hw/s390/css: avoid taking address members in packed structs
[qemu/ar7.git] / target / mips / msa_helper.c
blobc74e3cdc65bb6c81c583598a17d4e13b7069b847
1 /*
2 * MIPS SIMD Architecture Module Instruction emulation helpers for QEMU.
4 * Copyright (c) 2014 Imagination Technologies
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.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
21 #include "cpu.h"
22 #include "internal.h"
23 #include "exec/exec-all.h"
24 #include "exec/helper-proto.h"
26 /* Data format min and max values */
27 #define DF_BITS(df) (1 << ((df) + 3))
29 #define DF_MAX_INT(df) (int64_t)((1LL << (DF_BITS(df) - 1)) - 1)
30 #define M_MAX_INT(m) (int64_t)((1LL << ((m) - 1)) - 1)
32 #define DF_MIN_INT(df) (int64_t)(-(1LL << (DF_BITS(df) - 1)))
33 #define M_MIN_INT(m) (int64_t)(-(1LL << ((m) - 1)))
35 #define DF_MAX_UINT(df) (uint64_t)(-1ULL >> (64 - DF_BITS(df)))
36 #define M_MAX_UINT(m) (uint64_t)(-1ULL >> (64 - (m)))
38 #define UNSIGNED(x, df) ((x) & DF_MAX_UINT(df))
39 #define SIGNED(x, df) \
40 ((((int64_t)x) << (64 - DF_BITS(df))) >> (64 - DF_BITS(df)))
42 /* Element-by-element access macros */
43 #define DF_ELEMENTS(df) (MSA_WRLEN / DF_BITS(df))
45 static inline void msa_move_v(wr_t *pwd, wr_t *pws)
47 uint32_t i;
49 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
50 pwd->d[i] = pws->d[i];
54 #define MSA_FN_IMM8(FUNC, DEST, OPERATION) \
55 void helper_msa_ ## FUNC(CPUMIPSState *env, uint32_t wd, uint32_t ws, \
56 uint32_t i8) \
57 { \
58 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
59 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
60 uint32_t i; \
61 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
62 DEST = OPERATION; \
63 } \
66 MSA_FN_IMM8(andi_b, pwd->b[i], pws->b[i] & i8)
67 MSA_FN_IMM8(ori_b, pwd->b[i], pws->b[i] | i8)
68 MSA_FN_IMM8(nori_b, pwd->b[i], ~(pws->b[i] | i8))
69 MSA_FN_IMM8(xori_b, pwd->b[i], pws->b[i] ^ i8)
71 #define BIT_MOVE_IF_NOT_ZERO(dest, arg1, arg2, df) \
72 UNSIGNED(((dest & (~arg2)) | (arg1 & arg2)), df)
73 MSA_FN_IMM8(bmnzi_b, pwd->b[i],
74 BIT_MOVE_IF_NOT_ZERO(pwd->b[i], pws->b[i], i8, DF_BYTE))
76 #define BIT_MOVE_IF_ZERO(dest, arg1, arg2, df) \
77 UNSIGNED((dest & arg2) | (arg1 & (~arg2)), df)
78 MSA_FN_IMM8(bmzi_b, pwd->b[i],
79 BIT_MOVE_IF_ZERO(pwd->b[i], pws->b[i], i8, DF_BYTE))
81 #define BIT_SELECT(dest, arg1, arg2, df) \
82 UNSIGNED((arg1 & (~dest)) | (arg2 & dest), df)
83 MSA_FN_IMM8(bseli_b, pwd->b[i],
84 BIT_SELECT(pwd->b[i], pws->b[i], i8, DF_BYTE))
86 #undef MSA_FN_IMM8
88 #define SHF_POS(i, imm) (((i) & 0xfc) + (((imm) >> (2 * ((i) & 0x03))) & 0x03))
90 void helper_msa_shf_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
91 uint32_t ws, uint32_t imm)
93 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
94 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
95 wr_t wx, *pwx = &wx;
96 uint32_t i;
98 switch (df) {
99 case DF_BYTE:
100 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) {
101 pwx->b[i] = pws->b[SHF_POS(i, imm)];
103 break;
104 case DF_HALF:
105 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) {
106 pwx->h[i] = pws->h[SHF_POS(i, imm)];
108 break;
109 case DF_WORD:
110 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
111 pwx->w[i] = pws->w[SHF_POS(i, imm)];
113 break;
114 default:
115 assert(0);
117 msa_move_v(pwd, pwx);
120 #define MSA_FN_VECTOR(FUNC, DEST, OPERATION) \
121 void helper_msa_ ## FUNC(CPUMIPSState *env, uint32_t wd, uint32_t ws, \
122 uint32_t wt) \
124 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
125 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
126 wr_t *pwt = &(env->active_fpu.fpr[wt].wr); \
127 uint32_t i; \
128 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
129 DEST = OPERATION; \
133 MSA_FN_VECTOR(and_v, pwd->d[i], pws->d[i] & pwt->d[i])
134 MSA_FN_VECTOR(or_v, pwd->d[i], pws->d[i] | pwt->d[i])
135 MSA_FN_VECTOR(nor_v, pwd->d[i], ~(pws->d[i] | pwt->d[i]))
136 MSA_FN_VECTOR(xor_v, pwd->d[i], pws->d[i] ^ pwt->d[i])
137 MSA_FN_VECTOR(bmnz_v, pwd->d[i],
138 BIT_MOVE_IF_NOT_ZERO(pwd->d[i], pws->d[i], pwt->d[i], DF_DOUBLE))
139 MSA_FN_VECTOR(bmz_v, pwd->d[i],
140 BIT_MOVE_IF_ZERO(pwd->d[i], pws->d[i], pwt->d[i], DF_DOUBLE))
141 MSA_FN_VECTOR(bsel_v, pwd->d[i],
142 BIT_SELECT(pwd->d[i], pws->d[i], pwt->d[i], DF_DOUBLE))
143 #undef BIT_MOVE_IF_NOT_ZERO
144 #undef BIT_MOVE_IF_ZERO
145 #undef BIT_SELECT
146 #undef MSA_FN_VECTOR
148 static inline int64_t msa_addv_df(uint32_t df, int64_t arg1, int64_t arg2)
150 return arg1 + arg2;
153 static inline int64_t msa_subv_df(uint32_t df, int64_t arg1, int64_t arg2)
155 return arg1 - arg2;
158 static inline int64_t msa_ceq_df(uint32_t df, int64_t arg1, int64_t arg2)
160 return arg1 == arg2 ? -1 : 0;
163 static inline int64_t msa_cle_s_df(uint32_t df, int64_t arg1, int64_t arg2)
165 return arg1 <= arg2 ? -1 : 0;
168 static inline int64_t msa_cle_u_df(uint32_t df, int64_t arg1, int64_t arg2)
170 uint64_t u_arg1 = UNSIGNED(arg1, df);
171 uint64_t u_arg2 = UNSIGNED(arg2, df);
172 return u_arg1 <= u_arg2 ? -1 : 0;
175 static inline int64_t msa_clt_s_df(uint32_t df, int64_t arg1, int64_t arg2)
177 return arg1 < arg2 ? -1 : 0;
180 static inline int64_t msa_clt_u_df(uint32_t df, int64_t arg1, int64_t arg2)
182 uint64_t u_arg1 = UNSIGNED(arg1, df);
183 uint64_t u_arg2 = UNSIGNED(arg2, df);
184 return u_arg1 < u_arg2 ? -1 : 0;
187 static inline int64_t msa_max_s_df(uint32_t df, int64_t arg1, int64_t arg2)
189 return arg1 > arg2 ? arg1 : arg2;
192 static inline int64_t msa_max_u_df(uint32_t df, int64_t arg1, int64_t arg2)
194 uint64_t u_arg1 = UNSIGNED(arg1, df);
195 uint64_t u_arg2 = UNSIGNED(arg2, df);
196 return u_arg1 > u_arg2 ? arg1 : arg2;
199 static inline int64_t msa_min_s_df(uint32_t df, int64_t arg1, int64_t arg2)
201 return arg1 < arg2 ? arg1 : arg2;
204 static inline int64_t msa_min_u_df(uint32_t df, int64_t arg1, int64_t arg2)
206 uint64_t u_arg1 = UNSIGNED(arg1, df);
207 uint64_t u_arg2 = UNSIGNED(arg2, df);
208 return u_arg1 < u_arg2 ? arg1 : arg2;
211 #define MSA_BINOP_IMM_DF(helper, func) \
212 void helper_msa_ ## helper ## _df(CPUMIPSState *env, uint32_t df, \
213 uint32_t wd, uint32_t ws, int32_t u5) \
215 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
216 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
217 uint32_t i; \
219 switch (df) { \
220 case DF_BYTE: \
221 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
222 pwd->b[i] = msa_ ## func ## _df(df, pws->b[i], u5); \
224 break; \
225 case DF_HALF: \
226 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
227 pwd->h[i] = msa_ ## func ## _df(df, pws->h[i], u5); \
229 break; \
230 case DF_WORD: \
231 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
232 pwd->w[i] = msa_ ## func ## _df(df, pws->w[i], u5); \
234 break; \
235 case DF_DOUBLE: \
236 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
237 pwd->d[i] = msa_ ## func ## _df(df, pws->d[i], u5); \
239 break; \
240 default: \
241 assert(0); \
245 MSA_BINOP_IMM_DF(addvi, addv)
246 MSA_BINOP_IMM_DF(subvi, subv)
247 MSA_BINOP_IMM_DF(ceqi, ceq)
248 MSA_BINOP_IMM_DF(clei_s, cle_s)
249 MSA_BINOP_IMM_DF(clei_u, cle_u)
250 MSA_BINOP_IMM_DF(clti_s, clt_s)
251 MSA_BINOP_IMM_DF(clti_u, clt_u)
252 MSA_BINOP_IMM_DF(maxi_s, max_s)
253 MSA_BINOP_IMM_DF(maxi_u, max_u)
254 MSA_BINOP_IMM_DF(mini_s, min_s)
255 MSA_BINOP_IMM_DF(mini_u, min_u)
256 #undef MSA_BINOP_IMM_DF
258 void helper_msa_ldi_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
259 int32_t s10)
261 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
262 uint32_t i;
264 switch (df) {
265 case DF_BYTE:
266 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) {
267 pwd->b[i] = (int8_t)s10;
269 break;
270 case DF_HALF:
271 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) {
272 pwd->h[i] = (int16_t)s10;
274 break;
275 case DF_WORD:
276 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
277 pwd->w[i] = (int32_t)s10;
279 break;
280 case DF_DOUBLE:
281 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
282 pwd->d[i] = (int64_t)s10;
284 break;
285 default:
286 assert(0);
290 /* Data format bit position and unsigned values */
291 #define BIT_POSITION(x, df) ((uint64_t)(x) % DF_BITS(df))
293 static inline int64_t msa_sll_df(uint32_t df, int64_t arg1, int64_t arg2)
295 int32_t b_arg2 = BIT_POSITION(arg2, df);
296 return arg1 << b_arg2;
299 static inline int64_t msa_sra_df(uint32_t df, int64_t arg1, int64_t arg2)
301 int32_t b_arg2 = BIT_POSITION(arg2, df);
302 return arg1 >> b_arg2;
305 static inline int64_t msa_srl_df(uint32_t df, int64_t arg1, int64_t arg2)
307 uint64_t u_arg1 = UNSIGNED(arg1, df);
308 int32_t b_arg2 = BIT_POSITION(arg2, df);
309 return u_arg1 >> b_arg2;
312 static inline int64_t msa_bclr_df(uint32_t df, int64_t arg1, int64_t arg2)
314 int32_t b_arg2 = BIT_POSITION(arg2, df);
315 return UNSIGNED(arg1 & (~(1LL << b_arg2)), df);
