?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge remote-tracking branch 'qemu/master'
[qemu/ar7.git] / target-mips / msa_helper.c
blob654a0d2728901533f6f1a36cf1c83acd66bc512d
1 /*
2 * MIPS SIMD Architecture Module Instruction emulation helpers for QEMU.
3 *
4 * Copyright (c) 2014 Imagination Technologies
5 *
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
10 *
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
18 */
19
20 #include "qemu/osdep.h"
21 #include "cpu.h"
22 #include "exec/helper-proto.h"
23
24 /* Data format min and max values */
25 #define DF_BITS(df) (1 << ((df) + 3))
26
27 #define DF_MAX_INT(df) (int64_t)((1LL << (DF_BITS(df) - 1)) - 1)
28 #define M_MAX_INT(m) (int64_t)((1LL << ((m) - 1)) - 1)
29
30 #define DF_MIN_INT(df) (int64_t)(-(1LL << (DF_BITS(df) - 1)))
31 #define M_MIN_INT(m) (int64_t)(-(1LL << ((m) - 1)))
32
33 #define DF_MAX_UINT(df) (uint64_t)(-1ULL >> (64 - DF_BITS(df)))
34 #define M_MAX_UINT(m) (uint64_t)(-1ULL >> (64 - (m)))
35
36 #define UNSIGNED(x, df) ((x) & DF_MAX_UINT(df))
37 #define SIGNED(x, df) \
38 ((((int64_t)x) << (64 - DF_BITS(df))) >> (64 - DF_BITS(df)))
39
40 /* Element-by-element access macros */
41 #define DF_ELEMENTS(df) (MSA_WRLEN / DF_BITS(df))
42
43 static inline void msa_move_v(wr_t *pwd, wr_t *pws)
44 {
45 uint32_t i;
46
47 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
48 pwd->d[i] = pws->d[i];
49 }
50 }
51
52 #define MSA_FN_IMM8(FUNC, DEST, OPERATION) \
53 void helper_msa_ ## FUNC(CPUMIPSState *env, uint32_t wd, uint32_t ws, \
54 uint32_t i8) \
55 { \
56 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
57 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
58 uint32_t i; \
59 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
60 DEST = OPERATION; \
61 } \
62 }
63
64 MSA_FN_IMM8(andi_b, pwd->b[i], pws->b[i] & i8)
65 MSA_FN_IMM8(ori_b, pwd->b[i], pws->b[i] | i8)
66 MSA_FN_IMM8(nori_b, pwd->b[i], ~(pws->b[i] | i8))
67 MSA_FN_IMM8(xori_b, pwd->b[i], pws->b[i] ^ i8)
68
69 #define BIT_MOVE_IF_NOT_ZERO(dest, arg1, arg2, df) \
70 UNSIGNED(((dest & (~arg2)) | (arg1 & arg2)), df)
71 MSA_FN_IMM8(bmnzi_b, pwd->b[i],
72 BIT_MOVE_IF_NOT_ZERO(pwd->b[i], pws->b[i], i8, DF_BYTE))
73
74 #define BIT_MOVE_IF_ZERO(dest, arg1, arg2, df) \
75 UNSIGNED((dest & arg2) | (arg1 & (~arg2)), df)
76 MSA_FN_IMM8(bmzi_b, pwd->b[i],
77 BIT_MOVE_IF_ZERO(pwd->b[i], pws->b[i], i8, DF_BYTE))
78
79 #define BIT_SELECT(dest, arg1, arg2, df) \
80 UNSIGNED((arg1 & (~dest)) | (arg2 & dest), df)
81 MSA_FN_IMM8(bseli_b, pwd->b[i],
82 BIT_SELECT(pwd->b[i], pws->b[i], i8, DF_BYTE))
83
84 #undef MSA_FN_IMM8
85
86 #define SHF_POS(i, imm) (((i) & 0xfc) + (((imm) >> (2 * ((i) & 0x03))) & 0x03))
87
88 void helper_msa_shf_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
89 uint32_t ws, uint32_t imm)
90 {
91 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
92 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
93 wr_t wx, *pwx = &wx;
94 uint32_t i;
95
96 switch (df) {
97 case DF_BYTE:
98 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) {
99 pwx->b[i] = pws->b[SHF_POS(i, imm)];
100 }
101 break;
102 case DF_HALF:
103 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) {
104 pwx->h[i] = pws->h[SHF_POS(i, imm)];
105 }
106 break;
107 case DF_WORD:
108 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
109 pwx->w[i] = pws->w[SHF_POS(i, imm)];
110 }
111 break;
112 default:
113 assert(0);
114 }
115 msa_move_v(pwd, pwx);
116 }
117
118 #define MSA_FN_VECTOR(FUNC, DEST, OPERATION) \
119 void helper_msa_ ## FUNC(CPUMIPSState *env, uint32_t wd, uint32_t ws, \
120 uint32_t wt) \
121 { \
122 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
123 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
124 wr_t *pwt = &(env->active_fpu.fpr[wt].wr); \
125 uint32_t i; \
126 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
127 DEST = OPERATION; \
128 } \
129 }
130
131 MSA_FN_VECTOR(and_v, pwd->d[i], pws->d[i] & pwt->d[i])
132 MSA_FN_VECTOR(or_v, pwd->d[i], pws->d[i] | pwt->d[i])
133 MSA_FN_VECTOR(nor_v, pwd->d[i], ~(pws->d[i] | pwt->d[i]))
134 MSA_FN_VECTOR(xor_v, pwd->d[i], pws->d[i] ^ pwt->d[i])
135 MSA_FN_VECTOR(bmnz_v, pwd->d[i],
136 BIT_MOVE_IF_NOT_ZERO(pwd->d[i], pws->d[i], pwt->d[i], DF_DOUBLE))
137 MSA_FN_VECTOR(bmz_v, pwd->d[i],
138 BIT_MOVE_IF_ZERO(pwd->d[i], pws->d[i], pwt->d[i], DF_DOUBLE))
139 MSA_FN_VECTOR(bsel_v, pwd->d[i],
140 BIT_SELECT(pwd->d[i], pws->d[i], pwt->d[i], DF_DOUBLE))
141 #undef BIT_MOVE_IF_NOT_ZERO
142 #undef BIT_MOVE_IF_ZERO
143 #undef BIT_SELECT
144 #undef MSA_FN_VECTOR
145
146 static inline int64_t msa_addv_df(uint32_t df, int64_t arg1, int64_t arg2)
147 {
148 return arg1 + arg2;
149 }
150
151 static inline int64_t msa_subv_df(uint32_t df, int64_t arg1, int64_t arg2)
152 {
153 return arg1 - arg2;
154 }
155
156 static inline int64_t msa_ceq_df(uint32_t df, int64_t arg1, int64_t arg2)
157 {
158 return arg1 == arg2 ? -1 : 0;
159 }
160
161 static inline int64_t msa_cle_s_df(uint32_t df, int64_t arg1, int64_t arg2)
162 {
163 return arg1 <= arg2 ? -1 : 0;
164 }
165
166 static inline int64_t msa_cle_u_df(uint32_t df, int64_t arg1, int64_t arg2)
167 {
168 uint64_t u_arg1 = UNSIGNED(arg1, df);
169 uint64_t u_arg2 = UNSIGNED(arg2, df);
170 return u_arg1 <= u_arg2 ? -1 : 0;
171 }
172
173 static inline int64_t msa_clt_s_df(uint32_t df, int64_t arg1, int64_t arg2)
174 {
175 return arg1 < arg2 ? -1 : 0;
176 }
177
178 static inline int64_t msa_clt_u_df(uint32_t df, int64_t arg1, int64_t arg2)
179 {
180 uint64_t u_arg1 = UNSIGNED(arg1, df);
181 uint64_t u_arg2 = UNSIGNED(arg2, df);
182 return u_arg1 < u_arg2 ? -1 : 0;
183 }
184
185 static inline int64_t msa_max_s_df(uint32_t df, int64_t arg1, int64_t arg2)
186 {
187 return arg1 > arg2 ? arg1 : arg2;
188 }
189
190 static inline int64_t msa_max_u_df(uint32_t df, int64_t arg1, int64_t arg2)
191 {
192 uint64_t u_arg1 = UNSIGNED(arg1, df);
193 uint64_t u_arg2 = UNSIGNED(arg2, df);
194 return u_arg1 > u_arg2 ? arg1 : arg2;
195 }
196
197 static inline int64_t msa_min_s_df(uint32_t df, int64_t arg1, int64_t arg2)
198 {
199 return arg1 < arg2 ? arg1 : arg2;
200 }
201
202 static inline int64_t msa_min_u_df(uint32_t df, int64_t arg1, int64_t arg2)
203 {
204 uint64_t u_arg1 = UNSIGNED(arg1, df);
205 uint64_t u_arg2 = UNSIGNED(arg2, df);
206 return u_arg1 < u_arg2 ? arg1 : arg2;
207 }
208
209 #define MSA_BINOP_IMM_DF(helper, func) \
210 void helper_msa_ ## helper ## _df(CPUMIPSState *env, uint32_t df, \
211 uint32_t wd, uint32_t ws, int32_t u5) \
212 { \
213 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
214 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
215 uint32_t i; \
216 \
217 switch (df) { \
218 case DF_BYTE: \
219 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
220 pwd->b[i] = msa_ ## func ## _df(df, pws->b[i], u5); \
221 } \
222 break; \
223 case DF_HALF: \
224 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
225 pwd->h[i] = msa_ ## func ## _df(df, pws->h[i], u5); \
226 } \
227 break; \
228 case DF_WORD: \
229 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
230 pwd->w[i] = msa_ ## func ## _df(df, pws->w[i], u5); \
231 } \
232 break; \
233 case DF_DOUBLE: \
234 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
235 pwd->d[i] = msa_ ## func ## _df(df, pws->d[i], u5); \
236 } \
237 break; \
238 default: \
239 assert(0); \
240 } \
241 }
242
243 MSA_BINOP_IMM_DF(addvi, addv)
244 MSA_BINOP_IMM_DF(subvi, subv)
245 MSA_BINOP_IMM_DF(ceqi, ceq)
246 MSA_BINOP_IMM_DF(clei_s, cle_s)
247 MSA_BINOP_IMM_DF(clei_u, cle_u)
248 MSA_BINOP_IMM_DF(clti_s, clt_s)
249 MSA_BINOP_IMM_DF(clti_u, clt_u)
250 MSA_BINOP_IMM_DF(maxi_s, max_s)
251 MSA_BINOP_IMM_DF(maxi_u, max_u)
252 MSA_BINOP_IMM_DF(mini_s, min_s)
253 MSA_BINOP_IMM_DF(mini_u, min_u)
254 #undef MSA_BINOP_IMM_DF
255
256 void helper_msa_ldi_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
257 int32_t s10)
258 {
259 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
260 uint32_t i;
261
262 switch (df) {
263 case DF_BYTE:
264 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) {
265 pwd->b[i] = (int8_t)s10;
266 }
267 break;
268 case DF_HALF:
269 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) {
270 pwd->h[i] = (int16_t)s10;
271 }
272 break;
273 case DF_WORD:
274 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
275 pwd->w[i] = (int32_t)s10;
276 }
277 break;
278 case DF_DOUBLE:
279 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
280 pwd->d[i] = (int64_t)s10;
281 }
282 break;
283 default:
284 assert(0);
285 }
286 }
287
288 /* Data format bit position and unsigned values */
289 #define BIT_POSITION(x, df) ((uint64_t)(x) % DF_BITS(df))
290
291 static inline int64_t msa_sll_df(uint32_t df, int64_t arg1, int64_t arg2)
292 {
293 int32_t b_arg2 = BIT_POSITION(arg2, df);
294 return arg1 << b_arg2;
295 }
296
297 static inline int64_t msa_sra_df(uint32_t df, int64_t arg1, int64_t arg2)
298 {
299 int32_t b_arg2 = BIT_POSITION(arg2, df);
300 return arg1 >> b_arg2;
301 }
302
303 static inline int64_t msa_srl_df(uint32_t df, int64_t arg1, int64_t arg2)
304 {
305 uint64_t u_arg1 = UNSIGNED(arg1, df);
306 int32_t b_arg2 = BIT_POSITION(arg2, df);
307 return u_arg1 >> b_arg2;
308 }
309
310 static inline int64_t msa_bclr_df(uint32_t df, int64_t arg1, int64_t arg2)
311 {
312 int32_t b_arg2 = BIT_POSITION(arg2, df);
313 return UNSIGNED(arg1 & (~(1LL << b_arg2)), df);
314 }
315
316 static inline int64_t msa_bset_df(uint32_t df, int64_t arg1,
317 int64_t arg2)
318 {
319 int32_t b_arg2 = BIT_POSITION(arg2, df);
320 return UNSIGNED(arg1 | (1LL << b_arg2), df);
321 }
322
323 static inline int64_t msa_bneg_df(uint32_t df, int64_t arg1, int64_t arg2)
324 {
325 int32_t b_arg2 = BIT_POSITION(arg2, df);
326 return UNSIGNED(arg1 ^ (1LL << b_arg2), df);
327 }
328
329 static inline int64_t msa_binsl_df(uint32_t df, int64_t dest, int64_t arg1,
330 int64_t arg2)
331 {
332 uint64_t u_arg1 = UNSIGNED(arg1, df);
333 uint64_t u_dest = UNSIGNED(dest, df);
334 int32_t sh_d = BIT_POSITION(arg2, df) + 1;
335 int32_t sh_a = DF_BITS(df) - sh_d;
336 if (sh_d == DF_BITS(df)) {
337 return u_arg1;
338 } else {
339 return UNSIGNED(UNSIGNED(u_dest << sh_d, df) >> sh_d, df) |
340 UNSIGNED(UNSIGNED(u_arg1 >> sh_a, df) << sh_a, df);
341 }
342 }
343
344 static inline int64_t msa_binsr_df(uint32_t df, int64_t dest, int64_t arg1,
345 int64_t arg2)
346 {
347 uint64_t u_arg1 = UNSIGNED(arg1, df);
348 uint64_t u_dest = UNSIGNED(dest, df);
349 int32_t sh_d = BIT_POSITION(arg2, df) + 1;
350 int32_t sh_a = DF_BITS(df) - sh_d;
351 if (sh_d == DF_BITS(df)) {
352 return u_arg1;
353 } else {
354 return UNSIGNED(UNSIGNED(u_dest >> sh_d, df) << sh_d, df) |
355 UNSIGNED(UNSIGNED(u_arg1 << sh_a, df) >> sh_a, df);
356 }
357 }
358
359 static inline int64_t msa_sat_s_df(uint32_t df, int64_t arg, uint32_t m)
360 {
361 return arg < M_MIN_INT(m+1) ? M_MIN_INT(m+1) :
362 arg > M_MAX_INT(m+1) ? M_MAX_INT(m+1) :
363 arg;
364 }
365
366 static inline int64_t msa_sat_u_df(uint32_t df, int64_t arg, uint32_t m)
367 {
368 uint64_t u_arg = UNSIGNED(arg, df);
369 return u_arg < M_MAX_UINT(m+1) ? u_arg :
370 M_MAX_UINT(m+1);
371 }
372
373 static inline int64_t msa_srar_df(uint32_t df, int64_t arg1, int64_t arg2)
374 {
375 int32_t b_arg2 = BIT_POSITION(arg2, df);
376 if (b_arg2 == 0) {
377 return arg1;
378 } else {
379 int64_t r_bit = (arg1 >> (b_arg2 - 1)) & 1;
380 return (arg1 >> b_arg2) + r_bit;
381 }
382 }
383
384 static inline int64_t msa_srlr_df(uint32_t df, int64_t arg1, int64_t arg2)
385 {
386 uint64_t u_arg1 = UNSIGNED(arg1, df);
387 int32_t b_arg2 = BIT_POSITION(arg2, df);
388 if (b_arg2 == 0) {
389 return u_arg1;
390 } else {
391 uint64_t r_bit = (u_arg1 >> (b_arg2 - 1)) & 1;
392 return (u_arg1 >> b_arg2) + r_bit;
393 }
394 }
395
396 #define MSA_BINOP_IMMU_DF(helper, func) \
397 void helper_msa_ ## helper ## _df(CPUMIPSState *env, uint32_t df, uint32_t wd, \
398 uint32_t ws, uint32_t u5) \
399 { \
400 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
401 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
402 uint32_t i; \
403 \
404 switch (df) { \
405 case DF_BYTE: \
406 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
407 pwd->b[i] = msa_ ## func ## _df(df, pws->b[i], u5); \
408 } \
409 break; \
410 case DF_HALF: \
411 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
412 pwd->h[i] = msa_ ## func ## _df(df, pws->h[i], u5); \
413 } \
414 break; \
415 case DF_WORD: \
416 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
