]>
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 PM |
32 | |
33 | /* | |
34 | * Check that VFP access is enabled. If it is, do the necessary | |
35 | * M-profile lazy-FP handling and then return true. | |
36 | * If not, emit code to generate an appropriate exception and | |
37 | * return false. | |
38 | * The ignore_vfp_enabled argument specifies that we should ignore | |
39 | * whether VFP is enabled via FPEXC[EN]: this should be true for FMXR/FMRX | |
40 | * accesses to FPSID, FPEXC, MVFR0, MVFR1, MVFR2, and false for all other insns. | |
41 | */ | |
42 | static bool full_vfp_access_check(DisasContext *s, bool ignore_vfp_enabled) | |
43 | { | |
44 | if (s->fp_excp_el) { | |
45 | if (arm_dc_feature(s, ARM_FEATURE_M)) { | |
46 | gen_exception_insn(s, 4, EXCP_NOCP, syn_uncategorized(), | |
47 | s->fp_excp_el); | |
48 | } else { | |
49 | gen_exception_insn(s, 4, EXCP_UDEF, | |
50 | syn_fp_access_trap(1, 0xe, false), | |
51 | s->fp_excp_el); | |
52 | } | |
53 | return false; | |
54 | } | |
55 | ||
56 | if (!s->vfp_enabled && !ignore_vfp_enabled) { | |
57 | assert(!arm_dc_feature(s, ARM_FEATURE_M)); | |
58 | gen_exception_insn(s, 4, EXCP_UDEF, syn_uncategorized(), | |
59 | default_exception_el(s)); | |
60 | return false; | |
61 | } | |
62 | ||
63 | if (arm_dc_feature(s, ARM_FEATURE_M)) { | |
64 | /* Handle M-profile lazy FP state mechanics */ | |
65 | ||
66 | /* Trigger lazy-state preservation if necessary */ | |
67 | if (s->v7m_lspact) { | |
68 | /* | |
69 | * Lazy state saving affects external memory and also the NVIC, | |
70 | * so we must mark it as an IO operation for icount. | |
71 | */ | |
72 | if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) { | |
73 | gen_io_start(); | |
74 | } | |
75 | gen_helper_v7m_preserve_fp_state(cpu_env); | |
76 | if (tb_cflags(s->base.tb) & CF_USE_ICOUNT) { | |
77 | gen_io_end(); | |
78 | } | |
79 | /* | |
80 | * If the preserve_fp_state helper doesn't throw an exception | |
81 | * then it will clear LSPACT; we don't need to repeat this for | |
82 | * any further FP insns in this TB. | |
83 | */ | |
84 | s->v7m_lspact = false; | |
85 | } | |
86 | ||
87 | /* Update ownership of FP context: set FPCCR.S to match current state */ | |
88 | if (s->v8m_fpccr_s_wrong) { | |
89 | TCGv_i32 tmp; | |
90 | ||
91 | tmp = load_cpu_field(v7m.fpccr[M_REG_S]); | |
92 | if (s->v8m_secure) { | |
93 | tcg_gen_ori_i32(tmp, tmp, R_V7M_FPCCR_S_MASK); | |
94 | } else { | |
95 | tcg_gen_andi_i32(tmp, tmp, ~R_V7M_FPCCR_S_MASK); | |
96 | } | |
97 | store_cpu_field(tmp, v7m.fpccr[M_REG_S]); | |
98 | /* Don't need to do this for any further FP insns in this TB */ | |
99 | s->v8m_fpccr_s_wrong = false; | |
100 | } | |
101 | ||
102 | if (s->v7m_new_fp_ctxt_needed) { | |
103 | /* | |
104 | * Create new FP context by updating CONTROL.FPCA, CONTROL.SFPA | |
105 | * and the FPSCR. | |
106 | */ | |
107 | TCGv_i32 control, fpscr; | |
108 | uint32_t bits = R_V7M_CONTROL_FPCA_MASK; | |
109 | ||
110 | fpscr = load_cpu_field(v7m.fpdscr[s->v8m_secure]); | |