318 static inline int64_t msa_bset_df(uint32_t df, int64_t arg1,
319 int64_t arg2)
321 int32_t b_arg2 = BIT_POSITION(arg2, df);
322 return UNSIGNED(arg1 | (1LL << b_arg2), df);
325 static inline int64_t msa_bneg_df(uint32_t df, int64_t arg1, int64_t arg2)
327 int32_t b_arg2 = BIT_POSITION(arg2, df);
328 return UNSIGNED(arg1 ^ (1LL << b_arg2), df);
331 static inline int64_t msa_binsl_df(uint32_t df, int64_t dest, int64_t arg1,
332 int64_t arg2)
334 uint64_t u_arg1 = UNSIGNED(arg1, df);
335 uint64_t u_dest = UNSIGNED(dest, df);
336 int32_t sh_d = BIT_POSITION(arg2, df) + 1;
337 int32_t sh_a = DF_BITS(df) - sh_d;
338 if (sh_d == DF_BITS(df)) {
339 return u_arg1;
340 } else {
341 return UNSIGNED(UNSIGNED(u_dest << sh_d, df) >> sh_d, df) |
342 UNSIGNED(UNSIGNED(u_arg1 >> sh_a, df) << sh_a, df);
346 static inline int64_t msa_binsr_df(uint32_t df, int64_t dest, int64_t arg1,
347 int64_t arg2)
349 uint64_t u_arg1 = UNSIGNED(arg1, df);
350 uint64_t u_dest = UNSIGNED(dest, df);
351 int32_t sh_d = BIT_POSITION(arg2, df) + 1;
352 int32_t sh_a = DF_BITS(df) - sh_d;
353 if (sh_d == DF_BITS(df)) {
354 return u_arg1;
355 } else {
356 return UNSIGNED(UNSIGNED(u_dest >> sh_d, df) << sh_d, df) |
357 UNSIGNED(UNSIGNED(u_arg1 << sh_a, df) >> sh_a, df);
361 static inline int64_t msa_sat_s_df(uint32_t df, int64_t arg, uint32_t m)
363 return arg < M_MIN_INT(m+1) ? M_MIN_INT(m+1) :
364 arg > M_MAX_INT(m+1) ? M_MAX_INT(m+1) :
365 arg;
368 static inline int64_t msa_sat_u_df(uint32_t df, int64_t arg, uint32_t m)
370 uint64_t u_arg = UNSIGNED(arg, df);
371 return u_arg < M_MAX_UINT(m+1) ? u_arg :
372 M_MAX_UINT(m+1);
375 static inline int64_t msa_srar_df(uint32_t df, int64_t arg1, int64_t arg2)
377 int32_t b_arg2 = BIT_POSITION(arg2, df);
378 if (b_arg2 == 0) {
379 return arg1;
380 } else {
381 int64_t r_bit = (arg1 >> (b_arg2 - 1)) & 1;
382 return (arg1 >> b_arg2) + r_bit;
386 static inline int64_t msa_srlr_df(uint32_t df, int64_t arg1, int64_t arg2)
388 uint64_t u_arg1 = UNSIGNED(arg1, df);
389 int32_t b_arg2 = BIT_POSITION(arg2, df);
390 if (b_arg2 == 0) {
391 return u_arg1;
392 } else {
393 uint64_t r_bit = (u_arg1 >> (b_arg2 - 1)) & 1;
394 return (u_arg1 >> b_arg2) + r_bit;
398 #define MSA_BINOP_IMMU_DF(helper, func) \
399 void helper_msa_ ## helper ## _df(CPUMIPSState *env, uint32_t df, uint32_t wd, \
400 uint32_t ws, uint32_t u5) \
402 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
403 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
404 uint32_t i; \
406 switch (df) { \
407 case DF_BYTE: \
408 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
409 pwd->b[i] = msa_ ## func ## _df(df, pws->b[i], u5); \
411 break; \
412 case DF_HALF: \
413 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
414 pwd->h[i] = msa_ ## func ## _df(df, pws->h[i], u5); \
416 break; \
417 case DF_WORD: \
418 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
419 pwd->w[i] = msa_ ## func ## _df(df, pws->w[i], u5); \
421 break; \
422 case DF_DOUBLE: \
423 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
424 pwd->d[i] = msa_ ## func ## _df(df, pws->d[i], u5); \
426 break; \
427 default: \
428 assert(0); \
432 MSA_BINOP_IMMU_DF(slli, sll)
433 MSA_BINOP_IMMU_DF(srai, sra)
434 MSA_BINOP_IMMU_DF(srli, srl)
435 MSA_BINOP_IMMU_DF(bclri, bclr)
436 MSA_BINOP_IMMU_DF(bseti, bset)
437 MSA_BINOP_IMMU_DF(bnegi, bneg)
438 MSA_BINOP_IMMU_DF(sat_s, sat_s)
439 MSA_BINOP_IMMU_DF(sat_u, sat_u)
440 MSA_BINOP_IMMU_DF(srari, srar)
441 MSA_BINOP_IMMU_DF(srlri, srlr)
442 #undef MSA_BINOP_IMMU_DF
444 #define MSA_TEROP_IMMU_DF(helper, func) \
445 void helper_msa_ ## helper ## _df(CPUMIPSState *env, uint32_t df, \
446 uint32_t wd, uint32_t ws, uint32_t u5) \
448 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
449 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
450 uint32_t i; \
452 switch (df) { \
453 case DF_BYTE: \
454 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
455 pwd->b[i] = msa_ ## func ## _df(df, pwd->b[i], pws->b[i], \
456 u5); \
458 break; \
459 case DF_HALF: \
460 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
461 pwd->h[i] = msa_ ## func ## _df(df, pwd->h[i], pws->h[i], \
462 u5); \
464 break; \
465 case DF_WORD: \
466 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
467 pwd->w[i] = msa_ ## func ## _df(df, pwd->w[i], pws->w[i], \
468 u5); \
470 break; \
471 case DF_DOUBLE: \
472 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
473 pwd->d[i] = msa_ ## func ## _df(df, pwd->d[i], pws->d[i], \
474 u5); \
476 break; \
477 default: \
478 assert(0); \
482 MSA_TEROP_IMMU_DF(binsli, binsl)
483 MSA_TEROP_IMMU_DF(binsri, binsr)
484 #undef MSA_TEROP_IMMU_DF
486 static inline int64_t msa_max_a_df(uint32_t df, int64_t arg1, int64_t arg2)
488 uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
489 uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
490 return abs_arg1 > abs_arg2 ? arg1 : arg2;
493 static inline int64_t msa_min_a_df(uint32_t df, int64_t arg1, int64_t arg2)
495 uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
496 uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
497 return abs_arg1 < abs_arg2 ? arg1 : arg2;
500 static inline int64_t msa_add_a_df(uint32_t df, int64_t arg1, int64_t arg2)
502 uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
503 uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
504 return abs_arg1 + abs_arg2;
507 static inline int64_t msa_adds_a_df(uint32_t df, int64_t arg1, int64_t arg2)
509 uint64_t max_int = (uint64_t)DF_MAX_INT(df);
510 uint64_t abs_arg1 = arg1 >= 0 ? arg1 : -arg1;
511 uint64_t abs_arg2 = arg2 >= 0 ? arg2 : -arg2;
512 if (abs_arg1 > max_int || abs_arg2 > max_int) {
513 return (int64_t)max_int;
514 } else {
515 return (abs_arg1 < max_int - abs_arg2) ? abs_arg1 + abs_arg2 : max_int;
519 static inline int64_t msa_adds_s_df(uint32_t df, int64_t arg1, int64_t arg2)
521 int64_t max_int = DF_MAX_INT(df);
522 int64_t min_int = DF_MIN_INT(df);
523 if (arg1 < 0) {
524 return (min_int - arg1 < arg2) ? arg1 + arg2 : min_int;
525 } else {
526 return (arg2 < max_int - arg1) ? arg1 + arg2 : max_int;
530 static inline uint64_t msa_adds_u_df(uint32_t df, uint64_t arg1, uint64_t arg2)
532 uint64_t max_uint = DF_MAX_UINT(df);
533 uint64_t u_arg1 = UNSIGNED(arg1, df);
534 uint64_t u_arg2 = UNSIGNED(arg2, df);
535 return (u_arg1 < max_uint - u_arg2) ? u_arg1 + u_arg2 : max_uint;
538 static inline int64_t msa_ave_s_df(uint32_t df, int64_t arg1, int64_t arg2)
540 /* signed shift */
541 return (arg1 >> 1) + (arg2 >> 1) + (arg1 & arg2 & 1);
544 static inline uint64_t msa_ave_u_df(uint32_t df, uint64_t arg1, uint64_t arg2)
546 uint64_t u_arg1 = UNSIGNED(arg1, df);
547 uint64_t u_arg2 = UNSIGNED(arg2, df);
548 /* unsigned shift */
549 return (u_arg1 >> 1) + (u_arg2 >> 1) + (u_arg1 & u_arg2 & 1);
552 static inline int64_t msa_aver_s_df(uint32_t df, int64_t arg1, int64_t arg2)
554 /* signed shift */
555 return (arg1 >> 1) + (arg2 >> 1) + ((arg1 | arg2) & 1);
558 static inline uint64_t msa_aver_u_df(uint32_t df, uint64_t arg1, uint64_t arg2)
560 uint64_t u_arg1 = UNSIGNED(arg1, df);
561 uint64_t u_arg2 = UNSIGNED(arg2, df);
562 /* unsigned shift */
563 return (u_arg1 >> 1) + (u_arg2 >> 1) + ((u_arg1 | u_arg2) & 1);
566 static inline int64_t msa_subs_s_df(uint32_t df, int64_t arg1, int64_t arg2)
568 int64_t max_int = DF_MAX_INT(df);
569 int64_t min_int = DF_MIN_INT(df);
570 if (arg2 > 0) {
571 return (min_int + arg2 < arg1) ? arg1 - arg2 : min_int;
572 } else {
573 return (arg1 < max_int + arg2) ? arg1 - arg2 : max_int;
577 static inline int64_t msa_subs_u_df(uint32_t df, int64_t arg1, int64_t arg2)
579 uint64_t u_arg1 = UNSIGNED(arg1, df);
580 uint64_t u_arg2 = UNSIGNED(arg2, df);
581 return (u_arg1 > u_arg2) ? u_arg1 - u_arg2 : 0;
584 static inline int64_t msa_subsus_u_df(uint32_t df, int64_t arg1, int64_t arg2)
586 uint64_t u_arg1 = UNSIGNED(arg1, df);
587 uint64_t max_uint = DF_MAX_UINT(df);
588 if (arg2 >= 0) {
589 uint64_t u_arg2 = (uint64_t)arg2;
590 return (u_arg1 > u_arg2) ?
591 (int64_t)(u_arg1 - u_arg2) :
593 } else {
594 uint64_t u_arg2 = (uint64_t)(-arg2);
595 return (u_arg1 < max_uint - u_arg2) ?
596 (int64_t)(u_arg1 + u_arg2) :
597 (int64_t)max_uint;
601 static inline int64_t msa_subsuu_s_df(uint32_t df, int64_t arg1, int64_t arg2)
603 uint64_t u_arg1 = UNSIGNED(arg1, df);
604 uint64_t u_arg2 = UNSIGNED(arg2, df);
605 int64_t max_int = DF_MAX_INT(df);
606 int64_t min_int = DF_MIN_INT(df);
607 if (u_arg1 > u_arg2) {
608 return u_arg1 - u_arg2 < (uint64_t)max_int ?
609 (int64_t)(u_arg1 - u_arg2) :
610 max_int;
611 } else {
612 return u_arg2 - u_arg1 < (uint64_t)(-min_int) ?
613 (int64_t)(u_arg1 - u_arg2) :
614 min_int;
618 static inline int64_t msa_asub_s_df(uint32_t df, int64_t arg1, int64_t arg2)
620 /* signed compare */
621 return (arg1 < arg2) ?
