?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
pnv_phb4.c: change TYPE_PNV_PHB4_ROOT_BUS name
[qemu.git] / target / ppc / fpu_helper.c
blobe5c29b53b8b55f2d09862b39ab6bbe5b77c5fd67
1 /*
2 * PowerPC floating point and SPE emulation helpers for QEMU.
3 *
4 * Copyright (c) 2003-2007 Jocelyn Mayer
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 2.1 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 #include "qemu/osdep.h"
20 #include "cpu.h"
21 #include "exec/helper-proto.h"
22 #include "exec/exec-all.h"
23 #include "internal.h"
24 #include "fpu/softfloat.h"
25
26 static inline float128 float128_snan_to_qnan(float128 x)
27 {
28 float128 r;
29
30 r.high = x.high | 0x0000800000000000;
31 r.low = x.low;
32 return r;
33 }
34
35 #define float64_snan_to_qnan(x) ((x) | 0x0008000000000000ULL)
36 #define float32_snan_to_qnan(x) ((x) | 0x00400000)
37 #define float16_snan_to_qnan(x) ((x) | 0x0200)
38
39 static inline bool fp_exceptions_enabled(CPUPPCState *env)
40 {
41 #ifdef CONFIG_USER_ONLY
42 return true;
43 #else
44 return (env->msr & ((1U << MSR_FE0) | (1U << MSR_FE1))) != 0;
45 #endif
46 }
47
48 /*****************************************************************************/
49 /* Floating point operations helpers */
50
51 /*
52 * This is the non-arithmatic conversion that happens e.g. on loads.
53 * In the Power ISA pseudocode, this is called DOUBLE.
54 */
55 uint64_t helper_todouble(uint32_t arg)
56 {
57 uint32_t abs_arg = arg & 0x7fffffff;
58 uint64_t ret;
59
60 if (likely(abs_arg >= 0x00800000)) {
61 if (unlikely(extract32(arg, 23, 8) == 0xff)) {
62 /* Inf or NAN. */
63 ret = (uint64_t)extract32(arg, 31, 1) << 63;
64 ret |= (uint64_t)0x7ff << 52;
65 ret |= (uint64_t)extract32(arg, 0, 23) << 29;
66 } else {
67 /* Normalized operand. */
68 ret = (uint64_t)extract32(arg, 30, 2) << 62;
69 ret |= ((extract32(arg, 30, 1) ^ 1) * (uint64_t)7) << 59;
70 ret |= (uint64_t)extract32(arg, 0, 30) << 29;
71 }
72 } else {
73 /* Zero or Denormalized operand. */
74 ret = (uint64_t)extract32(arg, 31, 1) << 63;
75 if (unlikely(abs_arg != 0)) {
76 /*
77 * Denormalized operand.
78 * Shift fraction so that the msb is in the implicit bit position.
79 * Thus, shift is in the range [1:23].
80 */
81 int shift = clz32(abs_arg) - 8;
82 /*
83 * The first 3 terms compute the float64 exponent. We then bias
84 * this result by -1 so that we can swallow the implicit bit below.
85 */
86 int exp = -126 - shift + 1023 - 1;
87
88 ret |= (uint64_t)exp << 52;
89 ret += (uint64_t)abs_arg << (52 - 23 + shift);
90 }
91 }
92 return ret;
93 }
94
95 /*
96 * This is the non-arithmatic conversion that happens e.g. on stores.
97 * In the Power ISA pseudocode, this is called SINGLE.
98 */
99 uint32_t helper_tosingle(uint64_t arg)
100 {
101 int exp = extract64(arg, 52, 11);
102 uint32_t ret;
103
104 if (likely(exp > 896)) {
105 /* No denormalization required (includes Inf, NaN). */
106 ret = extract64(arg, 62, 2) << 30;
107 ret |= extract64(arg, 29, 30);
108 } else {
109 /*
110 * Zero or Denormal result. If the exponent is in bounds for
111 * a single-precision denormal result, extract the proper
112 * bits. If the input is not zero, and the exponent is out of
113 * bounds, then the result is undefined; this underflows to
114 * zero.
115 */
116 ret = extract64(arg, 63, 1) << 31;
117 if (unlikely(exp >= 874)) {
118 /* Denormal result. */
119 ret |= ((1ULL << 52) | extract64(arg, 0, 52)) >> (896 + 30 - exp);
120 }
121 }
122 return ret;
123 }
124
125 static inline int ppc_float32_get_unbiased_exp(float32 f)
126 {
127 return ((f >> 23) & 0xFF) - 127;
128 }
129
130 static inline int ppc_float64_get_unbiased_exp(float64 f)
131 {
132 return ((f >> 52) & 0x7FF) - 1023;
133 }
134
135 /* Classify a floating-point number. */
136 enum {
137 is_normal = 1,
138 is_zero = 2,
139 is_denormal = 4,
140 is_inf = 8,
141 is_qnan = 16,
142 is_snan = 32,
143 is_neg = 64,
144 };
145
146 #define COMPUTE_CLASS(tp) \
147 static int tp##_classify(tp arg) \
148 { \
149 int ret = tp##_is_neg(arg) * is_neg; \
150 if (unlikely(tp##_is_any_nan(arg))) { \
151 float_status dummy = { }; /* snan_bit_is_one = 0 */ \
152 ret |= (tp##_is_signaling_nan(arg, &dummy) \
153 ? is_snan : is_qnan); \
154 } else if (unlikely(tp##_is_infinity(arg))) { \
155 ret |= is_inf; \
156 } else if (tp##_is_zero(arg)) { \
157 ret |= is_zero; \
158 } else if (tp##_is_zero_or_denormal(arg)) { \
159 ret |= is_denormal; \
160 } else { \
161 ret |= is_normal; \
162 } \
163 return ret; \
164 }
165
166 COMPUTE_CLASS(float16)
167 COMPUTE_CLASS(float32)
168 COMPUTE_CLASS(float64)
169 COMPUTE_CLASS(float128)
170
171 static void set_fprf_from_class(CPUPPCState *env, int class)
172 {
173 static const uint8_t fprf[6][2] = {
174 { 0x04, 0x08 }, /* normalized */
175 { 0x02, 0x12 }, /* zero */
176 { 0x14, 0x18 }, /* denormalized */
177 { 0x05, 0x09 }, /* infinity */
178 { 0x11, 0x11 }, /* qnan */
179 { 0x00, 0x00 }, /* snan -- flags are undefined */
180 };
181 bool isneg = class & is_neg;
182
183 env->fpscr &= ~FP_FPRF;
184 env->fpscr |= fprf[ctz32(class)][isneg] << FPSCR_FPRF;
185 }
186
187 #define COMPUTE_FPRF(tp) \
188 void helper_compute_fprf_##tp(CPUPPCState *env, tp arg) \
189 { \
190 set_fprf_from_class(env, tp##_classify(arg)); \
191 }
192
193 COMPUTE_FPRF(float16)
194 COMPUTE_FPRF(float32)
195 COMPUTE_FPRF(float64)
196 COMPUTE_FPRF(float128)
197
198 /* Floating-point invalid operations exception */
199 static void finish_invalid_op_excp(CPUPPCState *env, int op, uintptr_t retaddr)
200 {
201 /* Update the floating-point invalid operation summary */
202 env->fpscr |= FP_VX;
203 /* Update the floating-point exception summary */
204 env->fpscr |= FP_FX;
205 if (fpscr_ve != 0) {
206 /* Update the floating-point enabled exception summary */
207 env->fpscr |= FP_FEX;
208 if (fp_exceptions_enabled(env)) {
209 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
210 POWERPC_EXCP_FP | op, retaddr);
211 }
212 }
213 }
214
215 static void finish_invalid_op_arith(CPUPPCState *env, int op,
216 bool set_fpcc, uintptr_t retaddr)
217 {
218 env->fpscr &= ~(FP_FR | FP_FI);
219 if (fpscr_ve == 0) {
220 if (set_fpcc) {
221 env->fpscr &= ~FP_FPCC;
222 env->fpscr |= (FP_C | FP_FU);
223 }
224 }
225 finish_invalid_op_excp(env, op, retaddr);
226 }
227
228 /* Signalling NaN */
229 static void float_invalid_op_vxsnan(CPUPPCState *env, uintptr_t retaddr)
230 {
231 env->fpscr |= FP_VXSNAN;
232 finish_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, retaddr);
233 }
234
235 /* Magnitude subtraction of infinities */
236 static void float_invalid_op_vxisi(CPUPPCState *env, bool set_fpcc,
237 uintptr_t retaddr)
238 {
239 env->fpscr |= FP_VXISI;
240 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXISI, set_fpcc, retaddr);
241 }
242
243 /* Division of infinity by infinity */
244 static void float_invalid_op_vxidi(CPUPPCState *env, bool set_fpcc,
245 uintptr_t retaddr)
246 {
247 env->fpscr |= FP_VXIDI;
248 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXIDI, set_fpcc, retaddr);
249 }
250
251 /* Division of zero by zero */
252 static void float_invalid_op_vxzdz(CPUPPCState *env, bool set_fpcc,
253 uintptr_t retaddr)
254 {
255 env->fpscr |= FP_VXZDZ;
256 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXZDZ, set_fpcc, retaddr);
257 }
258
259 /* Multiplication of zero by infinity */
260 static void float_invalid_op_vximz(CPUPPCState *env, bool set_fpcc,
261 uintptr_t retaddr)
262 {
263 env->fpscr |= FP_VXIMZ;
264 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXIMZ, set_fpcc, retaddr);
265 }
266
267 /* Square root of a negative number */
268 static void float_invalid_op_vxsqrt(CPUPPCState *env, bool set_fpcc,
269 uintptr_t retaddr)
270 {
271 env->fpscr |= FP_VXSQRT;
272 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXSQRT, set_fpcc, retaddr);
273 }
274
275 /* Ordered comparison of NaN */
276 static void float_invalid_op_vxvc(CPUPPCState *env, bool set_fpcc,
277 uintptr_t retaddr)
278 {
279 env->fpscr |= FP_VXVC;
280 if (set_fpcc) {
281 env->fpscr &= ~FP_FPCC;
282 env->fpscr |= (FP_C | FP_FU);
283 }
284 /* Update the floating-point invalid operation summary */
285 env->fpscr |= FP_VX;
286 /* Update the floating-point exception summary */
287 env->fpscr |= FP_FX;
288 /* We must update the target FPR before raising the exception */
289 if (fpscr_ve != 0) {
290 CPUState *cs = env_cpu(env);
291
292 cs->exception_index = POWERPC_EXCP_PROGRAM;
293 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_VXVC;
294 /* Update the floating-point enabled exception summary */
295 env->fpscr |= FP_FEX;
296 /* Exception is deferred */
297 }
298 }
299
300 /* Invalid conversion */
301 static void float_invalid_op_vxcvi(CPUPPCState *env, bool set_fpcc,
302 uintptr_t retaddr)
303 {
304 env->fpscr |= FP_VXCVI;
305 env->fpscr &= ~(FP_FR | FP_FI);
306 if (fpscr_ve == 0) {
307 if (set_fpcc) {
308 env->fpscr &= ~FP_FPCC;
309 env->fpscr |= (FP_C | FP_FU);
310 }
311 }
312 finish_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, retaddr);
313 }
314
315 static inline void float_zero_divide_excp(CPUPPCState *env, uintptr_t raddr)
316 {
317 env->fpscr |= FP_ZX;
318 env->fpscr &= ~(FP_FR | FP_FI);
319 /* Update the floating-point exception summary */
320 env->fpscr |= FP_FX;
321 if (fpscr_ze != 0) {
322 /* Update the floating-point enabled exception summary */
323 env->fpscr |= FP_FEX;
324 if (fp_exceptions_enabled(env)) {
325 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
326 POWERPC_EXCP_FP | POWERPC_EXCP_FP_ZX,