111 | gen_helper_vfp_set_fpscr(cpu_env, fpscr); | |
112 | tcg_temp_free_i32(fpscr); | |
113 | /* | |
114 | * We don't need to arrange to end the TB, because the only | |
115 | * parts of FPSCR which we cache in the TB flags are the VECLEN | |
116 | * and VECSTRIDE, and those don't exist for M-profile. | |
117 | */ | |
118 | ||
119 | if (s->v8m_secure) { | |
120 | bits |= R_V7M_CONTROL_SFPA_MASK; | |
121 | } | |
122 | control = load_cpu_field(v7m.control[M_REG_S]); | |
123 | tcg_gen_ori_i32(control, control, bits); | |
124 | store_cpu_field(control, v7m.control[M_REG_S]); | |
125 | /* Don't need to do this for any further FP insns in this TB */ | |
126 | s->v7m_new_fp_ctxt_needed = false; | |
127 | } | |
128 | } | |
129 | ||
130 | return true; | |
131 | } | |
b3ff4b87 PM |
132 | |
133 | /* | |
134 | * The most usual kind of VFP access check, for everything except | |
135 | * FMXR/FMRX to the always-available special registers. | |
136 | */ | |
137 | static bool vfp_access_check(DisasContext *s) | |
138 | { | |
139 | return full_vfp_access_check(s, false); | |
140 | } | |
f7bbb8f3 PM |
141 | |
142 | static bool trans_VSEL(DisasContext *s, arg_VSEL *a) | |
143 | { | |
144 | uint32_t rd, rn, rm; | |
145 | bool dp = a->dp; | |
146 | ||
147 | if (!dc_isar_feature(aa32_vsel, s)) { | |
148 | return false; | |
149 | } | |
150 | ||
151 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
152 | if (dp && !dc_isar_feature(aa32_fp_d32, s) && | |
153 | ((a->vm | a->vn | a->vd) & 0x10)) { | |
154 | return false; | |
155 | } | |
156 | rd = a->vd; | |
157 | rn = a->vn; | |
158 | rm = a->vm; | |
159 | ||
160 | if (!vfp_access_check(s)) { | |
161 | return true; | |
162 | } | |
163 | ||
164 | if (dp) { | |
165 | TCGv_i64 frn, frm, dest; | |
166 | TCGv_i64 tmp, zero, zf, nf, vf; | |
167 | ||
168 | zero = tcg_const_i64(0); | |
169 | ||
170 | frn = tcg_temp_new_i64(); | |
171 | frm = tcg_temp_new_i64(); | |
172 | dest = tcg_temp_new_i64(); | |
173 | ||
174 | zf = tcg_temp_new_i64(); | |
175 | nf = tcg_temp_new_i64(); | |
176 | vf = tcg_temp_new_i64(); | |
177 | ||
178 | tcg_gen_extu_i32_i64(zf, cpu_ZF); | |
179 | tcg_gen_ext_i32_i64(nf, cpu_NF); | |
180 | tcg_gen_ext_i32_i64(vf, cpu_VF); | |
181 | ||
182 | tcg_gen_ld_f64(frn, cpu_env, vfp_reg_offset(dp, rn)); | |
183 | tcg_gen_ld_f64(frm, cpu_env, vfp_reg_offset(dp, rm)); | |
184 | switch (a->cc) { | |
185 | case 0: /* eq: Z */ | |
186 | tcg_gen_movcond_i64(TCG_COND_EQ, dest, zf, zero, | |
187 | frn, frm); | |
188 | break; | |
189 | case 1: /* vs: V */ | |
190 | tcg_gen_movcond_i64(TCG_COND_LT, dest, vf, zero, | |
191 | frn, frm); | |
192 | break; | |
193 | case 2: /* ge: N == V -> N ^ V == 0 */ | |
194 | tmp = tcg_temp_new_i64(); | |
195 | tcg_gen_xor_i64(tmp, vf, nf); | |
196 | tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero, | |
197 | frn, frm); | |
198 | tcg_temp_free_i64(tmp); | |
199 | break; | |
200 | case 3: /* gt: !Z && N == V */ | |
201 | tcg_gen_movcond_i64(TCG_COND_NE, dest, zf, zero, | |