622 (uint64_t)(arg2 - arg1) : (uint64_t)(arg1 - arg2);
625 static inline uint64_t msa_asub_u_df(uint32_t df, uint64_t arg1, uint64_t arg2)
627 uint64_t u_arg1 = UNSIGNED(arg1, df);
628 uint64_t u_arg2 = UNSIGNED(arg2, df);
629 /* unsigned compare */
630 return (u_arg1 < u_arg2) ?
631 (uint64_t)(u_arg2 - u_arg1) : (uint64_t)(u_arg1 - u_arg2);
634 static inline int64_t msa_mulv_df(uint32_t df, int64_t arg1, int64_t arg2)
636 return arg1 * arg2;
639 static inline int64_t msa_div_s_df(uint32_t df, int64_t arg1, int64_t arg2)
641 if (arg1 == DF_MIN_INT(df) && arg2 == -1) {
642 return DF_MIN_INT(df);
644 return arg2 ? arg1 / arg2 : 0;
647 static inline int64_t msa_div_u_df(uint32_t df, int64_t arg1, int64_t arg2)
649 uint64_t u_arg1 = UNSIGNED(arg1, df);
650 uint64_t u_arg2 = UNSIGNED(arg2, df);
651 return u_arg2 ? u_arg1 / u_arg2 : 0;
654 static inline int64_t msa_mod_s_df(uint32_t df, int64_t arg1, int64_t arg2)
656 if (arg1 == DF_MIN_INT(df) && arg2 == -1) {
657 return 0;
659 return arg2 ? arg1 % arg2 : 0;
662 static inline int64_t msa_mod_u_df(uint32_t df, int64_t arg1, int64_t arg2)
664 uint64_t u_arg1 = UNSIGNED(arg1, df);
665 uint64_t u_arg2 = UNSIGNED(arg2, df);
666 return u_arg2 ? u_arg1 % u_arg2 : 0;
669 #define SIGNED_EVEN(a, df) \
670 ((((int64_t)(a)) << (64 - DF_BITS(df)/2)) >> (64 - DF_BITS(df)/2))
672 #define UNSIGNED_EVEN(a, df) \
673 ((((uint64_t)(a)) << (64 - DF_BITS(df)/2)) >> (64 - DF_BITS(df)/2))
675 #define SIGNED_ODD(a, df) \
676 ((((int64_t)(a)) << (64 - DF_BITS(df))) >> (64 - DF_BITS(df)/2))
678 #define UNSIGNED_ODD(a, df) \
679 ((((uint64_t)(a)) << (64 - DF_BITS(df))) >> (64 - DF_BITS(df)/2))
681 #define SIGNED_EXTRACT(e, o, a, df) \
682 do { \
683 e = SIGNED_EVEN(a, df); \
684 o = SIGNED_ODD(a, df); \
685 } while (0)
687 #define UNSIGNED_EXTRACT(e, o, a, df) \
688 do { \
689 e = UNSIGNED_EVEN(a, df); \
690 o = UNSIGNED_ODD(a, df); \
691 } while (0)
693 static inline int64_t msa_dotp_s_df(uint32_t df, int64_t arg1, int64_t arg2)
695 int64_t even_arg1;
696 int64_t even_arg2;
697 int64_t odd_arg1;
698 int64_t odd_arg2;
699 SIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
700 SIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
701 return (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
704 static inline int64_t msa_dotp_u_df(uint32_t df, int64_t arg1, int64_t arg2)
706 int64_t even_arg1;
707 int64_t even_arg2;
708 int64_t odd_arg1;
709 int64_t odd_arg2;
710 UNSIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
711 UNSIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
712 return (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
715 #define CONCATENATE_AND_SLIDE(s, k) \
716 do { \
717 for (i = 0; i < s; i++) { \
718 v[i] = pws->b[s * k + i]; \
719 v[i + s] = pwd->b[s * k + i]; \
721 for (i = 0; i < s; i++) { \
722 pwd->b[s * k + i] = v[i + n]; \
724 } while (0)
726 static inline void msa_sld_df(uint32_t df, wr_t *pwd,
727 wr_t *pws, target_ulong rt)
729 uint32_t n = rt % DF_ELEMENTS(df);
730 uint8_t v[64];
731 uint32_t i, k;
733 switch (df) {
734 case DF_BYTE:
735 CONCATENATE_AND_SLIDE(DF_ELEMENTS(DF_BYTE), 0);
736 break;
737 case DF_HALF:
738 for (k = 0; k < 2; k++) {
739 CONCATENATE_AND_SLIDE(DF_ELEMENTS(DF_HALF), k);
741 break;
742 case DF_WORD:
743 for (k = 0; k < 4; k++) {
744 CONCATENATE_AND_SLIDE(DF_ELEMENTS(DF_WORD), k);
746 break;
747 case DF_DOUBLE:
748 for (k = 0; k < 8; k++) {
749 CONCATENATE_AND_SLIDE(DF_ELEMENTS(DF_DOUBLE), k);
751 break;
752 default:
753 assert(0);
757 static inline int64_t msa_hadd_s_df(uint32_t df, int64_t arg1, int64_t arg2)
759 return SIGNED_ODD(arg1, df) + SIGNED_EVEN(arg2, df);
762 static inline int64_t msa_hadd_u_df(uint32_t df, int64_t arg1, int64_t arg2)
764 return UNSIGNED_ODD(arg1, df) + UNSIGNED_EVEN(arg2, df);
767 static inline int64_t msa_hsub_s_df(uint32_t df, int64_t arg1, int64_t arg2)
769 return SIGNED_ODD(arg1, df) - SIGNED_EVEN(arg2, df);
772 static inline int64_t msa_hsub_u_df(uint32_t df, int64_t arg1, int64_t arg2)
774 return UNSIGNED_ODD(arg1, df) - UNSIGNED_EVEN(arg2, df);
777 static inline int64_t msa_mul_q_df(uint32_t df, int64_t arg1, int64_t arg2)
779 int64_t q_min = DF_MIN_INT(df);
780 int64_t q_max = DF_MAX_INT(df);
782 if (arg1 == q_min && arg2 == q_min) {
783 return q_max;
785 return (arg1 * arg2) >> (DF_BITS(df) - 1);
788 static inline int64_t msa_mulr_q_df(uint32_t df, int64_t arg1, int64_t arg2)
790 int64_t q_min = DF_MIN_INT(df);
791 int64_t q_max = DF_MAX_INT(df);
792 int64_t r_bit = 1 << (DF_BITS(df) - 2);
794 if (arg1 == q_min && arg2 == q_min) {
795 return q_max;
797 return (arg1 * arg2 + r_bit) >> (DF_BITS(df) - 1);
800 #define MSA_BINOP_DF(func) \
801 void helper_msa_ ## func ## _df(CPUMIPSState *env, uint32_t df, \
802 uint32_t wd, uint32_t ws, uint32_t wt) \
804 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
805 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
806 wr_t *pwt = &(env->active_fpu.fpr[wt].wr); \
807 uint32_t i; \
809 switch (df) { \
810 case DF_BYTE: \
811 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
812 pwd->b[i] = msa_ ## func ## _df(df, pws->b[i], pwt->b[i]); \
814 break; \
815 case DF_HALF: \
816 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
817 pwd->h[i] = msa_ ## func ## _df(df, pws->h[i], pwt->h[i]); \
819 break; \
820 case DF_WORD: \
821 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
822 pwd->w[i] = msa_ ## func ## _df(df, pws->w[i], pwt->w[i]); \
824 break; \
825 case DF_DOUBLE: \
826 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
827 pwd->d[i] = msa_ ## func ## _df(df, pws->d[i], pwt->d[i]); \
829 break; \
830 default: \
831 assert(0); \
835 MSA_BINOP_DF(sll)
836 MSA_BINOP_DF(sra)
837 MSA_BINOP_DF(srl)
838 MSA_BINOP_DF(bclr)
839 MSA_BINOP_DF(bset)
840 MSA_BINOP_DF(bneg)
841 MSA_BINOP_DF(addv)
842 MSA_BINOP_DF(subv)
843 MSA_BINOP_DF(max_s)
844 MSA_BINOP_DF(max_u)
845 MSA_BINOP_DF(min_s)
846 MSA_BINOP_DF(min_u)
847 MSA_BINOP_DF(max_a)
848 MSA_BINOP_DF(min_a)
849 MSA_BINOP_DF(ceq)
850 MSA_BINOP_DF(clt_s)
851 MSA_BINOP_DF(clt_u)
852 MSA_BINOP_DF(cle_s)
853 MSA_BINOP_DF(cle_u)
854 MSA_BINOP_DF(add_a)
855 MSA_BINOP_DF(adds_a)
856 MSA_BINOP_DF(adds_s)
857 MSA_BINOP_DF(adds_u)
858 MSA_BINOP_DF(ave_s)
859 MSA_BINOP_DF(ave_u)
860 MSA_BINOP_DF(aver_s)
861 MSA_BINOP_DF(aver_u)
862 MSA_BINOP_DF(subs_s)
863 MSA_BINOP_DF(subs_u)
864 MSA_BINOP_DF(subsus_u)
865 MSA_BINOP_DF(subsuu_s)
866 MSA_BINOP_DF(asub_s)
867 MSA_BINOP_DF(asub_u)
868 MSA_BINOP_DF(mulv)
869 MSA_BINOP_DF(div_s)
870 MSA_BINOP_DF(div_u)
871 MSA_BINOP_DF(mod_s)
872 MSA_BINOP_DF(mod_u)
873 MSA_BINOP_DF(dotp_s)
874 MSA_BINOP_DF(dotp_u)
875 MSA_BINOP_DF(srar)
876 MSA_BINOP_DF(srlr)
877 MSA_BINOP_DF(hadd_s)
878 MSA_BINOP_DF(hadd_u)
879 MSA_BINOP_DF(hsub_s)
880 MSA_BINOP_DF(hsub_u)
882 MSA_BINOP_DF(mul_q)
883 MSA_BINOP_DF(mulr_q)
884 #undef MSA_BINOP_DF
886 void helper_msa_sld_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
887 uint32_t ws, uint32_t rt)
889 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
890 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
892 msa_sld_df(df, pwd, pws, env->active_tc.gpr[rt]);
895 static inline int64_t msa_maddv_df(uint32_t df, int64_t dest, int64_t arg1,
896 int64_t arg2)
898 return dest + arg1 * arg2;
901 static inline int64_t msa_msubv_df(uint32_t df, int64_t dest, int64_t arg1,
902 int64_t arg2)
904 return dest - arg1 * arg2;
907 static inline int64_t msa_dpadd_s_df(uint32_t df, int64_t dest, int64_t arg1,
908 int64_t arg2)
910 int64_t even_arg1;
911 int64_t even_arg2;
912 int64_t odd_arg1;
913 int64_t odd_arg2;
914 SIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
915 SIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
916 return dest + (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
919 static inline int64_t msa_dpadd_u_df(uint32_t df, int64_t dest, int64_t arg1,
920 int64_t arg2)
922 int64_t even_arg1;
923 int64_t even_arg2;
924 int64_t odd_arg1;
925 int64_t odd_arg2;
926 UNSIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
927 UNSIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
928 return dest + (even_arg1 * even_arg2) + (odd_arg1 * odd_arg2);
931 static inline int64_t msa_dpsub_s_df(uint32_t df, int64_t dest, int64_t arg1,
932 int64_t arg2)
934 int64_t even_arg1;
935 int64_t even_arg2;
936 int64_t odd_arg1;
937 int64_t odd_arg2;
938 SIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
939 SIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
940 return dest - ((even_arg1 * even_arg2) + (odd_arg1 * odd_arg2));
943 static inline int64_t msa_dpsub_u_df(uint32_t df, int64_t dest, int64_t arg1,
944 int64_t arg2)
946 int64_t even_arg1;
947 int64_t even_arg2;
948 int64_t odd_arg1;
949 int64_t odd_arg2;
950 UNSIGNED_EXTRACT(even_arg1, odd_arg1, arg1, df);
951 UNSIGNED_EXTRACT(even_arg2, odd_arg2, arg2, df);
952 return dest - ((even_arg1 * even_arg2) + (odd_arg1 * odd_arg2));
955 static inline int64_t msa_madd_q_df(uint32_t df, int64_t dest, int64_t arg1,
956 int64_t arg2)
958 int64_t q_prod, q_ret;
960 int64_t q_max = DF_MAX_INT(df);
961 int64_t q_min = DF_MIN_INT(df);
963 q_prod = arg1 * arg2;
964 q_ret = ((dest << (DF_BITS(df) - 1)) + q_prod) >> (DF_BITS(df) - 1);
966 return (q_ret < q_min) ? q_min : (q_max < q_ret) ? q_max : q_ret;
969 static inline int64_t msa_msub_q_df(uint32_t df, int64_t dest, int64_t arg1,
970 int64_t arg2)
972 int64_t q_prod, q_ret;
974 int64_t q_max = DF_MAX_INT(df);
975 int64_t q_min = DF_MIN_INT(df);
977 q_prod = arg1 * arg2;
978 q_ret = ((dest << (DF_BITS(df) - 1)) - q_prod) >> (DF_BITS(df) - 1);
980 return (q_ret < q_min) ? q_min : (q_max < q_ret) ? q_max : q_ret;
983 static inline int64_t msa_maddr_q_df(uint32_t df, int64_t dest, int64_t arg1,
984 int64_t arg2)
986 int64_t q_prod, q_ret;
988 int64_t q_max = DF_MAX_INT(df);
989 int64_t q_min = DF_MIN_INT(df);
990 int64_t r_bit = 1 << (DF_BITS(df) - 2);
992 q_prod = arg1 * arg2;
993 q_ret = ((dest << (DF_BITS(df) - 1)) + q_prod + r_bit) >> (DF_BITS(df) - 1);
995 return (q_ret < q_min) ? q_min : (q_max < q_ret) ? q_max : q_ret;
998 static inline int64_t msa_msubr_q_df(uint32_t df, int64_t dest, int64_t arg1,
999 int64_t arg2)
1001 int64_t q_prod, q_ret;
1003 int64_t q_max = DF_MAX_INT(df);
1004 int64_t q_min = DF_MIN_INT(df);
1005 int64_t r_bit = 1 << (DF_BITS(df) - 2);
1007 q_prod = arg1 * arg2;