327 raddr);
328 }
329 }
330 }
331
332 static inline void float_overflow_excp(CPUPPCState *env)
333 {
334 CPUState *cs = env_cpu(env);
335
336 env->fpscr |= FP_OX;
337 /* Update the floating-point exception summary */
338 env->fpscr |= FP_FX;
339 if (fpscr_oe != 0) {
340 /* XXX: should adjust the result */
341 /* Update the floating-point enabled exception summary */
342 env->fpscr |= FP_FEX;
343 /* We must update the target FPR before raising the exception */
344 cs->exception_index = POWERPC_EXCP_PROGRAM;
345 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_OX;
346 } else {
347 env->fpscr |= FP_XX;
348 env->fpscr |= FP_FI;
349 }
350 }
351
352 static inline void float_underflow_excp(CPUPPCState *env)
353 {
354 CPUState *cs = env_cpu(env);
355
356 env->fpscr |= FP_UX;
357 /* Update the floating-point exception summary */
358 env->fpscr |= FP_FX;
359 if (fpscr_ue != 0) {
360 /* XXX: should adjust the result */
361 /* Update the floating-point enabled exception summary */
362 env->fpscr |= FP_FEX;
363 /* We must update the target FPR before raising the exception */
364 cs->exception_index = POWERPC_EXCP_PROGRAM;
365 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_UX;
366 }
367 }
368
369 static inline void float_inexact_excp(CPUPPCState *env)
370 {
371 CPUState *cs = env_cpu(env);
372
373 env->fpscr |= FP_FI;
374 env->fpscr |= FP_XX;
375 /* Update the floating-point exception summary */
376 env->fpscr |= FP_FX;
377 if (fpscr_xe != 0) {
378 /* Update the floating-point enabled exception summary */
379 env->fpscr |= FP_FEX;
380 /* We must update the target FPR before raising the exception */
381 cs->exception_index = POWERPC_EXCP_PROGRAM;
382 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_XX;
383 }
384 }
385
386 void helper_fpscr_clrbit(CPUPPCState *env, uint32_t bit)
387 {
388 uint32_t mask = 1u << bit;
389 if (env->fpscr & mask) {
390 ppc_store_fpscr(env, env->fpscr & ~(target_ulong)mask);
391 }
392 }
393
394 void helper_fpscr_setbit(CPUPPCState *env, uint32_t bit)
395 {
396 uint32_t mask = 1u << bit;
397 if (!(env->fpscr & mask)) {
398 ppc_store_fpscr(env, env->fpscr | mask);
399 }
400 }
401
402 void helper_store_fpscr(CPUPPCState *env, uint64_t val, uint32_t nibbles)
403 {
404 target_ulong mask = 0;
405 int i;
406
407 /* TODO: push this extension back to translation time */
408 for (i = 0; i < sizeof(target_ulong) * 2; i++) {
409 if (nibbles & (1 << i)) {
410 mask |= (target_ulong) 0xf << (4 * i);
411 }
412 }
413 val = (val & mask) | (env->fpscr & ~mask);
414 ppc_store_fpscr(env, val);
415 }
416
417 void helper_fpscr_check_status(CPUPPCState *env)
418 {
419 CPUState *cs = env_cpu(env);
420 target_ulong fpscr = env->fpscr;
421 int error = 0;
422
423 if ((fpscr & FP_OX) && (fpscr & FP_OE)) {
424 error = POWERPC_EXCP_FP_OX;
425 } else if ((fpscr & FP_UX) && (fpscr & FP_UE)) {
426 error = POWERPC_EXCP_FP_UX;
427 } else if ((fpscr & FP_XX) && (fpscr & FP_XE)) {
428 error = POWERPC_EXCP_FP_XX;
429 } else if ((fpscr & FP_ZX) && (fpscr & FP_ZE)) {
430 error = POWERPC_EXCP_FP_ZX;
431 } else if (fpscr & FP_VE) {
432 if (fpscr & FP_VXSOFT) {
433 error = POWERPC_EXCP_FP_VXSOFT;
434 } else if (fpscr & FP_VXSNAN) {
435 error = POWERPC_EXCP_FP_VXSNAN;
436 } else if (fpscr & FP_VXISI) {
437 error = POWERPC_EXCP_FP_VXISI;
438 } else if (fpscr & FP_VXIDI) {
439 error = POWERPC_EXCP_FP_VXIDI;
440 } else if (fpscr & FP_VXZDZ) {
441 error = POWERPC_EXCP_FP_VXZDZ;
442 } else if (fpscr & FP_VXIMZ) {
443 error = POWERPC_EXCP_FP_VXIMZ;
444 } else if (fpscr & FP_VXVC) {
445 error = POWERPC_EXCP_FP_VXVC;
446 } else if (fpscr & FP_VXSQRT) {
447 error = POWERPC_EXCP_FP_VXSQRT;
448 } else if (fpscr & FP_VXCVI) {
449 error = POWERPC_EXCP_FP_VXCVI;
450 } else {
451 return;
452 }
453 } else {
454 return;
455 }
456 cs->exception_index = POWERPC_EXCP_PROGRAM;
457 env->error_code = error | POWERPC_EXCP_FP;
458 /* Deferred floating-point exception after target FPSCR update */
459 if (fp_exceptions_enabled(env)) {
460 raise_exception_err_ra(env, cs->exception_index,
461 env->error_code, GETPC());
462 }
463 }
464
465 static void do_float_check_status(CPUPPCState *env, uintptr_t raddr)
466 {
467 CPUState *cs = env_cpu(env);
468 int status = get_float_exception_flags(&env->fp_status);
469
470 if (status & float_flag_overflow) {
471 float_overflow_excp(env);
472 } else if (status & float_flag_underflow) {
473 float_underflow_excp(env);
474 }
475 if (status & float_flag_inexact) {
476 float_inexact_excp(env);
477 } else {
478 env->fpscr &= ~FP_FI; /* clear the FPSCR[FI] bit */
479 }
480
481 if (cs->exception_index == POWERPC_EXCP_PROGRAM &&
482 (env->error_code & POWERPC_EXCP_FP)) {
483 /* Deferred floating-point exception after target FPR update */
484 if (fp_exceptions_enabled(env)) {
485 raise_exception_err_ra(env, cs->exception_index,
486 env->error_code, raddr);
487 }
488 }
489 }
490
491 void helper_float_check_status(CPUPPCState *env)
492 {
493 do_float_check_status(env, GETPC());
494 }
495
496 void helper_reset_fpstatus(CPUPPCState *env)
497 {
498 set_float_exception_flags(0, &env->fp_status);
499 }
500
501 static void float_invalid_op_addsub(CPUPPCState *env, int flags,
502 bool set_fpcc, uintptr_t retaddr)
503 {
504 if (flags & float_flag_invalid_isi) {
505 float_invalid_op_vxisi(env, set_fpcc, retaddr);
506 } else if (flags & float_flag_invalid_snan) {
507 float_invalid_op_vxsnan(env, retaddr);
508 }
509 }
510
511 /* fadd - fadd. */
512 float64 helper_fadd(CPUPPCState *env, float64 arg1, float64 arg2)
513 {
514 float64 ret = float64_add(arg1, arg2, &env->fp_status);
515 int flags = get_float_exception_flags(&env->fp_status);
516
517 if (unlikely(flags & float_flag_invalid)) {
518 float_invalid_op_addsub(env, flags, 1, GETPC());
519 }
520
521 return ret;
522 }
523
524 /* fadds - fadds. */
525 float64 helper_fadds(CPUPPCState *env, float64 arg1, float64 arg2)
526 {
527 float64 ret = float64r32_add(arg1, arg2, &env->fp_status);
528 int flags = get_float_exception_flags(&env->fp_status);
529
530 if (unlikely(flags & float_flag_invalid)) {
531 float_invalid_op_addsub(env, flags, 1, GETPC());
532 }
533 return ret;
534 }
535
536 /* fsub - fsub. */
537 float64 helper_fsub(CPUPPCState *env, float64 arg1, float64 arg2)
538 {
539 float64 ret = float64_sub(arg1, arg2, &env->fp_status);
540 int flags = get_float_exception_flags(&env->fp_status);
541
542 if (unlikely(flags & float_flag_invalid)) {
543 float_invalid_op_addsub(env, flags, 1, GETPC());
544 }
545
546 return ret;
547 }
548
549 /* fsubs - fsubs. */
550 float64 helper_fsubs(CPUPPCState *env, float64 arg1, float64 arg2)
551 {
552 float64 ret = float64r32_sub(arg1, arg2, &env->fp_status);
553 int flags = get_float_exception_flags(&env->fp_status);
554
555 if (unlikely(flags & float_flag_invalid)) {
556 float_invalid_op_addsub(env, flags, 1, GETPC());
557 }
558 return ret;
559 }
560
561 static void float_invalid_op_mul(CPUPPCState *env, int flags,
562 bool set_fprc, uintptr_t retaddr)
563 {
564 if (flags & float_flag_invalid_imz) {
565 float_invalid_op_vximz(env, set_fprc, retaddr);
566 } else if (flags & float_flag_invalid_snan) {
567 float_invalid_op_vxsnan(env, retaddr);
568 }
569 }
570
571 /* fmul - fmul. */
572 float64 helper_fmul(CPUPPCState *env, float64 arg1, float64 arg2)
573 {
574 float64 ret = float64_mul(arg1, arg2, &env->fp_status);
575 int flags = get_float_exception_flags(&env->fp_status);
576
577 if (unlikely(flags & float_flag_invalid)) {
578 float_invalid_op_mul(env, flags, 1, GETPC());
579 }
580
581 return ret;
582 }
583
584 /* fmuls - fmuls. */
585 float64 helper_fmuls(CPUPPCState *env, float64 arg1, float64 arg2)
586 {
587 float64 ret = float64r32_mul(arg1, arg2, &env->fp_status);
588 int flags = get_float_exception_flags(&env->fp_status);
589
590 if (unlikely(flags & float_flag_invalid)) {
591 float_invalid_op_mul(env, flags, 1, GETPC());
592 }
593 return ret;
594 }
595
596 static void float_invalid_op_div(CPUPPCState *env, int flags,
597 bool set_fprc, uintptr_t retaddr)
598 {
599 if (flags & float_flag_invalid_idi) {
600 float_invalid_op_vxidi(env, set_fprc, retaddr);
601 } else if (flags & float_flag_invalid_zdz) {
602 float_invalid_op_vxzdz(env, set_fprc, retaddr);
603 } else if (flags & float_flag_invalid_snan) {
604 float_invalid_op_vxsnan(env, retaddr);
605 }
606 }
607
608 /* fdiv - fdiv. */
609 float64 helper_fdiv(CPUPPCState *env, float64 arg1, float64 arg2)
610 {
611 float64 ret = float64_div(arg1, arg2, &env->fp_status);
612 int flags = get_float_exception_flags(&env->fp_status);
613
614 if (unlikely(flags & float_flag_invalid)) {
615 float_invalid_op_div(env, flags, 1, GETPC());
616 }
617 if (unlikely(flags & float_flag_divbyzero)) {
618 float_zero_divide_excp(env, GETPC());
619 }
620
621 return ret;
622 }
623
624 /* fdivs - fdivs. */
625 float64 helper_fdivs(CPUPPCState *env, float64 arg1, float64 arg2)
626 {
627 float64 ret = float64r32_div(arg1, arg2, &env->fp_status);
628 int flags = get_float_exception_flags(&env->fp_status);
629
630 if (unlikely(flags & float_flag_invalid)) {
631 float_invalid_op_div(env, flags, 1, GETPC());
632 }
633 if (unlikely(flags & float_flag_divbyzero)) {
634 float_zero_divide_excp(env, GETPC());
635 }
636
637 return ret;
638 }
639
640 static uint64_t float_invalid_cvt(CPUPPCState *env, int flags,
641 uint64_t ret, uint64_t ret_nan,
642 bool set_fprc, uintptr_t retaddr)
643 {
644 /*
645 * VXCVI is different from most in that it sets two exception bits,
646 * VXCVI and VXSNAN for an SNaN input.