202 | frn, frm); | |
203 | tmp = tcg_temp_new_i64(); | |
204 | tcg_gen_xor_i64(tmp, vf, nf); | |
205 | tcg_gen_movcond_i64(TCG_COND_GE, dest, tmp, zero, | |
206 | dest, frm); | |
207 | tcg_temp_free_i64(tmp); | |
208 | break; | |
209 | } | |
210 | tcg_gen_st_f64(dest, cpu_env, vfp_reg_offset(dp, rd)); | |
211 | tcg_temp_free_i64(frn); | |
212 | tcg_temp_free_i64(frm); | |
213 | tcg_temp_free_i64(dest); | |
214 | ||
215 | tcg_temp_free_i64(zf); | |
216 | tcg_temp_free_i64(nf); | |
217 | tcg_temp_free_i64(vf); | |
218 | ||
219 | tcg_temp_free_i64(zero); | |
220 | } else { | |
221 | TCGv_i32 frn, frm, dest; | |
222 | TCGv_i32 tmp, zero; | |
223 | ||
224 | zero = tcg_const_i32(0); | |
225 | ||
226 | frn = tcg_temp_new_i32(); | |
227 | frm = tcg_temp_new_i32(); | |
228 | dest = tcg_temp_new_i32(); | |
229 | tcg_gen_ld_f32(frn, cpu_env, vfp_reg_offset(dp, rn)); | |
230 | tcg_gen_ld_f32(frm, cpu_env, vfp_reg_offset(dp, rm)); | |
231 | switch (a->cc) { | |
232 | case 0: /* eq: Z */ | |
233 | tcg_gen_movcond_i32(TCG_COND_EQ, dest, cpu_ZF, zero, | |
234 | frn, frm); | |
235 | break; | |
236 | case 1: /* vs: V */ | |
237 | tcg_gen_movcond_i32(TCG_COND_LT, dest, cpu_VF, zero, | |
238 | frn, frm); | |
239 | break; | |
240 | case 2: /* ge: N == V -> N ^ V == 0 */ | |
241 | tmp = tcg_temp_new_i32(); | |
242 | tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF); | |
243 | tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero, | |
244 | frn, frm); | |
245 | tcg_temp_free_i32(tmp); | |
246 | break; | |
247 | case 3: /* gt: !Z && N == V */ | |
248 | tcg_gen_movcond_i32(TCG_COND_NE, dest, cpu_ZF, zero, | |
249 | frn, frm); | |
250 | tmp = tcg_temp_new_i32(); | |
251 | tcg_gen_xor_i32(tmp, cpu_VF, cpu_NF); | |
252 | tcg_gen_movcond_i32(TCG_COND_GE, dest, tmp, zero, | |
253 | dest, frm); | |
254 | tcg_temp_free_i32(tmp); | |
255 | break; | |
256 | } | |
257 | tcg_gen_st_f32(dest, cpu_env, vfp_reg_offset(dp, rd)); | |
258 | tcg_temp_free_i32(frn); | |
259 | tcg_temp_free_i32(frm); | |
260 | tcg_temp_free_i32(dest); | |
261 | ||
262 | tcg_temp_free_i32(zero); | |
263 | } | |
264 | ||
265 | return true; | |
266 | } | |
267 | ||
268 | static bool trans_VMINMAXNM(DisasContext *s, arg_VMINMAXNM *a) | |
269 | { | |
270 | uint32_t rd, rn, rm; | |
271 | bool dp = a->dp; | |
272 | bool vmin = a->op; | |
273 | TCGv_ptr fpst; | |
274 | ||
275 | if (!dc_isar_feature(aa32_vminmaxnm, s)) { | |
276 | return false; | |
277 | } | |
278 | ||
279 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
280 | if (dp && !dc_isar_feature(aa32_fp_d32, s) && | |
281 | ((a->vm | a->vn | a->vd) & 0x10)) { | |
282 | return false; | |
283 | } | |
284 | rd = a->vd; | |
285 | rn = a->vn; | |
286 | rm = a->vm; | |
287 | ||
288 | if (!vfp_access_check(s)) { | |
289 | return true; | |
290 | } | |
291 | ||
292 | fpst = get_fpstatus_ptr(0); | |
293 | ||
294 | if (dp) { | |
295 | TCGv_i64 frn, frm, dest; | |
296 | ||
297 | frn = tcg_temp_new_i64(); | |
298 | frm = tcg_temp_new_i64(); | |