1008 q_ret = ((dest << (DF_BITS(df) - 1)) - q_prod + r_bit) >> (DF_BITS(df) - 1);
1010 return (q_ret < q_min) ? q_min : (q_max < q_ret) ? q_max : q_ret;
1013 #define MSA_TEROP_DF(func) \
1014 void helper_msa_ ## func ## _df(CPUMIPSState *env, uint32_t df, uint32_t wd, \
1015 uint32_t ws, uint32_t wt) \
1017 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
1018 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
1019 wr_t *pwt = &(env->active_fpu.fpr[wt].wr); \
1020 uint32_t i; \
1022 switch (df) { \
1023 case DF_BYTE: \
1024 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
1025 pwd->b[i] = msa_ ## func ## _df(df, pwd->b[i], pws->b[i], \
1026 pwt->b[i]); \
1028 break; \
1029 case DF_HALF: \
1030 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
1031 pwd->h[i] = msa_ ## func ## _df(df, pwd->h[i], pws->h[i], \
1032 pwt->h[i]); \
1034 break; \
1035 case DF_WORD: \
1036 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
1037 pwd->w[i] = msa_ ## func ## _df(df, pwd->w[i], pws->w[i], \
1038 pwt->w[i]); \
1040 break; \
1041 case DF_DOUBLE: \
1042 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
1043 pwd->d[i] = msa_ ## func ## _df(df, pwd->d[i], pws->d[i], \
1044 pwt->d[i]); \
1046 break; \
1047 default: \
1048 assert(0); \
1052 MSA_TEROP_DF(maddv)
1053 MSA_TEROP_DF(msubv)
1054 MSA_TEROP_DF(dpadd_s)
1055 MSA_TEROP_DF(dpadd_u)
1056 MSA_TEROP_DF(dpsub_s)
1057 MSA_TEROP_DF(dpsub_u)
1058 MSA_TEROP_DF(binsl)
1059 MSA_TEROP_DF(binsr)
1060 MSA_TEROP_DF(madd_q)
1061 MSA_TEROP_DF(msub_q)
1062 MSA_TEROP_DF(maddr_q)
1063 MSA_TEROP_DF(msubr_q)
1064 #undef MSA_TEROP_DF
1066 static inline void msa_splat_df(uint32_t df, wr_t *pwd,
1067 wr_t *pws, target_ulong rt)
1069 uint32_t n = rt % DF_ELEMENTS(df);
1070 uint32_t i;
1072 switch (df) {
1073 case DF_BYTE:
1074 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) {
1075 pwd->b[i] = pws->b[n];
1077 break;
1078 case DF_HALF:
1079 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) {
1080 pwd->h[i] = pws->h[n];
1082 break;
1083 case DF_WORD:
1084 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
1085 pwd->w[i] = pws->w[n];
1087 break;
1088 case DF_DOUBLE:
1089 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
1090 pwd->d[i] = pws->d[n];
1092 break;
1093 default:
1094 assert(0);
1098 void helper_msa_splat_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1099 uint32_t ws, uint32_t rt)
1101 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1102 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1104 msa_splat_df(df, pwd, pws, env->active_tc.gpr[rt]);
1107 #define MSA_DO_B MSA_DO(b)
1108 #define MSA_DO_H MSA_DO(h)
1109 #define MSA_DO_W MSA_DO(w)
1110 #define MSA_DO_D MSA_DO(d)
1112 #define MSA_LOOP_B MSA_LOOP(B)
1113 #define MSA_LOOP_H MSA_LOOP(H)
1114 #define MSA_LOOP_W MSA_LOOP(W)
1115 #define MSA_LOOP_D MSA_LOOP(D)
1117 #define MSA_LOOP_COND_B MSA_LOOP_COND(DF_BYTE)
1118 #define MSA_LOOP_COND_H MSA_LOOP_COND(DF_HALF)
1119 #define MSA_LOOP_COND_W MSA_LOOP_COND(DF_WORD)
1120 #define MSA_LOOP_COND_D MSA_LOOP_COND(DF_DOUBLE)
1122 #define MSA_LOOP(DF) \
1123 do { \
1124 for (i = 0; i < (MSA_LOOP_COND_ ## DF) ; i++) { \
1125 MSA_DO_ ## DF; \
1127 } while (0)
1129 #define MSA_FN_DF(FUNC) \
1130 void helper_msa_##FUNC(CPUMIPSState *env, uint32_t df, uint32_t wd, \
1131 uint32_t ws, uint32_t wt) \
1133 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
1134 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
1135 wr_t *pwt = &(env->active_fpu.fpr[wt].wr); \
1136 wr_t wx, *pwx = &wx; \
1137 uint32_t i; \
1138 switch (df) { \
1139 case DF_BYTE: \
1140 MSA_LOOP_B; \
1141 break; \
1142 case DF_HALF: \
1143 MSA_LOOP_H; \
1144 break; \
1145 case DF_WORD: \
1146 MSA_LOOP_W; \
1147 break; \
1148 case DF_DOUBLE: \
1149 MSA_LOOP_D; \
1150 break; \
1151 default: \
1152 assert(0); \
1154 msa_move_v(pwd, pwx); \
1157 #define MSA_LOOP_COND(DF) \
1158 (DF_ELEMENTS(DF) / 2)
1160 #define Rb(pwr, i) (pwr->b[i])
1161 #define Lb(pwr, i) (pwr->b[i + DF_ELEMENTS(DF_BYTE)/2])
1162 #define Rh(pwr, i) (pwr->h[i])
1163 #define Lh(pwr, i) (pwr->h[i + DF_ELEMENTS(DF_HALF)/2])
1164 #define Rw(pwr, i) (pwr->w[i])
1165 #define Lw(pwr, i) (pwr->w[i + DF_ELEMENTS(DF_WORD)/2])
1166 #define Rd(pwr, i) (pwr->d[i])
1167 #define Ld(pwr, i) (pwr->d[i + DF_ELEMENTS(DF_DOUBLE)/2])
1169 #define MSA_DO(DF) \
1170 do { \
1171 R##DF(pwx, i) = pwt->DF[2*i]; \
1172 L##DF(pwx, i) = pws->DF[2*i]; \
1173 } while (0)
1174 MSA_FN_DF(pckev_df)
1175 #undef MSA_DO
1177 #define MSA_DO(DF) \
1178 do { \
1179 R##DF(pwx, i) = pwt->DF[2*i+1]; \
1180 L##DF(pwx, i) = pws->DF[2*i+1]; \
1181 } while (0)
1182 MSA_FN_DF(pckod_df)
1183 #undef MSA_DO
1185 #define MSA_DO(DF) \
1186 do { \
1187 pwx->DF[2*i] = L##DF(pwt, i); \
1188 pwx->DF[2*i+1] = L##DF(pws, i); \
1189 } while (0)
1190 MSA_FN_DF(ilvl_df)
1191 #undef MSA_DO
1193 #define MSA_DO(DF) \
1194 do { \
1195 pwx->DF[2*i] = R##DF(pwt, i); \
1196 pwx->DF[2*i+1] = R##DF(pws, i); \
1197 } while (0)
1198 MSA_FN_DF(ilvr_df)
1199 #undef MSA_DO
1201 #define MSA_DO(DF) \
1202 do { \
1203 pwx->DF[2*i] = pwt->DF[2*i]; \
1204 pwx->DF[2*i+1] = pws->DF[2*i]; \
1205 } while (0)
1206 MSA_FN_DF(ilvev_df)
1207 #undef MSA_DO
1209 #define MSA_DO(DF) \
1210 do { \
1211 pwx->DF[2*i] = pwt->DF[2*i+1]; \
1212 pwx->DF[2*i+1] = pws->DF[2*i+1]; \
1213 } while (0)
1214 MSA_FN_DF(ilvod_df)
1215 #undef MSA_DO
1216 #undef MSA_LOOP_COND
1218 #define MSA_LOOP_COND(DF) \
1219 (DF_ELEMENTS(DF))
1221 #define MSA_DO(DF) \
1222 do { \
1223 uint32_t n = DF_ELEMENTS(df); \
1224 uint32_t k = (pwd->DF[i] & 0x3f) % (2 * n); \
1225 pwx->DF[i] = \
1226 (pwd->DF[i] & 0xc0) ? 0 : k < n ? pwt->DF[k] : pws->DF[k - n]; \
1227 } while (0)
1228 MSA_FN_DF(vshf_df)
1229 #undef MSA_DO
1230 #undef MSA_LOOP_COND
1231 #undef MSA_FN_DF
1233 void helper_msa_sldi_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1234 uint32_t ws, uint32_t n)
1236 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1237 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1239 msa_sld_df(df, pwd, pws, n);
1242 void helper_msa_splati_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1243 uint32_t ws, uint32_t n)
1245 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1246 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1248 msa_splat_df(df, pwd, pws, n);
1251 void helper_msa_copy_s_df(CPUMIPSState *env, uint32_t df, uint32_t rd,
1252 uint32_t ws, uint32_t n)
1254 n %= DF_ELEMENTS(df);
1256 switch (df) {
1257 case DF_BYTE:
1258 env->active_tc.gpr[rd] = (int8_t)env->active_fpu.fpr[ws].wr.b[n];
1259 break;
1260 case DF_HALF:
1261 env->active_tc.gpr[rd] = (int16_t)env->active_fpu.fpr[ws].wr.h[n];
1262 break;
1263 case DF_WORD:
1264 env->active_tc.gpr[rd] = (int32_t)env->active_fpu.fpr[ws].wr.w[n];
1265 break;
1266 #ifdef TARGET_MIPS64
1267 case DF_DOUBLE:
1268 env->active_tc.gpr[rd] = (int64_t)env->active_fpu.fpr[ws].wr.d[n];
1269 break;
1270 #endif
1271 default:
1272 assert(0);
1276 void helper_msa_copy_u_df(CPUMIPSState *env, uint32_t df, uint32_t rd,
1277 uint32_t ws, uint32_t n)
1279 n %= DF_ELEMENTS(df);
1281 switch (df) {
1282 case DF_BYTE:
1283 env->active_tc.gpr[rd] = (uint8_t)env->active_fpu.fpr[ws].wr.b[n];
1284 break;
1285 case DF_HALF:
1286 env->active_tc.gpr[rd] = (uint16_t)env->active_fpu.fpr[ws].wr.h[n];
1287 break;
1288 case DF_WORD:
1289 env->active_tc.gpr[rd] = (uint32_t)env->active_fpu.fpr[ws].wr.w[n];
1290 break;
1291 #ifdef TARGET_MIPS64
1292 case DF_DOUBLE:
1293 env->active_tc.gpr[rd] = (uint64_t)env->active_fpu.fpr[ws].wr.d[n];
1294 break;
1295 #endif
1296 default:
1297 assert(0);
1301 void helper_msa_insert_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1302 uint32_t rs_num, uint32_t n)
1304 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1305 target_ulong rs = env->active_tc.gpr[rs_num];
1307 switch (df) {
1308 case DF_BYTE:
1309 pwd->b[n] = (int8_t)rs;
1310 break;
1311 case DF_HALF:
1312 pwd->h[n] = (int16_t)rs;
1313 break;
1314 case DF_WORD:
1315 pwd->w[n] = (int32_t)rs;
1316 break;
1317 case DF_DOUBLE:
1318 pwd->d[n] = (int64_t)rs;
1319 break;
1320 default:
1321 assert(0);
1325 void helper_msa_insve_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1326 uint32_t ws, uint32_t n)
1328 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1329 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1331 switch (df) {
1332 case DF_BYTE:
1333 pwd->b[n] = (int8_t)pws->b[0];
1334 break;
1335 case DF_HALF:
1336 pwd->h[n] = (int16_t)pws->h[0];
1337 break;
1338 case DF_WORD:
1339 pwd->w[n] = (int32_t)pws->w[0];
1340 break;
1341 case DF_DOUBLE:
1342 pwd->d[n] = (int64_t)pws->d[0];
1343 break;
1344 default:
1345 assert(0);
1349 void helper_msa_ctcmsa(CPUMIPSState *env, target_ulong elm, uint32_t cd)
1351 switch (cd) {
1352 case 0:
1353 break;
1354 case 1:
1355 env->active_tc.msacsr = (int32_t)elm & MSACSR_MASK;
1356 restore_msa_fp_status(env);
1357 /* check exception */
1358 if ((GET_FP_ENABLE(env->active_tc.msacsr) | FP_UNIMPLEMENTED)
1359 & GET_FP_CAUSE(env->active_tc.msacsr)) {
1360 do_raise_exception(env, EXCP_MSAFPE, GETPC());
1362 break;
1366 target_ulong helper_msa_cfcmsa(CPUMIPSState *env, uint32_t cs)
1368 switch (cs) {
1369 case 0:
1370 return env->msair;
1371 case 1:
1372 return env->active_tc.msacsr & MSACSR_MASK;
1374 return 0;
1377 void helper_msa_move_v(CPUMIPSState *env, uint32_t wd, uint32_t ws)
1379 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1380 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
1382 msa_move_v(pwd, pws);
1385 static inline int64_t msa_pcnt_df(uint32_t df, int64_t arg)
1387 uint64_t x;
1389 x = UNSIGNED(arg, df);
1391 x = (x & 0x5555555555555555ULL) + ((x >> 1) & 0x5555555555555555ULL);
1392 x = (x & 0x3333333333333333ULL) + ((x >> 2) & 0x3333333333333333ULL);
1393 x = (x & 0x0F0F0F0F0F0F0F0FULL) + ((x >> 4) & 0x0F0F0F0F0F0F0F0FULL);
1394 x = (x & 0x00FF00FF00FF00FFULL) + ((x >> 8) & 0x00FF00FF00FF00FFULL);
1395 x = (x & 0x0000FFFF0000FFFFULL) + ((x >> 16) & 0x0000FFFF0000FFFFULL);
1396 x = (x & 0x00000000FFFFFFFFULL) + ((x >> 32));
1398 return x;
1401 static inline int64_t msa_nlzc_df(uint32_t df, int64_t arg)
1403 uint64_t x, y;
1404 int n, c;
1406 x = UNSIGNED(arg, df);
1407 n = DF_BITS(df);
1408 c = DF_BITS(df) / 2;
1410 do {
1411 y = x >> c;
1412 if (y != 0) {
1413 n = n - c;
1414 x = y;
1416 c = c >> 1;