647 */
648 if (flags & float_flag_invalid_snan) {
649 env->fpscr |= FP_VXSNAN;
650 }
651 float_invalid_op_vxcvi(env, set_fprc, retaddr);
652
653 return flags & float_flag_invalid_cvti ? ret : ret_nan;
654 }
655
656 #define FPU_FCTI(op, cvt, nanval) \
657 uint64_t helper_##op(CPUPPCState *env, float64 arg) \
658 { \
659 uint64_t ret = float64_to_##cvt(arg, &env->fp_status); \
660 int flags = get_float_exception_flags(&env->fp_status); \
661 if (unlikely(flags & float_flag_invalid)) { \
662 ret = float_invalid_cvt(env, flags, ret, nanval, 1, GETPC()); \
663 } \
664 return ret; \
665 }
666
667 FPU_FCTI(fctiw, int32, 0x80000000U)
668 FPU_FCTI(fctiwz, int32_round_to_zero, 0x80000000U)
669 FPU_FCTI(fctiwu, uint32, 0x00000000U)
670 FPU_FCTI(fctiwuz, uint32_round_to_zero, 0x00000000U)
671 FPU_FCTI(fctid, int64, 0x8000000000000000ULL)
672 FPU_FCTI(fctidz, int64_round_to_zero, 0x8000000000000000ULL)
673 FPU_FCTI(fctidu, uint64, 0x0000000000000000ULL)
674 FPU_FCTI(fctiduz, uint64_round_to_zero, 0x0000000000000000ULL)
675
676 #define FPU_FCFI(op, cvtr, is_single) \
677 uint64_t helper_##op(CPUPPCState *env, uint64_t arg) \
678 { \
679 CPU_DoubleU farg; \
680 \
681 if (is_single) { \
682 float32 tmp = cvtr(arg, &env->fp_status); \
683 farg.d = float32_to_float64(tmp, &env->fp_status); \
684 } else { \
685 farg.d = cvtr(arg, &env->fp_status); \
686 } \
687 do_float_check_status(env, GETPC()); \
688 return farg.ll; \
689 }
690
691 FPU_FCFI(fcfid, int64_to_float64, 0)
692 FPU_FCFI(fcfids, int64_to_float32, 1)
693 FPU_FCFI(fcfidu, uint64_to_float64, 0)
694 FPU_FCFI(fcfidus, uint64_to_float32, 1)
695
696 static uint64_t do_fri(CPUPPCState *env, uint64_t arg,
697 FloatRoundMode rounding_mode)
698 {
699 FloatRoundMode old_rounding_mode = get_float_rounding_mode(&env->fp_status);
700 int flags;
701
702 set_float_rounding_mode(rounding_mode, &env->fp_status);
703 arg = float64_round_to_int(arg, &env->fp_status);
704 set_float_rounding_mode(old_rounding_mode, &env->fp_status);
705
706 flags = get_float_exception_flags(&env->fp_status);
707 if (flags & float_flag_invalid_snan) {
708 float_invalid_op_vxsnan(env, GETPC());
709 }
710
711 /* fri* does not set FPSCR[XX] */
712 set_float_exception_flags(flags & ~float_flag_inexact, &env->fp_status);
713 do_float_check_status(env, GETPC());
714
715 return arg;
716 }
717
718 uint64_t helper_frin(CPUPPCState *env, uint64_t arg)
719 {
720 return do_fri(env, arg, float_round_ties_away);
721 }
722
723 uint64_t helper_friz(CPUPPCState *env, uint64_t arg)
724 {
725 return do_fri(env, arg, float_round_to_zero);
726 }
727
728 uint64_t helper_frip(CPUPPCState *env, uint64_t arg)
729 {
730 return do_fri(env, arg, float_round_up);
731 }
732
733 uint64_t helper_frim(CPUPPCState *env, uint64_t arg)
734 {
735 return do_fri(env, arg, float_round_down);
736 }
737
738 static void float_invalid_op_madd(CPUPPCState *env, int flags,
739 bool set_fpcc, uintptr_t retaddr)
740 {
741 if (flags & float_flag_invalid_imz) {
742 float_invalid_op_vximz(env, set_fpcc, retaddr);
743 } else {
744 float_invalid_op_addsub(env, flags, set_fpcc, retaddr);
745 }
746 }
747
748 static float64 do_fmadd(CPUPPCState *env, float64 a, float64 b,
749 float64 c, int madd_flags, uintptr_t retaddr)
750 {
751 float64 ret = float64_muladd(a, b, c, madd_flags, &env->fp_status);
752 int flags = get_float_exception_flags(&env->fp_status);
753
754 if (unlikely(flags & float_flag_invalid)) {
755 float_invalid_op_madd(env, flags, 1, retaddr);
756 }
757 return ret;
758 }
759
760 static uint64_t do_fmadds(CPUPPCState *env, float64 a, float64 b,
761 float64 c, int madd_flags, uintptr_t retaddr)
762 {
763 float64 ret = float64r32_muladd(a, b, c, madd_flags, &env->fp_status);
764 int flags = get_float_exception_flags(&env->fp_status);
765
766 if (unlikely(flags & float_flag_invalid)) {
767 float_invalid_op_madd(env, flags, 1, retaddr);
768 }
769 return ret;
770 }
771
772 #define FPU_FMADD(op, madd_flags) \
773 uint64_t helper_##op(CPUPPCState *env, uint64_t arg1, \
774 uint64_t arg2, uint64_t arg3) \
775 { return do_fmadd(env, arg1, arg2, arg3, madd_flags, GETPC()); } \
776 uint64_t helper_##op##s(CPUPPCState *env, uint64_t arg1, \
777 uint64_t arg2, uint64_t arg3) \
778 { return do_fmadds(env, arg1, arg2, arg3, madd_flags, GETPC()); }
779
780 #define MADD_FLGS 0
781 #define MSUB_FLGS float_muladd_negate_c
782 #define NMADD_FLGS float_muladd_negate_result
783 #define NMSUB_FLGS (float_muladd_negate_c | float_muladd_negate_result)
784
785 FPU_FMADD(fmadd, MADD_FLGS)
786 FPU_FMADD(fnmadd, NMADD_FLGS)
787 FPU_FMADD(fmsub, MSUB_FLGS)
788 FPU_FMADD(fnmsub, NMSUB_FLGS)
789
790 /* frsp - frsp. */
791 static uint64_t do_frsp(CPUPPCState *env, uint64_t arg, uintptr_t retaddr)
792 {
793 float32 f32 = float64_to_float32(arg, &env->fp_status);
794 int flags = get_float_exception_flags(&env->fp_status);
795
796 if (unlikely(flags & float_flag_invalid_snan)) {
797 float_invalid_op_vxsnan(env, retaddr);
798 }
799 return helper_todouble(f32);
800 }
801
802 uint64_t helper_frsp(CPUPPCState *env, uint64_t arg)
803 {
804 return do_frsp(env, arg, GETPC());
805 }
806
807 static void float_invalid_op_sqrt(CPUPPCState *env, int flags,
808 bool set_fpcc, uintptr_t retaddr)
809 {
810 if (unlikely(flags & float_flag_invalid_sqrt)) {
811 float_invalid_op_vxsqrt(env, set_fpcc, retaddr);
812 } else if (unlikely(flags & float_flag_invalid_snan)) {
813 float_invalid_op_vxsnan(env, retaddr);
814 }
815 }
816
817 /* fsqrt - fsqrt. */
818 float64 helper_fsqrt(CPUPPCState *env, float64 arg)
819 {
820 float64 ret = float64_sqrt(arg, &env->fp_status);
821 int flags = get_float_exception_flags(&env->fp_status);
822
823 if (unlikely(flags & float_flag_invalid)) {
824 float_invalid_op_sqrt(env, flags, 1, GETPC());
825 }
826
827 return ret;
828 }
829
830 /* fsqrts - fsqrts. */
831 float64 helper_fsqrts(CPUPPCState *env, float64 arg)
832 {
833 float64 ret = float64r32_sqrt(arg, &env->fp_status);
834 int flags = get_float_exception_flags(&env->fp_status);
835
836 if (unlikely(flags & float_flag_invalid)) {
837 float_invalid_op_sqrt(env, flags, 1, GETPC());
838 }
839 return ret;
840 }
841
842 /* fre - fre. */
843 float64 helper_fre(CPUPPCState *env, float64 arg)
844 {
845 /* "Estimate" the reciprocal with actual division. */
846 float64 ret = float64_div(float64_one, arg, &env->fp_status);
847 int flags = get_float_exception_flags(&env->fp_status);
848
849 if (unlikely(flags & float_flag_invalid_snan)) {
850 float_invalid_op_vxsnan(env, GETPC());
851 }
852 if (unlikely(flags & float_flag_divbyzero)) {
853 float_zero_divide_excp(env, GETPC());
854 /* For FPSCR.ZE == 0, the result is 1/2. */
855 ret = float64_set_sign(float64_half, float64_is_neg(arg));
856 }
857
858 return ret;
859 }
860
861 /* fres - fres. */
862 uint64_t helper_fres(CPUPPCState *env, uint64_t arg)
863 {
864 /* "Estimate" the reciprocal with actual division. */
865 float64 ret = float64r32_div(float64_one, arg, &env->fp_status);
866 int flags = get_float_exception_flags(&env->fp_status);
867
868 if (unlikely(flags & float_flag_invalid_snan)) {
869 float_invalid_op_vxsnan(env, GETPC());
870 }
871 if (unlikely(flags & float_flag_divbyzero)) {
872 float_zero_divide_excp(env, GETPC());
873 /* For FPSCR.ZE == 0, the result is 1/2. */
874 ret = float64_set_sign(float64_half, float64_is_neg(arg));
875 }
876
877 return ret;
878 }
879
880 /* frsqrte - frsqrte. */
881 float64 helper_frsqrte(CPUPPCState *env, float64 arg)
882 {
883 /* "Estimate" the reciprocal with actual division. */
884 float64 rets = float64_sqrt(arg, &env->fp_status);
885 float64 retd = float64_div(float64_one, rets, &env->fp_status);
886 int flags = get_float_exception_flags(&env->fp_status);
887
888 if (unlikely(flags & float_flag_invalid)) {
889 float_invalid_op_sqrt(env, flags, 1, GETPC());
890 }
891 if (unlikely(flags & float_flag_divbyzero)) {
892 /* Reciprocal of (square root of) zero. */
893 float_zero_divide_excp(env, GETPC());
894 }
895
896 return retd;
897 }
898
899 /* frsqrtes - frsqrtes. */
900 float64 helper_frsqrtes(CPUPPCState *env, float64 arg)
901 {
902 /* "Estimate" the reciprocal with actual division. */
903 float64 rets = float64_sqrt(arg, &env->fp_status);
904 float64 retd = float64r32_div(float64_one, rets, &env->fp_status);
905 int flags = get_float_exception_flags(&env->fp_status);
906
907 if (unlikely(flags & float_flag_invalid)) {
908 float_invalid_op_sqrt(env, flags, 1, GETPC());
909 }
910 if (unlikely(flags & float_flag_divbyzero)) {
911 /* Reciprocal of (square root of) zero. */
912 float_zero_divide_excp(env, GETPC());
913 }
914
915 return retd;
916 }
917
918 /* fsel - fsel. */
919 uint64_t helper_fsel(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
920 uint64_t arg3)
921 {
922 CPU_DoubleU farg1;
923
924 farg1.ll = arg1;
925
926 if ((!float64_is_neg(farg1.d) || float64_is_zero(farg1.d)) &&