299 | dest = tcg_temp_new_i64(); | |
300 | ||
301 | tcg_gen_ld_f64(frn, cpu_env, vfp_reg_offset(dp, rn)); | |
302 | tcg_gen_ld_f64(frm, cpu_env, vfp_reg_offset(dp, rm)); | |
303 | if (vmin) { | |
304 | gen_helper_vfp_minnumd(dest, frn, frm, fpst); | |
305 | } else { | |
306 | gen_helper_vfp_maxnumd(dest, frn, frm, fpst); | |
307 | } | |
308 | tcg_gen_st_f64(dest, cpu_env, vfp_reg_offset(dp, rd)); | |
309 | tcg_temp_free_i64(frn); | |
310 | tcg_temp_free_i64(frm); | |
311 | tcg_temp_free_i64(dest); | |
312 | } else { | |
313 | TCGv_i32 frn, frm, dest; | |
314 | ||
315 | frn = tcg_temp_new_i32(); | |
316 | frm = tcg_temp_new_i32(); | |
317 | dest = tcg_temp_new_i32(); | |
318 | ||
319 | tcg_gen_ld_f32(frn, cpu_env, vfp_reg_offset(dp, rn)); | |
320 | tcg_gen_ld_f32(frm, cpu_env, vfp_reg_offset(dp, rm)); | |
321 | if (vmin) { | |
322 | gen_helper_vfp_minnums(dest, frn, frm, fpst); | |
323 | } else { | |
324 | gen_helper_vfp_maxnums(dest, frn, frm, fpst); | |
325 | } | |
326 | tcg_gen_st_f32(dest, cpu_env, vfp_reg_offset(dp, rd)); | |
327 | tcg_temp_free_i32(frn); | |
328 | tcg_temp_free_i32(frm); | |
329 | tcg_temp_free_i32(dest); | |
330 | } | |
331 | ||
332 | tcg_temp_free_ptr(fpst); | |
333 | return true; | |
334 | } | |
335 | ||
336 | /* | |
337 | * Table for converting the most common AArch32 encoding of | |
338 | * rounding mode to arm_fprounding order (which matches the | |
339 | * common AArch64 order); see ARM ARM pseudocode FPDecodeRM(). | |
340 | */ | |
341 | static const uint8_t fp_decode_rm[] = { | |
342 | FPROUNDING_TIEAWAY, | |
343 | FPROUNDING_TIEEVEN, | |
344 | FPROUNDING_POSINF, | |
345 | FPROUNDING_NEGINF, | |
346 | }; | |
347 | ||
348 | static bool trans_VRINT(DisasContext *s, arg_VRINT *a) | |
349 | { | |
350 | uint32_t rd, rm; | |
351 | bool dp = a->dp; | |
352 | TCGv_ptr fpst; | |
353 | TCGv_i32 tcg_rmode; | |
354 | int rounding = fp_decode_rm[a->rm]; | |
355 | ||
356 | if (!dc_isar_feature(aa32_vrint, s)) { | |
357 | return false; | |
358 | } | |
359 | ||
360 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
361 | if (dp && !dc_isar_feature(aa32_fp_d32, s) && | |
362 | ((a->vm | a->vd) & 0x10)) { | |
363 | return false; | |
364 | } | |
365 | rd = a->vd; | |
366 | rm = a->vm; | |
367 | ||
368 | if (!vfp_access_check(s)) { | |
369 | return true; | |
370 | } | |
371 | ||
372 | fpst = get_fpstatus_ptr(0); | |
373 | ||
374 | tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding)); | |
375 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
376 | ||
377 | if (dp) { | |
378 | TCGv_i64 tcg_op; | |
379 | TCGv_i64 tcg_res; | |
380 | tcg_op = tcg_temp_new_i64(); | |
381 | tcg_res = tcg_temp_new_i64(); | |
382 | tcg_gen_ld_f64(tcg_op, cpu_env, vfp_reg_offset(dp, rm)); | |
383 | gen_helper_rintd(tcg_res, tcg_op, fpst); | |
384 | tcg_gen_st_f64(tcg_res, cpu_env, vfp_reg_offset(dp, rd)); | |
385 | tcg_temp_free_i64(tcg_op); | |
386 | tcg_temp_free_i64(tcg_res); | |
387 | } else { | |
388 | TCGv_i32 tcg_op; | |
389 | TCGv_i32 tcg_res; | |