417 pwd->w[i] = msa_ ## func ## _df(df, pws->w[i], u5); \
418 } \
419 break; \
420 case DF_DOUBLE: \
421 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
422 pwd->d[i] = msa_ ## func ## _df(df, pws->d[i], u5); \
423 } \
424 break; \
425 default: \
426 assert(0); \
427 } \
428 }
429
430 MSA_BINOP_IMMU_DF(slli, sll)
431 MSA_BINOP_IMMU_DF(srai, sra)
432 MSA_BINOP_IMMU_DF(srli, srl)
433 MSA_BINOP_IMMU_DF(bclri, bclr)
434 MSA_BINOP_IMMU_DF(bseti, bset)
435 MSA_BINOP_IMMU_DF(bnegi, bneg)
436 MSA_BINOP_IMMU_DF(sat_s, sat_s)
437 MSA_BINOP_IMMU_DF(sat_u, sat_u)
438 MSA_BINOP_IMMU_DF(srari, srar)
439 MSA_BINOP_IMMU_DF(srlri, srlr)
440 #undef MSA_BINOP_IMMU_DF
441
442 #define MSA_TEROP_IMMU_DF(helper, func) \
443 void helper_msa_ ## helper ## _df(CPUMIPSState *env, uint32_t df, \
444 uint32_t wd, uint32_t ws, uint32_t u5) \
445 { \
446 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
447 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
448 uint32_t i; \
449 \
450 switch (df) { \
451 case DF_BYTE: \
452 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
453 pwd->b[i] = msa_ ## func ## _df(df, pwd->b[i], pws->b[i], \
454 u5); \
455 } \
456 break; \
457 case DF_HALF: \
458 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
459 pwd->h[i] = msa_ ## func ## _df(df, pwd->h[i], pws->h[i], \
460 u5); \
461 } \
462 break; \
463 case DF_WORD: \
464 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
465 pwd->w[i] = msa_ ## func ## _df(df, pwd->w[i], pws->w[i], \
466 u5); \
467 } \
468 break; \
469 case DF_DOUBLE: \
470 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
471 pwd->d[i] = msa_ ## func ## _df(df, pwd->d[i], pws->d[i], \
472 u5); \
473 } \
474 break; \
475 default: \
476 assert(0); \
477 } \
478 }
479
480 MSA_TEROP_IMMU_DF(binsli, binsl)
481 MSA_TEROP_IMMU_DF(binsri, binsr)
482 #undef MSA_TEROP_IMMU_DF
483
484 static inline int64_t msa_max_a_df(uint32_t df, int64_t arg1, int64_t arg2)
485 {
486 uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
487 uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
488 return abs_arg1 > abs_arg2 ? arg1 : arg2;
489 }
490
491 static inline int64_t msa_min_a_df(uint32_t df, int64_t arg1, int64_t arg2)
492 {
493 uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
494 uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
495 return abs_arg1 < abs_arg2 ? arg1 : arg2;
496 }
497
498 static inline int64_t msa_add_a_df(uint32_t df, int64_t arg1, int64_t arg2)
499 {
500 uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
501 uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
502 return abs_arg1 + abs_arg2;
503 }
504
505 static inline int64_t msa_adds_a_df(uint32_t df, int64_t arg1, int64_t arg2)
506 {
507 uint64_t max_int = (uint64_t)DF_MAX_INT(df);
508 uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
509 uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
510 if (abs_arg1 > max_int || abs_arg2 > max_int) {
511 return (int64_t)max_int;
512 } else {
513 return (abs_arg1 < max_int - abs_arg2) ? abs_arg1 + abs_arg2 : max_int;
514 }
515 }
516
517 static inline int64_t msa_adds_s_df(uint32_t df, int64_t arg1, int64_t arg2)
518 {
519 int64_t max_int = DF_MAX_INT(df);
520 int64_t min_int = DF_MIN_INT(df);
521 if (arg1 < 0) {
522 return (min_int - arg1 < arg2) ? arg1 + arg2 : min_int;
523 } else {
524 return (arg2 < max_int - arg1) ? arg1 + arg2 : max_int;
525 }
526 }
527
528 static inline uint64_t msa_adds_u_df(uint32_t df, uint64_t arg1, uint64_t arg2)
529 {
530 uint64_t max_uint = DF_MAX_UINT(df);
531 uint64_t u_arg1 = UNSIGNED(arg1, df);
532 uint64_t u_arg2 = UNSIGNED(arg2, df);
533 return (u_arg1 < max_uint - u_arg2) ? u_arg1 + u_arg2 : max_uint;
534 }
535
536 static inline int64_t msa_ave_s_df(uint32_t df, int64_t arg1, int64_t arg2)
537 {
538 /* signed shift */
539 return (arg1 >> 1) + (arg2 >> 1) + (arg1 & arg2 & 1);
540 }
541
542 static inline uint64_t msa_ave_u_df(uint32_t df, uint64_t arg1, uint64_t arg2)
543 {
544 uint64_t u_arg1 = UNSIGNED(arg1, df);
545 uint64_t u_arg2 = UNSIGNED(arg2, df);
546 /* unsigned shift */
547 return (u_arg1 >> 1) + (u_arg2 >> 1) + (u_arg1 & u_arg2 & 1);
548 }
549
550 static inline int64_t msa_aver_s_df(uint32_t df, int64_t arg1, int64_t arg2)
551 {
552 /* signed shift */
553 return (arg1 >> 1) + (arg2 >> 1) + ((arg1 | arg2) & 1);
554 }
555
556 static inline uint64_t msa_aver_u_df(uint32_t df, uint64_t arg1, uint64_t arg2)
557 {
558 uint64_t u_arg1 = UNSIGNED(arg1, df);
559 uint64_t u_arg2 = UNSIGNED(arg2, df);
560 /* unsigned shift */
561 return (u_arg1 >> 1) + (u_arg2 >> 1) + ((u_arg1 | u_arg2) & 1);
562 }
563
564 static inline int64_t msa_subs_s_df(uint32_t df, int64_t arg1, int64_t arg2)
565 {
566 int64_t max_int = DF_MAX_INT(df);
567 int64_t min_int = DF_MIN_INT(df);
568 if (arg2 > 0) {
569 return (min_int + arg2 < arg1) ? arg1 - arg2 : min_int;
570 } else {
571 return (arg1 < max_int + arg2) ? arg1 - arg2 : max_int;
572 }
573 }
574
575 static inline int64_t msa_subs_u_df(uint32_t df, int64_t arg1, int64_t arg2)
576 {
577 uint64_t u_arg1 = UNSIGNED(arg1, df);
578 uint64_t u_arg2 = UNSIGNED(arg2, df);
579 return (u_arg1 > u_arg2) ? u_arg1 - u_arg2 : 0;
580 }
581
582 static inline int64_t msa_subsus_u_df(uint32_t df, int64_t arg1, int64_t arg2)
583 {
584 uint64_t u_arg1 = UNSIGNED(arg1, df);
585 uint64_t max_uint = DF_MAX_UINT(df);
586 if (arg2 >= 0) {
587 uint64_t u_arg2 = (uint64_t)arg2;
588 return (u_arg1 > u_arg2) ?
589 (int64_t)(u_arg1 - u_arg2) :
590 0;
591 } else {
592 uint64_t u_arg2 = (uint64_t)(-arg2);
593 return (u_arg1 < max_uint - u_arg2) ?
594 (int64_t)(u_arg1 + u_arg2) :
595 (int64_t)max_uint;
596 }
597 }
598
599 static inline int64_t msa_subsuu_s_df(uint32_t df, int64_t arg1, int64_t arg2)
600 {
601 uint64_t u_arg1 = UNSIGNED(arg1, df);
602 uint64_t u_arg2 = UNSIGNED(arg2, df);
603 int64_t max_int = DF_MAX_INT(df);
604 int64_t min_int = DF_MIN_INT(df);
605 if (u_arg1 > u_arg2) {
606 return u_arg1 - u_arg2 < (uint64_t)max_int ?
607 (int64_t)(u_arg1 - u_arg2) :
608 max_int;
609 } else {
610 return u_arg2 - u_arg1 < (uint64_t)(-min_int) ?
611 (int64_t)(u_arg1 - u_arg2) :
612 min_int;
613 }
614 }
615
616 static inline int64_t msa_asub_s_df(uint32_t df, int64_t arg1, int64_t arg2)
617 {
618 /* signed compare */
619 return (arg1 < arg2) ?
620 (uint64_t)(arg2 - arg1) : (uint64_t)(arg1 - arg2);
621 }
622
623 static inline uint64_t msa_asub_u_df(uint32_t df, uint64_t arg1, uint64_t arg2)
624 {
625 uint64_t u_arg1 = UNSIGNED(arg1, df);
626 uint64_t u_arg2 = UNSIGNED(arg2, df);
627 /* unsigned compare */
628 return (u_arg1 < u_arg2) ?
629 (uint64_t)(u_arg2 - u_arg1) : (uint64_t)(u_arg1 - u_arg2);
630 }
631
632 static inline int64_t msa_mulv_df(uint32_t df, int64_t arg1, int64_t arg2)
633 {
634 return arg1 * arg2;
635 }
636
637 static inline int64_t msa_div_s_df(uint32_t df, int64_t arg1, int64_t arg2)
638 {
639 if (arg1 == DF_MIN_INT(df) && arg2 == -1) {
640 return DF_MIN_INT(df);
641 }
642 return arg2 ? arg1 / arg2 : 0;
643 }
644
645 static inline int64_t msa_div_u_df(uint32_t df, int64_t arg1, int64_t arg2)
646 {
647 uint64_t u_arg1 = UNSIGNED(arg1, df);
648 uint64_t u_arg2 = UNSIGNED(arg2, df);
649 return u_arg2 ? u_arg1 / u_arg2 : 0;
650 }
651
652 static inline int64_t msa_mod_s_df(uint32_t df, int64_t arg1, int64_t arg2)
653 {
654 if (arg1 == DF_MIN_INT(df) && arg2 == -1) {
655 return 0;
656 }
657 return arg2 ? arg1 % arg2 : 0;
658 }
659
660 static inline int64_t msa_mod_u_df(uint32_t df, int64_t arg1, int64_t arg2)
661 {
662 uint64_t u_arg1 = UNSIGNED(arg1, df);
663 uint64_t u_arg2 = UNSIGNED(arg2, df);
664 return u_arg2 ? u_arg1 % u_arg2 : 0;
665 }
666
667 #define SIGNED_EVEN(a, df) \
668 ((((int64_t)(a)) << (64 - DF_BITS(df)/2)) >> (64 - DF_BITS(df)/2))
669
670 #define UNSIGNED_EVEN(a, df) \
671 ((((uint64_t)(a)) << (64 - DF_BITS(df)/2)) >> (64 - DF_BITS(df)/2))
672
673 #define SIGNED_ODD(a, df) \
674 ((((int64_t)(a)) << (64 - DF_BITS(df))) >> (64 - DF_BITS(df)/2))
675
676 #define UNSIGNED_ODD(a, df) \
677 ((((uint64_t)(a)) << (64 - DF_BITS(df))) >> (64 - DF_BITS(df)/2))
678
679 #define SIGNED_EXTRACT(e, o, a, df) \
680 do { \
681 e = SIGNED_EVEN(a, df); \
682 o = SIGNED_ODD(a, df); \
683 } while (0);
684
685 #define UNSIGNED_EXTRACT(e, o, a, df) \
686 do { \
687 e = UNSIGNED_EVEN(a, df); \
688 o = UNSIGNED_ODD(a, df); \
689 } while (0);
690
691 static inline int64_t msa_dotp_s_df(uint32_t df, int64_t arg1, int64_t arg2)
692 {
693 int64_t even_arg1;
694 int64_t even_arg2;
695 int64_t odd_arg1;
696 int64_t odd_arg2;
697 SIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
698 SIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
699 return (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
700 }
701
702 static inline int64_t msa_dotp_u_df(uint32_t df, int64_t arg1, int64_t arg2)
703 {
704 int64_t even_arg1;
705 int64_t even_arg2;
706 int64_t odd_arg1;
707 int64_t odd_arg2;
708 UNSIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
709 UNSIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
710 return (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
711 }
712
713 #define CONCATENATE_AND_SLIDE(s, k) \
714 do { \
715 for (i = 0; i < s; i++) { \
716 v[i] = pws->b[s * k + i]; \
717 v[i + s] = pwd->b[s * k + i]; \
718 } \
719 for (i = 0; i < s; i++) { \
720 pwd->b[s * k + i] = v[i + n]; \
721 } \
722 } while (0)
723
724 static inline void msa_sld_df(uint32_t df, wr_t *pwd,
725 wr_t *pws, target_ulong rt)
726 {
727 uint32_t n = rt % DF_ELEMENTS(df);
728 uint8_t v[64];
729 uint32_t i, k;
730
731 switch (df) {
732 case DF_BYTE:
733 CONCATENATE_AND_SLIDE(DF_ELEMENTS(DF_BYTE), 0);
734 break;
735 case DF_HALF:
736 for (k = 0; k < 2; k++) {
737 CONCATENATE_AND_SLIDE(DF_ELEMENTS(DF_HALF), k);
738 }
739 break;
740 case DF_WORD:
741 for (k = 0; k < 4; k++) {
742 CONCATENATE_AND_SLIDE(DF_ELEMENTS(DF_WORD), k);
743 }
744 break;
745 case DF_DOUBLE:
746 for (k = 0; k < 8; k++) {
747 CONCATENATE_AND_SLIDE(DF_ELEMENTS(DF_DOUBLE), k);
748 }
749 break;
750 default:
751 assert(0);
752 }
753 }
754
755 static inline int64_t msa_hadd_s_df(uint32_t df, int64_t arg1, int64_t arg2)
756 {
757 return SIGNED_ODD(arg1, df) + SIGNED_EVEN(arg2, df);
758 }
759
760 static inline int64_t msa_hadd_u_df(uint32_t df, int64_t arg1, int64_t arg2)
761 {
762 return UNSIGNED_ODD(arg1, df) + UNSIGNED_EVEN(arg2, df);
763 }
764
765 static inline int64_t msa_hsub_s_df(uint32_t df, int64_t arg1, int64_t arg2)
766 {
767 return SIGNED_ODD(arg1, df) - SIGNED_EVEN(arg2, df);
768 }
769
770 static inline int64_t msa_hsub_u_df(uint32_t df, int64_t arg1, int64_t arg2)
771 {
772 return UNSIGNED_ODD(arg1, df) - UNSIGNED_EVEN(arg2, df);
773 }
774
775 static inline int64_t msa_mul_q_df(uint32_t df, int64_t arg1, int64_t arg2)
776 {
777 int64_t q_min = DF_MIN_INT(df);
778 int64_t q_max = DF_MAX_INT(df);
779
780 if (arg1 == q_min && arg2 == q_min) {
781 return q_max;
782 }
783 return (arg1 * arg2) >> (DF_BITS(df) - 1);
784 }
785
786 static inline int64_t msa_mulr_q_df(uint32_t df, int64_t arg1, int64_t arg2)
787 {
788 int64_t q_min = DF_MIN_INT(df);
789 int64_t q_max = DF_MAX_INT(df);
790 int64_t r_bit = 1 << (DF_BITS(df) - 2);
791
792 if (arg1 == q_min && arg2 == q_min) {
793 return q_max;
794 }
795 return (arg1 * arg2 + r_bit) >> (DF_BITS(df) - 1);
796 }
797
798 #define MSA_BINOP_DF(func) \
799 void helper_msa_ ## func ## _df(CPUMIPSState *env, uint32_t df, \
800 uint32_t wd, uint32_t ws, uint32_t wt) \
801 { \
802 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
803 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
804 wr_t *pwt = &(env->active_fpu.fpr[wt].wr); \
805 uint32_t i; \
806 \
807 switch (df) { \
808 case DF_BYTE: \
809 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
810 pwd->b[i] = msa_ ## func ## _df(df, pws->b[i], pwt->b[i]); \
811 } \
812 break; \
813 case DF_HALF: \
814 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
815 pwd->h[i] = msa_ ## func ## _df(df, pws->h[i], pwt->h[i]); \
816 } \
817 break; \
818 case DF_WORD: \
819 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
820 pwd->w[i] = msa_ ## func ## _df(df, pws->w[i], pwt->w[i]); \
821 } \
822 break; \
823 case DF_DOUBLE: \
824 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
825 pwd->d[i] = msa_ ## func ## _df(df, pws->d[i], pwt->d[i]); \
826 } \
827 break; \
828 default: \
829 assert(0); \
830 } \
831 }
832
833 MSA_BINOP_DF(sll)
834 MSA_BINOP_DF(sra)
835 MSA_BINOP_DF(srl)