927 !float64_is_any_nan(farg1.d)) {
928 return arg2;
929 } else {
930 return arg3;
931 }
932 }
933
934 uint32_t helper_ftdiv(uint64_t fra, uint64_t frb)
935 {
936 int fe_flag = 0;
937 int fg_flag = 0;
938
939 if (unlikely(float64_is_infinity(fra) ||
940 float64_is_infinity(frb) ||
941 float64_is_zero(frb))) {
942 fe_flag = 1;
943 fg_flag = 1;
944 } else {
945 int e_a = ppc_float64_get_unbiased_exp(fra);
946 int e_b = ppc_float64_get_unbiased_exp(frb);
947
948 if (unlikely(float64_is_any_nan(fra) ||
949 float64_is_any_nan(frb))) {
950 fe_flag = 1;
951 } else if ((e_b <= -1022) || (e_b >= 1021)) {
952 fe_flag = 1;
953 } else if (!float64_is_zero(fra) &&
954 (((e_a - e_b) >= 1023) ||
955 ((e_a - e_b) <= -1021) ||
956 (e_a <= -970))) {
957 fe_flag = 1;
958 }
959
960 if (unlikely(float64_is_zero_or_denormal(frb))) {
961 /* XB is not zero because of the above check and */
962 /* so must be denormalized. */
963 fg_flag = 1;
964 }
965 }
966
967 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
968 }
969
970 uint32_t helper_ftsqrt(uint64_t frb)
971 {
972 int fe_flag = 0;
973 int fg_flag = 0;
974
975 if (unlikely(float64_is_infinity(frb) || float64_is_zero(frb))) {
976 fe_flag = 1;
977 fg_flag = 1;
978 } else {
979 int e_b = ppc_float64_get_unbiased_exp(frb);
980
981 if (unlikely(float64_is_any_nan(frb))) {
982 fe_flag = 1;
983 } else if (unlikely(float64_is_zero(frb))) {
984 fe_flag = 1;
985 } else if (unlikely(float64_is_neg(frb))) {
986 fe_flag = 1;
987 } else if (!float64_is_zero(frb) && (e_b <= (-1022 + 52))) {
988 fe_flag = 1;
989 }
990
991 if (unlikely(float64_is_zero_or_denormal(frb))) {
992 /* XB is not zero because of the above check and */
993 /* therefore must be denormalized. */
994 fg_flag = 1;
995 }
996 }
997
998 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
999 }
1000
1001 void helper_fcmpu(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
1002 uint32_t crfD)
1003 {
1004 CPU_DoubleU farg1, farg2;
1005 uint32_t ret = 0;
1006
1007 farg1.ll = arg1;
1008 farg2.ll = arg2;
1009
1010 if (unlikely(float64_is_any_nan(farg1.d) ||
1011 float64_is_any_nan(farg2.d))) {
1012 ret = 0x01UL;
1013 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) {
1014 ret = 0x08UL;
1015 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) {
1016 ret = 0x04UL;
1017 } else {
1018 ret = 0x02UL;
1019 }
1020
1021 env->fpscr &= ~FP_FPCC;
1022 env->fpscr |= ret << FPSCR_FPCC;
1023 env->crf[crfD] = ret;
1024 if (unlikely(ret == 0x01UL
1025 && (float64_is_signaling_nan(farg1.d, &env->fp_status) ||
1026 float64_is_signaling_nan(farg2.d, &env->fp_status)))) {
1027 /* sNaN comparison */
1028 float_invalid_op_vxsnan(env, GETPC());
1029 }
1030 }
1031
1032 void helper_fcmpo(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
1033 uint32_t crfD)
1034 {
1035 CPU_DoubleU farg1, farg2;
1036 uint32_t ret = 0;
1037
1038 farg1.ll = arg1;
1039 farg2.ll = arg2;
1040
1041 if (unlikely(float64_is_any_nan(farg1.d) ||
1042 float64_is_any_nan(farg2.d))) {
1043 ret = 0x01UL;
1044 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) {
1045 ret = 0x08UL;
1046 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) {
1047 ret = 0x04UL;
1048 } else {
1049 ret = 0x02UL;
1050 }
1051
1052 env->fpscr &= ~FP_FPCC;
1053 env->fpscr |= ret << FPSCR_FPCC;
1054 env->crf[crfD] = (uint32_t) ret;
1055 if (unlikely(ret == 0x01UL)) {
1056 float_invalid_op_vxvc(env, 1, GETPC());
1057 if (float64_is_signaling_nan(farg1.d, &env->fp_status) ||
1058 float64_is_signaling_nan(farg2.d, &env->fp_status)) {
1059 /* sNaN comparison */
1060 float_invalid_op_vxsnan(env, GETPC());
1061 }
1062 }
1063 }
1064
1065 /* Single-precision floating-point conversions */
1066 static inline uint32_t efscfsi(CPUPPCState *env, uint32_t val)
1067 {
1068 CPU_FloatU u;
1069
1070 u.f = int32_to_float32(val, &env->vec_status);
1071
1072 return u.l;
1073 }
1074
1075 static inline uint32_t efscfui(CPUPPCState *env, uint32_t val)
1076 {
1077 CPU_FloatU u;
1078
1079 u.f = uint32_to_float32(val, &env->vec_status);
1080
1081 return u.l;
1082 }
1083
1084 static inline int32_t efsctsi(CPUPPCState *env, uint32_t val)
1085 {
1086 CPU_FloatU u;
1087
1088 u.l = val;
1089 /* NaN are not treated the same way IEEE 754 does */
1090 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1091 return 0;
1092 }
1093
1094 return float32_to_int32(u.f, &env->vec_status);
1095 }
1096
1097 static inline uint32_t efsctui(CPUPPCState *env, uint32_t val)
1098 {
1099 CPU_FloatU u;
1100
1101 u.l = val;
1102 /* NaN are not treated the same way IEEE 754 does */
1103 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1104 return 0;
1105 }
1106
1107 return float32_to_uint32(u.f, &env->vec_status);
1108 }
1109
1110 static inline uint32_t efsctsiz(CPUPPCState *env, uint32_t val)
1111 {
1112 CPU_FloatU u;
1113
1114 u.l = val;
1115 /* NaN are not treated the same way IEEE 754 does */
1116 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1117 return 0;
1118 }
1119
1120 return float32_to_int32_round_to_zero(u.f, &env->vec_status);
1121 }
1122
1123 static inline uint32_t efsctuiz(CPUPPCState *env, uint32_t val)
1124 {
1125 CPU_FloatU u;
1126
1127 u.l = val;
1128 /* NaN are not treated the same way IEEE 754 does */
1129 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1130 return 0;
1131 }
1132
1133 return float32_to_uint32_round_to_zero(u.f, &env->vec_status);
1134 }
1135
1136 static inline uint32_t efscfsf(CPUPPCState *env, uint32_t val)
1137 {
1138 CPU_FloatU u;
1139 float32 tmp;
1140
1141 u.f = int32_to_float32(val, &env->vec_status);
1142 tmp = int64_to_float32(1ULL << 32, &env->vec_status);
1143 u.f = float32_div(u.f, tmp, &env->vec_status);
1144
1145 return u.l;
1146 }
1147
1148 static inline uint32_t efscfuf(CPUPPCState *env, uint32_t val)
1149 {
1150 CPU_FloatU u;
1151 float32 tmp;
1152
1153 u.f = uint32_to_float32(val, &env->vec_status);
1154 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1155 u.f = float32_div(u.f, tmp, &env->vec_status);
1156
1157 return u.l;
1158 }
1159
1160 static inline uint32_t efsctsf(CPUPPCState *env, uint32_t val)
1161 {
1162 CPU_FloatU u;
1163 float32 tmp;
1164
1165 u.l = val;
1166 /* NaN are not treated the same way IEEE 754 does */
1167 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1168 return 0;
1169 }
1170 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1171 u.f = float32_mul(u.f, tmp, &env->vec_status);
1172
1173 return float32_to_int32(u.f, &env->vec_status);
1174 }
1175
1176 static inline uint32_t efsctuf(CPUPPCState *env, uint32_t val)
1177 {
1178 CPU_FloatU u;
1179 float32 tmp;
1180
1181 u.l = val;
1182 /* NaN are not treated the same way IEEE 754 does */
1183 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1184 return 0;
1185 }
1186 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1187 u.f = float32_mul(u.f, tmp, &env->vec_status);
1188
1189 return float32_to_uint32(u.f, &env->vec_status);
1190 }
1191
1192 #define HELPER_SPE_SINGLE_CONV(name) \
1193 uint32_t helper_e##name(CPUPPCState *env, uint32_t val) \
1194 { \
1195 return e##name(env, val); \
1196 }
1197 /* efscfsi */
1198 HELPER_SPE_SINGLE_CONV(fscfsi);
1199 /* efscfui */
1200 HELPER_SPE_SINGLE_CONV(fscfui);
1201 /* efscfuf */
1202 HELPER_SPE_SINGLE_CONV(fscfuf);
1203 /* efscfsf */
1204 HELPER_SPE_SINGLE_CONV(fscfsf);
1205 /* efsctsi */
1206 HELPER_SPE_SINGLE_CONV(fsctsi);
1207 /* efsctui */
1208 HELPER_SPE_SINGLE_CONV(fsctui);
1209 /* efsctsiz */
1210 HELPER_SPE_SINGLE_CONV(fsctsiz);
1211 /* efsctuiz */
1212 HELPER_SPE_SINGLE_CONV(fsctuiz);
1213 /* efsctsf */
1214 HELPER_SPE_SINGLE_CONV(fsctsf);
1215 /* efsctuf */
1216 HELPER_SPE_SINGLE_CONV(fsctuf);
1217
1218 #define HELPER_SPE_VECTOR_CONV(name) \
1219 uint64_t helper_ev##name(CPUPPCState *env, uint64_t val) \
1220 { \
1221 return ((uint64_t)e##name(env, val >> 32) << 32) | \
1222 (uint64_t)e##name(env, val); \
1223 }
1224 /* evfscfsi */
1225 HELPER_SPE_VECTOR_CONV(fscfsi);
1226 /* evfscfui */
1227 HELPER_SPE_VECTOR_CONV(fscfui);
1228 /* evfscfuf */
1229 HELPER_SPE_VECTOR_CONV(fscfuf);
1230 /* evfscfsf */
1231 HELPER_SPE_VECTOR_CONV(fscfsf);
1232 /* evfsctsi */
1233 HELPER_SPE_VECTOR_CONV(fsctsi);
1234 /* evfsctui */
1235 HELPER_SPE_VECTOR_CONV(fsctui);
1236 /* evfsctsiz */
1237 HELPER_SPE_VECTOR_CONV(fsctsiz);
1238 /* evfsctuiz */
1239 HELPER_SPE_VECTOR_CONV(fsctuiz);
1240 /* evfsctsf */
1241 HELPER_SPE_VECTOR_CONV(fsctsf);
1242 /* evfsctuf */
1243 HELPER_SPE_VECTOR_CONV(fsctuf);
1244
1245 /* Single-precision floating-point arithmetic */
1246 static inline uint32_t efsadd(CPUPPCState *env, uint32_t op1, uint32_t op2)
1247 {
1248 CPU_FloatU u1, u2;
1249
1250 u1.l = op1;
1251 u2.l = op2;
1252 u1.f = float32_add(u1.f, u2.f, &env->vec_status);
1253 return u1.l;
1254 }
1255
1256 static inline uint32_t efssub(CPUPPCState *env, uint32_t op1, uint32_t op2)
1257 {
1258 CPU_FloatU u1, u2;
