uint32_t delta_m = 0; | |
1440 | uint32_t delta_d = 0; | |
90287e22 PM |
1441 | int veclen = s->vec_len; |
1442 | TCGv_i64 f0, fd; | |
1443 | ||
1444 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
1445 | if (!dc_isar_feature(aa32_fp_d32, s) && ((vd | vm) & 0x10)) { | |
1446 | return false; | |
1447 | } | |
1448 | ||
1449 | if (!dc_isar_feature(aa32_fpshvec, s) && | |
1450 | (veclen != 0 || s->vec_stride != 0)) { | |
1451 | return false; | |
1452 | } | |
1453 | ||
1454 | if (!vfp_access_check(s)) { | |
1455 | return true; | |
1456 | } | |
1457 | ||
1458 | if (veclen > 0) { | |
90287e22 | 1459 | /* Figure out what type of vector operation this is. */ |
18cf951a | 1460 | if (vfp_dreg_is_scalar(vd)) { |
90287e22 PM |
1461 | /* scalar */ |
1462 | veclen = 0; | |
1463 | } else { | |
1464 | delta_d = (s->vec_stride >> 1) + 1; | |
1465 | ||
18cf951a | 1466 | if (vfp_dreg_is_scalar(vm)) { |
90287e22 PM |
1467 | /* mixed scalar/vector */ |
1468 | delta_m = 0; | |
1469 | } else { | |
1470 | /* vector */ | |
1471 | delta_m = delta_d; | |
1472 | } | |
1473 | } | |
1474 | } | |
1475 | ||
1476 | f0 = tcg_temp_new_i64(); | |
1477 | fd = tcg_temp_new_i64(); | |
1478 | ||
1479 | neon_load_reg64(f0, vm); | |
1480 | ||
1481 | for (;;) { | |
1482 | fn(fd, f0); | |
1483 | neon_store_reg64(fd, vd); | |
1484 | ||
1485 | if (veclen == 0) { | |
1486 | break; | |
1487 | } | |
1488 | ||
1489 | if (delta_m == 0) { | |
1490 | /* single source one-many */ | |
1491 | while (veclen--) { | |
18cf951a | 1492 | vd = vfp_advance_dreg(vd, delta_d); |
90287e22 PM |
1493 | neon_store_reg64(fd, vd); |
1494 | } | |
1495 | break; | |
1496 | } | |
1497 | ||
1498 | /* Set up the operands for the next iteration */ | |
1499 | veclen--; | |
18cf951a PM |
1500 | vd = vfp_advance_dreg(vd, delta_d); |
1501 | vd = vfp_advance_dreg(vm, delta_m); | |
90287e22 PM |
1502 | neon_load_reg64(f0, vm); |
1503 | } | |
1504 | ||
1505 | tcg_temp_free_i64(f0); | |
1506 | tcg_temp_free_i64(fd); | |
1507 | ||
1508 | return true; | |
1509 | } | |
1510 | ||
266bd25c PM |
1511 | static void gen_VMLA_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst) |
1512 | { | |
1513 | /* Note that order of inputs to the add matters for NaNs */ | |
1514 | TCGv_i32 tmp = tcg_temp_new_i32(); | |
1515 | ||
1516 | gen_helper_vfp_muls(tmp, vn, vm, fpst); | |
1517 | gen_helper_vfp_adds(vd, vd, tmp, fpst); | |
1518 | tcg_temp_free_i32(tmp); | |
1519 | } | |
1520 | ||
1521 | static bool trans_VMLA_sp(DisasContext *s, arg_VMLA_sp *a) | |
1522 | { | |
1523 | return do_vfp_3op_sp(s, gen_VMLA_sp, a->vd, a->vn, a->vm, true); | |
1524 | } | |
1525 | ||
1526 | static void gen_VMLA_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst) | |
1527 | { | |
1528 | /* Note that order of inputs to the add matters for NaNs */ | |
1529 | TCGv_i64 tmp = tcg_temp_new_i64(); | |
1530 | ||
1531 | gen_helper_vfp_muld(tmp, vn, vm, fpst); | |
1532 | gen_helper_vfp_addd(vd, vd, tmp, fpst); | |
1533 | tcg_temp_free_i64(tmp); | |
1534 | } | |
1535 | ||
1536 | static bool trans_VMLA_dp(DisasContext *s, arg_VMLA_sp *a) | |
1537 | { | |
1538 | return do_vfp_3op_dp(s, gen_VMLA_dp, a->vd, a->vn, a->vm, true); | |
1539 | } | |
e7258280 PM |
1540 | |
1541 | static void gen_VMLS_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst) | |
1542 | { | |
1543 | /* | |
1544 | * VMLS: vd = vd + -(vn * vm) | |
1545 | * Note that order of inputs to the add matters for NaNs. | |
1546 | */ | |
1547 | TCGv_i32 tmp = tcg_temp_new_i32(); | |
1548 | ||
1549 | gen_helper_vfp_muls(tmp, vn, vm, fpst); | |
1550 | gen_helper_vfp_negs(tmp, tmp); | |
1551 | gen_helper_vfp_adds(vd, vd, tmp, fpst); | |
1552 | tcg_temp_free_i32(tmp); | |
1553 | } | |
1554 | ||
1555 | static bool trans_VMLS_sp(DisasContext *s, arg_VMLS_sp *a) | |
1556 | { | |
1557 | return do_vfp_3op_sp(s, gen_VMLS_sp, a->vd, a->vn, a->vm, true); | |
1558 | } | |
1559 | ||
1560 | static void gen_VMLS_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst) | |
1561 | { | |
1562 | /* | |
1563 | * VMLS: vd = vd + -(vn * vm) | |
1564 | * Note that order of inputs to the add matters for NaNs. | |
1565 | */ | |
1566 | TCGv_i64 tmp = tcg_temp_new_i64(); | |
1567 | ||
1568 | gen_helper_vfp_muld(tmp, vn, vm, fpst); | |
1569 | gen_helper_vfp_negd(tmp, tmp); | |
1570 | gen_helper_vfp_addd(vd, vd, tmp, fpst); | |
1571 | tcg_temp_free_i64(tmp); | |
1572 | } | |
1573 | ||
1574 | static bool trans_VMLS_dp(DisasContext *s, arg_VMLS_sp *a) | |
1575 | { | |
1576 | return do_vfp_3op_dp(s, gen_VMLS_dp, a->vd, a->vn, a->vm, true); | |
1577 | } | |
c54a416c PM |
1578 | |
1579 | static void gen_VNMLS_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst) | |
1580 | { | |
1581 | /* | |
1582 | * VNMLS: -fd + (fn * fm) | |
1583 | * Note that it isn't valid to replace (-A + B) with (B - A) or similar | |
1584 | * plausible looking simplifications because this will give wrong results | |
1585 | * for NaNs. | |
1586 | */ | |
1587 | TCGv_i32 tmp = tcg_temp_new_i32(); | |