836 MSA_BINOP_DF(bclr)
837 MSA_BINOP_DF(bset)
838 MSA_BINOP_DF(bneg)
839 MSA_BINOP_DF(addv)
840 MSA_BINOP_DF(subv)
841 MSA_BINOP_DF(max_s)
842 MSA_BINOP_DF(max_u)
843 MSA_BINOP_DF(min_s)
844 MSA_BINOP_DF(min_u)
845 MSA_BINOP_DF(max_a)
846 MSA_BINOP_DF(min_a)
847 MSA_BINOP_DF(ceq)
848 MSA_BINOP_DF(clt_s)
849 MSA_BINOP_DF(clt_u)
850 MSA_BINOP_DF(cle_s)
851 MSA_BINOP_DF(cle_u)
852 MSA_BINOP_DF(add_a)
853 MSA_BINOP_DF(adds_a)
854 MSA_BINOP_DF(adds_s)
855 MSA_BINOP_DF(adds_u)
856 MSA_BINOP_DF(ave_s)
857 MSA_BINOP_DF(ave_u)
858 MSA_BINOP_DF(aver_s)
859 MSA_BINOP_DF(aver_u)
860 MSA_BINOP_DF(subs_s)
861 MSA_BINOP_DF(subs_u)
862 MSA_BINOP_DF(subsus_u)
863 MSA_BINOP_DF(subsuu_s)
864 MSA_BINOP_DF(asub_s)
865 MSA_BINOP_DF(asub_u)
866 MSA_BINOP_DF(mulv)
867 MSA_BINOP_DF(div_s)
868 MSA_BINOP_DF(div_u)
869 MSA_BINOP_DF(mod_s)
870 MSA_BINOP_DF(mod_u)
871 MSA_BINOP_DF(dotp_s)
872 MSA_BINOP_DF(dotp_u)
873 MSA_BINOP_DF(srar)
874 MSA_BINOP_DF(srlr)
875 MSA_BINOP_DF(hadd_s)
876 MSA_BINOP_DF(hadd_u)
877 MSA_BINOP_DF(hsub_s)
878 MSA_BINOP_DF(hsub_u)
879
880 MSA_BINOP_DF(mul_q)
881 MSA_BINOP_DF(mulr_q)
882 #undef MSA_BINOP_DF
883
884 void helper_msa_sld_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
885 uint32_t ws, uint32_t rt)
886 {
887 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
888 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
889
890 msa_sld_df(df, pwd, pws, env->active_tc.gpr[rt]);
891 }
892
893 static inline int64_t msa_maddv_df(uint32_t df, int64_t dest, int64_t arg1,
894 int64_t arg2)
895 {
896 return dest + arg1 * arg2;
897 }
898
899 static inline int64_t msa_msubv_df(uint32_t df, int64_t dest, int64_t arg1,
900 int64_t arg2)
901 {
902 return dest - arg1 * arg2;
903 }
904
905 static inline int64_t msa_dpadd_s_df(uint32_t df, int64_t dest, int64_t arg1,
906 int64_t arg2)
907 {
908 int64_t even_arg1;
909 int64_t even_arg2;
910 int64_t odd_arg1;
911 int64_t odd_arg2;
912 SIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
913 SIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
914 return dest + (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
915 }
916
917 static inline int64_t msa_dpadd_u_df(uint32_t df, int64_t dest, int64_t arg1,
918 int64_t arg2)
919 {
920 int64_t even_arg1;
921 int64_t even_arg2;
922 int64_t odd_arg1;
923 int64_t odd_arg2;
924 UNSIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
925 UNSIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
926 return dest + (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
927 }
928
929 static inline int64_t msa_dpsub_s_df(uint32_t df, int64_t dest, int64_t arg1,
930 int64_t arg2)
931 {
932 int64_t even_arg1;
933 int64_t even_arg2;
934 int64_t odd_arg1;
935 int64_t odd_arg2;
936 SIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
937 SIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
938 return dest - ((even_arg1 * even_arg2) + (odd_arg1 * odd_arg2));
939 }
940
941 static inline int64_t msa_dpsub_u_df(uint32_t df, int64_t dest, int64_t arg1,
942 int64_t arg2)
943 {
944 int64_t even_arg1;
945 int64_t even_arg2;
946 int64_t odd_arg1;
947 int64_t odd_arg2;
948 UNSIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
949 UNSIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
950 return dest - ((even_arg1 * even_arg2) + (odd_arg1 * odd_arg2));
951 }
952
953 static inline int64_t msa_madd_q_df(uint32_t df, int64_t dest, int64_t arg1,
954 int64_t arg2)
955 {
956 int64_t q_prod, q_ret;
957
958 int64_t q_max = DF_MAX_INT(df);
959 int64_t q_min = DF_MIN_INT(df);
960
961 q_prod = arg1 * arg2;
962 q_ret = ((dest << (DF_BITS(df) - 1)) + q_prod) >> (DF_BITS(df) - 1);
963
964 return (q_ret < q_min) ? q_min : (q_max < q_ret) ? q_max : q_ret;
965 }
966
967 static inline int64_t msa_msub_q_df(uint32_t df, int64_t dest, int64_t arg1,
968 int64_t arg2)
969 {
970 int64_t q_prod, q_ret;
971
972 int64_t q_max = DF_MAX_INT(df);
973 int64_t q_min = DF_MIN_INT(df);
974
975 q_prod = arg1 * arg2;
976 q_ret = ((dest << (DF_BITS(df) - 1)) - q_prod) >> (DF_BITS(df) - 1);
977
978 return (q_ret < q_min) ? q_min : (q_max < q_ret) ? q_max : q_ret;
979 }
980
981 static inline int64_t msa_maddr_q_df(uint32_t df, int64_t dest, int64_t arg1,
982 int64_t arg2)
983 {
984 int64_t q_prod, q_ret;
985
986 int64_t q_max = DF_MAX_INT(df);
987 int64_t q_min = DF_MIN_INT(df);
988 int64_t r_bit = 1 << (DF_BITS(df) - 2);
989
990 q_prod = arg1 * arg2;
991 q_ret = ((dest << (DF_BITS(df) - 1)) + q_prod + r_bit) >> (DF_BITS(df) - 1);
992
993 return (q_ret < q_min) ? q_min : (q_max < q_ret) ? q_max : q_ret;
994 }
995
996 static inline int64_t msa_msubr_q_df(uint32_t df, int64_t dest, int64_t arg1,
997 int64_t arg2)
998 {
999 int64_t q_prod, q_ret;
1000
1001 int64_t q_max = DF_MAX_INT(df);
1002 int64_t q_min = DF_MIN_INT(df);
1003 int64_t r_bit = 1 << (DF_BITS(df) - 2);
1004
1005 q_prod = arg1 * arg2;
1006 q_ret = ((dest << (DF_BITS(df) - 1)) - q_prod + r_bit) >> (DF_BITS(df) - 1);
1007
1008 return (q_ret < q_min) ? q_min : (q_max < q_ret) ? q_max : q_ret;
1009 }
1010
1011 #define MSA_TEROP_DF(func) \
1012 void helper_msa_ ## func ## _df(CPUMIPSState *env, uint32_t df, uint32_t wd, \
1013 uint32_t ws, uint32_t wt) \
1014 { \
1015 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
1016 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
1017 wr_t *pwt = &(env->active_fpu.fpr[wt].wr); \
1018 uint32_t i; \
1019 \
1020 switch (df) { \
1021 case DF_BYTE: \
1022 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
1023 pwd->b[i] = msa_ ## func ## _df(df, pwd->b[i], pws->b[i], \
1024 pwt->b[i]); \
1025 } \
1026 break; \
1027 case DF_HALF: \
1028 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
1029 pwd->h[i] = msa_ ## func ## _df(df, pwd->h[i], pws->h[i], \
1030 pwt->h[i]); \
1031 } \
1032 break; \
1033 case DF_WORD: \
1034 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
1035 pwd->w[i] = msa_ ## func ## _df(df, pwd->w[i], pws->w[i], \
1036 pwt->w[i]); \
1037 } \
1038 break; \
1039 case DF_DOUBLE: \
1040 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
1041 pwd->d[i] = msa_ ## func ## _df(df, pwd->d[i], pws->d[i], \
1042 pwt->d[i]); \
1043 } \
1044 break; \
1045 default: \
1046 assert(0); \
1047 } \
1048 }
1049
1050 MSA_TEROP_DF(maddv)
1051 MSA_TEROP_DF(msubv)
1052 MSA_TEROP_DF(dpadd_s)
1053 MSA_TEROP_DF(dpadd_u)
1054 MSA_TEROP_DF(dpsub_s)
1055 MSA_TEROP_DF(dpsub_u)
1056 MSA_TEROP_DF(binsl)
1057 MSA_TEROP_DF(binsr)
1058 MSA_TEROP_DF(madd_q)
1059 MSA_TEROP_DF(msub_q)
1060 MSA_TEROP_DF(maddr_q)
1061 MSA_TEROP_DF(msubr_q)
1062 #undef MSA_TEROP_DF
1063
1064 static inline void msa_splat_df(uint32_t df, wr_t *pwd,
1065 wr_t *pws, target_ulong rt)
1066 {
1067 uint32_t n = rt % DF_ELEMENTS(df);
1068 uint32_t i;
1069
1070 switch (df) {
1071 case DF_BYTE:
1072 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) {
1073 pwd->b[i] = pws->b[n];
1074 }
1075 break;
1076 case DF_HALF:
1077 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) {
1078 pwd->h[i] = pws->h[n];
1079 }
1080 break;
1081 case DF_WORD:
1082 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
1083 pwd->w[i] = pws->w[n];
1084 }
1085 break;
1086 case DF_DOUBLE:
1087 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
1088 pwd->d[i] = pws->d[n];
1089 }
1090 break;
1091 default:
1092 assert(0);
1093 }
1094 }
1095
1096 void helper_msa_splat_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1097 uint32_t ws, uint32_t rt)
1098 {
1099 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1100 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1101
1102 msa_splat_df(df, pwd, pws, env->active_tc.gpr[rt]);
1103 }
1104
1105 #define MSA_DO_B MSA_DO(b)
1106 #define MSA_DO_H MSA_DO(h)
1107 #define MSA_DO_W MSA_DO(w)
1108 #define MSA_DO_D MSA_DO(d)
1109
1110 #define MSA_LOOP_B MSA_LOOP(B)
1111 #define MSA_LOOP_H MSA_LOOP(H)
1112 #define MSA_LOOP_W MSA_LOOP(W)
1113 #define MSA_LOOP_D MSA_LOOP(D)
1114
1115 #define MSA_LOOP_COND_B MSA_LOOP_COND(DF_BYTE)
1116 #define MSA_LOOP_COND_H MSA_LOOP_COND(DF_HALF)
1117 #define MSA_LOOP_COND_W MSA_LOOP_COND(DF_WORD)
1118 #define MSA_LOOP_COND_D MSA_LOOP_COND(DF_DOUBLE)
1119
1120 #define MSA_LOOP(DF) \
1121 for (i = 0; i < (MSA_LOOP_COND_ ## DF) ; i++) { \
1122 MSA_DO_ ## DF \
1123 }
1124
1125 #define MSA_FN_DF(FUNC) \
1126 void helper_msa_##FUNC(CPUMIPSState *env, uint32_t df, uint32_t wd, \
1127 uint32_t ws, uint32_t wt) \
1128 { \
1129 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
1130 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
1131 wr_t *pwt = &(env->active_fpu.fpr[wt].wr); \
1132 wr_t wx, *pwx = &wx; \
1133 uint32_t i; \
1134 switch (df) { \
1135 case DF_BYTE: \
1136 MSA_LOOP_B \
1137 break; \
1138 case DF_HALF: \
1139 MSA_LOOP_H \
1140 break; \
1141 case DF_WORD: \
1142 MSA_LOOP_W \
1143 break; \
1144 case DF_DOUBLE: \
1145 MSA_LOOP_D \
1146 break; \
1147 default: \
1148 assert(0); \
1149 } \
1150 msa_move_v(pwd, pwx); \
1151 }
1152
1153 #define MSA_LOOP_COND(DF) \
1154 (DF_ELEMENTS(DF) / 2)
1155
1156 #define Rb(pwr, i) (pwr->b[i])
1157 #define Lb(pwr, i) (pwr->b[i + DF_ELEMENTS(DF_BYTE)/2])
1158 #define Rh(pwr, i) (pwr->h[i])
1159 #define Lh(pwr, i) (pwr->h[i + DF_ELEMENTS(DF_HALF)/2])
1160 #define Rw(pwr, i) (pwr->w[i])
1161 #define Lw(pwr, i) (pwr->w[i + DF_ELEMENTS(DF_WORD)/2])
1162 #define Rd(pwr, i) (pwr->d[i])
1163 #define Ld(pwr, i) (pwr->d[i + DF_ELEMENTS(DF_DOUBLE)/2])
1164
1165 #define MSA_DO(DF) \
1166 do { \
1167 R##DF(pwx, i) = pwt->DF[2*i]; \
1168 L##DF(pwx, i) = pws->DF[2*i]; \
1169 } while (0);
1170 MSA_FN_DF(pckev_df)
1171 #undef MSA_DO
1172
1173 #define MSA_DO(DF) \
1174 do { \
1175 R##DF(pwx, i) = pwt->DF[2*i+1]; \
1176 L##DF(pwx, i) = pws->DF[2*i+1]; \
1177 } while (0);
1178 MSA_FN_DF(pckod_df)
1179 #undef MSA_DO
1180
1181 #define MSA_DO(DF) \
1182 do { \
1183 pwx->DF[2*i] = L##DF(pwt, i); \
1184 pwx->DF[2*i+1] = L##DF(pws, i); \
1185 } while (0);
1186 MSA_FN_DF(ilvl_df)
1187 #undef MSA_DO
1188
1189 #define MSA_DO(DF) \
1190 do { \
1191 pwx->DF[2*i] = R##DF(pwt, i); \
1192 pwx->DF[2*i+1] = R##DF(pws, i); \
1193 } while (0);
1194 MSA_FN_DF(ilvr_df)
1195 #undef MSA_DO
1196
1197 #define MSA_DO(DF) \
1198 do { \
1199 pwx->DF[2*i] = pwt->DF[2*i]; \
1200 pwx->DF[2*i+1] = pws->DF[2*i]; \
1201 } while (0);
1202 MSA_FN_DF(ilvev_df)
1203 #undef MSA_DO
1204
1205 #define MSA_DO(DF) \
1206 do { \
1207 pwx->DF[2*i] = pwt->DF[2*i+1]; \
1208 pwx->DF[2*i+1] = pws->DF[2*i+1]; \
1209 } while (0);
1210 MSA_FN_DF(ilvod_df)
1211 #undef MSA_DO
1212 #undef MSA_LOOP_COND
1213
1214 #define MSA_LOOP_COND(DF) \
1215 (DF_ELEMENTS(DF))
1216
1217 #define MSA_DO(DF) \
1218 do { \
1219 uint32_t n = DF_ELEMENTS(df); \
1220 uint32_t k = (pwd->DF[i] & 0x3f) % (2 * n); \
1221 pwx->DF[i] = \
1222 (pwd->DF[i] & 0xc0) ? 0 : k < n ? pwt->DF[k] : pws->DF[k - n]; \
1223 } while (0);
1224 MSA_FN_DF(vshf_df)
1225 #undef MSA_DO
1226 #undef MSA_LOOP_COND
1227 #undef MSA_FN_DF
1228
1229 void helper_msa_sldi_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1230 uint32_t ws, uint32_t n)
1231 {
1232 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1233 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1234
1235 msa_sld_df(df, pwd, pws, n);
1236 }
1237
1238 void helper_msa_splati_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1239 uint32_t ws, uint32_t n)
1240 {
1241 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1242 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1243
1244 msa_splat_df(df, pwd, pws, n);
1245 }
1246
1247 void helper_msa_copy_s_df(CPUMIPSState *env, uint32_t df, uint32_t rd,
1248 uint32_t ws, uint32_t n)
1249 {
1250 n %= DF_ELEMENTS(df);
1251
1252 switch (df) {
1253 case DF_BYTE:
1254 env->active_tc.gpr[rd] = (int8_t)env->active_fpu.fpr[ws].wr.b[n];
1255 break;
1256 case DF_HALF:
1257 env->active_tc.gpr[rd] = (int16_t)env->active_fpu.fpr[ws].wr.h[n];
1258 break;
1259 case DF_WORD:
1260 env->active_tc.gpr[rd] = (int32_t)env->active_fpu.fpr[ws].wr.w[n];
1261 break;
1262 #ifdef TARGET_MIPS64
1263 case DF_DOUBLE:
1264 env->active_tc.gpr[rd] = (int64_t)env->active_fpu.fpr[ws].wr.d[n];
1265 break;
1266 #endif
1267 default:
1268 assert(0);
1269 }
1270 }
1271
1272 void helper_msa_copy_u_df(CPUMIPSState *env, uint32_t df, uint32_t rd,
1273 uint32_t ws, uint32_t n)
1274 {
1275 n %= DF_ELEMENTS(df);
1276
1277 switch (df) {
1278 case DF_BYTE:
1279 env->active_tc.gpr[rd] = (uint8_t)env->active_fpu.fpr[ws].wr.b[n];
1280 break;
1281 case DF_HALF:
1282 env->active_tc.gpr[rd] = (uint16_t)env->active_fpu.fpr[ws].wr.h[n];