1259
1260 u1.l = op1;
1261 u2.l = op2;
1262 u1.f = float32_sub(u1.f, u2.f, &env->vec_status);
1263 return u1.l;
1264 }
1265
1266 static inline uint32_t efsmul(CPUPPCState *env, uint32_t op1, uint32_t op2)
1267 {
1268 CPU_FloatU u1, u2;
1269
1270 u1.l = op1;
1271 u2.l = op2;
1272 u1.f = float32_mul(u1.f, u2.f, &env->vec_status);
1273 return u1.l;
1274 }
1275
1276 static inline uint32_t efsdiv(CPUPPCState *env, uint32_t op1, uint32_t op2)
1277 {
1278 CPU_FloatU u1, u2;
1279
1280 u1.l = op1;
1281 u2.l = op2;
1282 u1.f = float32_div(u1.f, u2.f, &env->vec_status);
1283 return u1.l;
1284 }
1285
1286 #define HELPER_SPE_SINGLE_ARITH(name) \
1287 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \
1288 { \
1289 return e##name(env, op1, op2); \
1290 }
1291 /* efsadd */
1292 HELPER_SPE_SINGLE_ARITH(fsadd);
1293 /* efssub */
1294 HELPER_SPE_SINGLE_ARITH(fssub);
1295 /* efsmul */
1296 HELPER_SPE_SINGLE_ARITH(fsmul);
1297 /* efsdiv */
1298 HELPER_SPE_SINGLE_ARITH(fsdiv);
1299
1300 #define HELPER_SPE_VECTOR_ARITH(name) \
1301 uint64_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \
1302 { \
1303 return ((uint64_t)e##name(env, op1 >> 32, op2 >> 32) << 32) | \
1304 (uint64_t)e##name(env, op1, op2); \
1305 }
1306 /* evfsadd */
1307 HELPER_SPE_VECTOR_ARITH(fsadd);
1308 /* evfssub */
1309 HELPER_SPE_VECTOR_ARITH(fssub);
1310 /* evfsmul */
1311 HELPER_SPE_VECTOR_ARITH(fsmul);
1312 /* evfsdiv */
1313 HELPER_SPE_VECTOR_ARITH(fsdiv);
1314
1315 /* Single-precision floating-point comparisons */
1316 static inline uint32_t efscmplt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1317 {
1318 CPU_FloatU u1, u2;
1319
1320 u1.l = op1;
1321 u2.l = op2;
1322 return float32_lt(u1.f, u2.f, &env->vec_status) ? 4 : 0;
1323 }
1324
1325 static inline uint32_t efscmpgt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1326 {
1327 CPU_FloatU u1, u2;
1328
1329 u1.l = op1;
1330 u2.l = op2;
1331 return float32_le(u1.f, u2.f, &env->vec_status) ? 0 : 4;
1332 }
1333
1334 static inline uint32_t efscmpeq(CPUPPCState *env, uint32_t op1, uint32_t op2)
1335 {
1336 CPU_FloatU u1, u2;
1337
1338 u1.l = op1;
1339 u2.l = op2;
1340 return float32_eq(u1.f, u2.f, &env->vec_status) ? 4 : 0;
1341 }
1342
1343 static inline uint32_t efststlt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1344 {
1345 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1346 return efscmplt(env, op1, op2);
1347 }
1348
1349 static inline uint32_t efststgt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1350 {
1351 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1352 return efscmpgt(env, op1, op2);
1353 }
1354
1355 static inline uint32_t efststeq(CPUPPCState *env, uint32_t op1, uint32_t op2)
1356 {
1357 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1358 return efscmpeq(env, op1, op2);
1359 }
1360
1361 #define HELPER_SINGLE_SPE_CMP(name) \
1362 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \
1363 { \
1364 return e##name(env, op1, op2); \
1365 }
1366 /* efststlt */
1367 HELPER_SINGLE_SPE_CMP(fststlt);
1368 /* efststgt */
1369 HELPER_SINGLE_SPE_CMP(fststgt);
1370 /* efststeq */
1371 HELPER_SINGLE_SPE_CMP(fststeq);
1372 /* efscmplt */
1373 HELPER_SINGLE_SPE_CMP(fscmplt);
1374 /* efscmpgt */
1375 HELPER_SINGLE_SPE_CMP(fscmpgt);
1376 /* efscmpeq */
1377 HELPER_SINGLE_SPE_CMP(fscmpeq);
1378
1379 static inline uint32_t evcmp_merge(int t0, int t1)
1380 {
1381 return (t0 << 3) | (t1 << 2) | ((t0 | t1) << 1) | (t0 & t1);
1382 }
1383
1384 #define HELPER_VECTOR_SPE_CMP(name) \
1385 uint32_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \
1386 { \
1387 return evcmp_merge(e##name(env, op1 >> 32, op2 >> 32), \
1388 e##name(env, op1, op2)); \
1389 }
1390 /* evfststlt */
1391 HELPER_VECTOR_SPE_CMP(fststlt);
1392 /* evfststgt */
1393 HELPER_VECTOR_SPE_CMP(fststgt);
1394 /* evfststeq */
1395 HELPER_VECTOR_SPE_CMP(fststeq);
1396 /* evfscmplt */
1397 HELPER_VECTOR_SPE_CMP(fscmplt);
1398 /* evfscmpgt */
1399 HELPER_VECTOR_SPE_CMP(fscmpgt);
1400 /* evfscmpeq */
1401 HELPER_VECTOR_SPE_CMP(fscmpeq);
1402
1403 /* Double-precision floating-point conversion */
1404 uint64_t helper_efdcfsi(CPUPPCState *env, uint32_t val)
1405 {
1406 CPU_DoubleU u;
1407
1408 u.d = int32_to_float64(val, &env->vec_status);
1409
1410 return u.ll;
1411 }
1412
1413 uint64_t helper_efdcfsid(CPUPPCState *env, uint64_t val)
1414 {
1415 CPU_DoubleU u;
1416
1417 u.d = int64_to_float64(val, &env->vec_status);
1418
1419 return u.ll;
1420 }
1421
1422 uint64_t helper_efdcfui(CPUPPCState *env, uint32_t val)
1423 {
1424 CPU_DoubleU u;
1425
1426 u.d = uint32_to_float64(val, &env->vec_status);
1427
1428 return u.ll;
1429 }
1430
1431 uint64_t helper_efdcfuid(CPUPPCState *env, uint64_t val)
1432 {
1433 CPU_DoubleU u;
1434
1435 u.d = uint64_to_float64(val, &env->vec_status);
1436
1437 return u.ll;
1438 }
1439
1440 uint32_t helper_efdctsi(CPUPPCState *env, uint64_t val)
1441 {
1442 CPU_DoubleU u;
1443
1444 u.ll = val;
1445 /* NaN are not treated the same way IEEE 754 does */
1446 if (unlikely(float64_is_any_nan(u.d))) {
1447 return 0;
1448 }
1449
1450 return float64_to_int32(u.d, &env->vec_status);
1451 }
1452
1453 uint32_t helper_efdctui(CPUPPCState *env, uint64_t val)
1454 {
1455 CPU_DoubleU u;
1456
1457 u.ll = val;
1458 /* NaN are not treated the same way IEEE 754 does */
1459 if (unlikely(float64_is_any_nan(u.d))) {
1460 return 0;
1461 }
1462
1463 return float64_to_uint32(u.d, &env->vec_status);
1464 }
1465
1466 uint32_t helper_efdctsiz(CPUPPCState *env, uint64_t val)
1467 {
1468 CPU_DoubleU u;
1469
1470 u.ll = val;
1471 /* NaN are not treated the same way IEEE 754 does */
1472 if (unlikely(float64_is_any_nan(u.d))) {
1473 return 0;
1474 }
1475
1476 return float64_to_int32_round_to_zero(u.d, &env->vec_status);
1477 }
1478
1479 uint64_t helper_efdctsidz(CPUPPCState *env, uint64_t val)
1480 {
1481 CPU_DoubleU u;
1482
1483 u.ll = val;
1484 /* NaN are not treated the same way IEEE 754 does */
1485 if (unlikely(float64_is_any_nan(u.d))) {
1486 return 0;
1487 }
1488
1489 return float64_to_int64_round_to_zero(u.d, &env->vec_status);
1490 }
1491
1492 uint32_t helper_efdctuiz(CPUPPCState *env, uint64_t val)
1493 {
1494 CPU_DoubleU u;
1495
1496 u.ll = val;
1497 /* NaN are not treated the same way IEEE 754 does */
1498 if (unlikely(float64_is_any_nan(u.d))) {
1499 return 0;
1500 }
1501
1502 return float64_to_uint32_round_to_zero(u.d, &env->vec_status);
1503 }
1504
1505 uint64_t helper_efdctuidz(CPUPPCState *env, uint64_t val)
1506 {
1507 CPU_DoubleU u;
1508
1509 u.ll = val;
1510 /* NaN are not treated the same way IEEE 754 does */
1511 if (unlikely(float64_is_any_nan(u.d))) {
1512 return 0;
1513 }
1514
1515 return float64_to_uint64_round_to_zero(u.d, &env->vec_status);
1516 }
1517
1518 uint64_t helper_efdcfsf(CPUPPCState *env, uint32_t val)
1519 {
1520 CPU_DoubleU u;
1521 float64 tmp;
1522
1523 u.d = int32_to_float64(val, &env->vec_status);
1524 tmp = int64_to_float64(1ULL << 32, &env->vec_status);
1525 u.d = float64_div(u.d, tmp, &env->vec_status);
1526
1527 return u.ll;
1528 }
1529
1530 uint64_t helper_efdcfuf(CPUPPCState *env, uint32_t val)
1531 {
1532 CPU_DoubleU u;
1533 float64 tmp;
1534
1535 u.d = uint32_to_float64(val, &env->vec_status);
1536 tmp = int64_to_float64(1ULL << 32, &env->vec_status);
1537 u.d = float64_div(u.d, tmp, &env->vec_status);
1538
1539 return u.ll;
1540 }
1541
1542 uint32_t helper_efdctsf(CPUPPCState *env, uint64_t val)
1543 {
1544 CPU_DoubleU u;
1545 float64 tmp;
1546
1547 u.ll = val;
1548 /* NaN are not treated the same way IEEE 754 does */
1549 if (unlikely(float64_is_any_nan(u.d))) {
1550 return 0;
1551 }
1552 tmp = uint64_to_float64(1ULL << 32, &env->vec_status);
1553 u.d = float64_mul(u.d, tmp, &env->vec_status);
1554
1555 return float64_to_int32(u.d, &env->vec_status);
1556 }
1557
1558 uint32_t helper_efdctuf(CPUPPCState *env, uint64_t val)
1559 {
1560 CPU_DoubleU u;
1561 float64 tmp;
1562
1563 u.ll = val;
1564 /* NaN are not treated the same way IEEE 754 does */
1565 if (unlikely(float64_is_any_nan(u.d))) {
1566 return 0;
1567 }
1568 tmp = uint64_to_float64(1ULL << 32, &env->vec_status);
1569 u.d = float64_mul(u.d, tmp, &env->vec_status);
1570
1571 return float64_to_uint32(u.d, &env->vec_status);
1572 }
1573
1574 uint32_t helper_efscfd(CPUPPCState *env, uint64_t val)
1575 {
1576 CPU_DoubleU u1;
1577 CPU_FloatU u2;
1578
1579 u1.ll = val;
1580 u2.f = float64_to_float32(u1.d, &env->vec_status);
1581
1582 return u2.l;
1583 }
1584
1585 uint64_t helper_efdcfs(CPUPPCState *env, uint32_t val)
1586 {
1587 CPU_DoubleU u2;
1588 CPU_FloatU u1;
1589
1590 u1.l = val;
1591 u2.d = float32_to_float64(u1.f, &env->vec_status);
1592
1593 return u2.ll;
1594 }
1595
1596 /* Double precision fixed-point arithmetic */
1597 uint64_t helper_efdadd(CPUPPCState *env, uint64_t op1, uint64_t op2)
1598 {
1599 CPU_DoubleU u1, u2;
1600
1601 u1.ll = op1;