1588 | ||
1589 | gen_helper_vfp_muls(tmp, vn, vm, fpst); | |
1590 | gen_helper_vfp_negs(vd, vd); | |
1591 | gen_helper_vfp_adds(vd, vd, tmp, fpst); | |
1592 | tcg_temp_free_i32(tmp); | |
1593 | } | |
1594 | ||
1595 | static bool trans_VNMLS_sp(DisasContext *s, arg_VNMLS_sp *a) | |
1596 | { | |
1597 | return do_vfp_3op_sp(s, gen_VNMLS_sp, a->vd, a->vn, a->vm, true); | |
1598 | } | |
1599 | ||
1600 | static void gen_VNMLS_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst) | |
1601 | { | |
1602 | /* | |
1603 | * VNMLS: -fd + (fn * fm) | |
1604 | * Note that it isn't valid to replace (-A + B) with (B - A) or similar | |
1605 | * plausible looking simplifications because this will give wrong results | |
1606 | * for NaNs. | |
1607 | */ | |
1608 | TCGv_i64 tmp = tcg_temp_new_i64(); | |
1609 | ||
1610 | gen_helper_vfp_muld(tmp, vn, vm, fpst); | |
1611 | gen_helper_vfp_negd(vd, vd); | |
1612 | gen_helper_vfp_addd(vd, vd, tmp, fpst); | |
1613 | tcg_temp_free_i64(tmp); | |
1614 | } | |
1615 | ||
1616 | static bool trans_VNMLS_dp(DisasContext *s, arg_VNMLS_sp *a) | |
1617 | { | |
1618 | return do_vfp_3op_dp(s, gen_VNMLS_dp, a->vd, a->vn, a->vm, true); | |
1619 | } | |
8a483533 PM |
1620 | |
1621 | static void gen_VNMLA_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst) | |
1622 | { | |
1623 | /* VNMLA: -fd + -(fn * fm) */ | |
1624 | TCGv_i32 tmp = tcg_temp_new_i32(); | |
1625 | ||
1626 | gen_helper_vfp_muls(tmp, vn, vm, fpst); | |
1627 | gen_helper_vfp_negs(tmp, tmp); | |
1628 | gen_helper_vfp_negs(vd, vd); | |
1629 | gen_helper_vfp_adds(vd, vd, tmp, fpst); | |
1630 | tcg_temp_free_i32(tmp); | |
1631 | } | |
1632 | ||
1633 | static bool trans_VNMLA_sp(DisasContext *s, arg_VNMLA_sp *a) | |
1634 | { | |
1635 | return do_vfp_3op_sp(s, gen_VNMLA_sp, a->vd, a->vn, a->vm, true); | |
1636 | } | |
1637 | ||
1638 | static void gen_VNMLA_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst) | |
1639 | { | |
1640 | /* VNMLA: -fd + (fn * fm) */ | |
1641 | TCGv_i64 tmp = tcg_temp_new_i64(); | |
1642 | ||
1643 | gen_helper_vfp_muld(tmp, vn, vm, fpst); | |
1644 | gen_helper_vfp_negd(tmp, tmp); | |
1645 | gen_helper_vfp_negd(vd, vd); | |
1646 | gen_helper_vfp_addd(vd, vd, tmp, fpst); | |
1647 | tcg_temp_free_i64(tmp); | |
1648 | } | |
1649 | ||
1650 | static bool trans_VNMLA_dp(DisasContext *s, arg_VNMLA_sp *a) | |
1651 | { | |
1652 | return do_vfp_3op_dp(s, gen_VNMLA_dp, a->vd, a->vn, a->vm, true); | |
1653 | } | |
88c5188c PM |
1654 | |
1655 | static bool trans_VMUL_sp(DisasContext *s, arg_VMUL_sp *a) | |
1656 | { | |
1657 | return do_vfp_3op_sp(s, gen_helper_vfp_muls, a->vd, a->vn, a->vm, false); | |
1658 | } | |
1659 | ||
1660 | static bool trans_VMUL_dp(DisasContext *s, arg_VMUL_sp *a) | |
1661 | { | |
1662 | return do_vfp_3op_dp(s, gen_helper_vfp_muld, a->vd, a->vn, a->vm, false); | |
1663 | } | |
43c4be12 PM |
1664 | |
1665 | static void gen_VNMUL_sp(TCGv_i32 vd, TCGv_i32 vn, TCGv_i32 vm, TCGv_ptr fpst) | |
1666 | { | |
1667 | /* VNMUL: -(fn * fm) */ | |
1668 | gen_helper_vfp_muls(vd, vn, vm, fpst); | |
1669 | gen_helper_vfp_negs(vd, vd); | |
1670 | } | |
1671 | ||
1672 | static bool trans_VNMUL_sp(DisasContext *s, arg_VNMUL_sp *a) | |
1673 | { | |
1674 | return do_vfp_3op_sp(s, gen_VNMUL_sp, a->vd, a->vn, a->vm, false); | |
1675 | } | |
1676 | ||
1677 | static void gen_VNMUL_dp(TCGv_i64 vd, TCGv_i64 vn, TCGv_i64 vm, TCGv_ptr fpst) | |
1678 | { | |
1679 | /* VNMUL: -(fn * fm) */ | |
1680 | gen_helper_vfp_muld(vd, vn, vm, fpst); | |
1681 | gen_helper_vfp_negd(vd, vd); | |
1682 | } | |
1683 | ||
1684 | static bool trans_VNMUL_dp(DisasContext *s, arg_VNMUL_sp *a) | |
1685 | { | |
1686 | return do_vfp_3op_dp(s, gen_VNMUL_dp, a->vd, a->vn, a->vm, false); | |
1687 | } | |
ce28b303 PM |
1688 | |
1689 | static bool trans_VADD_sp(DisasContext *s, arg_VADD_sp *a) | |
1690 | { | |
1691 | return do_vfp_3op_sp(s, gen_helper_vfp_adds, a->vd, a->vn, a->vm, false); | |
1692 | } | |
1693 | ||
1694 | static bool trans_VADD_dp(DisasContext *s, arg_VADD_sp *a) | |
1695 | { | |
1696 | return do_vfp_3op_dp(s, gen_helper_vfp_addd, a->vd, a->vn, a->vm, false); | |
1697 | } | |
8fec9a11 PM |
1698 | |
1699 | static bool trans_VSUB_sp(DisasContext *s, arg_VSUB_sp *a) | |
1700 | { | |
1701 | return do_vfp_3op_sp(s, gen_helper_vfp_subs, a->vd, a->vn, a->vm, false); | |
1702 | } | |
1703 | ||
1704 | static bool trans_VSUB_dp(DisasContext *s, arg_VSUB_sp *a) | |
1705 | { | |
1706 | return do_vfp_3op_dp(s, gen_helper_vfp_subd, a->vd, a->vn, a->vm, false); | |
1707 | } | |
519ee7ae PM |
1708 | |
1709 | static bool trans_VDIV_sp(DisasContext *s, arg_VDIV_sp *a) | |
1710 | { | |
1711 | return do_vfp_3op_sp(s, gen_helper_vfp_divs, a->vd, a->vn, a->vm, false); | |
1712 | } | |
1713 | ||
1714 | static bool trans_VDIV_dp(DisasContext *s, arg_VDIV_sp *a) | |
1715 | { | |
1716 | return do_vfp_3op_dp(s, gen_helper_vfp_divd, a->vd, a->vn, a->vm, false); | |
1717 | } | |
d4893b01 PM |
1718 | |
1719 | static bool trans_VFM_sp(DisasContext *s, arg_VFM_sp *a) | |