1283 break;
1284 case DF_WORD:
1285 env->active_tc.gpr[rd] = (uint32_t)env->active_fpu.fpr[ws].wr.w[n];
1286 break;
1287 #ifdef TARGET_MIPS64
1288 case DF_DOUBLE:
1289 env->active_tc.gpr[rd] = (uint64_t)env->active_fpu.fpr[ws].wr.d[n];
1290 break;
1291 #endif
1292 default:
1293 assert(0);
1294 }
1295 }
1296
1297 void helper_msa_insert_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1298 uint32_t rs_num, uint32_t n)
1299 {
1300 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1301 target_ulong rs = env->active_tc.gpr[rs_num];
1302
1303 switch (df) {
1304 case DF_BYTE:
1305 pwd->b[n] = (int8_t)rs;
1306 break;
1307 case DF_HALF:
1308 pwd->h[n] = (int16_t)rs;
1309 break;
1310 case DF_WORD:
1311 pwd->w[n] = (int32_t)rs;
1312 break;
1313 case DF_DOUBLE:
1314 pwd->d[n] = (int64_t)rs;
1315 break;
1316 default:
1317 assert(0);
1318 }
1319 }
1320
1321 void helper_msa_insve_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1322 uint32_t ws, uint32_t n)
1323 {
1324 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1325 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1326
1327 switch (df) {
1328 case DF_BYTE:
1329 pwd->b[n] = (int8_t)pws->b[0];
1330 break;
1331 case DF_HALF:
1332 pwd->h[n] = (int16_t)pws->h[0];
1333 break;
1334 case DF_WORD:
1335 pwd->w[n] = (int32_t)pws->w[0];
1336 break;
1337 case DF_DOUBLE:
1338 pwd->d[n] = (int64_t)pws->d[0];
1339 break;
1340 default:
1341 assert(0);
1342 }
1343 }
1344
1345 void helper_msa_ctcmsa(CPUMIPSState *env, target_ulong elm, uint32_t cd)
1346 {
1347 switch (cd) {
1348 case 0:
1349 break;
1350 case 1:
1351 env->active_tc.msacsr = (int32_t)elm & MSACSR_MASK;
1352 restore_msa_fp_status(env);
1353 /* check exception */
1354 if ((GET_FP_ENABLE(env->active_tc.msacsr) | FP_UNIMPLEMENTED)
1355 & GET_FP_CAUSE(env->active_tc.msacsr)) {
1356 do_raise_exception(env, EXCP_MSAFPE, GETPC());
1357 }
1358 break;
1359 }
1360 }
1361
1362 target_ulong helper_msa_cfcmsa(CPUMIPSState *env, uint32_t cs)
1363 {
1364 switch (cs) {
1365 case 0:
1366 return env->msair;
1367 case 1:
1368 return env->active_tc.msacsr & MSACSR_MASK;
1369 }
1370 return 0;
1371 }
1372
1373 void helper_msa_move_v(CPUMIPSState *env, uint32_t wd, uint32_t ws)
1374 {
1375 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1376 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1377
1378 msa_move_v(pwd, pws);
1379 }
1380
1381 static inline int64_t msa_pcnt_df(uint32_t df, int64_t arg)
1382 {
1383 uint64_t x;
1384
1385 x = UNSIGNED(arg, df);
1386
1387 x = (x & 0x5555555555555555ULL) + ((x >> 1) & 0x5555555555555555ULL);
1388 x = (x & 0x3333333333333333ULL) + ((x >> 2) & 0x3333333333333333ULL);
1389 x = (x & 0x0F0F0F0F0F0F0F0FULL) + ((x >> 4) & 0x0F0F0F0F0F0F0F0FULL);
1390 x = (x & 0x00FF00FF00FF00FFULL) + ((x >> 8) & 0x00FF00FF00FF00FFULL);
1391 x = (x & 0x0000FFFF0000FFFFULL) + ((x >> 16) & 0x0000FFFF0000FFFFULL);
1392 x = (x & 0x00000000FFFFFFFFULL) + ((x >> 32));
1393
1394 return x;
1395 }
1396
1397 static inline int64_t msa_nlzc_df(uint32_t df, int64_t arg)
1398 {
1399 uint64_t x, y;
1400 int n, c;
1401
1402 x = UNSIGNED(arg, df);
1403 n = DF_BITS(df);
1404 c = DF_BITS(df) / 2;
1405
1406 do {
1407 y = x >> c;
1408 if (y != 0) {
1409 n = n - c;
1410 x = y;
1411 }
1412 c = c >> 1;
1413 } while (c != 0);
1414
1415 return n - x;
1416 }
1417
1418 static inline int64_t msa_nloc_df(uint32_t df, int64_t arg)
1419 {
1420 return msa_nlzc_df(df, UNSIGNED((~arg), df));
1421 }
1422
1423 void helper_msa_fill_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1424 uint32_t rs)
1425 {
1426 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1427 uint32_t i;
1428
1429 switch (df) {
1430 case DF_BYTE:
1431 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) {
1432 pwd->b[i] = (int8_t)env->active_tc.gpr[rs];
1433 }
1434 break;
1435 case DF_HALF:
1436 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) {
1437 pwd->h[i] = (int16_t)env->active_tc.gpr[rs];
1438 }
1439 break;
1440 case DF_WORD:
1441 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
1442 pwd->w[i] = (int32_t)env->active_tc.gpr[rs];
1443 }
1444 break;
1445 case DF_DOUBLE:
1446 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
1447 pwd->d[i] = (int64_t)env->active_tc.gpr[rs];
1448 }
1449 break;
1450 default:
1451 assert(0);
1452 }
1453 }
1454
1455 #define MSA_UNOP_DF(func) \
1456 void helper_msa_ ## func ## _df(CPUMIPSState *env, uint32_t df, \
1457 uint32_t wd, uint32_t ws) \
1458 { \
1459 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
1460 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
1461 uint32_t i; \
1462 \
1463 switch (df) { \
1464 case DF_BYTE: \
1465 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
1466 pwd->b[i] = msa_ ## func ## _df(df, pws->b[i]); \
1467 } \
1468 break; \
1469 case DF_HALF: \
1470 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
1471 pwd->h[i] = msa_ ## func ## _df(df, pws->h[i]); \
1472 } \
1473 break; \
1474 case DF_WORD: \
1475 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
1476 pwd->w[i] = msa_ ## func ## _df(df, pws->w[i]); \
1477 } \
1478 break; \
1479 case DF_DOUBLE: \
1480 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
1481 pwd->d[i] = msa_ ## func ## _df(df, pws->d[i]); \
1482 } \
1483 break; \
1484 default: \
1485 assert(0); \
1486 } \
1487 }
1488
1489 MSA_UNOP_DF(nlzc)
1490 MSA_UNOP_DF(nloc)
1491 MSA_UNOP_DF(pcnt)
1492 #undef MSA_UNOP_DF
1493
1494 #define FLOAT_ONE32 make_float32(0x3f8 << 20)
1495 #define FLOAT_ONE64 make_float64(0x3ffULL << 52)
1496
1497 #define FLOAT_SNAN16 (float16_default_nan ^ 0x0220)
1498 /* 0x7c20 */
1499 #define FLOAT_SNAN32 (float32_default_nan ^ 0x00400020)
1500 /* 0x7f800020 */
1501 #define FLOAT_SNAN64 (float64_default_nan ^ 0x0008000000000020ULL)
1502 /* 0x7ff0000000000020 */
1503
1504 static inline void clear_msacsr_cause(CPUMIPSState *env)
1505 {
1506 SET_FP_CAUSE(env->active_tc.msacsr, 0);
1507 }
1508
1509 static inline void check_msacsr_cause(CPUMIPSState *env, uintptr_t retaddr)
1510 {
1511 if ((GET_FP_CAUSE(env->active_tc.msacsr) &
1512 (GET_FP_ENABLE(env->active_tc.msacsr) | FP_UNIMPLEMENTED)) == 0) {
1513 UPDATE_FP_FLAGS(env->active_tc.msacsr,
1514 GET_FP_CAUSE(env->active_tc.msacsr));
1515 } else {
1516 do_raise_exception(env, EXCP_MSAFPE, retaddr);
1517 }
1518 }
1519
1520 /* Flush-to-zero use cases for update_msacsr() */
1521 #define CLEAR_FS_UNDERFLOW 1
1522 #define CLEAR_IS_INEXACT 2
1523 #define RECIPROCAL_INEXACT 4
1524
1525 static inline int update_msacsr(CPUMIPSState *env, int action, int denormal)
1526 {
1527 int ieee_ex;
1528
1529 int c;
1530 int cause;
1531 int enable;
1532
1533 ieee_ex = get_float_exception_flags(&env->active_tc.msa_fp_status);
1534
1535 /* QEMU softfloat does not signal all underflow cases */
1536 if (denormal) {
1537 ieee_ex |= float_flag_underflow;
1538 }
1539
1540 c = ieee_ex_to_mips(ieee_ex);
1541 enable = GET_FP_ENABLE(env->active_tc.msacsr) | FP_UNIMPLEMENTED;
1542
1543 /* Set Inexact (I) when flushing inputs to zero */
1544 if ((ieee_ex & float_flag_input_denormal) &&
1545 (env->active_tc.msacsr & MSACSR_FS_MASK) != 0) {
1546 if (action & CLEAR_IS_INEXACT) {
1547 c &= ~FP_INEXACT;
1548 } else {
1549 c |= FP_INEXACT;
1550 }
1551 }
1552
1553 /* Set Inexact (I) and Underflow (U) when flushing outputs to zero */
1554 if ((ieee_ex & float_flag_output_denormal) &&
1555 (env->active_tc.msacsr & MSACSR_FS_MASK) != 0) {
1556 c |= FP_INEXACT;
1557 if (action & CLEAR_FS_UNDERFLOW) {
1558 c &= ~FP_UNDERFLOW;
1559 } else {
1560 c |= FP_UNDERFLOW;
1561 }
1562 }
1563
1564 /* Set Inexact (I) when Overflow (O) is not enabled */
1565 if ((c & FP_OVERFLOW) != 0 && (enable & FP_OVERFLOW) == 0) {
1566 c |= FP_INEXACT;
1567 }
1568
1569 /* Clear Exact Underflow when Underflow (U) is not enabled */
1570 if ((c & FP_UNDERFLOW) != 0 && (enable & FP_UNDERFLOW) == 0 &&
1571 (c & FP_INEXACT) == 0) {
1572 c &= ~FP_UNDERFLOW;
1573 }
1574
1575 /* Reciprocal operations set only Inexact when valid and not
1576 divide by zero */
1577 if ((action & RECIPROCAL_INEXACT) &&
1578 (c & (FP_INVALID | FP_DIV0)) == 0) {
1579 c = FP_INEXACT;
1580 }
1581
1582 cause = c & enable; /* all current enabled exceptions */
1583
1584 if (cause == 0) {
1585 /* No enabled exception, update the MSACSR Cause
1586 with all current exceptions */
1587 SET_FP_CAUSE(env->active_tc.msacsr,
1588 (GET_FP_CAUSE(env->active_tc.msacsr) | c));
1589 } else {
1590 /* Current exceptions are enabled */
1591 if ((env->active_tc.msacsr & MSACSR_NX_MASK) == 0) {
1592 /* Exception(s) will trap, update MSACSR Cause
1593 with all enabled exceptions */
1594 SET_FP_CAUSE(env->active_tc.msacsr,
1595 (GET_FP_CAUSE(env->active_tc.msacsr) | c));
1596 }
1597 }
1598
1599 return c;
1600 }
1601
1602 static inline int get_enabled_exceptions(const CPUMIPSState *env, int c)
1603 {
1604 int enable = GET_FP_ENABLE(env->active_tc.msacsr) | FP_UNIMPLEMENTED;
1605 return c & enable;
1606 }
1607
1608 static inline float16 float16_from_float32(int32_t a, flag ieee,
1609 float_status *status)
1610 {
1611 float16 f_val;
1612
1613 f_val = float32_to_float16((float32)a, ieee, status);
1614 f_val = float16_maybe_silence_nan(f_val);
1615
1616 return a < 0 ? (f_val | (1 << 15)) : f_val;
1617 }
1618
1619 static inline float32 float32_from_float64(int64_t a, float_status *status)
1620 {
1621 float32 f_val;
1622
1623 f_val = float64_to_float32((float64)a, status);
1624 f_val = float32_maybe_silence_nan(f_val);
1625
1626 return a < 0 ? (f_val | (1 << 31)) : f_val;
1627 }
1628
1629 static inline float32 float32_from_float16(int16_t a, flag ieee,
1630 float_status *status)
1631 {
1632 float32 f_val;
1633
1634 f_val = float16_to_float32((float16)a, ieee, status);
1635 f_val = float32_maybe_silence_nan(f_val);
1636
1637 return a < 0 ? (f_val | (1 << 31)) : f_val;
1638 }
1639
1640 static inline float64 float64_from_float32(int32_t a, float_status *status)
1641 {
1642 float64 f_val;
1643
1644 f_val = float32_to_float64((float64)a, status);
1645 f_val = float64_maybe_silence_nan(f_val);
1646
1647 return a < 0 ? (f_val | (1ULL << 63)) : f_val;
1648 }
1649
1650 static inline float32 float32_from_q16(int16_t a, float_status *status)
1651 {
1652 float32 f_val;
1653
1654 /* conversion as integer and scaling */
1655 f_val = int32_to_float32(a, status);
1656 f_val = float32_scalbn(f_val, -15, status);
1657
1658 return f_val;
1659 }
1660
1661 static inline float64 float64_from_q32(int32_t a, float_status *status)
1662 {
1663 float64 f_val;
1664
1665 /* conversion as integer and scaling */
1666 f_val = int32_to_float64(a, status);
1667 f_val = float64_scalbn(f_val, -31, status);
1668
1669 return f_val;
1670 }
1671
1672 static inline int16_t float32_to_q16(float32 a, float_status *status)
1673 {
1674 int32_t q_val;
1675 int32_t q_min = 0xffff8000;
1676 int32_t q_max = 0x00007fff;
1677
1678 int ieee_ex;
1679
1680 if (float32_is_any_nan(a)) {
1681 float_raise(float_flag_invalid, status);
1682 return 0;
1683 }
1684
1685 /* scaling */
1686 a = float32_scalbn(a, 15, status);
1687
1688 ieee_ex = get_float_exception_flags(status);
1689 set_float_exception_flags(ieee_ex & (~float_flag_underflow)
1690 , status);
1691
1692 if (ieee_ex & float_flag_overflow) {
1693 float_raise(float_flag_inexact, status);
1694 return (int32_t)a < 0 ? q_min : q_max;
1695 }
1696
1697 /* conversion to int */
1698 q_val = float32_to_int32(a, status);
1699
1700 ieee_ex = get_float_exception_flags(status);
1701 set_float_exception_flags(ieee_ex & (~float_flag_underflow)
1702 , status);
1703
1704 if (ieee_ex & float_flag_invalid) {
1705 set_float_exception_flags(ieee_ex & (~float_flag_invalid)
1706 , status);
1707 float_raise(float_flag_overflow | float_flag_inexact, status);
1708 return (int32_t)a < 0 ? q_min : q_max;
1709 }
1710
1711 if (q_val < q_min) {
1712 float_raise(float_flag_overflow | float_flag_inexact, status);
1713 return (int16_t)q_min;
1714 }
1715
1716 if (q_max < q_val) {
1717 float_raise(float_flag_overflow | float_flag_inexact, status);
1718 return (int16_t)q_max;
1719 }
1720
1721 return (int16_t)q_val;
1722 }
1723
1724 static inline int32_t float64_to_q32(float64 a, float_status *status)
1725 {
1726 int64_t q_val;
1727 int64_t q_min = 0xffffffff80000000LL;
1728 int64_t q_max = 0x000000007fffffffLL;
1729
1730 int ieee_ex;
1731
1732 if (float64_is_any_nan(a)) {
1733 float_raise(float_flag_invalid, status);
1734 return 0;
1735 }
1736
1737 /* scaling */