1602 u2.ll = op2;
1603 u1.d = float64_add(u1.d, u2.d, &env->vec_status);
1604 return u1.ll;
1605 }
1606
1607 uint64_t helper_efdsub(CPUPPCState *env, uint64_t op1, uint64_t op2)
1608 {
1609 CPU_DoubleU u1, u2;
1610
1611 u1.ll = op1;
1612 u2.ll = op2;
1613 u1.d = float64_sub(u1.d, u2.d, &env->vec_status);
1614 return u1.ll;
1615 }
1616
1617 uint64_t helper_efdmul(CPUPPCState *env, uint64_t op1, uint64_t op2)
1618 {
1619 CPU_DoubleU u1, u2;
1620
1621 u1.ll = op1;
1622 u2.ll = op2;
1623 u1.d = float64_mul(u1.d, u2.d, &env->vec_status);
1624 return u1.ll;
1625 }
1626
1627 uint64_t helper_efddiv(CPUPPCState *env, uint64_t op1, uint64_t op2)
1628 {
1629 CPU_DoubleU u1, u2;
1630
1631 u1.ll = op1;
1632 u2.ll = op2;
1633 u1.d = float64_div(u1.d, u2.d, &env->vec_status);
1634 return u1.ll;
1635 }
1636
1637 /* Double precision floating point helpers */
1638 uint32_t helper_efdtstlt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1639 {
1640 CPU_DoubleU u1, u2;
1641
1642 u1.ll = op1;
1643 u2.ll = op2;
1644 return float64_lt(u1.d, u2.d, &env->vec_status) ? 4 : 0;
1645 }
1646
1647 uint32_t helper_efdtstgt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1648 {
1649 CPU_DoubleU u1, u2;
1650
1651 u1.ll = op1;
1652 u2.ll = op2;
1653 return float64_le(u1.d, u2.d, &env->vec_status) ? 0 : 4;
1654 }
1655
1656 uint32_t helper_efdtsteq(CPUPPCState *env, uint64_t op1, uint64_t op2)
1657 {
1658 CPU_DoubleU u1, u2;
1659
1660 u1.ll = op1;
1661 u2.ll = op2;
1662 return float64_eq_quiet(u1.d, u2.d, &env->vec_status) ? 4 : 0;
1663 }
1664
1665 uint32_t helper_efdcmplt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1666 {
1667 /* XXX: TODO: test special values (NaN, infinites, ...) */
1668 return helper_efdtstlt(env, op1, op2);
1669 }
1670
1671 uint32_t helper_efdcmpgt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1672 {
1673 /* XXX: TODO: test special values (NaN, infinites, ...) */
1674 return helper_efdtstgt(env, op1, op2);
1675 }
1676
1677 uint32_t helper_efdcmpeq(CPUPPCState *env, uint64_t op1, uint64_t op2)
1678 {
1679 /* XXX: TODO: test special values (NaN, infinites, ...) */
1680 return helper_efdtsteq(env, op1, op2);
1681 }
1682
1683 #define float64_to_float64(x, env) x
1684
1685
1686 /*
1687 * VSX_ADD_SUB - VSX floating point add/subtract
1688 * name - instruction mnemonic
1689 * op - operation (add or sub)
1690 * nels - number of elements (1, 2 or 4)
1691 * tp - type (float32 or float64)
1692 * fld - vsr_t field (VsrD(*) or VsrW(*))
1693 * sfprf - set FPRF
1694 */
1695 #define VSX_ADD_SUB(name, op, nels, tp, fld, sfprf, r2sp) \
1696 void helper_##name(CPUPPCState *env, ppc_vsr_t *xt, \
1697 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1698 { \
1699 ppc_vsr_t t = *xt; \
1700 int i; \
1701 \
1702 helper_reset_fpstatus(env); \
1703 \
1704 for (i = 0; i < nels; i++) { \
1705 float_status tstat = env->fp_status; \
1706 set_float_exception_flags(0, &tstat); \
1707 t.fld = tp##_##op(xa->fld, xb->fld, &tstat); \
1708 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1709 \
1710 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1711 float_invalid_op_addsub(env, tstat.float_exception_flags, \
1712 sfprf, GETPC()); \
1713 } \
1714 \
1715 if (r2sp) { \
1716 t.fld = do_frsp(env, t.fld, GETPC()); \
1717 } \
1718 \
1719 if (sfprf) { \
1720 helper_compute_fprf_float64(env, t.fld); \
1721 } \
1722 } \
1723 *xt = t; \
1724 do_float_check_status(env, GETPC()); \
1725 }
1726
1727 VSX_ADD_SUB(xsadddp, add, 1, float64, VsrD(0), 1, 0)
1728 VSX_ADD_SUB(xsaddsp, add, 1, float64, VsrD(0), 1, 1)
1729 VSX_ADD_SUB(xvadddp, add, 2, float64, VsrD(i), 0, 0)
1730 VSX_ADD_SUB(xvaddsp, add, 4, float32, VsrW(i), 0, 0)
1731 VSX_ADD_SUB(xssubdp, sub, 1, float64, VsrD(0), 1, 0)
1732 VSX_ADD_SUB(xssubsp, sub, 1, float64, VsrD(0), 1, 1)
1733 VSX_ADD_SUB(xvsubdp, sub, 2, float64, VsrD(i), 0, 0)
1734 VSX_ADD_SUB(xvsubsp, sub, 4, float32, VsrW(i), 0, 0)
1735
1736 void helper_xsaddqp(CPUPPCState *env, uint32_t opcode,
1737 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb)
1738 {
1739 ppc_vsr_t t = *xt;
1740 float_status tstat;
1741
1742 helper_reset_fpstatus(env);
1743
1744 tstat = env->fp_status;
1745 if (unlikely(Rc(opcode) != 0)) {
1746 tstat.float_rounding_mode = float_round_to_odd;
1747 }
1748
1749 set_float_exception_flags(0, &tstat);
1750 t.f128 = float128_add(xa->f128, xb->f128, &tstat);
1751 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1752
1753 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1754 float_invalid_op_addsub(env, tstat.float_exception_flags, 1, GETPC());
1755 }
1756
1757 helper_compute_fprf_float128(env, t.f128);
1758
1759 *xt = t;
1760 do_float_check_status(env, GETPC());
1761 }
1762
1763 /*
1764 * VSX_MUL - VSX floating point multiply
1765 * op - instruction mnemonic
1766 * nels - number of elements (1, 2 or 4)
1767 * tp - type (float32 or float64)
1768 * fld - vsr_t field (VsrD(*) or VsrW(*))
1769 * sfprf - set FPRF
1770 */
1771 #define VSX_MUL(op, nels, tp, fld, sfprf, r2sp) \
1772 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
1773 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1774 { \
1775 ppc_vsr_t t = *xt; \
1776 int i; \
1777 \
1778 helper_reset_fpstatus(env); \
1779 \
1780 for (i = 0; i < nels; i++) { \
1781 float_status tstat = env->fp_status; \
1782 set_float_exception_flags(0, &tstat); \
1783 t.fld = tp##_mul(xa->fld, xb->fld, &tstat); \
1784 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1785 \
1786 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1787 float_invalid_op_mul(env, tstat.float_exception_flags, \
1788 sfprf, GETPC()); \
1789 } \
1790 \
1791 if (r2sp) { \
1792 t.fld = do_frsp(env, t.fld, GETPC()); \
1793 } \
1794 \
1795 if (sfprf) { \
1796 helper_compute_fprf_float64(env, t.fld); \
1797 } \
1798 } \
1799 \
1800 *xt = t; \
1801 do_float_check_status(env, GETPC()); \
1802 }
1803
1804 VSX_MUL(xsmuldp, 1, float64, VsrD(0), 1, 0)
1805 VSX_MUL(xsmulsp, 1, float64, VsrD(0), 1, 1)
1806 VSX_MUL(xvmuldp, 2, float64, VsrD(i), 0, 0)
1807 VSX_MUL(xvmulsp, 4, float32, VsrW(i), 0, 0)
1808
1809 void helper_xsmulqp(CPUPPCState *env, uint32_t opcode,
1810 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb)
1811 {
1812 ppc_vsr_t t = *xt;
1813 float_status tstat;
1814
1815 helper_reset_fpstatus(env);
1816 tstat = env->fp_status;
1817 if (unlikely(Rc(opcode) != 0)) {
1818 tstat.float_rounding_mode = float_round_to_odd;
1819 }
1820
1821 set_float_exception_flags(0, &tstat);
1822 t.f128 = float128_mul(xa->f128, xb->f128, &tstat);
1823 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1824
1825 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1826 float_invalid_op_mul(env, tstat.float_exception_flags, 1, GETPC());
1827 }
1828 helper_compute_fprf_float128(env, t.f128);
1829
1830 *xt = t;
1831 do_float_check_status(env, GETPC());
1832 }
1833
1834 /*
1835 * VSX_DIV - VSX floating point divide
1836 * op - instruction mnemonic
1837 * nels - number of elements (1, 2 or 4)
1838 * tp - type (float32 or float64)
1839 * fld - vsr_t field (VsrD(*) or VsrW(*))
1840 * sfprf - set FPRF
1841 */
1842 #define VSX_DIV(op, nels, tp, fld, sfprf, r2sp) \
1843 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
1844 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1845 { \
1846 ppc_vsr_t t = *xt; \
1847 int i; \
1848 \
1849 helper_reset_fpstatus(env); \
1850 \
1851 for (i = 0; i < nels; i++) { \
1852 float_status tstat = env->fp_status; \
1853 set_float_exception_flags(0, &tstat); \
1854 t.fld = tp##_div(xa->fld, xb->fld, &tstat); \
1855 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1856 \
1857 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1858 float_invalid_op_div(env, tstat.float_exception_flags, \
1859 sfprf, GETPC()); \
1860 } \
1861 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) { \
1862 float_zero_divide_excp(env, GETPC()); \
1863 } \
1864 \
1865 if (r2sp) { \
1866 t.fld = do_frsp(env, t.fld, GETPC()); \
1867 } \
1868 \
1869 if (sfprf) { \
1870 helper_compute_fprf_float64(env, t.fld); \
1871 } \
1872 } \
1873 \
1874 *xt = t; \
1875 do_float_check_status(env, GETPC()); \
1876 }
1877
1878 VSX_DIV(xsdivdp, 1, float64, VsrD(0), 1, 0)
1879 VSX_DIV(xsdivsp, 1, float64, VsrD(0), 1, 1)
1880 VSX_DIV(xvdivdp, 2, float64, VsrD(i), 0, 0)
1881 VSX_DIV(xvdivsp, 4, float32, VsrW(i), 0, 0)
1882
1883 void helper_xsdivqp(CPUPPCState *env, uint32_t opcode,
1884 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb)
1885 {
1886 ppc_vsr_t t = *xt;
1887 float_status tstat;
1888
1889 helper_reset_fpstatus(env);
1890 tstat = env->fp_status;
1891 if (unlikely(Rc(opcode) != 0)) {
1892 tstat.float_rounding_mode = float_round_to_odd;
1893 }
1894
1895 set_float_exception_flags(0, &tstat);
1896 t.f128 = float128_div(xa->f128, xb->f128, &tstat);
1897 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1898