1417 } while (c != 0);
1419 return n - x;
1422 static inline int64_t msa_nloc_df(uint32_t df, int64_t arg)
1424 return msa_nlzc_df(df, UNSIGNED((~arg), df));
1427 void helper_msa_fill_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
1428 uint32_t rs)
1430 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
1431 uint32_t i;
1433 switch (df) {
1434 case DF_BYTE:
1435 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) {
1436 pwd->b[i] = (int8_t)env->active_tc.gpr[rs];
1438 break;
1439 case DF_HALF:
1440 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) {
1441 pwd->h[i] = (int16_t)env->active_tc.gpr[rs];
1443 break;
1444 case DF_WORD:
1445 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
1446 pwd->w[i] = (int32_t)env->active_tc.gpr[rs];
1448 break;
1449 case DF_DOUBLE:
1450 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
1451 pwd->d[i] = (int64_t)env->active_tc.gpr[rs];
1453 break;
1454 default:
1455 assert(0);
1459 #define MSA_UNOP_DF(func) \
1460 void helper_msa_ ## func ## _df(CPUMIPSState *env, uint32_t df, \
1461 uint32_t wd, uint32_t ws) \
1463 wr_t *pwd = &(env->active_fpu.fpr[wd].wr); \
1464 wr_t *pws = &(env->active_fpu.fpr[ws].wr); \
1465 uint32_t i; \
1467 switch (df) { \
1468 case DF_BYTE: \
1469 for (i = 0; i < DF_ELEMENTS(DF_BYTE); i++) { \
1470 pwd->b[i] = msa_ ## func ## _df(df, pws->b[i]); \
1472 break; \
1473 case DF_HALF: \
1474 for (i = 0; i < DF_ELEMENTS(DF_HALF); i++) { \
1475 pwd->h[i] = msa_ ## func ## _df(df, pws->h[i]); \
1477 break; \
1478 case DF_WORD: \
1479 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) { \
1480 pwd->w[i] = msa_ ## func ## _df(df, pws->w[i]); \
1482 break; \
1483 case DF_DOUBLE: \
1484 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) { \
1485 pwd->d[i] = msa_ ## func ## _df(df, pws->d[i]); \
1487 break; \
1488 default: \
1489 assert(0); \
1493 MSA_UNOP_DF(nlzc)
1494 MSA_UNOP_DF(nloc)
1495 MSA_UNOP_DF(pcnt)
1496 #undef MSA_UNOP_DF
1498 #define FLOAT_ONE32 make_float32(0x3f8 << 20)
1499 #define FLOAT_ONE64 make_float64(0x3ffULL << 52)
1501 #define FLOAT_SNAN16(s) (float16_default_nan(s) ^ 0x0220)
1502 /* 0x7c20 */
1503 #define FLOAT_SNAN32(s) (float32_default_nan(s) ^ 0x00400020)
1504 /* 0x7f800020 */
1505 #define FLOAT_SNAN64(s) (float64_default_nan(s) ^ 0x0008000000000020ULL)
1506 /* 0x7ff0000000000020 */
1508 static inline void clear_msacsr_cause(CPUMIPSState *env)
1510 SET_FP_CAUSE(env->active_tc.msacsr, 0);
1513 static inline void check_msacsr_cause(CPUMIPSState *env, uintptr_t retaddr)
1515 if ((GET_FP_CAUSE(env->active_tc.msacsr) &
1516 (GET_FP_ENABLE(env->active_tc.msacsr) | FP_UNIMPLEMENTED)) == 0) {
1517 UPDATE_FP_FLAGS(env->active_tc.msacsr,
1518 GET_FP_CAUSE(env->active_tc.msacsr));
1519 } else {
1520 do_raise_exception(env, EXCP_MSAFPE, retaddr);
1524 /* Flush-to-zero use cases for update_msacsr() */
1525 #define CLEAR_FS_UNDERFLOW 1
1526 #define CLEAR_IS_INEXACT 2
1527 #define RECIPROCAL_INEXACT 4
1529 static inline int update_msacsr(CPUMIPSState *env, int action, int denormal)
1531 int ieee_ex;
1533 int c;
1534 int cause;
1535 int enable;
1537 ieee_ex = get_float_exception_flags(&env->active_tc.msa_fp_status);
1539 /* QEMU softfloat does not signal all underflow cases */
1540 if (denormal) {
1541 ieee_ex |= float_flag_underflow;
1544 c = ieee_ex_to_mips(ieee_ex);
1545 enable = GET_FP_ENABLE(env->active_tc.msacsr) | FP_UNIMPLEMENTED;
1547 /* Set Inexact (I) when flushing inputs to zero */
1548 if ((ieee_ex & float_flag_input_denormal) &&
1549 (env->active_tc.msacsr & MSACSR_FS_MASK) != 0) {
1550 if (action & CLEAR_IS_INEXACT) {
1551 c &= ~FP_INEXACT;
1552 } else {
1553 c |= FP_INEXACT;
1557 /* Set Inexact (I) and Underflow (U) when flushing outputs to zero */
1558 if ((ieee_ex & float_flag_output_denormal) &&
1559 (env->active_tc.msacsr & MSACSR_FS_MASK) != 0) {
1560 c |= FP_INEXACT;
1561 if (action & CLEAR_FS_UNDERFLOW) {
1562 c &= ~FP_UNDERFLOW;
1563 } else {
1564 c |= FP_UNDERFLOW;
1568 /* Set Inexact (I) when Overflow (O) is not enabled */
1569 if ((c & FP_OVERFLOW) != 0 && (enable & FP_OVERFLOW) == 0) {
1570 c |= FP_INEXACT;
1573 /* Clear Exact Underflow when Underflow (U) is not enabled */
1574 if ((c & FP_UNDERFLOW) != 0 && (enable & FP_UNDERFLOW) == 0 &&
1575 (c & FP_INEXACT) == 0) {
1576 c &= ~FP_UNDERFLOW;
1579 /* Reciprocal operations set only Inexact when valid and not
1580 divide by zero */
1581 if ((action & RECIPROCAL_INEXACT) &&
1582 (c & (FP_INVALID | FP_DIV0)) == 0) {
1583 c = FP_INEXACT;
1586 cause = c & enable; /* all current enabled exceptions */
1588 if (cause == 0) {
1589 /* No enabled exception, update the MSACSR Cause
1590 with all current exceptions */
1591 SET_FP_CAUSE(env->active_tc.msacsr,
1592 (GET_FP_CAUSE(env->active_tc.msacsr) | c));
1593 } else {
1594 /* Current exceptions are enabled */
1595 if ((env->active_tc.msacsr & MSACSR_NX_MASK) == 0) {
1596 /* Exception(s) will trap, update MSACSR Cause
1597 with all enabled exceptions */
1598 SET_FP_CAUSE(env->active_tc.msacsr,
1599 (GET_FP_CAUSE(env->active_tc.msacsr) | c));
1603 return c;
1606 static inline int get_enabled_exceptions(const CPUMIPSState *env, int c)
1608 int enable = GET_FP_ENABLE(env->active_tc.msacsr) | FP_UNIMPLEMENTED;
1609 return c & enable;
1612 static inline float16 float16_from_float32(int32_t a, flag ieee,
1613 float_status *status)
1615 float16 f_val;
1617 f_val = float32_to_float16((float32)a, ieee, status);
1619 return a < 0 ? (f_val | (1 << 15)) : f_val;
1622 static inline float32 float32_from_float64(int64_t a, float_status *status)
1624 float32 f_val;
1626 f_val = float64_to_float32((float64)a, status);
1628 return a < 0 ? (f_val | (1 << 31)) : f_val;
1631 static inline float32 float32_from_float16(int16_t a, flag ieee,
1632 float_status *status)
1634 float32 f_val;
1636 f_val = float16_to_float32((float16)a, ieee, status);
1638 return a < 0 ? (f_val | (1 << 31)) : f_val;
1641 static inline float64 float64_from_float32(int32_t a, float_status *status)
1643 float64 f_val;
1645 f_val = float32_to_float64((float64)a, status);
1647 return a < 0 ? (f_val | (1ULL << 63)) : f_val;
1650 static inline float32 float32_from_q16(int16_t a, float_status *status)
1652 float32 f_val;
1654 /* conversion as integer and scaling */
1655 f_val = int32_to_float32(a, status);
1656 f_val = float32_scalbn(f_val, -15, status);
1658 return f_val;
1661 static inline float64 float64_from_q32(int32_t a, float_status *status)
1663 float64 f_val;
1665 /* conversion as integer and scaling */
1666 f_val = int32_to_float64(a, status);
1667 f_val = float64_scalbn(f_val, -31, status);
1669 return f_val;
1672 static inline int16_t float32_to_q16(float32 a, float_status *status)
1674 int32_t q_val;
1675 int32_t q_min = 0xffff8000;
1676 int32_t q_max = 0x00007fff;
1678 int ieee_ex;
1680 if (float32_is_any_nan(a)) {
1681 float_raise(float_flag_invalid, status);
1682 return 0;
1685 /* scaling */
1686 a = float32_scalbn(a, 15, status);
1688 ieee_ex = get_float_exception_flags(status);
1689 set_float_exception_flags(ieee_ex & (~float_flag_underflow)
1690 , status);
1692 if (ieee_ex & float_flag_overflow) {
1693 float_raise(float_flag_inexact, status);
1694 return (int32_t)a < 0 ? q_min : q_max;
1697 /* conversion to int */
1698 q_val = float32_to_int32(a, status);
1700 ieee_ex = get_float_exception_flags(status);
1701 set_float_exception_flags(ieee_ex & (~float_flag_underflow)
1702 , status);
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;
1711 if (q_val < q_min) {
1712 float_raise(float_flag_overflow | float_flag_inexact, status);
1713 return (int16_t)q_min;
1716 if (q_max < q_val) {
1717 float_raise(float_flag_overflow | float_flag_inexact, status);
1718 return (int16_t)q_max;
1721 return (int16_t)q_val;
1724 static inline int32_t float64_to_q32(float64 a, float_status *status)
1726 int64_t q_val;
1727 int64_t q_min = 0xffffffff80000000LL;
1728 int64_t q_max = 0x000000007fffffffLL;
1730 int ieee_ex;
1732 if (float64_is_any_nan(a)) {
1733 float_raise(float_flag_invalid, status);
1734 return 0;
1737 /* scaling */
1738 a = float64_scalbn(a, 31, status);
1740 ieee_ex = get_float_exception_flags(status);
1741 set_float_exception_flags(ieee_ex & (~float_flag_underflow)
1742 , status);
1744 if (ieee_ex & float_flag_overflow) {
1745 float_raise(float_flag_inexact, status);
1746 return (int64_t)a < 0 ? q_min : q_max;
1749 /* conversion to integer */
1750 q_val = float64_to_int64(a, status);
1752 ieee_ex = get_float_exception_flags(status);
1753 set_float_exception_flags(ieee_ex & (~float_flag_underflow)
1754 , status);
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;
1763 if (q_val < q_min) {
1764 float_raise(float_flag_overflow | float_flag_inexact, status);
1765 return (int32_t)q_min;
1768 if (q_max < q_val) {
1769 float_raise(float_flag_overflow | float_flag_inexact, status);
1770 return (int32_t)q_max;
1773 return (int32_t)q_val;
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); \
1787 DEST = cond ? M_MAX_UINT(BITS) : 0; \
1788 c = update_msacsr(env, CLEAR_IS_INEXACT, 0); \
1790 if (get_enabled_exceptions(env, c)) { \
1791 DEST = ((FLOAT_SNAN ## BITS(status) >> 6) << 6) | c; \
1793 } while (0)
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; \
1801 } while (0)
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); \
1809 } while (0)
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); \
1817 } while (0)
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); \
1828 } while (0)
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); \
1836 } while (0)
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); \
1844 } while (0)
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); \
1852 } while (0)
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)
1858 wr_t wx, *pwx = &wx;
1859 uint32_t i;
1861 clear_msacsr_cause(env);
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);
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);
1873 break;
1874 default:
1875 assert(0);
1878 check_msacsr_cause(env, retaddr);
1880 msa_move_v(pwd, pwx);
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)
1887 wr_t wx, *pwx = &wx;
1888 uint32_t i;
1890 clear_msacsr_cause(env);
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);
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);
1904 break;
1905 default:
1906 assert(0);
1909 check_msacsr_cause(env, retaddr);
1911 msa_move_v(pwd, pwx);
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)
1918 wr_t wx, *pwx = &wx;
1919 uint32_t i;
1921 clear_msacsr_cause(env);
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);
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);
1933 break;
1934 default:
1935 assert(0);
1938 check_msacsr_cause(env, retaddr);
1940 msa_move_v(pwd, pwx);
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)
1947 wr_t wx, *pwx = &wx;
1948 uint32_t i;
1950 clear_msacsr_cause(env);
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);
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);
1962 break;
1963 default:
1964 assert(0);
1967 check_msacsr_cause(env, retaddr);
1969 msa_move_v(pwd, pwx);
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)
1976 wr_t wx, *pwx = &wx;
1977 uint32_t i;
1979 clear_msacsr_cause(env);
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);
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);
1991 break;
1992 default:
1993 assert(0);
1996 check_msacsr_cause(env, retaddr);
1998 msa_move_v(pwd, pwx);
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)
2005 wr_t wx, *pwx = &wx;
2006 uint32_t i;
2008 clear_msacsr_cause(env);
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);
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);
2020 break;
2021 default:
2022 assert(0);
2025 check_msacsr_cause(env, retaddr);
2027 msa_move_v(pwd, pwx);
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)
2034 wr_t wx, *pwx = &wx;
2035 uint32_t i;
2037 clear_msacsr_cause(env);
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);
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);
2049 break;
2050 default:
2051 assert(0);
2054 check_msacsr_cause(env, retaddr);
2056 msa_move_v(pwd, pwx);
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)
2063 wr_t wx, *pwx = &wx;
2064 uint32_t i;
2066 clear_msacsr_cause(env);
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);
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);
2078 break;
2079 default:
2080 assert(0);
2083 check_msacsr_cause(env, retaddr);
2085 msa_move_v(pwd, pwx);
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)
2092 wr_t wx, *pwx = &wx;
2093 uint32_t i;
2095 clear_msacsr_cause(env);
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);
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);
2107 break;
2108 default:
2109 assert(0);
2112 check_msacsr_cause(env, retaddr);
2114 msa_move_v(pwd, pwx);
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)
2121 wr_t wx, *pwx = &wx;
2122 uint32_t i;
2124 clear_msacsr_cause(env);
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);
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);
2136 break;
2137 default:
2138 assert(0);
2141 check_msacsr_cause(env, retaddr);
2143 msa_move_v(pwd, pwx);
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)
2150 wr_t wx, *pwx = &wx;
2151 uint32_t i;
2153 clear_msacsr_cause(env);
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);
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);
2165 break;
2166 default:
2167 assert(0);
2170 check_msacsr_cause(env, retaddr);
2172 msa_move_v(pwd, pwx);
2175 void helper_msa_fcaf_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2176 uint32_t ws, uint32_t wt)
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());
2184 void helper_msa_fcun_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2185 uint32_t ws, uint32_t wt)
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());
2193 void helper_msa_fceq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2194 uint32_t ws, uint32_t wt)
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());
2202 void helper_msa_fcueq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2203 uint32_t ws, uint32_t wt)
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());
2211 void helper_msa_fclt_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2212 uint32_t ws, uint32_t wt)
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());
2220 void helper_msa_fcult_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2221 uint32_t ws, uint32_t wt)
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());
2229 void helper_msa_fcle_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2230 uint32_t ws, uint32_t wt)
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());
2238 void helper_msa_fcule_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2239 uint32_t ws, uint32_t wt)
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());
2247 void helper_msa_fsaf_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2248 uint32_t ws, uint32_t wt)
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());
2256 void helper_msa_fsun_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2257 uint32_t ws, uint32_t wt)
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());
2265 void helper_msa_fseq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2266 uint32_t ws, uint32_t wt)
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());
2274 void helper_msa_fsueq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2275 uint32_t ws, uint32_t wt)
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());
2283 void helper_msa_fslt_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2284 uint32_t ws, uint32_t wt)
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());
2292 void helper_msa_fsult_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2293 uint32_t ws, uint32_t wt)
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());
2301 void helper_msa_fsle_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2302 uint32_t ws, uint32_t wt)
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());
2310 void helper_msa_fsule_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2311 uint32_t ws, uint32_t wt)
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());
2319 void helper_msa_fcor_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2320 uint32_t ws, uint32_t wt)
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());
2328 void helper_msa_fcune_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2329 uint32_t ws, uint32_t wt)
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());
2337 void helper_msa_fcne_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2338 uint32_t ws, uint32_t wt)
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());
2346 void helper_msa_fsor_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2347 uint32_t ws, uint32_t wt)
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());
2355 void helper_msa_fsune_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2356 uint32_t ws, uint32_t wt)
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());
2364 void helper_msa_fsne_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2365 uint32_t ws, uint32_t wt)
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());
2373 #define float16_is_zero(ARG) 0
2374 #define float16_is_zero_or_denormal(ARG) 0
2376 #define IS_DENORMAL(ARG, BITS) \
2377 (!float ## BITS ## _is_zero(ARG) \
2378 && float ## BITS ## _is_zero_or_denormal(ARG))
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; \
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)); \
2389 if (get_enabled_exceptions(env, c)) { \
2390 DEST = ((FLOAT_SNAN ## BITS(status) >> 6) << 6) | c; \
2392 } while (0)
2394 void helper_msa_fadd_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2395 uint32_t ws, uint32_t wt)
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;
2403 clear_msacsr_cause(env);
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);
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);
2415 break;
2416 default:
2417 assert(0);
2420 check_msacsr_cause(env, GETPC());
2421 msa_move_v(pwd, pwx);
2424 void helper_msa_fsub_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2425 uint32_t ws, uint32_t wt)
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;
2433 clear_msacsr_cause(env);
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);
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);
2445 break;
2446 default:
2447 assert(0);
2450 check_msacsr_cause(env, GETPC());
2451 msa_move_v(pwd, pwx);
2454 void helper_msa_fmul_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2455 uint32_t ws, uint32_t wt)
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;
2463 clear_msacsr_cause(env);
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);
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);
2475 break;
2476 default:
2477 assert(0);
2480 check_msacsr_cause(env, GETPC());
2482 msa_move_v(pwd, pwx);
2485 void helper_msa_fdiv_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2486 uint32_t ws, uint32_t wt)
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;
2494 clear_msacsr_cause(env);
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);
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);
2506 break;
2507 default:
2508 assert(0);
2511 check_msacsr_cause(env, GETPC());
2513 msa_move_v(pwd, pwx);
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; \
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)); \
2525 if (get_enabled_exceptions(env, c)) { \
2526 DEST = ((FLOAT_SNAN ## BITS(status) >> 6) << 6) | c; \
2528 } while (0)
2530 void helper_msa_fmadd_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2531 uint32_t ws, uint32_t wt)
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;
2539 clear_msacsr_cause(env);
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);
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);
2553 break;
2554 default:
2555 assert(0);
2558 check_msacsr_cause(env, GETPC());
2560 msa_move_v(pwd, pwx);
2563 void helper_msa_fmsub_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2564 uint32_t ws, uint32_t wt)
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;
2572 clear_msacsr_cause(env);
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);
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);
2588 break;
2589 default:
2590 assert(0);
2593 check_msacsr_cause(env, GETPC());
2595 msa_move_v(pwd, pwx);
2598 void helper_msa_fexp2_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2599 uint32_t ws, uint32_t wt)
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;
2607 clear_msacsr_cause(env);