1720 | { | |
1721 | /* | |
1722 | * VFNMA : fd = muladd(-fd, fn, fm) | |
1723 | * VFNMS : fd = muladd(-fd, -fn, fm) | |
1724 | * VFMA : fd = muladd( fd, fn, fm) | |
1725 | * VFMS : fd = muladd( fd, -fn, fm) | |
1726 | * | |
1727 | * These are fused multiply-add, and must be done as one floating | |
1728 | * point operation with no rounding between the multiplication and | |
1729 | * addition steps. NB that doing the negations here as separate | |
1730 | * steps is correct : an input NaN should come out with its sign | |
1731 | * bit flipped if it is a negated-input. | |
1732 | */ | |
1733 | TCGv_ptr fpst; | |
1734 | TCGv_i32 vn, vm, vd; | |
1735 | ||
1736 | /* | |
1737 | * Present in VFPv4 only. | |
1738 | * In v7A, UNPREDICTABLE with non-zero vector length/stride; from | |
1739 | * v8A, must UNDEF. We choose to UNDEF for both v7A and v8A. | |
1740 | */ | |
1741 | if (!arm_dc_feature(s, ARM_FEATURE_VFP4) || | |
1742 | (s->vec_len != 0 || s->vec_stride != 0)) { | |
1743 | return false; | |
1744 | } | |
1745 | ||
1746 | if (!vfp_access_check(s)) { | |
1747 | return true; | |
1748 | } | |
1749 | ||
1750 | vn = tcg_temp_new_i32(); | |
1751 | vm = tcg_temp_new_i32(); | |
1752 | vd = tcg_temp_new_i32(); | |
1753 | ||
1754 | neon_load_reg32(vn, a->vn); | |
1755 | neon_load_reg32(vm, a->vm); | |
1756 | if (a->o2) { | |
1757 | /* VFNMS, VFMS */ | |
1758 | gen_helper_vfp_negs(vn, vn); | |
1759 | } | |
1760 | neon_load_reg32(vd, a->vd); | |
1761 | if (a->o1 & 1) { | |
1762 | /* VFNMA, VFNMS */ | |
1763 | gen_helper_vfp_negs(vd, vd); | |
1764 | } | |
1765 | fpst = get_fpstatus_ptr(0); | |
1766 | gen_helper_vfp_muladds(vd, vn, vm, vd, fpst); | |
1767 | neon_store_reg32(vd, a->vd); | |
1768 | ||
1769 | tcg_temp_free_ptr(fpst); | |
1770 | tcg_temp_free_i32(vn); | |
1771 | tcg_temp_free_i32(vm); | |
1772 | tcg_temp_free_i32(vd); | |
1773 | ||
1774 | return true; | |
1775 | } | |
1776 | ||
1777 | static bool trans_VFM_dp(DisasContext *s, arg_VFM_sp *a) | |
1778 | { | |
1779 | /* | |
1780 | * VFNMA : fd = muladd(-fd, fn, fm) | |
1781 | * VFNMS : fd = muladd(-fd, -fn, fm) | |
1782 | * VFMA : fd = muladd( fd, fn, fm) | |
1783 | * VFMS : fd = muladd( fd, -fn, fm) | |
1784 | * | |
1785 | * These are fused multiply-add, and must be done as one floating | |
1786 | * point operation with no rounding between the multiplication and | |
1787 | * addition steps. NB that doing the negations here as separate | |
1788 | * steps is correct : an input NaN should come out with its sign | |
1789 | * bit flipped if it is a negated-input. | |
1790 | */ | |
1791 | TCGv_ptr fpst; | |
1792 | TCGv_i64 vn, vm, vd; | |
1793 | ||
1794 | /* | |
1795 | * Present in VFPv4 only. | |
1796 | * In v7A, UNPREDICTABLE with non-zero vector length/stride; from | |
1797 | * v8A, must UNDEF. We choose to UNDEF for both v7A and v8A. | |
1798 | */ | |
1799 | if (!arm_dc_feature(s, ARM_FEATURE_VFP4) || | |
1800 | (s->vec_len != 0 || s->vec_stride != 0)) { | |
1801 | return false; | |
1802 | } | |
1803 | ||
1804 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
1805 | if (!dc_isar_feature(aa32_fp_d32, s) && ((a->vd | a->vn | a->vm) & 0x10)) { | |
1806 | return false; | |
1807 | } | |
1808 | ||
1809 | if (!vfp_access_check(s)) { | |
1810 | return true; | |
1811 | } | |
1812 | ||
1813 | vn = tcg_temp_new_i64(); | |
1814 | vm = tcg_temp_new_i64(); | |
1815 | vd = tcg_temp_new_i64(); | |
1816 | ||
1817 | neon_load_reg64(vn, a->vn); | |
1818 | neon_load_reg64(vm, a->vm); | |
1819 | if (a->o2) { | |
1820 | /* VFNMS, VFMS */ | |
1821 | gen_helper_vfp_negd(vn, vn); | |
1822 | } | |
1823 | neon_load_reg64(vd, a->vd); | |
1824 | if (a->o1 & 1) { | |
1825 | /* VFNMA, VFNMS */ | |
1826 | gen_helper_vfp_negd(vd, vd); | |
1827 | } | |
1828 | fpst = get_fpstatus_ptr(0); | |
1829 | gen_helper_vfp_muladdd(vd, vn, vm, vd, fpst); | |
1830 | neon_store_reg64(vd, a->vd); | |
1831 | ||
1832 | tcg_temp_free_ptr(fpst); | |
1833 | tcg_temp_free_i64(vn); | |
1834 | tcg_temp_free_i64(vm); | |
1835 | tcg_temp_free_i64(vd); | |
1836 | ||
1837 | return true; | |
1838 | } | |
b518c753 PM |
1839 | |
1840 | static bool trans_VMOV_imm_sp(DisasContext *s, arg_VMOV_imm_sp *a) | |
1841 | { | |
1842 | uint32_t delta_d = 0; | |
b518c753 PM |
1843 | int veclen = s->vec_len; |
1844 | TCGv_i32 fd; | |
9bee50b4 | 1845 | uint32_t vd; |
b518c753 PM |
1846 | |
1847 | vd = a->vd; | |
1848 | ||
1849 | if (!dc_isar_feature(aa32_fpshvec, s) && | |
1850 | (veclen != 0 || s->vec_stride != 0)) { | |
1851 | return false; | |
1852 | } | |
1853 | ||
1854 | if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) { | |
1855 | return false; | |
1856 | } | |
1857 | ||
1858 | if (!vfp_access_check(s)) { | |
1859 | return true; | |
1860 | } | |
1861 | ||
1862 | if (veclen > 0) { | |
b518c753 | 1863 | /* Figure out what type of vector operation this is. */ |
18cf951a | 1864 | if (vfp_sreg_is_scalar(vd)) { |
b518c753 PM |
1865 | /* scalar */ |
1866 | veclen = 0; | |