1899 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1900 float_invalid_op_div(env, tstat.float_exception_flags, 1, GETPC());
1901 }
1902 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) {
1903 float_zero_divide_excp(env, GETPC());
1904 }
1905
1906 helper_compute_fprf_float128(env, t.f128);
1907 *xt = t;
1908 do_float_check_status(env, GETPC());
1909 }
1910
1911 /*
1912 * VSX_RE - VSX floating point reciprocal estimate
1913 * op - instruction mnemonic
1914 * nels - number of elements (1, 2 or 4)
1915 * tp - type (float32 or float64)
1916 * fld - vsr_t field (VsrD(*) or VsrW(*))
1917 * sfprf - set FPRF
1918 */
1919 #define VSX_RE(op, nels, tp, fld, sfprf, r2sp) \
1920 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
1921 { \
1922 ppc_vsr_t t = *xt; \
1923 int i; \
1924 \
1925 helper_reset_fpstatus(env); \
1926 \
1927 for (i = 0; i < nels; i++) { \
1928 if (unlikely(tp##_is_signaling_nan(xb->fld, &env->fp_status))) { \
1929 float_invalid_op_vxsnan(env, GETPC()); \
1930 } \
1931 t.fld = tp##_div(tp##_one, xb->fld, &env->fp_status); \
1932 \
1933 if (r2sp) { \
1934 t.fld = do_frsp(env, t.fld, GETPC()); \
1935 } \
1936 \
1937 if (sfprf) { \
1938 helper_compute_fprf_float64(env, t.fld); \
1939 } \
1940 } \
1941 \
1942 *xt = t; \
1943 do_float_check_status(env, GETPC()); \
1944 }
1945
1946 VSX_RE(xsredp, 1, float64, VsrD(0), 1, 0)
1947 VSX_RE(xsresp, 1, float64, VsrD(0), 1, 1)
1948 VSX_RE(xvredp, 2, float64, VsrD(i), 0, 0)
1949 VSX_RE(xvresp, 4, float32, VsrW(i), 0, 0)
1950
1951 /*
1952 * VSX_SQRT - VSX floating point square root
1953 * op - instruction mnemonic
1954 * nels - number of elements (1, 2 or 4)
1955 * tp - type (float32 or float64)
1956 * fld - vsr_t field (VsrD(*) or VsrW(*))
1957 * sfprf - set FPRF
1958 */
1959 #define VSX_SQRT(op, nels, tp, fld, sfprf, r2sp) \
1960 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
1961 { \
1962 ppc_vsr_t t = *xt; \
1963 int i; \
1964 \
1965 helper_reset_fpstatus(env); \
1966 \
1967 for (i = 0; i < nels; i++) { \
1968 float_status tstat = env->fp_status; \
1969 set_float_exception_flags(0, &tstat); \
1970 t.fld = tp##_sqrt(xb->fld, &tstat); \
1971 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1972 \
1973 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1974 float_invalid_op_sqrt(env, tstat.float_exception_flags, \
1975 sfprf, GETPC()); \
1976 } \
1977 \
1978 if (r2sp) { \
1979 t.fld = do_frsp(env, t.fld, GETPC()); \
1980 } \
1981 \
1982 if (sfprf) { \
1983 helper_compute_fprf_float64(env, t.fld); \
1984 } \
1985 } \
1986 \
1987 *xt = t; \
1988 do_float_check_status(env, GETPC()); \
1989 }
1990
1991 VSX_SQRT(xssqrtdp, 1, float64, VsrD(0), 1, 0)
1992 VSX_SQRT(xssqrtsp, 1, float64, VsrD(0), 1, 1)
1993 VSX_SQRT(xvsqrtdp, 2, float64, VsrD(i), 0, 0)
1994 VSX_SQRT(xvsqrtsp, 4, float32, VsrW(i), 0, 0)
1995
1996 /*
1997 *VSX_RSQRTE - VSX floating point reciprocal square root estimate
1998 * op - instruction mnemonic
1999 * nels - number of elements (1, 2 or 4)
2000 * tp - type (float32 or float64)
2001 * fld - vsr_t field (VsrD(*) or VsrW(*))
2002 * sfprf - set FPRF
2003 */
2004 #define VSX_RSQRTE(op, nels, tp, fld, sfprf, r2sp) \
2005 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2006 { \
2007 ppc_vsr_t t = *xt; \
2008 int i; \
2009 \
2010 helper_reset_fpstatus(env); \
2011 \
2012 for (i = 0; i < nels; i++) { \
2013 float_status tstat = env->fp_status; \
2014 set_float_exception_flags(0, &tstat); \
2015 t.fld = tp##_sqrt(xb->fld, &tstat); \
2016 t.fld = tp##_div(tp##_one, t.fld, &tstat); \
2017 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
2018 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
2019 float_invalid_op_sqrt(env, tstat.float_exception_flags, \
2020 sfprf, GETPC()); \
2021 } \
2022 if (r2sp) { \
2023 t.fld = do_frsp(env, t.fld, GETPC()); \
2024 } \
2025 \
2026 if (sfprf) { \
2027 helper_compute_fprf_float64(env, t.fld); \
2028 } \
2029 } \
2030 \
2031 *xt = t; \
2032 do_float_check_status(env, GETPC()); \
2033 }
2034
2035 VSX_RSQRTE(xsrsqrtedp, 1, float64, VsrD(0), 1, 0)
2036 VSX_RSQRTE(xsrsqrtesp, 1, float64, VsrD(0), 1, 1)
2037 VSX_RSQRTE(xvrsqrtedp, 2, float64, VsrD(i), 0, 0)
2038 VSX_RSQRTE(xvrsqrtesp, 4, float32, VsrW(i), 0, 0)
2039
2040 /*
2041 * VSX_TDIV - VSX floating point test for divide
2042 * op - instruction mnemonic
2043 * nels - number of elements (1, 2 or 4)
2044 * tp - type (float32 or float64)
2045 * fld - vsr_t field (VsrD(*) or VsrW(*))
2046 * emin - minimum unbiased exponent
2047 * emax - maximum unbiased exponent
2048 * nbits - number of fraction bits
2049 */
2050 #define VSX_TDIV(op, nels, tp, fld, emin, emax, nbits) \
2051 void helper_##op(CPUPPCState *env, uint32_t opcode, \
2052 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2053 { \
2054 int i; \
2055 int fe_flag = 0; \
2056 int fg_flag = 0; \
2057 \
2058 for (i = 0; i < nels; i++) { \
2059 if (unlikely(tp##_is_infinity(xa->fld) || \
2060 tp##_is_infinity(xb->fld) || \
2061 tp##_is_zero(xb->fld))) { \
2062 fe_flag = 1; \
2063 fg_flag = 1; \
2064 } else { \
2065 int e_a = ppc_##tp##_get_unbiased_exp(xa->fld); \
2066 int e_b = ppc_##tp##_get_unbiased_exp(xb->fld); \
2067 \
2068 if (unlikely(tp##_is_any_nan(xa->fld) || \
2069 tp##_is_any_nan(xb->fld))) { \
2070 fe_flag = 1; \
2071 } else if ((e_b <= emin) || (e_b >= (emax - 2))) { \
2072 fe_flag = 1; \
2073 } else if (!tp##_is_zero(xa->fld) && \
2074 (((e_a - e_b) >= emax) || \
2075 ((e_a - e_b) <= (emin + 1)) || \
2076 (e_a <= (emin + nbits)))) { \
2077 fe_flag = 1; \
2078 } \
2079 \
2080 if (unlikely(tp##_is_zero_or_denormal(xb->fld))) { \
2081 /* \
2082 * XB is not zero because of the above check and so \
2083 * must be denormalized. \
2084 */ \
2085 fg_flag = 1; \
2086 } \
2087 } \
2088 } \
2089 \
2090 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
2091 }
2092
2093 VSX_TDIV(xstdivdp, 1, float64, VsrD(0), -1022, 1023, 52)
2094 VSX_TDIV(xvtdivdp, 2, float64, VsrD(i), -1022, 1023, 52)
2095 VSX_TDIV(xvtdivsp, 4, float32, VsrW(i), -126, 127, 23)
2096
2097 /*
2098 * VSX_TSQRT - VSX floating point test for square root
2099 * op - instruction mnemonic
2100 * nels - number of elements (1, 2 or 4)
2101 * tp - type (float32 or float64)
2102 * fld - vsr_t field (VsrD(*) or VsrW(*))
2103 * emin - minimum unbiased exponent
2104 * emax - maximum unbiased exponent
2105 * nbits - number of fraction bits
2106 */
2107 #define VSX_TSQRT(op, nels, tp, fld, emin, nbits) \
2108 void helper_##op(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xb) \
2109 { \
2110 int i; \
2111 int fe_flag = 0; \
2112 int fg_flag = 0; \
2113 \
2114 for (i = 0; i < nels; i++) { \
2115 if (unlikely(tp##_is_infinity(xb->fld) || \
2116 tp##_is_zero(xb->fld))) { \
2117 fe_flag = 1; \
2118 fg_flag = 1; \
2119 } else { \
2120 int e_b = ppc_##tp##_get_unbiased_exp(xb->fld); \
2121 \
2122 if (unlikely(tp##_is_any_nan(xb->fld))) { \
2123 fe_flag = 1; \
2124 } else if (unlikely(tp##_is_zero(xb->fld))) { \
2125 fe_flag = 1; \
2126 } else if (unlikely(tp##_is_neg(xb->fld))) { \
2127 fe_flag = 1; \
2128 } else if (!tp##_is_zero(xb->fld) && \
2129 (e_b <= (emin + nbits))) { \
2130 fe_flag = 1; \
2131 } \
2132 \
2133 if (unlikely(tp##_is_zero_or_denormal(xb->fld))) { \
2134 /* \
2135 * XB is not zero because of the above check and \
2136 * therefore must be denormalized. \
2137 */ \
2138 fg_flag = 1; \
2139 } \
2140 } \
2141 } \
2142 \
2143 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
2144 }
2145
2146 VSX_TSQRT(xstsqrtdp, 1, float64, VsrD(0), -1022, 52)
2147 VSX_TSQRT(xvtsqrtdp, 2, float64, VsrD(i), -1022, 52)
2148 VSX_TSQRT(xvtsqrtsp, 4, float32, VsrW(i), -126, 23)
2149
2150 /*
2151 * VSX_MADD - VSX floating point muliply/add variations
2152 * op - instruction mnemonic
2153 * nels - number of elements (1, 2 or 4)
2154 * tp - type (float32 or float64)
2155 * fld - vsr_t field (VsrD(*) or VsrW(*))
2156 * maddflgs - flags for the float*muladd routine that control the
2157 * various forms (madd, msub, nmadd, nmsub)
2158 * sfprf - set FPRF
2159 */
2160 #define VSX_MADD(op, nels, tp, fld, maddflgs, sfprf, r2sp) \
2161 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2162 ppc_vsr_t *xa, ppc_vsr_t *b, ppc_vsr_t *c) \
2163 { \
2164 ppc_vsr_t t = *xt; \
2165 int i; \
2166 \
2167 helper_reset_fpstatus(env); \
2168 \
2169 for (i = 0; i < nels; i++) { \
2170 float_status tstat = env->fp_status; \
2171 set_float_exception_flags(0, &tstat); \
2172 if (r2sp && (tstat.float_rounding_mode == float_round_nearest_even)) {\
2173 /* \
2174 * Avoid double rounding errors by rounding the intermediate \
2175 * result to odd. \
2176 */ \
2177 set_float_rounding_mode(float_round_to_zero, &tstat); \
2178 t.fld = tp##_muladd(xa->fld, b->fld, c->fld, \
2179 maddflgs, &tstat); \
2180 t.fld |= (get_float_exception_flags(&tstat) & \
2181 float_flag_inexact) != 0; \
2182 } else { \
2183 t.fld = tp##_muladd(xa->fld, b->fld, c->fld, \