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);
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);
2625 break;
2626 default:
2627 assert(0);
2630 check_msacsr_cause(env, GETPC());
2632 msa_move_v(pwd, pwx);
2635 #define MSA_FLOAT_UNOP(DEST, OP, ARG, BITS) \
2636 do { \
2637 float_status *status = &env->active_tc.msa_fp_status; \
2638 int c; \
2640 set_float_exception_flags(0, status); \
2641 DEST = float ## BITS ## _ ## OP(ARG, status); \
2642 c = update_msacsr(env, 0, IS_DENORMAL(DEST, BITS)); \
2644 if (get_enabled_exceptions(env, c)) { \
2645 DEST = ((FLOAT_SNAN ## BITS(status) >> 6) << 6) | c; \
2647 } while (0)
2649 void helper_msa_fexdo_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2650 uint32_t ws, uint32_t wt)
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;
2658 clear_msacsr_cause(env);
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;
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);
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);
2677 break;
2678 default:
2679 assert(0);
2682 check_msacsr_cause(env, GETPC());
2683 msa_move_v(pwd, pwx);
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; \
2691 set_float_exception_flags(0, status); \
2692 DEST = float ## BITS ## _ ## OP(ARG, status); \
2693 c = update_msacsr(env, CLEAR_FS_UNDERFLOW, 0); \
2695 if (get_enabled_exceptions(env, c)) { \
2696 DEST = ((FLOAT_SNAN ## XBITS(status) >> 6) << 6) | c; \
2698 } while (0)
2700 void helper_msa_ftq_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2701 uint32_t ws, uint32_t wt)
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;
2709 clear_msacsr_cause(env);
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);
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);
2723 break;
2724 default:
2725 assert(0);
2728 check_msacsr_cause(env, GETPC());
2730 msa_move_v(pwd, pwx);
2733 #define NUMBER_QNAN_PAIR(ARG1, ARG2, BITS, STATUS) \
2734 !float ## BITS ## _is_any_nan(ARG1) \
2735 && float ## BITS ## _is_quiet_nan(ARG2, STATUS)
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; \
2742 set_float_exception_flags(0, status); \
2743 DEST = float ## BITS ## _ ## OP(ARG1, ARG2, status); \
2744 c = update_msacsr(env, 0, 0); \
2746 if (get_enabled_exceptions(env, c)) { \
2747 DEST = ((FLOAT_SNAN ## BITS(status) >> 6) << 6) | c; \
2749 } while (0)
2751 #define FMAXMIN_A(F, G, X, _S, _T, BITS, STATUS) \
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, STATUS)) { \
2756 T = S; \
2758 else if (NUMBER_QNAN_PAIR(T, S, BITS, STATUS)) { \
2759 S = T; \
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)
2769 void helper_msa_fmin_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2770 uint32_t ws, uint32_t wt)
2772 float_status *status = &env->active_tc.msa_fp_status;
2773 wr_t wx, *pwx = &wx;
2774 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2775 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2776 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2777 uint32_t i;
2779 clear_msacsr_cause(env);
2781 switch (df) {
2782 case DF_WORD:
2783 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2784 if (NUMBER_QNAN_PAIR(pws->w[i], pwt->w[i], 32, status)) {
2785 MSA_FLOAT_MAXOP(pwx->w[i], min, pws->w[i], pws->w[i], 32);
2786 } else if (NUMBER_QNAN_PAIR(pwt->w[i], pws->w[i], 32, status)) {
2787 MSA_FLOAT_MAXOP(pwx->w[i], min, pwt->w[i], pwt->w[i], 32);
2788 } else {
2789 MSA_FLOAT_MAXOP(pwx->w[i], min, pws->w[i], pwt->w[i], 32);
2792 break;
2793 case DF_DOUBLE:
2794 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2795 if (NUMBER_QNAN_PAIR(pws->d[i], pwt->d[i], 64, status)) {
2796 MSA_FLOAT_MAXOP(pwx->d[i], min, pws->d[i], pws->d[i], 64);
2797 } else if (NUMBER_QNAN_PAIR(pwt->d[i], pws->d[i], 64, status)) {
2798 MSA_FLOAT_MAXOP(pwx->d[i], min, pwt->d[i], pwt->d[i], 64);
2799 } else {
2800 MSA_FLOAT_MAXOP(pwx->d[i], min, pws->d[i], pwt->d[i], 64);
2803 break;
2804 default:
2805 assert(0);
2808 check_msacsr_cause(env, GETPC());
2810 msa_move_v(pwd, pwx);
2813 void helper_msa_fmin_a_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2814 uint32_t ws, uint32_t wt)
2816 float_status *status = &env->active_tc.msa_fp_status;
2817 wr_t wx, *pwx = &wx;
2818 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2819 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2820 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2821 uint32_t i;
2823 clear_msacsr_cause(env);
2825 switch (df) {
2826 case DF_WORD:
2827 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2828 FMAXMIN_A(min, max, pwx->w[i], pws->w[i], pwt->w[i], 32, status);
2830 break;
2831 case DF_DOUBLE:
2832 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2833 FMAXMIN_A(min, max, pwx->d[i], pws->d[i], pwt->d[i], 64, status);
2835 break;
2836 default:
2837 assert(0);
2840 check_msacsr_cause(env, GETPC());
2842 msa_move_v(pwd, pwx);
2845 void helper_msa_fmax_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2846 uint32_t ws, uint32_t wt)
2848 float_status *status = &env->active_tc.msa_fp_status;
2849 wr_t wx, *pwx = &wx;
2850 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2851 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2852 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2853 uint32_t i;
2855 clear_msacsr_cause(env);
2857 switch (df) {
2858 case DF_WORD:
2859 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2860 if (NUMBER_QNAN_PAIR(pws->w[i], pwt->w[i], 32, status)) {
2861 MSA_FLOAT_MAXOP(pwx->w[i], max, pws->w[i], pws->w[i], 32);
2862 } else if (NUMBER_QNAN_PAIR(pwt->w[i], pws->w[i], 32, status)) {
2863 MSA_FLOAT_MAXOP(pwx->w[i], max, pwt->w[i], pwt->w[i], 32);
2864 } else {
2865 MSA_FLOAT_MAXOP(pwx->w[i], max, pws->w[i], pwt->w[i], 32);
2868 break;
2869 case DF_DOUBLE:
2870 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2871 if (NUMBER_QNAN_PAIR(pws->d[i], pwt->d[i], 64, status)) {
2872 MSA_FLOAT_MAXOP(pwx->d[i], max, pws->d[i], pws->d[i], 64);
2873 } else if (NUMBER_QNAN_PAIR(pwt->d[i], pws->d[i], 64, status)) {
2874 MSA_FLOAT_MAXOP(pwx->d[i], max, pwt->d[i], pwt->d[i], 64);
2875 } else {
2876 MSA_FLOAT_MAXOP(pwx->d[i], max, pws->d[i], pwt->d[i], 64);
2879 break;
2880 default:
2881 assert(0);
2884 check_msacsr_cause(env, GETPC());
2886 msa_move_v(pwd, pwx);
2889 void helper_msa_fmax_a_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2890 uint32_t ws, uint32_t wt)
2892 float_status *status = &env->active_tc.msa_fp_status;
2893 wr_t wx, *pwx = &wx;
2894 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2895 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2896 wr_t *pwt = &(env->active_fpu.fpr[wt].wr);
2897 uint32_t i;
2899 clear_msacsr_cause(env);
2901 switch (df) {
2902 case DF_WORD:
2903 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2904 FMAXMIN_A(max, min, pwx->w[i], pws->w[i], pwt->w[i], 32, status);
2906 break;
2907 case DF_DOUBLE:
2908 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2909 FMAXMIN_A(max, min, pwx->d[i], pws->d[i], pwt->d[i], 64, status);
2911 break;
2912 default:
2913 assert(0);
2916 check_msacsr_cause(env, GETPC());
2918 msa_move_v(pwd, pwx);
2921 void helper_msa_fclass_df(CPUMIPSState *env, uint32_t df,
2922 uint32_t wd, uint32_t ws)
2924 float_status* status = &env->active_tc.msa_fp_status;
2926 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2927 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2928 if (df == DF_WORD) {
2929 pwd->w[0] = float_class_s(pws->w[0], status);
2930 pwd->w[1] = float_class_s(pws->w[1], status);
2931 pwd->w[2] = float_class_s(pws->w[2], status);
2932 pwd->w[3] = float_class_s(pws->w[3], status);
2933 } else {
2934 pwd->d[0] = float_class_d(pws->d[0], status);
2935 pwd->d[1] = float_class_d(pws->d[1], status);
2939 #define MSA_FLOAT_UNOP0(DEST, OP, ARG, BITS) \
2940 do { \
2941 float_status *status = &env->active_tc.msa_fp_status; \
2942 int c; \
2944 set_float_exception_flags(0, status); \
2945 DEST = float ## BITS ## _ ## OP(ARG, status); \
2946 c = update_msacsr(env, CLEAR_FS_UNDERFLOW, 0); \
2948 if (get_enabled_exceptions(env, c)) { \
2949 DEST = ((FLOAT_SNAN ## BITS(status) >> 6) << 6) | c; \
2950 } else if (float ## BITS ## _is_any_nan(ARG)) { \
2951 DEST = 0; \
2953 } while (0)
2955 void helper_msa_ftrunc_s_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2956 uint32_t ws)
2958 wr_t wx, *pwx = &wx;
2959 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2960 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2961 uint32_t i;
2963 clear_msacsr_cause(env);
2965 switch (df) {
2966 case DF_WORD:
2967 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2968 MSA_FLOAT_UNOP0(pwx->w[i], to_int32_round_to_zero, pws->w[i], 32);
2970 break;
2971 case DF_DOUBLE:
2972 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
2973 MSA_FLOAT_UNOP0(pwx->d[i], to_int64_round_to_zero, pws->d[i], 64);
2975 break;
2976 default:
2977 assert(0);
2980 check_msacsr_cause(env, GETPC());
2982 msa_move_v(pwd, pwx);
2985 void helper_msa_ftrunc_u_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
2986 uint32_t ws)
2988 wr_t wx, *pwx = &wx;
2989 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
2990 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
2991 uint32_t i;
2993 clear_msacsr_cause(env);
2995 switch (df) {
2996 case DF_WORD:
2997 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
2998 MSA_FLOAT_UNOP0(pwx->w[i], to_uint32_round_to_zero, pws->w[i], 32);
3000 break;
3001 case DF_DOUBLE:
3002 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3003 MSA_FLOAT_UNOP0(pwx->d[i], to_uint64_round_to_zero, pws->d[i], 64);
3005 break;
3006 default:
3007 assert(0);
3010 check_msacsr_cause(env, GETPC());
3012 msa_move_v(pwd, pwx);
3015 void helper_msa_fsqrt_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3016 uint32_t ws)
3018 wr_t wx, *pwx = &wx;
3019 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3020 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3021 uint32_t i;
3023 clear_msacsr_cause(env);
3025 switch (df) {
3026 case DF_WORD:
3027 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3028 MSA_FLOAT_UNOP(pwx->w[i], sqrt, pws->w[i], 32);
3030 break;
3031 case DF_DOUBLE:
3032 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3033 MSA_FLOAT_UNOP(pwx->d[i], sqrt, pws->d[i], 64);
3035 break;
3036 default:
3037 assert(0);