1867 | } else { | |
1868 | delta_d = s->vec_stride + 1; | |
1869 | } | |
1870 | } | |
1871 | ||
9bee50b4 | 1872 | fd = tcg_const_i32(vfp_expand_imm(MO_32, a->imm)); |
b518c753 PM |
1873 | |
1874 | for (;;) { | |
1875 | neon_store_reg32(fd, vd); | |
1876 | ||
1877 | if (veclen == 0) { | |
1878 | break; | |
1879 | } | |
1880 | ||
1881 | /* Set up the operands for the next iteration */ | |
1882 | veclen--; | |
18cf951a | 1883 | vd = vfp_advance_sreg(vd, delta_d); |
b518c753 PM |
1884 | } |
1885 | ||
1886 | tcg_temp_free_i32(fd); | |
1887 | return true; | |
1888 | } | |
1889 | ||
1890 | static bool trans_VMOV_imm_dp(DisasContext *s, arg_VMOV_imm_dp *a) | |
1891 | { | |
1892 | uint32_t delta_d = 0; | |
b518c753 PM |
1893 | int veclen = s->vec_len; |
1894 | TCGv_i64 fd; | |
9bee50b4 | 1895 | uint32_t vd; |
b518c753 PM |
1896 | |
1897 | vd = a->vd; | |
1898 | ||
1899 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
1900 | if (!dc_isar_feature(aa32_fp_d32, s) && (vd & 0x10)) { | |
1901 | return false; | |
1902 | } | |
1903 | ||
1904 | if (!dc_isar_feature(aa32_fpshvec, s) && | |
1905 | (veclen != 0 || s->vec_stride != 0)) { | |
1906 | return false; | |
1907 | } | |
1908 | ||
1909 | if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) { | |
1910 | return false; | |
1911 | } | |
1912 | ||
1913 | if (!vfp_access_check(s)) { | |
1914 | return true; | |
1915 | } | |
1916 | ||
1917 | if (veclen > 0) { | |
b518c753 | 1918 | /* Figure out what type of vector operation this is. */ |
18cf951a | 1919 | if (vfp_dreg_is_scalar(vd)) { |
b518c753 PM |
1920 | /* scalar */ |
1921 | veclen = 0; | |
1922 | } else { | |
1923 | delta_d = (s->vec_stride >> 1) + 1; | |
1924 | } | |
1925 | } | |
1926 | ||
9bee50b4 | 1927 | fd = tcg_const_i64(vfp_expand_imm(MO_64, a->imm)); |
b518c753 PM |
1928 | |
1929 | for (;;) { | |
1930 | neon_store_reg64(fd, vd); | |
1931 | ||
1932 | if (veclen == 0) { | |
1933 | break; | |
1934 | } | |
1935 | ||
1936 | /* Set up the operands for the next iteration */ | |
1937 | veclen--; | |
18cf951a | 1938 | vfp_advance_dreg(vd, delta_d); |
b518c753 PM |
1939 | } |
1940 | ||
1941 | tcg_temp_free_i64(fd); | |
1942 | return true; | |
1943 | } | |
90287e22 | 1944 | |
17552b97 PM |
1945 | static bool trans_VMOV_reg_sp(DisasContext *s, arg_VMOV_reg_sp *a) |
1946 | { | |
1947 | return do_vfp_2op_sp(s, tcg_gen_mov_i32, a->vd, a->vm); | |
1948 | } | |
1949 | ||
1950 | static bool trans_VMOV_reg_dp(DisasContext *s, arg_VMOV_reg_dp *a) | |
1951 | { | |
1952 | return do_vfp_2op_dp(s, tcg_gen_mov_i64, a->vd, a->vm); | |
1953 | } | |
1954 | ||
90287e22 PM |
1955 | static bool trans_VABS_sp(DisasContext *s, arg_VABS_sp *a) |
1956 | { | |
1957 | return do_vfp_2op_sp(s, gen_helper_vfp_abss, a->vd, a->vm); | |
1958 | } | |
1959 | ||
1960 | static bool trans_VABS_dp(DisasContext *s, arg_VABS_dp *a) | |
1961 | { | |
1962 | return do_vfp_2op_dp(s, gen_helper_vfp_absd, a->vd, a->vm); | |
1963 | } | |
1882651a PM |
1964 | |
1965 | static bool trans_VNEG_sp(DisasContext *s, arg_VNEG_sp *a) | |
1966 | { | |
1967 | return do_vfp_2op_sp(s, gen_helper_vfp_negs, a->vd, a->vm); | |
1968 | } | |
1969 | ||
1970 | static bool trans_VNEG_dp(DisasContext *s, arg_VNEG_dp *a) | |
1971 | { | |
1972 | return do_vfp_2op_dp(s, gen_helper_vfp_negd, a->vd, a->vm); | |
1973 | } | |
b8474540 PM |
1974 | |
1975 | static void gen_VSQRT_sp(TCGv_i32 vd, TCGv_i32 vm) | |
1976 | { | |
1977 | gen_helper_vfp_sqrts(vd, vm, cpu_env); | |
1978 | } | |
1979 | ||
1980 | static bool trans_VSQRT_sp(DisasContext *s, arg_VSQRT_sp *a) | |
1981 | { | |
1982 | return do_vfp_2op_sp(s, gen_VSQRT_sp, a->vd, a->vm); | |
1983 | } | |
1984 | ||
1985 | static void gen_VSQRT_dp(TCGv_i64 vd, TCGv_i64 vm) | |
1986 | { | |
1987 | gen_helper_vfp_sqrtd(vd, vm, cpu_env); | |
1988 | } | |
1989 | ||
1990 | static bool trans_VSQRT_dp(DisasContext *s, arg_VSQRT_dp *a) | |
1991 | { | |
1992 | return do_vfp_2op_dp(s, gen_VSQRT_dp, a->vd, a->vm); | |
1993 | } | |
386bba23 PM |
1994 | |
1995 | static bool trans_VCMP_sp(DisasContext *s, arg_VCMP_sp *a) | |
1996 | { | |
1997 | TCGv_i32 vd, vm; | |
1998 | ||
1999 | /* Vm/M bits must be zero for the Z variant */ | |
2000 | if (a->z && a->vm != 0) { | |
2001 | return false; | |
2002 | } | |
2003 | ||
2004 | if (!vfp_access_check(s)) { | |
2005 | return true; | |
2006 | } | |
2007 | ||
2008 | vd = tcg_temp_new_i32(); | |
2009 | vm = tcg_temp_new_i32(); | |
2010 | ||
2011 | neon_load_reg32(vd, a->vd); | |
2012 | if (a->z) { | |
2013 | tcg_gen_movi_i32(vm, 0); | |
2014 | } else { | |
2015 | neon_load_reg32(vm, a->vm); | |
2016 | } | |
2017 | ||
2018 | if (a->e) { | |
2019 | gen_helper_vfp_cmpes(vd, vm, cpu_env); | |
2020 | } else { | |
2021 | gen_helper_vfp_cmps(vd, vm, cpu_env); | |
2022 | } | |
2023 | ||
2024 | tcg_temp_free_i32(vd); | |
2025 | tcg_temp_free_i32(vm); | |
2026 | ||
2027 | return true; | |
2028 | } | |
2029 | ||
2030 | static bool trans_VCMP_dp(DisasContext *s, arg_VCMP_dp *a) | |
2031 | { | |