390 | tcg_op = tcg_temp_new_i32(); | |
391 | tcg_res = tcg_temp_new_i32(); | |
392 | tcg_gen_ld_f32(tcg_op, cpu_env, vfp_reg_offset(dp, rm)); | |
393 | gen_helper_rints(tcg_res, tcg_op, fpst); | |
394 | tcg_gen_st_f32(tcg_res, cpu_env, vfp_reg_offset(dp, rd)); | |
395 | tcg_temp_free_i32(tcg_op); | |
396 | tcg_temp_free_i32(tcg_res); | |
397 | } | |
398 | ||
399 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
400 | tcg_temp_free_i32(tcg_rmode); | |
401 | ||
402 | tcg_temp_free_ptr(fpst); | |
403 | return true; | |
404 | } | |
405 | ||
406 | static bool trans_VCVT(DisasContext *s, arg_VCVT *a) | |
407 | { | |
408 | uint32_t rd, rm; | |
409 | bool dp = a->dp; | |
410 | TCGv_ptr fpst; | |
411 | TCGv_i32 tcg_rmode, tcg_shift; | |
412 | int rounding = fp_decode_rm[a->rm]; | |
413 | bool is_signed = a->op; | |
414 | ||
415 | if (!dc_isar_feature(aa32_vcvt_dr, s)) { | |
416 | return false; | |
417 | } | |
418 | ||
419 | /* UNDEF accesses to D16-D31 if they don't exist */ | |
420 | if (dp && !dc_isar_feature(aa32_fp_d32, s) && (a->vm & 0x10)) { | |
421 | return false; | |
422 | } | |
423 | rd = a->vd; | |
424 | rm = a->vm; | |
425 | ||
426 | if (!vfp_access_check(s)) { | |
427 | return true; | |
428 | } | |
429 | ||
430 | fpst = get_fpstatus_ptr(0); | |
431 | ||
432 | tcg_shift = tcg_const_i32(0); | |
433 | ||
434 | tcg_rmode = tcg_const_i32(arm_rmode_to_sf(rounding)); | |
435 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
436 | ||
437 | if (dp) { | |
438 | TCGv_i64 tcg_double, tcg_res; | |
439 | TCGv_i32 tcg_tmp; | |
440 | tcg_double = tcg_temp_new_i64(); | |
441 | tcg_res = tcg_temp_new_i64(); | |
442 | tcg_tmp = tcg_temp_new_i32(); | |
443 | tcg_gen_ld_f64(tcg_double, cpu_env, vfp_reg_offset(1, rm)); | |
444 | if (is_signed) { | |
445 | gen_helper_vfp_tosld(tcg_res, tcg_double, tcg_shift, fpst); | |
446 | } else { | |
447 | gen_helper_vfp_tould(tcg_res, tcg_double, tcg_shift, fpst); | |
448 | } | |
449 | tcg_gen_extrl_i64_i32(tcg_tmp, tcg_res); | |
450 | tcg_gen_st_f32(tcg_tmp, cpu_env, vfp_reg_offset(0, rd)); | |
451 | tcg_temp_free_i32(tcg_tmp); | |
452 | tcg_temp_free_i64(tcg_res); | |
453 | tcg_temp_free_i64(tcg_double); | |
454 | } else { | |
455 | TCGv_i32 tcg_single, tcg_res; | |
456 | tcg_single = tcg_temp_new_i32(); | |
457 | tcg_res = tcg_temp_new_i32(); | |
458 | tcg_gen_ld_f32(tcg_single, cpu_env, vfp_reg_offset(0, rm)); | |
459 | if (is_signed) { | |
460 | gen_helper_vfp_tosls(tcg_res, tcg_single, tcg_shift, fpst); | |
461 | } else { | |
462 | gen_helper_vfp_touls(tcg_res, tcg_single, tcg_shift, fpst); | |
463 | } | |
464 | tcg_gen_st_f32(tcg_res, cpu_env, vfp_reg_offset(0, rd)); | |
465 | tcg_temp_free_i32(tcg_res); | |
466 | tcg_temp_free_i32(tcg_single); | |
467 | } | |
468 | ||
469 | gen_helper_set_rmode(tcg_rmode, tcg_rmode, fpst); | |
470 | tcg_temp_free_i32(tcg_rmode); | |
471 | ||
472 | tcg_temp_free_i32(tcg_shift); | |
473 | ||
474 | tcg_temp_free_ptr(fpst); | |
475 | ||
476 | return true; | |
477 | } |