3040 check_msacsr_cause(env, GETPC());
3042 msa_move_v(pwd, pwx);
3045 #define MSA_FLOAT_RECIPROCAL(DEST, ARG, BITS) \
3046 do { \
3047 float_status *status = &env->active_tc.msa_fp_status; \
3048 int c; \
3050 set_float_exception_flags(0, status); \
3051 DEST = float ## BITS ## _ ## div(FLOAT_ONE ## BITS, ARG, status); \
3052 c = update_msacsr(env, float ## BITS ## _is_infinity(ARG) || \
3053 float ## BITS ## _is_quiet_nan(DEST, status) ? \
3054 0 : RECIPROCAL_INEXACT, \
3055 IS_DENORMAL(DEST, BITS)); \
3057 if (get_enabled_exceptions(env, c)) { \
3058 DEST = ((FLOAT_SNAN ## BITS(status) >> 6) << 6) | c; \
3060 } while (0)
3062 void helper_msa_frsqrt_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3063 uint32_t ws)
3065 wr_t wx, *pwx = &wx;
3066 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3067 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3068 uint32_t i;
3070 clear_msacsr_cause(env);
3072 switch (df) {
3073 case DF_WORD:
3074 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3075 MSA_FLOAT_RECIPROCAL(pwx->w[i], float32_sqrt(pws->w[i],
3076 &env->active_tc.msa_fp_status), 32);
3078 break;
3079 case DF_DOUBLE:
3080 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3081 MSA_FLOAT_RECIPROCAL(pwx->d[i], float64_sqrt(pws->d[i],
3082 &env->active_tc.msa_fp_status), 64);
3084 break;
3085 default:
3086 assert(0);
3089 check_msacsr_cause(env, GETPC());
3091 msa_move_v(pwd, pwx);
3094 void helper_msa_frcp_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3095 uint32_t ws)
3097 wr_t wx, *pwx = &wx;
3098 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3099 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3100 uint32_t i;
3102 clear_msacsr_cause(env);
3104 switch (df) {
3105 case DF_WORD:
3106 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3107 MSA_FLOAT_RECIPROCAL(pwx->w[i], pws->w[i], 32);
3109 break;
3110 case DF_DOUBLE:
3111 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3112 MSA_FLOAT_RECIPROCAL(pwx->d[i], pws->d[i], 64);
3114 break;
3115 default:
3116 assert(0);
3119 check_msacsr_cause(env, GETPC());
3121 msa_move_v(pwd, pwx);
3124 void helper_msa_frint_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3125 uint32_t ws)
3127 wr_t wx, *pwx = &wx;
3128 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3129 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3130 uint32_t i;
3132 clear_msacsr_cause(env);
3134 switch (df) {
3135 case DF_WORD:
3136 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3137 MSA_FLOAT_UNOP(pwx->w[i], round_to_int, pws->w[i], 32);
3139 break;
3140 case DF_DOUBLE:
3141 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3142 MSA_FLOAT_UNOP(pwx->d[i], round_to_int, pws->d[i], 64);
3144 break;
3145 default:
3146 assert(0);
3149 check_msacsr_cause(env, GETPC());
3151 msa_move_v(pwd, pwx);
3154 #define MSA_FLOAT_LOGB(DEST, ARG, BITS) \
3155 do { \
3156 float_status *status = &env->active_tc.msa_fp_status; \
3157 int c; \
3159 set_float_exception_flags(0, status); \
3160 set_float_rounding_mode(float_round_down, status); \
3161 DEST = float ## BITS ## _ ## log2(ARG, status); \
3162 DEST = float ## BITS ## _ ## round_to_int(DEST, status); \
3163 set_float_rounding_mode(ieee_rm[(env->active_tc.msacsr & \
3164 MSACSR_RM_MASK) >> MSACSR_RM], \
3165 status); \
3167 set_float_exception_flags(get_float_exception_flags(status) & \
3168 (~float_flag_inexact), \
3169 status); \
3171 c = update_msacsr(env, 0, IS_DENORMAL(DEST, BITS)); \
3173 if (get_enabled_exceptions(env, c)) { \
3174 DEST = ((FLOAT_SNAN ## BITS(status) >> 6) << 6) | c; \
3176 } while (0)
3178 void helper_msa_flog2_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3179 uint32_t ws)
3181 wr_t wx, *pwx = &wx;
3182 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3183 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3184 uint32_t i;
3186 clear_msacsr_cause(env);
3188 switch (df) {
3189 case DF_WORD:
3190 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3191 MSA_FLOAT_LOGB(pwx->w[i], pws->w[i], 32);
3193 break;
3194 case DF_DOUBLE:
3195 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3196 MSA_FLOAT_LOGB(pwx->d[i], pws->d[i], 64);
3198 break;
3199 default:
3200 assert(0);
3203 check_msacsr_cause(env, GETPC());
3205 msa_move_v(pwd, pwx);
3208 void helper_msa_fexupl_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3209 uint32_t ws)
3211 wr_t wx, *pwx = &wx;
3212 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3213 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3214 uint32_t i;
3216 clear_msacsr_cause(env);
3218 switch (df) {
3219 case DF_WORD:
3220 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3221 /* Half precision floats come in two formats: standard
3222 IEEE and "ARM" format. The latter gains extra exponent
3223 range by omitting the NaN/Inf encodings. */
3224 flag ieee = 1;
3226 MSA_FLOAT_BINOP(pwx->w[i], from_float16, Lh(pws, i), ieee, 32);
3228 break;
3229 case DF_DOUBLE:
3230 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3231 MSA_FLOAT_UNOP(pwx->d[i], from_float32, Lw(pws, i), 64);
3233 break;
3234 default:
3235 assert(0);
3238 check_msacsr_cause(env, GETPC());
3239 msa_move_v(pwd, pwx);
3242 void helper_msa_fexupr_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3243 uint32_t ws)
3245 wr_t wx, *pwx = &wx;
3246 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3247 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3248 uint32_t i;
3250 clear_msacsr_cause(env);
3252 switch (df) {
3253 case DF_WORD:
3254 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3255 /* Half precision floats come in two formats: standard
3256 IEEE and "ARM" format. The latter gains extra exponent
3257 range by omitting the NaN/Inf encodings. */
3258 flag ieee = 1;
3260 MSA_FLOAT_BINOP(pwx->w[i], from_float16, Rh(pws, i), ieee, 32);
3262 break;
3263 case DF_DOUBLE:
3264 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3265 MSA_FLOAT_UNOP(pwx->d[i], from_float32, Rw(pws, i), 64);
3267 break;
3268 default:
3269 assert(0);
3272 check_msacsr_cause(env, GETPC());
3273 msa_move_v(pwd, pwx);
3276 void helper_msa_ffql_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3277 uint32_t ws)
3279 wr_t wx, *pwx = &wx;
3280 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3281 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3282 uint32_t i;
3284 switch (df) {
3285 case DF_WORD:
3286 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3287 MSA_FLOAT_UNOP(pwx->w[i], from_q16, Lh(pws, i), 32);
3289 break;
3290 case DF_DOUBLE:
3291 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3292 MSA_FLOAT_UNOP(pwx->d[i], from_q32, Lw(pws, i), 64);
3294 break;
3295 default:
3296 assert(0);
3299 msa_move_v(pwd, pwx);
3302 void helper_msa_ffqr_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3303 uint32_t ws)
3305 wr_t wx, *pwx = &wx;
3306 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3307 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3308 uint32_t i;
3310 switch (df) {
3311 case DF_WORD:
3312 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3313 MSA_FLOAT_UNOP(pwx->w[i], from_q16, Rh(pws, i), 32);
3315 break;
3316 case DF_DOUBLE:
3317 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3318 MSA_FLOAT_UNOP(pwx->d[i], from_q32, Rw(pws, i), 64);
3320 break;
3321 default:
3322 assert(0);
3325 msa_move_v(pwd, pwx);
3328 void helper_msa_ftint_s_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3329 uint32_t ws)
3331 wr_t wx, *pwx = &wx;
3332 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3333 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3334 uint32_t i;
3336 clear_msacsr_cause(env);
3338 switch (df) {
3339 case DF_WORD:
3340 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3341 MSA_FLOAT_UNOP0(pwx->w[i], to_int32, pws->w[i], 32);
3343 break;
3344 case DF_DOUBLE:
3345 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3346 MSA_FLOAT_UNOP0(pwx->d[i], to_int64, pws->d[i], 64);
3348 break;
3349 default:
3350 assert(0);
3353 check_msacsr_cause(env, GETPC());
3355 msa_move_v(pwd, pwx);
3358 void helper_msa_ftint_u_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3359 uint32_t ws)
3361 wr_t wx, *pwx = &wx;
3362 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3363 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3364 uint32_t i;
3366 clear_msacsr_cause(env);
3368 switch (df) {
3369 case DF_WORD:
3370 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3371 MSA_FLOAT_UNOP0(pwx->w[i], to_uint32, pws->w[i], 32);
3373 break;
3374 case DF_DOUBLE:
3375 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3376 MSA_FLOAT_UNOP0(pwx->d[i], to_uint64, pws->d[i], 64);
3378 break;
3379 default:
3380 assert(0);
3383 check_msacsr_cause(env, GETPC());
3385 msa_move_v(pwd, pwx);
3388 #define float32_from_int32 int32_to_float32
3389 #define float32_from_uint32 uint32_to_float32
3391 #define float64_from_int64 int64_to_float64
3392 #define float64_from_uint64 uint64_to_float64
3394 void helper_msa_ffint_s_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3395 uint32_t ws)
3397 wr_t wx, *pwx = &wx;
3398 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3399 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3400 uint32_t i;
3402 clear_msacsr_cause(env);
3404 switch (df) {
3405 case DF_WORD:
3406 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3407 MSA_FLOAT_UNOP(pwx->w[i], from_int32, pws->w[i], 32);
3409 break;
3410 case DF_DOUBLE:
3411 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3412 MSA_FLOAT_UNOP(pwx->d[i], from_int64, pws->d[i], 64);
3414 break;
3415 default:
3416 assert(0);
3419 check_msacsr_cause(env, GETPC());
3421 msa_move_v(pwd, pwx);
3424 void helper_msa_ffint_u_df(CPUMIPSState *env, uint32_t df, uint32_t wd,
3425 uint32_t ws)
3427 wr_t wx, *pwx = &wx;
3428 wr_t *pwd = &(env->active_fpu.fpr[wd].wr);
3429 wr_t *pws = &(env->active_fpu.fpr[ws].wr);
3430 uint32_t i;
3432 clear_msacsr_cause(env);
3434 switch (df) {
3435 case DF_WORD:
3436 for (i = 0; i < DF_ELEMENTS(DF_WORD); i++) {
3437 MSA_FLOAT_UNOP(pwx->w[i], from_uint32, pws->w[i], 32);
3439 break;
3440 case DF_DOUBLE:
3441 for (i = 0; i < DF_ELEMENTS(DF_DOUBLE); i++) {
3442 MSA_FLOAT_UNOP(pwx->d[i], from_uint64, pws->d[i], 64);
3444 break;
3445 default:
3446 assert(0);
3449 check_msacsr_cause(env, GETPC());
3451 msa_move_v(pwd, pwx);