2032 | TCGv_i64 vd, vm; | |
2033 | ||
2034 | /* Vm/M bits must be zero for the Z variant */ | |
2035 | if (a->z && a->vm != 0) { | |
2036 | return false; | |
2037 | } | |
2038 | ||
2039 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2040 | if (!dc_isar_feature(aa32_fp_d32, s) && ((a->vd | a->vm) & 0x10)) { | |
2041 | return false; | |
2042 | } | |
2043 | ||
2044 | if (!vfp_access_check(s)) { | |
2045 | return true; | |
2046 | } | |
2047 | ||
2048 | vd = tcg_temp_new_i64(); | |
2049 | vm = tcg_temp_new_i64(); | |
2050 | ||
2051 | neon_load_reg64(vd, a->vd); | |
2052 | if (a->z) { | |
2053 | tcg_gen_movi_i64(vm, 0); | |
2054 | } else { | |
2055 | neon_load_reg64(vm, a->vm); | |
2056 | } | |
2057 | ||
2058 | if (a->e) { | |
2059 | gen_helper_vfp_cmped(vd, vm, cpu_env); | |
2060 | } else { | |
2061 | gen_helper_vfp_cmpd(vd, vm, cpu_env); | |
2062 | } | |
2063 | ||
2064 | tcg_temp_free_i64(vd); | |
2065 | tcg_temp_free_i64(vm); | |
2066 | ||
2067 | return true; | |
2068 | } | |
b623d803 PM |
2069 | |
2070 | static bool trans_VCVT_f32_f16(DisasContext *s, arg_VCVT_f32_f16 *a) | |
2071 | { | |
2072 | TCGv_ptr fpst; | |
2073 | TCGv_i32 ahp_mode; | |
2074 | TCGv_i32 tmp; | |
2075 | ||
2076 | if (!dc_isar_feature(aa32_fp16_spconv, s)) { | |
2077 | return false; | |
2078 | } | |
2079 | ||
2080 | if (!vfp_access_check(s)) { | |
2081 | return true; | |
2082 | } | |
2083 | ||
2084 | fpst = get_fpstatus_ptr(false); | |
2085 | ahp_mode = get_ahp_flag(); | |
2086 | tmp = tcg_temp_new_i32(); | |
2087 | /* The T bit tells us if we want the low or high 16 bits of Vm */ | |
2088 | tcg_gen_ld16u_i32(tmp, cpu_env, vfp_f16_offset(a->vm, a->t)); | |
2089 | gen_helper_vfp_fcvt_f16_to_f32(tmp, tmp, fpst, ahp_mode); | |
2090 | neon_store_reg32(tmp, a->vd); | |
2091 | tcg_temp_free_i32(ahp_mode); | |
2092 | tcg_temp_free_ptr(fpst); | |
2093 | tcg_temp_free_i32(tmp); | |
2094 | return true; | |
2095 | } | |
2096 | ||
2097 | static bool trans_VCVT_f64_f16(DisasContext *s, arg_VCVT_f64_f16 *a) | |
2098 | { | |
2099 | TCGv_ptr fpst; | |
2100 | TCGv_i32 ahp_mode; | |
2101 | TCGv_i32 tmp; | |
2102 | TCGv_i64 vd; | |
2103 | ||
2104 | if (!dc_isar_feature(aa32_fp16_dpconv, s)) { | |
2105 | return false; | |
2106 | } | |
2107 | ||
2108 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2109 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vd & 0x10)) { | |
2110 | return false; | |
2111 | } | |
2112 | ||
2113 | if (!vfp_access_check(s)) { | |
2114 | return true; | |
2115 | } | |
2116 | ||
2117 | fpst = get_fpstatus_ptr(false); | |
2118 | ahp_mode = get_ahp_flag(); | |
2119 | tmp = tcg_temp_new_i32(); | |
2120 | /* The T bit tells us if we want the low or high 16 bits of Vm */ | |
2121 | tcg_gen_ld16u_i32(tmp, cpu_env, vfp_f16_offset(a->vm, a->t)); | |
2122 | vd = tcg_temp_new_i64(); | |
2123 | gen_helper_vfp_fcvt_f16_to_f64(vd, tmp, fpst, ahp_mode); | |
2124 | neon_store_reg64(vd, a->vd); | |
2125 | tcg_temp_free_i32(ahp_mode); | |
2126 | tcg_temp_free_ptr(fpst); | |
2127 | tcg_temp_free_i32(tmp); | |
2128 | tcg_temp_free_i64(vd); | |
2129 | return true; | |
2130 | } | |
cdfd14e8 PM |
2131 | |
2132 | static bool trans_VCVT_f16_f32(DisasContext *s, arg_VCVT_f16_f32 *a) | |
2133 | { | |
2134 | TCGv_ptr fpst; | |
2135 | TCGv_i32 ahp_mode; | |
2136 | TCGv_i32 tmp; | |
2137 | ||
2138 | if (!dc_isar_feature(aa32_fp16_spconv, s)) { | |
2139 | return false; | |
2140 | } | |
2141 | ||
2142 | if (!vfp_access_check(s)) { | |
2143 | return true; | |
2144 | } | |
2145 | ||
2146 | fpst = get_fpstatus_ptr(false); | |
2147 | ahp_mode = get_ahp_flag(); | |
2148 | tmp = tcg_temp_new_i32(); | |
2149 | ||
2150 | neon_load_reg32(tmp, a->vm); | |
2151 | gen_helper_vfp_fcvt_f32_to_f16(tmp, tmp, fpst, ahp_mode); | |
2152 | tcg_gen_st16_i32(tmp, cpu_env, vfp_f16_offset(a->vd, a->t)); | |
2153 | tcg_temp_free_i32(ahp_mode); | |
2154 | tcg_temp_free_ptr(fpst); | |
2155 | tcg_temp_free_i32(tmp); | |
2156 | return true; | |
2157 | } | |
2158 | ||
2159 | static bool trans_VCVT_f16_f64(DisasContext *s, arg_VCVT_f16_f64 *a) | |
2160 | { | |
2161 | TCGv_ptr fpst; | |
2162 | TCGv_i32 ahp_mode; | |
2163 | TCGv_i32 tmp; | |
2164 | TCGv_i64 vm; | |
2165 | ||
2166 | if (!dc_isar_feature(aa32_fp16_dpconv, s)) { | |
2167 | return false; | |
2168 | } | |
2169 | ||
2170 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2171 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vm & 0x10)) { | |
2172 | return false; | |
2173 | } | |
2174 | ||
2175 | if (!vfp_access_check(s)) { | |
2176 | return true; | |
2177 | } | |
2178 | ||
2179 | fpst = get_fpstatus_ptr(false); | |
2180 | ahp_mode = get_ahp_flag(); | |
2181 | tmp = tcg_temp_new_i32(); | |
2182 | vm = tcg_temp_new_i64(); | |
2183 | ||
2184 | neon_load_reg64(vm, a->vm); | |
2185 | gen_helper_vfp_fcvt_f64_to_f16(tmp, vm, fpst, ahp_mode); | |
2186 | tcg_temp_free_i64(vm); | |
2187 | tcg_gen_st16_i32(tmp, cpu_env, vfp_f16_offset(a->vd, a->t)); | |
2188 | tcg_temp_free_i32(ahp_mode); | |
2189 | tcg_temp_free_ptr(fpst); | |