1738 a = float64_scalbn(a, 31, status);
1739
1740 ieee_ex = get_float_exception_flags(status);
1741 set_float_exception_flags(ieee_ex & (~float_flag_underflow)
1742 , status);
1743
1744 if (ieee_ex & float_flag_overflow) {
1745 float_raise(float_flag_inexact, status);
1746 return (int64_t)a < 0 ? q_min : q_max;
1747 }
1748
1749 /* conversion to integer */
1750 q_val = float64_to_int64(a, status);
1751
1752 ieee_ex = get_float_exception_flags(status);
1753 set_float_exception_flags(ieee_ex & (~float_flag_underflow)
1754 , status);
1755
1756 if (ieee_ex & float_flag_invalid) {
1757 set_float_exception_flags(ieee_ex & (~float_flag_invalid)
1758 , status);
1759 float_raise(float_flag_overflow | float_flag_inexact, status);
1760 return (int64_t)a < 0 ? q_min : q_max;
1761 }
1762
1763 if (q_val < q_min) {
1764 float_raise(float_flag_overflow | float_flag_inexact, status);
1765 return (int32_t)q_min;
1766 }
1767
1768 if (q_max < q_val) {
1769 float_raise(float_flag_overflow | float_flag_inexact, status);
1770 return (int32_t)q_max;
1771 }
1772
1773 return (int32_t)q_val;
1774 }
1775
1776 #define MSA_FLOAT_COND(DEST, OP, ARG1, ARG2, BITS, QUIET) \
1777 do { \
1778 float_status *status = &env->active_tc.msa_fp_status; \
1779 int c; \
1780 int64_t cond; \
1781 set_float_exception_flags(0, status); \
1782 if (!QUIET) { \
1783 cond = float ## BITS ## _ ## OP(ARG1, ARG2, status); \
1784 } else { \
1785 cond = float ## BITS ## _ ## OP ## _quiet(ARG1, ARG2, status); \
1786 } \
1787 DEST = cond ? M_MAX_UINT(BITS) : 0; \
1788 c = update_msacsr(env, CLEAR_IS_INEXACT, 0); \
1789 \
1790 if (get_enabled_exceptions(env, c)) { \
1791 DEST = ((FLOAT_SNAN ## BITS >> 6) << 6) | c; \
1792 } \
1793 } while (0)
1794
1795 #define MSA_FLOAT_AF(DEST, ARG1, ARG2, BITS, QUIET) \
1796 do { \
1797 MSA_FLOAT_COND(DEST, eq, ARG1, ARG2, BITS, QUIET); \
1798 if ((DEST & M_MAX_UINT(BITS)) == M_MAX_UINT(BITS)) { \
1799 DEST = 0; \
1800 } \
1801 } while (0)
1802
1803 #define MSA_FLOAT_UEQ(DEST, ARG1, ARG2, BITS, QUIET) \
1804 do { \
1805 MSA_FLOAT_COND(DEST, unordered, ARG1, ARG2, BITS, QUIET); \
1806 if (DEST == 0) { \
1807 MSA_FLOAT_COND(DEST, eq, ARG1, ARG2, BITS, QUIET); \
1808 } \
1809 } while (0)
1810
1811 #define MSA_FLOAT_NE(DEST, ARG1, ARG2, BITS, QUIET) \
1812 do { \
1813 MSA_FLOAT_COND(DEST, lt, ARG1, ARG2, BITS, QUIET); \
1814 if (DEST == 0) { \
1815 MSA_FLOAT_COND(DEST, lt, ARG2, ARG1, BITS, QUIET); \
1816 } \
1817 } while (0)
1818
1819 #define MSA_FLOAT_UNE(DEST, ARG1, ARG2, BITS, QUIET) \
1820 do { \
1821 MSA_FLOAT_COND(DEST, unordered, ARG1, ARG2, BITS, QUIET); \
1822 if (DEST == 0) { \
1823 MSA_FLOAT_COND(DEST, lt, ARG1, ARG2, BITS, QUIET); \
1824 if (DEST == 0) { \
1825 MSA_FLOAT_COND(DEST, lt, ARG2, ARG1, BITS, QUIET); \
1826 } \
1827 } \
1828 } while (0)
1829
1830 #define MSA_FLOAT_ULE(DEST, ARG1, ARG2, BITS, QUIET) \
1831 do { \
1832 MSA_FLOAT_COND(DEST, unordered, ARG1, ARG2, BITS, QUIET); \
1833 if (DEST == 0) { \
1834 MSA_FLOAT_COND(DEST, le, ARG1, ARG2, BITS, QUIET); \
1835 } \
1836 } while (0)
1837
1838 #define MSA_FLOAT_ULT(DEST, ARG1, ARG2, BITS, QUIET) \
1839 do { \
1840 MSA_FLOAT_COND(DEST, unordered, ARG1, ARG2, BITS, QUIET); \
1841 if (DEST == 0) { \
1842 MSA_FLOAT_COND(DEST, lt, ARG1, ARG2, BITS, QUIET); \
1843 } \
1844 } while (0)
1845
1846 #define MSA_FLOAT_OR(DEST, ARG1, ARG2, BITS, QUIET) \
1847 do { \
1848 MSA_FLOAT_COND(DEST, le, ARG1, ARG2, BITS, QUIET); \
1849 if (DEST == 0) { \
1850 MSA_FLOAT_COND(DEST, le, ARG2, ARG1, BITS, QUIET); \
1851 } \
1852 } while (0)
1853
1854 static inline void compare_af(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
1855 wr_t *pwt, uint32_t df, int quiet,
1856 uintptr_t retaddr)
1857 {
1858 wr_t wx, *pwx = &wx;
1859 uint32_t i;
1860
1861 clear_msacsr_cause(env);
1862
1863 switch (df) {
1864 case DF_WORD:
1865 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
1866 MSA_FLOAT_AF(pwx->w[i], pws->w[i], pwt->w[i], 32, quiet);
1867 }
1868 break;
1869 case DF_DOUBLE:
1870 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
1871 MSA_FLOAT_AF(pwx->d[i], pws->d[i], pwt->d[i], 64, quiet);
1872 }
1873 break;
1874 default:
1875 assert(0);
1876 }
1877
1878 check_msacsr_cause(env, retaddr);
1879
1880 msa_move_v(pwd, pwx);
1881 }
1882
1883 static inline void compare_un(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
1884 wr_t *pwt, uint32_t df, int quiet,
1885 uintptr_t retaddr)
1886 {
1887 wr_t wx, *pwx = &wx;
1888 uint32_t i;
1889
1890 clear_msacsr_cause(env);
1891
1892 switch (df) {
1893 case DF_WORD:
1894 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
1895 MSA_FLOAT_COND(pwx->w[i], unordered, pws->w[i], pwt->w[i], 32,
1896 quiet);
1897 }
1898 break;
1899 case DF_DOUBLE:
1900 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
1901 MSA_FLOAT_COND(pwx->d[i], unordered, pws->d[i], pwt->d[i], 64,
1902 quiet);
1903 }
1904 break;
1905 default:
1906 assert(0);
1907 }
1908
1909 check_msacsr_cause(env, retaddr);
1910
1911 msa_move_v(pwd, pwx);
1912 }
1913
1914 static inline void compare_eq(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
1915 wr_t *pwt, uint32_t df, int quiet,
1916 uintptr_t retaddr)
1917 {
1918 wr_t wx, *pwx = &wx;
1919 uint32_t i;
1920
1921 clear_msacsr_cause(env);
1922
1923 switch (df) {
1924 case DF_WORD:
1925 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
1926 MSA_FLOAT_COND(pwx->w[i], eq, pws->w[i], pwt->w[i], 32, quiet);
1927 }
1928 break;
1929 case DF_DOUBLE:
1930 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
1931 MSA_FLOAT_COND(pwx->d[i], eq, pws->d[i], pwt->d[i], 64, quiet);
1932 }
1933 break;
1934 default:
1935 assert(0);
1936 }
1937
1938 check_msacsr_cause(env, retaddr);
1939
1940 msa_move_v(pwd, pwx);
1941 }
1942
1943 static inline void compare_ueq(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
1944 wr_t *pwt, uint32_t df, int quiet,
1945 uintptr_t retaddr)
1946 {
1947 wr_t wx, *pwx = &wx;
1948 uint32_t i;
1949
1950 clear_msacsr_cause(env);
1951
1952 switch (df) {
1953 case DF_WORD:
1954 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
1955 MSA_FLOAT_UEQ(pwx->w[i], pws->w[i], pwt->w[i], 32, quiet);
1956 }
1957 break;
1958 case DF_DOUBLE:
1959 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
1960 MSA_FLOAT_UEQ(pwx->d[i], pws->d[i], pwt->d[i], 64, quiet);
1961 }
1962 break;
1963 default:
1964 assert(0);
1965 }
1966
1967 check_msacsr_cause(env, retaddr);
1968
1969 msa_move_v(pwd, pwx);
1970 }
1971
1972 static inline void compare_lt(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
1973 wr_t *pwt, uint32_t df, int quiet,
1974 uintptr_t retaddr)
1975 {
1976 wr_t wx, *pwx = &wx;
1977 uint32_t i;
1978
1979 clear_msacsr_cause(env);
1980
1981 switch (df) {
1982 case DF_WORD:
1983 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
1984 MSA_FLOAT_COND(pwx->w[i], lt, pws->w[i], pwt->w[i], 32, quiet);
1985 }
1986 break;
1987 case DF_DOUBLE:
1988 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
1989 MSA_FLOAT_COND(pwx->d[i], lt, pws->d[i], pwt->d[i], 64, quiet);
1990 }
1991 break;
1992 default:
1993 assert(0);
1994 }
1995
1996 check_msacsr_cause(env, retaddr);
1997
1998 msa_move_v(pwd, pwx);
1999 }
2000
2001 static inline void compare_ult(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
2002 wr_t *pwt, uint32_t df, int quiet,
2003 uintptr_t retaddr)
2004 {
2005 wr_t wx, *pwx = &wx;
2006 uint32_t i;
2007
2008 clear_msacsr_cause(env);
2009
2010 switch (df) {
2011 case DF_WORD:
2012 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2013 MSA_FLOAT_ULT(pwx->w[i], pws->w[i], pwt->w[i], 32, quiet);
2014 }
2015 break;
2016 case DF_DOUBLE:
2017 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2018 MSA_FLOAT_ULT(pwx->d[i], pws->d[i], pwt->d[i], 64, quiet);
2019 }
2020 break;
2021 default:
2022 assert(0);
2023 }
2024
2025 check_msacsr_cause(env, retaddr);
2026
2027 msa_move_v(pwd, pwx);
2028 }
2029
2030 static inline void compare_le(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
2031 wr_t *pwt, uint32_t df, int quiet,
2032 uintptr_t retaddr)
2033 {
2034 wr_t wx, *pwx = &wx;
2035 uint32_t i;
2036
2037 clear_msacsr_cause(env);
2038
2039 switch (df) {
2040 case DF_WORD:
2041 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2042 MSA_FLOAT_COND(pwx->w[i], le, pws->w[i], pwt->w[i], 32, quiet);
2043 }
2044 break;
2045 case DF_DOUBLE:
2046 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2047 MSA_FLOAT_COND(pwx->d[i], le, pws->d[i], pwt->d[i], 64, quiet);
2048 }
2049 break;
2050 default:
2051 assert(0);
2052 }
2053
2054 check_msacsr_cause(env, retaddr);
2055
2056 msa_move_v(pwd, pwx);
2057 }
2058
2059 static inline void compare_ule(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
2060 wr_t *pwt, uint32_t df, int quiet,
2061 uintptr_t retaddr)
2062 {
2063 wr_t wx, *pwx = &wx;
2064 uint32_t i;
2065
2066 clear_msacsr_cause(env);
2067
2068 switch (df) {
2069 case DF_WORD:
2070 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2071 MSA_FLOAT_ULE(pwx->w[i], pws->w[i], pwt->w[i], 32, quiet);
2072 }
2073 break;
2074 case DF_DOUBLE:
2075 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2076 MSA_FLOAT_ULE(pwx->d[i], pws->d[i], pwt->d[i], 64, quiet);
2077 }
2078 break;
2079 default:
2080 assert(0);
2081 }
2082
2083 check_msacsr_cause(env, retaddr);
2084
2085 msa_move_v(pwd, pwx);
2086 }
2087
2088 static inline void compare_or(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
2089 wr_t *pwt, uint32_t df, int quiet,
2090 uintptr_t retaddr)
2091 {
2092 wr_t wx, *pwx = &wx;
2093 uint32_t i;
2094
2095 clear_msacsr_cause(env);
2096
2097 switch (df) {
2098 case DF_WORD:
2099 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2100 MSA_FLOAT_OR(pwx->w[i], pws->w[i], pwt->w[i], 32, quiet);
2101 }
2102 break;
2103 case DF_DOUBLE:
2104 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2105 MSA_FLOAT_OR(pwx->d[i], pws->d[i], pwt->d[i], 64, quiet);
2106 }
2107 break;
2108 default:
2109 assert(0);
2110 }
2111
2112 check_msacsr_cause(env, retaddr);
2113
2114 msa_move_v(pwd, pwx);
2115 }
2116
2117 static inline void compare_une(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
2118 wr_t *pwt, uint32_t df, int quiet,
2119 uintptr_t retaddr)
2120 {
2121 wr_t wx, *pwx = &wx;
2122 uint32_t i;
2123
2124 clear_msacsr_cause(env);
2125
2126 switch (df) {
2127 case DF_WORD:
2128 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2129 MSA_FLOAT_UNE(pwx->w[i], pws->w[i], pwt->w[i], 32, quiet);
2130 }
2131 break;
2132 case DF_DOUBLE:
2133 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2134 MSA_FLOAT_UNE(pwx->d[i], pws->d[i], pwt->d[i], 64, quiet);
2135 }
2136 break;
2137 default:
2138 assert(0);
2139 }
2140
2141 check_msacsr_cause(env, retaddr);
2142
2143 msa_move_v(pwd, pwx);
2144 }
2145
2146 static inline void compare_ne(CPUMIPSState *env, wr_t *pwd, wr_t *pws,
2147 wr_t *pwt, uint32_t df, int quiet,
2148 uintptr_t retaddr)
2149 {
2150 wr_t wx, *pwx = &wx;
2151 uint32_t i;
2152
2153 clear_msacsr_cause(env);
2154
2155 switch (df) {
2156 case DF_WORD:
2157 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2158 MSA_FLOAT_NE(pwx->w[i], pws->w[i], pwt->w[i], 32, quiet);
2159 }
2160 break;
2161 case DF_DOUBLE:
2162 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2163 MSA_FLOAT_NE(pwx->d[i], pws->d[i], pwt->d[i], 64, quiet);
2164 }
2165 break;
2166 default:
2167 assert(0);
2168 }
2169
2170 check_msacsr_cause(env, retaddr);
2171
2172 msa_move_v(pwd, pwx);
2173 }
2174
2175 void helper_msa_fcaf_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2176 uint32_t ws, uint32_t wt)
2177 {
2178 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2179 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2180 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2181 compare_af(env, pwd, pws, pwt, df, 1, GETPC());
2182 }
2183
2184 void helper_msa_fcun_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2185 uint32_t ws, uint32_t wt)
2186 {
2187 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2188 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2189 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2190 compare_un(env, pwd, pws, pwt, df, 1, GETPC());
2191 }
2192
2193 void helper_msa_fceq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2194 uint32_t ws, uint32_t wt)
2195 {
2196 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2197 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2198 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2199 compare_eq(env, pwd, pws, pwt, df, 1, GETPC());