2190 | tcg_temp_free_i32(tmp); | |
2191 | return true; | |
2192 | } | |
e25155f5 PM |
2193 | |
2194 | static bool trans_VRINTR_sp(DisasContext *s, arg_VRINTR_sp *a) | |
2195 | { | |
2196 | TCGv_ptr fpst; | |
2197 | TCGv_i32 tmp; | |
2198 | ||
2199 | if (!dc_isar_feature(aa32_vrint, s)) { | |
2200 | return false; | |
2201 | } | |
2202 | ||
2203 | if (!vfp_access_check(s)) { | |
2204 | return true; | |
2205 | } | |
2206 | ||
2207 | tmp = tcg_temp_new_i32(); | |
2208 | neon_load_reg32(tmp, a->vm); | |
2209 | fpst = get_fpstatus_ptr(false); | |
2210 | gen_helper_rints(tmp, tmp, fpst); | |
2211 | neon_store_reg32(tmp, a->vd); | |
2212 | tcg_temp_free_ptr(fpst); | |
2213 | tcg_temp_free_i32(tmp); | |
2214 | return true; | |
2215 | } | |
2216 | ||
2217 | static bool trans_VRINTR_dp(DisasContext *s, arg_VRINTR_sp *a) | |
2218 | { | |
2219 | TCGv_ptr fpst; | |
2220 | TCGv_i64 tmp; | |
2221 | ||
2222 | if (!dc_isar_feature(aa32_vrint, s)) { | |
2223 | return false; | |
2224 | } | |
2225 | ||
2226 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2227 | if (!dc_isar_feature(aa32_fp_d32, s) && ((a->vd | a->vm) & 0x10)) { | |
2228 | return false; | |
2229 | } | |
2230 | ||
2231 | if (!vfp_access_check(s)) { | |
2232 | return true; | |
2233 | } | |
2234 | ||
2235 | tmp = tcg_temp_new_i64(); | |
2236 | neon_load_reg64(tmp, a->vm); | |
2237 | fpst = get_fpstatus_ptr(false); | |
2238 | gen_helper_rintd(tmp, tmp, fpst); | |
2239 | neon_store_reg64(tmp, a->vd); | |
2240 | tcg_temp_free_ptr(fpst); | |
2241 | tcg_temp_free_i64(tmp); | |
2242 | return true; | |
2243 | } | |
2244 | ||
2245 | static bool trans_VRINTZ_sp(DisasContext *s, arg_VRINTZ_sp *a) | |
2246 | { | |
2247 | TCGv_ptr fpst; | |
2248 | TCGv_i32 tmp; | |
2249 | TCGv_i32 tcg_rmode; | |
2250 | ||
2251 | if (!dc_isar_feature(aa32_vrint, s)) { | |
2252 | return false; | |
2253 | } | |
2254 | ||
2255 | if (!vfp_access_check(s)) { | |
2256 | return true; | |
2257 | } | |
2258 | ||
2259 | tmp = tcg_temp_new_i32(); | |
2260 | neon_load_reg32(tmp, a->vm); | |
2261 | fpst = get_fpstatus_ptr(false); | |
2262 | tcg_rmode = tcg_const_i32(float_round_to_zero); | |
2263 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
2264 | gen_helper_rints(tmp, tmp, fpst); | |
2265 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
2266 | neon_store_reg32(tmp, a->vd); | |
2267 | tcg_temp_free_ptr(fpst); | |
2268 | tcg_temp_free_i32(tcg_rmode); | |
2269 | tcg_temp_free_i32(tmp); | |
2270 | return true; | |
2271 | } | |
2272 | ||
2273 | static bool trans_VRINTZ_dp(DisasContext *s, arg_VRINTZ_sp *a) | |
2274 | { | |
2275 | TCGv_ptr fpst; | |
2276 | TCGv_i64 tmp; | |
2277 | TCGv_i32 tcg_rmode; | |
2278 | ||
2279 | if (!dc_isar_feature(aa32_vrint, s)) { | |
2280 | return false; | |
2281 | } | |
2282 | ||
2283 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2284 | if (!dc_isar_feature(aa32_fp_d32, s) && ((a->vd | a->vm) & 0x10)) { | |
2285 | return false; | |
2286 | } | |
2287 | ||
2288 | if (!vfp_access_check(s)) { | |
2289 | return true; | |
2290 | } | |
2291 | ||
2292 | tmp = tcg_temp_new_i64(); | |
2293 | neon_load_reg64(tmp, a->vm); | |
2294 | fpst = get_fpstatus_ptr(false); | |
2295 | tcg_rmode = tcg_const_i32(float_round_to_zero); | |
2296 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
2297 | gen_helper_rintd(tmp, tmp, fpst); | |
2298 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
2299 | neon_store_reg64(tmp, a->vd); | |
2300 | tcg_temp_free_ptr(fpst); | |
2301 | tcg_temp_free_i64(tmp); | |
2302 | tcg_temp_free_i32(tcg_rmode); | |
2303 | return true; | |
2304 | } | |
2305 | ||
2306 | static bool trans_VRINTX_sp(DisasContext *s, arg_VRINTX_sp *a) | |
2307 | { | |
2308 | TCGv_ptr fpst; | |
2309 | TCGv_i32 tmp; | |
2310 | ||
2311 | if (!dc_isar_feature(aa32_vrint, s)) { | |
2312 | return false; | |
2313 | } | |
2314 | ||
2315 | if (!vfp_access_check(s)) { | |
2316 | return true; | |
2317 | } | |
2318 | ||
2319 | tmp = tcg_temp_new_i32(); | |
2320 | neon_load_reg32(tmp, a->vm); | |
2321 | fpst = get_fpstatus_ptr(false); | |
2322 | gen_helper_rints_exact(tmp, tmp, fpst); | |
2323 | neon_store_reg32(tmp, a->vd); | |
2324 | tcg_temp_free_ptr(fpst); | |
2325 | tcg_temp_free_i32(tmp); | |
2326 | return true; | |
2327 | } | |
2328 | ||
2329 | static bool trans_VRINTX_dp(DisasContext *s, arg_VRINTX_dp *a) | |
2330 | { | |
2331 | TCGv_ptr fpst; | |
2332 | TCGv_i64 tmp; | |
2333 | ||
2334 | if (!dc_isar_feature(aa32_vrint, s)) { | |
2335 | return false; | |
2336 | } | |
2337 | ||
2338 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2339 | if (!dc_isar_feature(aa32_fp_d32, s) && ((a->vd | a->vm) & 0x10)) { | |
2340 | return false; | |
2341 | } | |
2342 | ||
2343 | if (!vfp_access_check(s)) { | |
2344 | return true; | |
2345 | } | |
2346 | ||
2347 | tmp = tcg_temp_new_i64(); | |
2348 | neon_load_reg64(tmp, a->vm); | |
2349 | fpst = get_fpstatus_ptr(false); | |
2350 | gen_helper_rintd_exact(tmp, tmp, fpst); | |
2351 | neon_store_reg64(tmp, a->vd); | |
2352 | tcg_temp_free_ptr(fpst); | |
2353 | tcg_temp_free_i64(tmp); | |
2354 | return true; | |
2355 | } | |
6ed7e49c PM |
2356 | |
2357 | static bool trans_VCVT_sp(DisasContext *s, arg_VCVT_sp *a) | |
2358 | { | |
2359 | TCGv_i64 vd; | |
2360 | TCGv_i32 vm; | |
2361 | ||
2362 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2363 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vd & 0x10)) { | |
2364 | return false; | |
2365 | } | |
2366 | ||
2367 | if (!vfp_access_check(s)) { | |
2368 | return true; | |
2369 | } | |
2370 | ||
2371 | vm = tcg_temp_new_i32(); | |
2372 | vd = tcg_temp_new_i64(); | |
2373 | neon_load_reg32(vm, a->vm); | |
2374 | gen_helper_vfp_fcvtds(vd, vm, cpu_env); | |
2375 | neon_store_reg64(vd, a->vd); | |
2376 | tcg_temp_free_i32(vm); | |
2377 | tcg_temp_free_i64(vd); | |
2378 | return true; | |
2379 | } | |
2380 | ||
2381 | static bool trans_VCVT_dp(DisasContext *s, arg_VCVT_dp *a) | |
2382 | { | |
2383 | TCGv_i64 vm; | |
2384 | TCGv_i32 vd; | |
2385 | ||
2386 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2387 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vm & 0x10)) { | |
2388 | return false; | |
2389 | } | |
2390 | ||
2391 | if (!vfp_access_check(s)) { | |
2392 | return true; | |
2393 | } | |
2394 | ||
2395 | vd = tcg_temp_new_i32(); | |
2396 | vm = tcg_temp_new_i64(); | |
2397 | neon_load_reg64(vm, a->vm); | |
2398 | gen_helper_vfp_fcvtsd(vd, vm, cpu_env); | |
2399 | neon_store_reg32(vd, a->vd); | |
2400 | tcg_temp_free_i32(vd); | |
2401 | tcg_temp_free_i64(vm); | |
2402 | return true; | |
2403 | } | |
8fc9d891 PM |
2404 | |
2405 | static bool trans_VCVT_int_sp(DisasContext *s, arg_VCVT_int_sp *a) | |
2406 | { | |
2407 | TCGv_i32 vm; | |
2408 | TCGv_ptr fpst; | |
2409 | ||
2410 | if (!vfp_access_check(s)) { | |
2411 | return true; | |
2412 | } | |
2413 | ||
2414 | vm = tcg_temp_new_i32(); | |
2415 | neon_load_reg32(vm, a->vm); | |
2416 | fpst = get_fpstatus_ptr(false); | |
2417 | if (a->s) { | |
2418 | /* i32 -> f32 */ | |
2419 | gen_helper_vfp_sitos(vm, vm, fpst); | |
2420 | } else { | |
2421 | /* u32 -> f32 */ | |
2422 | gen_helper_vfp_uitos(vm, vm, fpst); | |
2423 | } | |
2424 | neon_store_reg32(vm, a->vd); | |
2425 | tcg_temp_free_i32(vm); | |
2426 | tcg_temp_free_ptr(fpst); | |
2427 | return true; | |
2428 | } | |
2429 | ||
2430 | static bool trans_VCVT_int_dp(DisasContext *s, arg_VCVT_int_dp *a) | |
2431 | { | |
2432 | TCGv_i32 vm; | |
2433 | TCGv_i64 vd; | |
2434 | TCGv_ptr fpst; | |
2435 | ||
2436 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2437 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vd & 0x10)) { | |
2438 | return false; | |
2439 | } | |
2440 | ||
2441 | if (!vfp_access_check(s)) { | |
2442 | return true; | |
2443 | } | |
2444 | ||
2445 | vm = tcg_temp_new_i32(); | |
2446 | vd = tcg_temp_new_i64(); | |
2447 | neon_load_reg32(vm, a->vm); | |
2448 | fpst = get_fpstatus_ptr(false); | |
2449 | if (a->s) { | |
2450 | /* i32 -> f64 */ | |
2451 | gen_helper_vfp_sitod(vd, vm, fpst); | |
2452 | } else { | |
2453 | /* u32 -> f64 */ | |
2454 | gen_helper_vfp_uitod(vd, vm, fpst); | |
2455 | } | |
2456 | neon_store_reg64(vd, a->vd); | |
2457 | tcg_temp_free_i32(vm); | |
2458 | tcg_temp_free_i64(vd); | |
2459 | tcg_temp_free_ptr(fpst); | |
2460 | return true; | |
2461 | } | |
92073e94 PM |
2462 | |
2463 | static bool trans_VJCVT(DisasContext *s, arg_VJCVT *a) | |
2464 | { | |
2465 | TCGv_i32 vd; | |
2466 | TCGv_i64 vm; | |
2467 | ||
2468 | if (!dc_isar_feature(aa32_jscvt, s)) { | |
2469 | return false; | |
2470 | } | |
2471 | ||
2472 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2473 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vm & 0x10)) { | |
2474 | return false; | |
2475 | } | |
2476 | ||
2477 | if (!vfp_access_check(s)) { | |
2478 | return true; | |
2479 | } | |
2480 | ||
2481 | vm = tcg_temp_new_i64(); | |
2482 | vd = tcg_temp_new_i32(); | |
2483 | neon_load_reg64(vm, a->vm); | |
2484 | gen_helper_vjcvt(vd, vm, cpu_env); | |
2485 | neon_store_reg32(vd, a->vd); | |
2486 | tcg_temp_free_i64(vm); | |
2487 | tcg_temp_free_i32(vd); | |
2488 | return true; | |
2489 | } | |
e3d6f429 PM |
2490 | |
2491 | static bool trans_VCVT_fix_sp(DisasContext *s, arg_VCVT_fix_sp *a) | |
2492 | { | |
2493 | TCGv_i32 vd, shift; | |
2494 | TCGv_ptr fpst; | |
2495 | int frac_bits; | |
2496 | ||
2497 | if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) { | |
2498 | return false; | |
2499 | } | |
2500 | ||
2501 | if (!vfp_access_check(s)) { | |
2502 | return true; | |
2503 | } | |
2504 | ||
2505 | frac_bits = (a->opc & 1) ? (32 - a->imm) : (16 - a->imm); | |
2506 | ||
2507 | vd = tcg_temp_new_i32(); | |
2508 | neon_load_reg32(vd, a->vd); | |
2509 | ||
2510 | fpst = get_fpstatus_ptr(false); | |
2511 | shift = tcg_const_i32(frac_bits); | |
2512 | ||
2513 | /* Switch on op:U:sx bits */ | |
2514 | switch (a->opc) { | |
2515 | case 0: | |
2516 | gen_helper_vfp_shtos(vd, vd, shift, fpst); | |
2517 | break; | |