2200 }
2201
2202 void helper_msa_fcueq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2203 uint32_t ws, uint32_t wt)
2204 {
2205 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2206 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2207 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2208 compare_ueq(env, pwd, pws, pwt, df, 1, GETPC());
2209 }
2210
2211 void helper_msa_fclt_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2212 uint32_t ws, uint32_t wt)
2213 {
2214 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2215 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2216 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2217 compare_lt(env, pwd, pws, pwt, df, 1, GETPC());
2218 }
2219
2220 void helper_msa_fcult_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2221 uint32_t ws, uint32_t wt)
2222 {
2223 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2224 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2225 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2226 compare_ult(env, pwd, pws, pwt, df, 1, GETPC());
2227 }
2228
2229 void helper_msa_fcle_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2230 uint32_t ws, uint32_t wt)
2231 {
2232 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2233 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2234 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2235 compare_le(env, pwd, pws, pwt, df, 1, GETPC());
2236 }
2237
2238 void helper_msa_fcule_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2239 uint32_t ws, uint32_t wt)
2240 {
2241 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2242 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2243 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2244 compare_ule(env, pwd, pws, pwt, df, 1, GETPC());
2245 }
2246
2247 void helper_msa_fsaf_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2248 uint32_t ws, uint32_t wt)
2249 {
2250 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2251 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2252 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2253 compare_af(env, pwd, pws, pwt, df, 0, GETPC());
2254 }
2255
2256 void helper_msa_fsun_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2257 uint32_t ws, uint32_t wt)
2258 {
2259 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2260 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2261 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2262 compare_un(env, pwd, pws, pwt, df, 0, GETPC());
2263 }
2264
2265 void helper_msa_fseq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2266 uint32_t ws, uint32_t wt)
2267 {
2268 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2269 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2270 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2271 compare_eq(env, pwd, pws, pwt, df, 0, GETPC());
2272 }
2273
2274 void helper_msa_fsueq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2275 uint32_t ws, uint32_t wt)
2276 {
2277 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2278 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2279 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2280 compare_ueq(env, pwd, pws, pwt, df, 0, GETPC());
2281 }
2282
2283 void helper_msa_fslt_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2284 uint32_t ws, uint32_t wt)
2285 {
2286 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2287 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2288 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2289 compare_lt(env, pwd, pws, pwt, df, 0, GETPC());
2290 }
2291
2292 void helper_msa_fsult_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2293 uint32_t ws, uint32_t wt)
2294 {
2295 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2296 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2297 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2298 compare_ult(env, pwd, pws, pwt, df, 0, GETPC());
2299 }
2300
2301 void helper_msa_fsle_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2302 uint32_t ws, uint32_t wt)
2303 {
2304 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2305 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2306 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2307 compare_le(env, pwd, pws, pwt, df, 0, GETPC());
2308 }
2309
2310 void helper_msa_fsule_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2311 uint32_t ws, uint32_t wt)
2312 {
2313 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2314 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2315 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2316 compare_ule(env, pwd, pws, pwt, df, 0, GETPC());
2317 }
2318
2319 void helper_msa_fcor_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2320 uint32_t ws, uint32_t wt)
2321 {
2322 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2323 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2324 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2325 compare_or(env, pwd, pws, pwt, df, 1, GETPC());
2326 }
2327
2328 void helper_msa_fcune_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2329 uint32_t ws, uint32_t wt)
2330 {
2331 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2332 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2333 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2334 compare_une(env, pwd, pws, pwt, df, 1, GETPC());
2335 }
2336
2337 void helper_msa_fcne_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2338 uint32_t ws, uint32_t wt)
2339 {
2340 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2341 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2342 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2343 compare_ne(env, pwd, pws, pwt, df, 1, GETPC());
2344 }
2345
2346 void helper_msa_fsor_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2347 uint32_t ws, uint32_t wt)
2348 {
2349 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2350 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2351 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2352 compare_or(env, pwd, pws, pwt, df, 0, GETPC());
2353 }
2354
2355 void helper_msa_fsune_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2356 uint32_t ws, uint32_t wt)
2357 {
2358 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2359 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2360 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2361 compare_une(env, pwd, pws, pwt, df, 0, GETPC());
2362 }
2363
2364 void helper_msa_fsne_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2365 uint32_t ws, uint32_t wt)
2366 {
2367 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2368 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2369 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2370 compare_ne(env, pwd, pws, pwt, df, 0, GETPC());
2371 }
2372
2373 #define float16_is_zero(ARG) 0
2374 #define float16_is_zero_or_denormal(ARG) 0
2375
2376 #define IS_DENORMAL(ARG, BITS) \
2377 (!float ## BITS ## _is_zero(ARG) \
2378 && float ## BITS ## _is_zero_or_denormal(ARG))
2379
2380 #define MSA_FLOAT_BINOP(DEST, OP, ARG1, ARG2, BITS) \
2381 do { \
2382 float_status *status = &env->active_tc.msa_fp_status; \
2383 int c; \
2384 \
2385 set_float_exception_flags(0, status); \
2386 DEST = float ## BITS ## _ ## OP(ARG1, ARG2, status); \
2387 c = update_msacsr(env, 0, IS_DENORMAL(DEST, BITS)); \
2388 \
2389 if (get_enabled_exceptions(env, c)) { \
2390 DEST = ((FLOAT_SNAN ## BITS >> 6) << 6) | c; \
2391 } \
2392 } while (0)
2393
2394 void helper_msa_fadd_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2395 uint32_t ws, uint32_t wt)
2396 {
2397 wr_t wx, *pwx = &wx;
2398 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2399 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2400 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2401 uint32_t i;
2402
2403 clear_msacsr_cause(env);
2404
2405 switch (df) {
2406 case DF_WORD:
2407 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2408 MSA_FLOAT_BINOP(pwx->w[i], add, pws->w[i], pwt->w[i], 32);
2409 }
2410 break;
2411 case DF_DOUBLE:
2412 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2413 MSA_FLOAT_BINOP(pwx->d[i], add, pws->d[i], pwt->d[i], 64);
2414 }
2415 break;
2416 default:
2417 assert(0);
2418 }
2419
2420 check_msacsr_cause(env, GETPC());
2421 msa_move_v(pwd, pwx);
2422 }
2423
2424 void helper_msa_fsub_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2425 uint32_t ws, uint32_t wt)
2426 {
2427 wr_t wx, *pwx = &wx;
2428 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2429 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2430 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2431 uint32_t i;
2432
2433 clear_msacsr_cause(env);
2434
2435 switch (df) {
2436 case DF_WORD:
2437 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2438 MSA_FLOAT_BINOP(pwx->w[i], sub, pws->w[i], pwt->w[i], 32);
2439 }
2440 break;
2441 case DF_DOUBLE:
2442 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2443 MSA_FLOAT_BINOP(pwx->d[i], sub, pws->d[i], pwt->d[i], 64);
2444 }
2445 break;
2446 default:
2447 assert(0);
2448 }
2449
2450 check_msacsr_cause(env, GETPC());
2451 msa_move_v(pwd, pwx);
2452 }
2453
2454 void helper_msa_fmul_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2455 uint32_t ws, uint32_t wt)
2456 {
2457 wr_t wx, *pwx = &wx;
2458 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2459 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2460 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2461 uint32_t i;
2462
2463 clear_msacsr_cause(env);
2464
2465 switch (df) {
2466 case DF_WORD:
2467 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2468 MSA_FLOAT_BINOP(pwx->w[i], mul, pws->w[i], pwt->w[i], 32);
2469 }
2470 break;
2471 case DF_DOUBLE:
2472 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2473 MSA_FLOAT_BINOP(pwx->d[i], mul, pws->d[i], pwt->d[i], 64);
2474 }
2475 break;
2476 default:
2477 assert(0);
2478 }
2479
2480 check_msacsr_cause(env, GETPC());
2481
2482 msa_move_v(pwd, pwx);
2483 }
2484
2485 void helper_msa_fdiv_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2486 uint32_t ws, uint32_t wt)
2487 {
2488 wr_t wx, *pwx = &wx;
2489 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2490 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2491 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2492 uint32_t i;
2493
2494 clear_msacsr_cause(env);
2495
2496 switch (df) {
2497 case DF_WORD:
2498 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2499 MSA_FLOAT_BINOP(pwx->w[i], div, pws->w[i], pwt->w[i], 32);
2500 }
2501 break;
2502 case DF_DOUBLE:
2503 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2504 MSA_FLOAT_BINOP(pwx->d[i], div, pws->d[i], pwt->d[i], 64);
2505 }
2506 break;
2507 default:
2508 assert(0);
2509 }
2510
2511 check_msacsr_cause(env, GETPC());
2512
2513 msa_move_v(pwd, pwx);
2514 }
2515
2516 #define MSA_FLOAT_MULADD(DEST, ARG1, ARG2, ARG3, NEGATE, BITS) \
2517 do { \
2518 float_status *status = &env->active_tc.msa_fp_status; \
2519 int c; \
2520 \
2521 set_float_exception_flags(0, status); \
2522 DEST = float ## BITS ## _muladd(ARG2, ARG3, ARG1, NEGATE, status); \
2523 c = update_msacsr(env, 0, IS_DENORMAL(DEST, BITS)); \
2524 \
2525 if (get_enabled_exceptions(env, c)) { \
2526 DEST = ((FLOAT_SNAN ## BITS >> 6) << 6) | c; \
2527 } \
2528 } while (0)
2529
2530 void helper_msa_fmadd_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2531 uint32_t ws, uint32_t wt)
2532 {
2533 wr_t wx, *pwx = &wx;
2534 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2535 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2536 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2537 uint32_t i;
2538
2539 clear_msacsr_cause(env);
2540
2541 switch (df) {
2542 case DF_WORD:
2543 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2544 MSA_FLOAT_MULADD(pwx->w[i], pwd->w[i],
2545 pws->w[i], pwt->w[i], 0, 32);
2546 }
2547 break;
2548 case DF_DOUBLE:
2549 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2550 MSA_FLOAT_MULADD(pwx->d[i], pwd->d[i],
2551 pws->d[i], pwt->d[i], 0, 64);
2552 }
2553 break;
2554 default:
2555 assert(0);
2556 }
2557
2558 check_msacsr_cause(env, GETPC());
2559
2560 msa_move_v(pwd, pwx);
2561 }
2562
2563 void helper_msa_fmsub_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2564 uint32_t ws, uint32_t wt)
2565 {
2566 wr_t wx, *pwx = &wx;