2518 | case 1: | |
2519 | gen_helper_vfp_sltos(vd, vd, shift, fpst); | |
2520 | break; | |
2521 | case 2: | |
2522 | gen_helper_vfp_uhtos(vd, vd, shift, fpst); | |
2523 | break; | |
2524 | case 3: | |
2525 | gen_helper_vfp_ultos(vd, vd, shift, fpst); | |
2526 | break; | |
2527 | case 4: | |
2528 | gen_helper_vfp_toshs_round_to_zero(vd, vd, shift, fpst); | |
2529 | break; | |
2530 | case 5: | |
2531 | gen_helper_vfp_tosls_round_to_zero(vd, vd, shift, fpst); | |
2532 | break; | |
2533 | case 6: | |
2534 | gen_helper_vfp_touhs_round_to_zero(vd, vd, shift, fpst); | |
2535 | break; | |
2536 | case 7: | |
2537 | gen_helper_vfp_touls_round_to_zero(vd, vd, shift, fpst); | |
2538 | break; | |
2539 | default: | |
2540 | g_assert_not_reached(); | |
2541 | } | |
2542 | ||
2543 | neon_store_reg32(vd, a->vd); | |
2544 | tcg_temp_free_i32(vd); | |
2545 | tcg_temp_free_i32(shift); | |
2546 | tcg_temp_free_ptr(fpst); | |
2547 | return true; | |
2548 | } | |
2549 | ||
2550 | static bool trans_VCVT_fix_dp(DisasContext *s, arg_VCVT_fix_dp *a) | |
2551 | { | |
2552 | TCGv_i64 vd; | |
2553 | TCGv_i32 shift; | |
2554 | TCGv_ptr fpst; | |
2555 | int frac_bits; | |
2556 | ||
2557 | if (!arm_dc_feature(s, ARM_FEATURE_VFP3)) { | |
2558 | return false; | |
2559 | } | |
2560 | ||
2561 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2562 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vd & 0x10)) { | |
2563 | return false; | |
2564 | } | |
2565 | ||
2566 | if (!vfp_access_check(s)) { | |
2567 | return true; | |
2568 | } | |
2569 | ||
2570 | frac_bits = (a->opc & 1) ? (32 - a->imm) : (16 - a->imm); | |
2571 | ||
2572 | vd = tcg_temp_new_i64(); | |
2573 | neon_load_reg64(vd, a->vd); | |
2574 | ||
2575 | fpst = get_fpstatus_ptr(false); | |
2576 | shift = tcg_const_i32(frac_bits); | |
2577 | ||
2578 | /* Switch on op:U:sx bits */ | |
2579 | switch (a->opc) { | |
2580 | case 0: | |
2581 | gen_helper_vfp_shtod(vd, vd, shift, fpst); | |
2582 | break; | |
2583 | case 1: | |
2584 | gen_helper_vfp_sltod(vd, vd, shift, fpst); | |
2585 | break; | |
2586 | case 2: | |
2587 | gen_helper_vfp_uhtod(vd, vd, shift, fpst); | |
2588 | break; | |
2589 | case 3: | |
2590 | gen_helper_vfp_ultod(vd, vd, shift, fpst); | |
2591 | break; | |
2592 | case 4: | |
2593 | gen_helper_vfp_toshd_round_to_zero(vd, vd, shift, fpst); | |
2594 | break; | |
2595 | case 5: | |
2596 | gen_helper_vfp_tosld_round_to_zero(vd, vd, shift, fpst); | |
2597 | break; | |
2598 | case 6: | |
2599 | gen_helper_vfp_touhd_round_to_zero(vd, vd, shift, fpst); | |
2600 | break; | |
2601 | case 7: | |
2602 | gen_helper_vfp_tould_round_to_zero(vd, vd, shift, fpst); | |
2603 | break; | |
2604 | default: | |
2605 | g_assert_not_reached(); | |
2606 | } | |
2607 | ||
2608 | neon_store_reg64(vd, a->vd); | |
2609 | tcg_temp_free_i64(vd); | |
2610 | tcg_temp_free_i32(shift); | |
2611 | tcg_temp_free_ptr(fpst); | |
2612 | return true; | |
2613 | } | |
3111bfc2 PM |
2614 | |
2615 | static bool trans_VCVT_sp_int(DisasContext *s, arg_VCVT_sp_int *a) | |
2616 | { | |
2617 | TCGv_i32 vm; | |
2618 | TCGv_ptr fpst; | |
2619 | ||
2620 | if (!vfp_access_check(s)) { | |
2621 | return true; | |
2622 | } | |
2623 | ||
2624 | fpst = get_fpstatus_ptr(false); | |
2625 | vm = tcg_temp_new_i32(); | |
2626 | neon_load_reg32(vm, a->vm); | |
2627 | ||
2628 | if (a->s) { | |
2629 | if (a->rz) { | |
2630 | gen_helper_vfp_tosizs(vm, vm, fpst); | |
2631 | } else { | |
2632 | gen_helper_vfp_tosis(vm, vm, fpst); | |
2633 | } | |
2634 | } else { | |
2635 | if (a->rz) { | |
2636 | gen_helper_vfp_touizs(vm, vm, fpst); | |
2637 | } else { | |
2638 | gen_helper_vfp_touis(vm, vm, fpst); | |
2639 | } | |
2640 | } | |
2641 | neon_store_reg32(vm, a->vd); | |
2642 | tcg_temp_free_i32(vm); | |
2643 | tcg_temp_free_ptr(fpst); | |
2644 | return true; | |
2645 | } | |
2646 | ||
2647 | static bool trans_VCVT_dp_int(DisasContext *s, arg_VCVT_dp_int *a) | |
2648 | { | |
2649 | TCGv_i32 vd; | |
2650 | TCGv_i64 vm; | |
2651 | TCGv_ptr fpst; | |
2652 | ||
2653 | /* UNDEF accesses to D16-D31 if they don't exist. */ | |
2654 | if (!dc_isar_feature(aa32_fp_d32, s) && (a->vm & 0x10)) { | |
2655 | return false; | |
2656 | } | |
2657 | ||
2658 | if (!vfp_access_check(s)) { | |
2659 | return true; | |
2660 | } | |
2661 | ||
2662 | fpst = get_fpstatus_ptr(false); | |
2663 | vm = tcg_temp_new_i64(); | |
2664 | vd = tcg_temp_new_i32(); | |
2665 | neon_load_reg64(vm, a->vm); | |
2666 | ||
2667 | if (a->s) { | |
2668 | if (a->rz) { | |
2669 | gen_helper_vfp_tosizd(vd, vm, fpst); | |
2670 | } else { | |
2671 | gen_helper_vfp_tosid(vd, vm, fpst); | |
2672 | } | |
2673 | } else { | |
2674 | if (a->rz) { | |
2675 | gen_helper_vfp_touizd(vd, vm, fpst); | |
2676 | } else { | |
2677 | gen_helper_vfp_touid(vd, vm, fpst); | |
2678 | } | |
2679 | } | |
2680 | neon_store_reg32(vd, a->vd); | |
2681 | tcg_temp_free_i32(vd); | |
2682 | tcg_temp_free_i64(vm); | |
2683 | tcg_temp_free_ptr(fpst); | |
2684 | return true; | |
2685 | } |