2567 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2568 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2569 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2570 uint32_t i;
2571
2572 clear_msacsr_cause(env);
2573
2574 switch (df) {
2575 case DF_WORD:
2576 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2577 MSA_FLOAT_MULADD(pwx->w[i], pwd->w[i],
2578 pws->w[i], pwt->w[i],
2579 float_muladd_negate_product, 32);
2580 }
2581 break;
2582 case DF_DOUBLE:
2583 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2584 MSA_FLOAT_MULADD(pwx->d[i], pwd->d[i],
2585 pws->d[i], pwt->d[i],
2586 float_muladd_negate_product, 64);
2587 }
2588 break;
2589 default:
2590 assert(0);
2591 }
2592
2593 check_msacsr_cause(env, GETPC());
2594
2595 msa_move_v(pwd, pwx);
2596 }
2597
2598 void helper_msa_fexp2_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2599 uint32_t ws, uint32_t wt)
2600 {
2601 wr_t wx, *pwx = &wx;
2602 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2603 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2604 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2605 uint32_t i;
2606
2607 clear_msacsr_cause(env);
2608
2609 switch (df) {
2610 case DF_WORD:
2611 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2612 MSA_FLOAT_BINOP(pwx->w[i], scalbn, pws->w[i],
2613 pwt->w[i] > 0x200 ? 0x200 :
2614 pwt->w[i] < -0x200 ? -0x200 : pwt->w[i],
2615 32);
2616 }
2617 break;
2618 case DF_DOUBLE:
2619 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2620 MSA_FLOAT_BINOP(pwx->d[i], scalbn, pws->d[i],
2621 pwt->d[i] > 0x1000 ? 0x1000 :
2622 pwt->d[i] < -0x1000 ? -0x1000 : pwt->d[i],
2623 64);
2624 }
2625 break;
2626 default:
2627 assert(0);
2628 }
2629
2630 check_msacsr_cause(env, GETPC());
2631
2632 msa_move_v(pwd, pwx);
2633 }
2634
2635 #define MSA_FLOAT_UNOP(DEST, OP, ARG, BITS) \
2636 do { \
2637 float_status *status = &env->active_tc.msa_fp_status; \
2638 int c; \
2639 \
2640 set_float_exception_flags(0, status); \
2641 DEST = float ## BITS ## _ ## OP(ARG, status); \
2642 c = update_msacsr(env, 0, IS_DENORMAL(DEST, BITS)); \
2643 \
2644 if (get_enabled_exceptions(env, c)) { \
2645 DEST = ((FLOAT_SNAN ## BITS >> 6) << 6) | c; \
2646 } \
2647 } while (0)
2648
2649 void helper_msa_fexdo_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2650 uint32_t ws, uint32_t wt)
2651 {
2652 wr_t wx, *pwx = &wx;
2653 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2654 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2655 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2656 uint32_t i;
2657
2658 clear_msacsr_cause(env);
2659
2660 switch (df) {
2661 case DF_WORD:
2662 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2663 /* Half precision floats come in two formats: standard
2664 IEEE and "ARM" format. The latter gains extra exponent
2665 range by omitting the NaN/Inf encodings. */
2666 flag ieee = 1;
2667
2668 MSA_FLOAT_BINOP(Lh(pwx, i), from_float32, pws->w[i], ieee, 16);
2669 MSA_FLOAT_BINOP(Rh(pwx, i), from_float32, pwt->w[i], ieee, 16);
2670 }
2671 break;
2672 case DF_DOUBLE:
2673 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2674 MSA_FLOAT_UNOP(Lw(pwx, i), from_float64, pws->d[i], 32);
2675 MSA_FLOAT_UNOP(Rw(pwx, i), from_float64, pwt->d[i], 32);
2676 }
2677 break;
2678 default:
2679 assert(0);
2680 }
2681
2682 check_msacsr_cause(env, GETPC());
2683 msa_move_v(pwd, pwx);
2684 }
2685
2686 #define MSA_FLOAT_UNOP_XD(DEST, OP, ARG, BITS, XBITS) \
2687 do { \
2688 float_status *status = &env->active_tc.msa_fp_status; \
2689 int c; \
2690 \
2691 set_float_exception_flags(0, status); \
2692 DEST = float ## BITS ## _ ## OP(ARG, status); \
2693 c = update_msacsr(env, CLEAR_FS_UNDERFLOW, 0); \
2694 \
2695 if (get_enabled_exceptions(env, c)) { \
2696 DEST = ((FLOAT_SNAN ## XBITS >> 6) << 6) | c; \
2697 } \
2698 } while (0)
2699
2700 void helper_msa_ftq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2701 uint32_t ws, uint32_t wt)
2702 {
2703 wr_t wx, *pwx = &wx;
2704 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2705 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2706 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2707 uint32_t i;
2708
2709 clear_msacsr_cause(env);
2710
2711 switch (df) {
2712 case DF_WORD:
2713 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2714 MSA_FLOAT_UNOP_XD(Lh(pwx, i), to_q16, pws->w[i], 32, 16);
2715 MSA_FLOAT_UNOP_XD(Rh(pwx, i), to_q16, pwt->w[i], 32, 16);
2716 }
2717 break;
2718 case DF_DOUBLE:
2719 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2720 MSA_FLOAT_UNOP_XD(Lw(pwx, i), to_q32, pws->d[i], 64, 32);
2721 MSA_FLOAT_UNOP_XD(Rw(pwx, i), to_q32, pwt->d[i], 64, 32);
2722 }
2723 break;
2724 default:
2725 assert(0);
2726 }
2727
2728 check_msacsr_cause(env, GETPC());
2729
2730 msa_move_v(pwd, pwx);
2731 }
2732
2733 #define NUMBER_QNAN_PAIR(ARG1, ARG2, BITS) \
2734 !float ## BITS ## _is_any_nan(ARG1) \
2735 && float ## BITS ## _is_quiet_nan(ARG2)
2736
2737 #define MSA_FLOAT_MAXOP(DEST, OP, ARG1, ARG2, BITS) \
2738 do { \
2739 float_status *status = &env->active_tc.msa_fp_status; \
2740 int c; \
2741 \
2742 set_float_exception_flags(0, status); \
2743 DEST = float ## BITS ## _ ## OP(ARG1, ARG2, status); \
2744 c = update_msacsr(env, 0, 0); \
2745 \
2746 if (get_enabled_exceptions(env, c)) { \
2747 DEST = ((FLOAT_SNAN ## BITS >> 6) << 6) | c; \
2748 } \
2749 } while (0)
2750
2751 #define FMAXMIN_A(F, G, X, _S, _T, BITS) \
2752 do { \
2753 uint## BITS ##_t S = _S, T = _T; \
2754 uint## BITS ##_t as, at, xs, xt, xd; \
2755 if (NUMBER_QNAN_PAIR(S, T, BITS)) { \
2756 T = S; \
2757 } \
2758 else if (NUMBER_QNAN_PAIR(T, S, BITS)) { \
2759 S = T; \
2760 } \
2761 as = float## BITS ##_abs(S); \
2762 at = float## BITS ##_abs(T); \
2763 MSA_FLOAT_MAXOP(xs, F, S, T, BITS); \
2764 MSA_FLOAT_MAXOP(xt, G, S, T, BITS); \
2765 MSA_FLOAT_MAXOP(xd, F, as, at, BITS); \
2766 X = (as == at || xd == float## BITS ##_abs(xs)) ? xs : xt; \
2767 } while (0)
2768
2769 void helper_msa_fmin_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2770 uint32_t ws, uint32_t wt)
2771 {
2772 wr_t wx, *pwx = &wx;
2773 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2774 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2775 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2776 uint32_t i;
2777
2778 clear_msacsr_cause(env);
2779
2780 switch (df) {
2781 case DF_WORD:
2782 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2783 if (NUMBER_QNAN_PAIR(pws->w[i], pwt->w[i], 32)) {
2784 MSA_FLOAT_MAXOP(pwx->w[i], min, pws->w[i], pws->w[i], 32);
2785 } else if (NUMBER_QNAN_PAIR(pwt->w[i], pws->w[i], 32)) {
2786 MSA_FLOAT_MAXOP(pwx->w[i], min, pwt->w[i], pwt->w[i], 32);
2787 } else {
2788 MSA_FLOAT_MAXOP(pwx->w[i], min, pws->w[i], pwt->w[i], 32);
2789 }
2790 }
2791 break;
2792 case DF_DOUBLE:
2793 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2794 if (NUMBER_QNAN_PAIR(pws->d[i], pwt->d[i], 64)) {
2795 MSA_FLOAT_MAXOP(pwx->d[i], min, pws->d[i], pws->d[i], 64);
2796 } else if (NUMBER_QNAN_PAIR(pwt->d[i], pws->d[i], 64)) {
2797 MSA_FLOAT_MAXOP(pwx->d[i], min, pwt->d[i], pwt->d[i], 64);
2798 } else {
2799 MSA_FLOAT_MAXOP(pwx->d[i], min, pws->d[i], pwt->d[i], 64);
2800 }
2801 }
2802 break;
2803 default:
2804 assert(0);
2805 }
2806
2807 check_msacsr_cause(env, GETPC());
2808
2809 msa_move_v(pwd, pwx);
2810 }
2811
2812 void helper_msa_fmin_a_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2813 uint32_t ws, uint32_t wt)
2814 {
2815 wr_t wx, *pwx = &wx;
2816 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2817 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2818 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2819 uint32_t i;
2820
2821 clear_msacsr_cause(env);
2822
2823 switch (df) {
2824 case DF_WORD:
2825 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2826 FMAXMIN_A(min, max, pwx->w[i], pws->w[i], pwt->w[i], 32);
2827 }
2828 break;
2829 case DF_DOUBLE:
2830 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2831 FMAXMIN_A(min, max, pwx->d[i], pws->d[i], pwt->d[i], 64);
2832 }
2833 break;
2834 default:
2835 assert(0);
2836 }
2837
2838 check_msacsr_cause(env, GETPC());
2839
2840 msa_move_v(pwd, pwx);
2841 }
2842
2843 void helper_msa_fmax_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2844 uint32_t ws, uint32_t wt)
2845 {
2846 wr_t wx, *pwx = &wx;
2847 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2848 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2849 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2850 uint32_t i;
2851
2852 clear_msacsr_cause(env);
2853
2854 switch (df) {
2855 case DF_WORD:
2856 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2857 if (NUMBER_QNAN_PAIR(pws->w[i], pwt->w[i], 32)) {
2858 MSA_FLOAT_MAXOP(pwx->w[i], max, pws->w[i], pws->w[i], 32);
2859 } else if (NUMBER_QNAN_PAIR(pwt->w[i], pws->w[i], 32)) {
2860 MSA_FLOAT_MAXOP(pwx->w[i], max, pwt->w[i], pwt->w[i], 32);
2861 } else {
2862 MSA_FLOAT_MAXOP(pwx->w[i], max, pws->w[i], pwt->w[i], 32);
2863 }
2864 }
2865 break;
2866 case DF_DOUBLE:
2867 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2868 if (NUMBER_QNAN_PAIR(pws->d[i], pwt->d[i], 64)) {
2869 MSA_FLOAT_MAXOP(pwx->d[i], max, pws->d[i], pws->d[i], 64);
2870 } else if (NUMBER_QNAN_PAIR(pwt->d[i], pws->d[i], 64)) {
2871 MSA_FLOAT_MAXOP(pwx->d[i], max, pwt->d[i], pwt->d[i], 64);
2872 } else {
2873 MSA_FLOAT_MAXOP(pwx->d[i], max, pws->d[i], pwt->d[i], 64);
2874 }
2875 }
2876 break;
2877 default:
2878 assert(0);
2879 }
2880
2881 check_msacsr_cause(env, GETPC());
2882
2883 msa_move_v(pwd, pwx);
2884 }
2885
2886 void helper_msa_fmax_a_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2887 uint32_t ws, uint32_t wt)
2888 {
2889 wr_t wx, *pwx = &wx;
2890 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2891 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2892 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2893 uint32_t i;
2894
2895 clear_msacsr_cause(env);
2896
2897 switch (df) {
2898 case DF_WORD:
2899 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2900 FMAXMIN_A(max, min, pwx->w[i], pws->w[i], pwt->w[i], 32);
2901 }
2902 break;
2903 case DF_DOUBLE:
2904 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2905 FMAXMIN_A(max, min, pwx->d[i], pws->d[i], pwt->d[i], 64);
2906 }
2907 break;
2908 default:
2909 assert(0);
2910 }
2911
2912 check_msacsr_cause(env, GETPC());
2913
2914 msa_move_v(pwd, pwx);
2915 }
2916
2917 void helper_msa_fclass_df(CPUMIPSState *env, uint32_t df,
2918 uint32_t wd, uint32_t ws)
2919 {
2920 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2921 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2922 if (df == DF_WORD) {
2923 pwd->w[0] = helper_float_class_s(pws->w[0]);
2924 pwd->w[1] = helper_float_class_s(pws->w[1]);
2925 pwd->w[2] = helper_float_class_s(pws->w[2]);
2926 pwd->w[3] = helper_float_class_s(pws->w[3]);
2927 } else {
2928 pwd->d[0] = helper_float_class_d(pws->d[0]);
2929 pwd->d[1] = helper_float_class_d(pws->d[1]);
2930 }
2931 }
2932
2933 #define MSA_FLOAT_UNOP0(DEST, OP, ARG, BITS) \
2934 do { \
2935 float_status *status = &env->active_tc.msa_fp_status; \
2936 int c; \
2937 \
2938 set_float_exception_flags(0, status); \
2939 DEST = float ## BITS ## _ ## OP(ARG, status); \
2940 c = update_msacsr(env, CLEAR_FS_UNDERFLOW, 0); \
2941 \
2942 if (get_enabled_exceptions(env, c)) { \
2943 DEST = ((FLOAT_SNAN ## BITS >> 6) << 6) | c; \
2944 } else if (float ## BITS ## _is_any_nan(ARG)) { \
2945 DEST = 0; \
2946 } \
2947 } while (0)
2948
2949 void helper_msa_ftrunc_s_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2950 uint32_t ws)
2951 {
2952 wr_t wx, *pwx = &wx;
2953 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2954 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2955 uint32_t i;
2956
2957 clear_msacsr_cause(env);
2958
2959 switch (df) {
2960 case DF_WORD:
2961 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2962 MSA_FLOAT_UNOP0(pwx->w[i], to_int32_round_to_zero, pws->w[i], 32);
2963 }
2964 break;
2965 case DF_DOUBLE:
2966 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2967 MSA_FLOAT_UNOP0(pwx->d[i], to_int64_round_to_zero, pws->d[i], 64);
2968 }
2969 break;
2970 default:
2971 assert(0);
2972 }
2973
2974 check_msacsr_cause(env, GETPC());
2975
2976 msa_move_v(pwd, pwx);
2977 }
2978
2979 void helper_msa_ftrunc_u_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2980 uint32_t ws)
2981 {
2982 wr_t wx, *pwx = &wx;
2983 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2984 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2985 uint32_t i;
2986
2987 clear_msacsr_cause(env);
2988
2989 switch (df) {
2990 case DF_WORD:
2991 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2992 MSA_FLOAT_UNOP0(pwx->w[i], to_uint32_round_to_zero, pws->w[i], 32);
2993 }
2994 break;
2995 case DF_DOUBLE:
2996 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2997 MSA_FLOAT_UNOP0(pwx->d[i], to_uint64_round_to_zero, pws->d[i], 64);
2998 }
2999 break;
3000 default:
3001 assert(0);
3002 }
3003
3004 check_msacsr_cause(env, GETPC());
3005
3006 msa_move_v(pwd, pwx);
3007 }
3008
3009 void helper_msa_fsqrt_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3010 uint32_t ws)
3011 {
3012 wr_t wx, *pwx = &wx;
3013 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3014 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3015 uint32_t i;
3016
3017 clear_msacsr_cause(env);
3018
3019 switch (df) {
3020 case DF_WORD:
3021 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3022 MSA_FLOAT_UNOP(pwx->w[i], sqrt, pws->w[i], 32);
3023 }
3024 break;
3025 case DF_DOUBLE:
3026 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3027 MSA_FLOAT_UNOP(pwx->d[i], sqrt, pws->d[i], 64);
3028 }
3029 break;
3030 default:
3031 assert(0);
3032 }
3033
3034 check_msacsr_cause(env, GETPC());
3035
3036 msa_move_v(pwd, pwx);
3037 }
3038
3039 #define MSA_FLOAT_RECIPROCAL(DEST, ARG, BITS) \
3040 do { \
3041 float_status *status = &env->active_tc.msa_fp_status; \
3042 int c; \
3043 \
3044 set_float_exception_flags(0, status); \
3045 DEST = float ## BITS ## _ ## div(FLOAT_ONE ## BITS, ARG, status); \
3046 c = update_msacsr(env, float ## BITS ## _is_infinity(ARG) || \
3047 float ## BITS ## _is_quiet_nan(DEST) ? \
3048 0 : RECIPROCAL_INEXACT, \
3049 IS_DENORMAL(DEST, BITS)); \
3050 \
3051 if (get_enabled_exceptions(env, c)) { \
3052 DEST = ((FLOAT_SNAN ## BITS >> 6) << 6) | c; \
3053 } \
3054 } while (0)
3055
3056 void helper_msa_frsqrt_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3057 uint32_t ws)
3058 {
3059 wr_t wx, *pwx = &wx;
3060 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3061 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3062 uint32_t i;
3063
3064 clear_msacsr_cause(env);
3065
3066 switch (df) {
3067 case DF_WORD:
3068 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3069 MSA_FLOAT_RECIPROCAL(pwx->w[i], float32_sqrt(pws->w[i],
3070 &env->active_tc.msa_fp_status), 32);
3071 }
3072 break;
3073 case DF_DOUBLE:
3074 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3075 MSA_FLOAT_RECIPROCAL(pwx->d[i], float64_sqrt(pws->d[i],
3076 &env->active_tc.msa_fp_status), 64);
3077 }
3078 break;
3079 default:
3080 assert(0);
3081 }
3082
3083 check_msacsr_cause(env, GETPC());
3084
3085 msa_move_v(pwd, pwx);
3086 }
3087
3088 void helper_msa_frcp_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3089 uint32_t ws)
3090 {
3091 wr_t wx, *pwx = &wx;
3092 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3093 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3094 uint32_t i;
3095
3096 clear_msacsr_cause(env);
3097
3098 switch (df) {
3099 case DF_WORD:
3100 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3101 MSA_FLOAT_RECIPROCAL(pwx->w[i], pws->w[i], 32);
3102 }
3103 break;
3104 case DF_DOUBLE:
3105 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3106 MSA_FLOAT_RECIPROCAL(pwx->d[i], pws->d[i], 64);
3107 }
3108 break;
3109 default:
3110 assert(0);
3111 }
3112
3113 check_msacsr_cause(env, GETPC());
3114
3115 msa_move_v(pwd, pwx);
3116 }
3117
3118 void helper_msa_frint_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3119 uint32_t ws)
3120 {
3121 wr_t wx, *pwx = &wx;
3122 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3123 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3124 uint32_t i;
3125
3126 clear_msacsr_cause(env);
3127
3128 switch (df) {
3129 case DF_WORD:
3130 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3131 MSA_FLOAT_UNOP(pwx->w[i], round_to_int, pws->w[i], 32);
3132 }
3133 break;
3134 case DF_DOUBLE:
3135 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3136 MSA_FLOAT_UNOP(pwx->d[i], round_to_int, pws->d[i], 64);
3137 }
3138 break;
3139 default:
3140 assert(0);
3141 }
3142
3143 check_msacsr_cause(env, GETPC());
3144
3145 msa_move_v(pwd, pwx);
3146 }
3147
3148 #define MSA_FLOAT_LOGB(DEST, ARG, BITS) \
3149 do { \
3150 float_status *status = &env->active_tc.msa_fp_status; \
3151 int c; \
3152 \
3153 set_float_exception_flags(0, status); \
3154 set_float_rounding_mode(float_round_down, status); \
3155 DEST = float ## BITS ## _ ## log2(ARG, status); \
3156 DEST = float ## BITS ## _ ## round_to_int(DEST, status); \
3157 set_float_rounding_mode(ieee_rm[(env->active_tc.msacsr & \
3158 MSACSR_RM_MASK) >> MSACSR_RM], \
3159 status); \
3160 \
3161 set_float_exception_flags(get_float_exception_flags(status) & \
3162 (~float_flag_inexact), \
3163 status); \
3164 \
3165 c = update_msacsr(env, 0, IS_DENORMAL(DEST, BITS)); \
3166 \
3167 if (get_enabled_exceptions(env, c)) { \
3168 DEST = ((FLOAT_SNAN ## BITS >> 6) << 6) | c; \
3169 } \
3170 } while (0)
3171
3172 void helper_msa_flog2_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3173 uint32_t ws)
3174 {
3175 wr_t wx, *pwx = &wx;
3176 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3177 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3178 uint32_t i;
3179
3180 clear_msacsr_cause(env);
3181
3182 switch (df) {
3183 case DF_WORD:
3184 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3185 MSA_FLOAT_LOGB(pwx->w[i], pws->w[i], 32);
3186 }
3187 break;
3188 case DF_DOUBLE:
3189 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3190 MSA_FLOAT_LOGB(pwx->d[i], pws->d[i], 64);
3191 }
3192 break;
3193 default:
3194 assert(0);
3195 }
3196
3197 check_msacsr_cause(env, GETPC());
3198
3199 msa_move_v(pwd, pwx);
3200 }
3201
3202 void helper_msa_fexupl_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3203 uint32_t ws)
3204 {
3205 wr_t wx, *pwx = &wx;
3206 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3207 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3208 uint32_t i;
3209
3210 clear_msacsr_cause(env);
3211
3212 switch (df) {
3213 case DF_WORD:
3214 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3215 /* Half precision floats come in two formats: standard
3216 IEEE and "ARM" format. The latter gains extra exponent
3217 range by omitting the NaN/Inf encodings. */
3218 flag ieee = 1;
3219
3220 MSA_FLOAT_BINOP(pwx->w[i], from_float16, Lh(pws, i), ieee, 32);
3221 }
3222 break;
3223 case DF_DOUBLE:
3224 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3225 MSA_FLOAT_UNOP(pwx->d[i], from_float32, Lw(pws, i), 64);
3226 }
3227 break;
3228 default:
3229 assert(0);
3230 }
3231
3232 check_msacsr_cause(env, GETPC());
3233 msa_move_v(pwd, pwx);
3234 }
3235
3236 void helper_msa_fexupr_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3237 uint32_t ws)
3238 {
3239 wr_t wx, *pwx = &wx;
3240 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3241 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3242 uint32_t i;
3243
3244 clear_msacsr_cause(env);
3245
3246 switch (df) {
3247 case DF_WORD:
3248 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3249 /* Half precision floats come in two formats: standard
3250 IEEE and "ARM" format. The latter gains extra exponent
3251 range by omitting the NaN/Inf encodings. */
3252 flag ieee = 1;
3253
3254 MSA_FLOAT_BINOP(pwx->w[i], from_float16, Rh(pws, i), ieee, 32);
3255 }
3256 break;
3257 case DF_DOUBLE:
3258 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3259 MSA_FLOAT_UNOP(pwx->d[i], from_float32, Rw(pws, i), 64);
3260 }
3261 break;
3262 default:
3263 assert(0);
3264 }
3265
3266 check_msacsr_cause(env, GETPC());
3267 msa_move_v(pwd, pwx);
3268 }
3269
3270 void helper_msa_ffql_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3271 uint32_t ws)
3272 {
3273 wr_t wx, *pwx = &wx;
3274 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3275 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3276 uint32_t i;
3277
3278 switch (df) {
3279 case DF_WORD:
3280 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3281 MSA_FLOAT_UNOP(pwx->w[i], from_q16, Lh(pws, i), 32);
3282 }
3283 break;
3284 case DF_DOUBLE:
3285 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3286 MSA_FLOAT_UNOP(pwx->d[i], from_q32, Lw(pws, i), 64);
3287 }
3288 break;
3289 default:
3290 assert(0);
3291 }
3292
3293 msa_move_v(pwd, pwx);
3294 }
3295
3296 void helper_msa_ffqr_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3297 uint32_t ws)
3298 {
3299 wr_t wx, *pwx = &wx;
3300 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3301 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3302 uint32_t i;
3303
3304 switch (df) {
3305 case DF_WORD:
3306 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3307 MSA_FLOAT_UNOP(pwx->w[i], from_q16, Rh(pws, i), 32);
3308 }
3309 break;
3310 case DF_DOUBLE:
3311 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3312 MSA_FLOAT_UNOP(pwx->d[i], from_q32, Rw(pws, i), 64);
3313 }
3314 break;
3315 default:
3316 assert(0);
3317 }
3318
3319 msa_move_v(pwd, pwx);
3320 }
3321
3322 void helper_msa_ftint_s_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3323 uint32_t ws)
3324 {
3325 wr_t wx, *pwx = &wx;
3326 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3327 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3328 uint32_t i;
3329
3330 clear_msacsr_cause(env);
3331
3332 switch (df) {
3333 case DF_WORD:
3334 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3335 MSA_FLOAT_UNOP0(pwx->w[i], to_int32, pws->w[i], 32);
3336 }
3337 break;
3338 case DF_DOUBLE:
3339 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3340 MSA_FLOAT_UNOP0(pwx->d[i], to_int64, pws->d[i], 64);
3341 }
3342 break;
3343 default:
3344 assert(0);
3345 }
3346
3347 check_msacsr_cause(env, GETPC());
3348
3349 msa_move_v(pwd, pwx);
3350 }
3351
3352 void helper_msa_ftint_u_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3353 uint32_t ws)
3354 {
3355 wr_t wx, *pwx = &wx;
3356 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3357 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3358 uint32_t i;
3359
3360 clear_msacsr_cause(env);
3361
3362 switch (df) {
3363 case DF_WORD:
3364 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3365 MSA_FLOAT_UNOP0(pwx->w[i], to_uint32, pws->w[i], 32);
3366 }
3367 break;
3368 case DF_DOUBLE:
3369 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3370 MSA_FLOAT_UNOP0(pwx->d[i], to_uint64, pws->d[i], 64);
3371 }
3372 break;
3373 default:
3374 assert(0);
3375 }
3376
3377 check_msacsr_cause(env, GETPC());
3378
3379 msa_move_v(pwd, pwx);
3380 }
3381
3382 #define float32_from_int32 int32_to_float32
3383 #define float32_from_uint32 uint32_to_float32
3384
3385 #define float64_from_int64 int64_to_float64
3386 #define float64_from_uint64 uint64_to_float64
3387
3388 void helper_msa_ffint_s_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3389 uint32_t ws)
3390 {
3391 wr_t wx, *pwx = &wx;
3392 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3393 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3394 uint32_t i;
3395
3396 clear_msacsr_cause(env);
3397
3398 switch (df) {
3399 case DF_WORD:
3400 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3401 MSA_FLOAT_UNOP(pwx->w[i], from_int32, pws->w[i], 32);
3402 }
3403 break;
3404 case DF_DOUBLE:
3405 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3406 MSA_FLOAT_UNOP(pwx->d[i], from_int64, pws->d[i], 64);
3407 }
3408 break;
3409 default:
3410 assert(0);
3411 }
3412
3413 check_msacsr_cause(env, GETPC());
3414
3415 msa_move_v(pwd, pwx);
3416 }
3417
3418 void helper_msa_ffint_u_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3419 uint32_t ws)
3420 {
3421 wr_t wx, *pwx = &wx;
3422 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3423 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3424 uint32_t i;
3425
3426 clear_msacsr_cause(env);
3427
3428 switch (df) {
3429 case DF_WORD:
3430 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3431 MSA_FLOAT_UNOP(pwx->w[i], from_uint32, pws->w[i], 32);
3432 }
3433 break;
3434 case DF_DOUBLE:
3435 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3436 MSA_FLOAT_UNOP(pwx->d[i], from_uint64, pws->d[i], 64);
3437 }
3438 break;
3439 default:
3440 assert(0);
3441 }
3442
3443 check_msacsr_cause(env, GETPC());
3444
3445 msa_move_v(pwd, pwx);
3446 }
AR7 emulation for QEMU