2184 maddflgs, &tstat); \
2185 } \
2186 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
2187 \
2188 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
2189 float_invalid_op_madd(env, tstat.float_exception_flags, \
2190 sfprf, GETPC()); \
2191 } \
2192 \
2193 if (r2sp) { \
2194 t.fld = do_frsp(env, t.fld, GETPC()); \
2195 } \
2196 \
2197 if (sfprf) { \
2198 helper_compute_fprf_float64(env, t.fld); \
2199 } \
2200 } \
2201 *xt = t; \
2202 do_float_check_status(env, GETPC()); \
2203 }
2204
2205 VSX_MADD(xsmadddp, 1, float64, VsrD(0), MADD_FLGS, 1, 0)
2206 VSX_MADD(xsmsubdp, 1, float64, VsrD(0), MSUB_FLGS, 1, 0)
2207 VSX_MADD(xsnmadddp, 1, float64, VsrD(0), NMADD_FLGS, 1, 0)
2208 VSX_MADD(xsnmsubdp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 0)
2209 VSX_MADD(xsmaddsp, 1, float64, VsrD(0), MADD_FLGS, 1, 1)
2210 VSX_MADD(xsmsubsp, 1, float64, VsrD(0), MSUB_FLGS, 1, 1)
2211 VSX_MADD(xsnmaddsp, 1, float64, VsrD(0), NMADD_FLGS, 1, 1)
2212 VSX_MADD(xsnmsubsp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 1)
2213
2214 VSX_MADD(xvmadddp, 2, float64, VsrD(i), MADD_FLGS, 0, 0)
2215 VSX_MADD(xvmsubdp, 2, float64, VsrD(i), MSUB_FLGS, 0, 0)
2216 VSX_MADD(xvnmadddp, 2, float64, VsrD(i), NMADD_FLGS, 0, 0)
2217 VSX_MADD(xvnmsubdp, 2, float64, VsrD(i), NMSUB_FLGS, 0, 0)
2218
2219 VSX_MADD(xvmaddsp, 4, float32, VsrW(i), MADD_FLGS, 0, 0)
2220 VSX_MADD(xvmsubsp, 4, float32, VsrW(i), MSUB_FLGS, 0, 0)
2221 VSX_MADD(xvnmaddsp, 4, float32, VsrW(i), NMADD_FLGS, 0, 0)
2222 VSX_MADD(xvnmsubsp, 4, float32, VsrW(i), NMSUB_FLGS, 0, 0)
2223
2224 /*
2225 * VSX_SCALAR_CMP_DP - VSX scalar floating point compare double precision
2226 * op - instruction mnemonic
2227 * cmp - comparison operation
2228 * exp - expected result of comparison
2229 * svxvc - set VXVC bit
2230 */
2231 #define VSX_SCALAR_CMP_DP(op, cmp, exp, svxvc) \
2232 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2233 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2234 { \
2235 ppc_vsr_t t = *xt; \
2236 bool vxsnan_flag = false, vxvc_flag = false, vex_flag = false; \
2237 \
2238 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \
2239 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2240 vxsnan_flag = true; \
2241 if (fpscr_ve == 0 && svxvc) { \
2242 vxvc_flag = true; \
2243 } \
2244 } else if (svxvc) { \
2245 vxvc_flag = float64_is_quiet_nan(xa->VsrD(0), &env->fp_status) || \
2246 float64_is_quiet_nan(xb->VsrD(0), &env->fp_status); \
2247 } \
2248 if (vxsnan_flag) { \
2249 float_invalid_op_vxsnan(env, GETPC()); \
2250 } \
2251 if (vxvc_flag) { \
2252 float_invalid_op_vxvc(env, 0, GETPC()); \
2253 } \
2254 vex_flag = fpscr_ve && (vxvc_flag || vxsnan_flag); \
2255 \
2256 if (!vex_flag) { \
2257 if (float64_##cmp(xb->VsrD(0), xa->VsrD(0), \
2258 &env->fp_status) == exp) { \
2259 t.VsrD(0) = -1; \
2260 t.VsrD(1) = 0; \
2261 } else { \
2262 t.VsrD(0) = 0; \
2263 t.VsrD(1) = 0; \
2264 } \
2265 } \
2266 *xt = t; \
2267 do_float_check_status(env, GETPC()); \
2268 }
2269
2270 VSX_SCALAR_CMP_DP(xscmpeqdp, eq, 1, 0)
2271 VSX_SCALAR_CMP_DP(xscmpgedp, le, 1, 1)
2272 VSX_SCALAR_CMP_DP(xscmpgtdp, lt, 1, 1)
2273 VSX_SCALAR_CMP_DP(xscmpnedp, eq, 0, 0)
2274
2275 void helper_xscmpexpdp(CPUPPCState *env, uint32_t opcode,
2276 ppc_vsr_t *xa, ppc_vsr_t *xb)
2277 {
2278 int64_t exp_a, exp_b;
2279 uint32_t cc;
2280
2281 exp_a = extract64(xa->VsrD(0), 52, 11);
2282 exp_b = extract64(xb->VsrD(0), 52, 11);
2283
2284 if (unlikely(float64_is_any_nan(xa->VsrD(0)) ||
2285 float64_is_any_nan(xb->VsrD(0)))) {
2286 cc = CRF_SO;
2287 } else {
2288 if (exp_a < exp_b) {
2289 cc = CRF_LT;
2290 } else if (exp_a > exp_b) {
2291 cc = CRF_GT;
2292 } else {
2293 cc = CRF_EQ;
2294 }
2295 }
2296
2297 env->fpscr &= ~FP_FPCC;
2298 env->fpscr |= cc << FPSCR_FPCC;
2299 env->crf[BF(opcode)] = cc;
2300
2301 do_float_check_status(env, GETPC());
2302 }
2303
2304 void helper_xscmpexpqp(CPUPPCState *env, uint32_t opcode,
2305 ppc_vsr_t *xa, ppc_vsr_t *xb)
2306 {
2307 int64_t exp_a, exp_b;
2308 uint32_t cc;
2309
2310 exp_a = extract64(xa->VsrD(0), 48, 15);
2311 exp_b = extract64(xb->VsrD(0), 48, 15);
2312
2313 if (unlikely(float128_is_any_nan(xa->f128) ||
2314 float128_is_any_nan(xb->f128))) {
2315 cc = CRF_SO;
2316 } else {
2317 if (exp_a < exp_b) {
2318 cc = CRF_LT;
2319 } else if (exp_a > exp_b) {
2320 cc = CRF_GT;
2321 } else {
2322 cc = CRF_EQ;
2323 }
2324 }
2325
2326 env->fpscr &= ~FP_FPCC;
2327 env->fpscr |= cc << FPSCR_FPCC;
2328 env->crf[BF(opcode)] = cc;
2329
2330 do_float_check_status(env, GETPC());
2331 }
2332
2333 static inline void do_scalar_cmp(CPUPPCState *env, ppc_vsr_t *xa, ppc_vsr_t *xb,
2334 int crf_idx, bool ordered)
2335 {
2336 uint32_t cc;
2337 bool vxsnan_flag = false, vxvc_flag = false;
2338
2339 helper_reset_fpstatus(env);
2340
2341 switch (float64_compare(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) {
2342 case float_relation_less:
2343 cc = CRF_LT;
2344 break;
2345 case float_relation_equal:
2346 cc = CRF_EQ;
2347 break;
2348 case float_relation_greater:
2349 cc = CRF_GT;
2350 break;
2351 case float_relation_unordered:
2352 cc = CRF_SO;
2353
2354 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) ||
2355 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) {
2356 vxsnan_flag = true;
2357 if (fpscr_ve == 0 && ordered) {
2358 vxvc_flag = true;
2359 }
2360 } else if (float64_is_quiet_nan(xa->VsrD(0), &env->fp_status) ||
2361 float64_is_quiet_nan(xb->VsrD(0), &env->fp_status)) {
2362 if (ordered) {
2363 vxvc_flag = true;
2364 }
2365 }
2366
2367 break;
2368 default:
2369 g_assert_not_reached();
2370 }
2371
2372 env->fpscr &= ~FP_FPCC;
2373 env->fpscr |= cc << FPSCR_FPCC;
2374 env->crf[crf_idx] = cc;
2375
2376 if (vxsnan_flag) {
2377 float_invalid_op_vxsnan(env, GETPC());
2378 }
2379 if (vxvc_flag) {
2380 float_invalid_op_vxvc(env, 0, GETPC());
2381 }
2382
2383 do_float_check_status(env, GETPC());
2384 }
2385
2386 void helper_xscmpodp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xa,
2387 ppc_vsr_t *xb)
2388 {
2389 do_scalar_cmp(env, xa, xb, BF(opcode), true);
2390 }
2391
2392 void helper_xscmpudp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xa,
2393 ppc_vsr_t *xb)
2394 {
2395 do_scalar_cmp(env, xa, xb, BF(opcode), false);
2396 }
2397
2398 static inline void do_scalar_cmpq(CPUPPCState *env, ppc_vsr_t *xa,
2399 ppc_vsr_t *xb, int crf_idx, bool ordered)
2400 {
2401 uint32_t cc;
2402 bool vxsnan_flag = false, vxvc_flag = false;
2403
2404 helper_reset_fpstatus(env);
2405
2406 switch (float128_compare(xa->f128, xb->f128, &env->fp_status)) {
2407 case float_relation_less:
2408 cc = CRF_LT;
2409 break;
2410 case float_relation_equal:
2411 cc = CRF_EQ;
2412 break;
2413 case float_relation_greater:
2414 cc = CRF_GT;
2415 break;
2416 case float_relation_unordered:
2417 cc = CRF_SO;
2418
2419 if (float128_is_signaling_nan(xa->f128, &env->fp_status) ||
2420 float128_is_signaling_nan(xb->f128, &env->fp_status)) {
2421 vxsnan_flag = true;
2422 if (fpscr_ve == 0 && ordered) {
2423 vxvc_flag = true;
2424 }
2425 } else if (float128_is_quiet_nan(xa->f128, &env->fp_status) ||
2426 float128_is_quiet_nan(xb->f128, &env->fp_status)) {
2427 if (ordered) {
2428 vxvc_flag = true;
2429 }
2430 }
2431
2432 break;
2433 default:
2434 g_assert_not_reached();
2435 }
2436
2437 env->fpscr &= ~FP_FPCC;
2438 env->fpscr |= cc << FPSCR_FPCC;
2439 env->crf[crf_idx] = cc;
2440
2441 if (vxsnan_flag) {
2442 float_invalid_op_vxsnan(env, GETPC());
2443 }
2444 if (vxvc_flag) {
2445 float_invalid_op_vxvc(env, 0, GETPC());
2446 }
2447
2448 do_float_check_status(env, GETPC());
2449 }
2450
2451 void helper_xscmpoqp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xa,
2452 ppc_vsr_t *xb)
2453 {
2454 do_scalar_cmpq(env, xa, xb, BF(opcode), true);
2455 }
2456
2457 void helper_xscmpuqp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xa,
2458 ppc_vsr_t *xb)
2459 {
2460 do_scalar_cmpq(env, xa, xb, BF(opcode), false);
2461 }
2462
2463 /*
2464 * VSX_MAX_MIN - VSX floating point maximum/minimum
2465 * name - instruction mnemonic
2466 * op - operation (max or min)
2467 * nels - number of elements (1, 2 or 4)
2468 * tp - type (float32 or float64)
2469 * fld - vsr_t field (VsrD(*) or VsrW(*))
2470 */
2471 #define VSX_MAX_MIN(name, op, nels, tp, fld) \
2472 void helper_##name(CPUPPCState *env, ppc_vsr_t *xt, \
2473 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2474 { \
2475 ppc_vsr_t t = *xt; \
2476 int i; \
2477 \
2478 for (i = 0; i < nels; i++) { \
2479 t.fld = tp##_##op(xa->fld, xb->fld, &env->fp_status); \
2480 if (unlikely(tp##_is_signaling_nan(xa->fld, &env->fp_status) || \
2481 tp##_is_signaling_nan(xb->fld, &env->fp_status))) { \
2482 float_invalid_op_vxsnan(env, GETPC()); \
2483 } \
2484 } \
2485 \
2486 *xt = t; \
2487 do_float_check_status(env, GETPC()); \
2488 }
2489
2490 VSX_MAX_MIN(xsmaxdp, maxnum, 1, float64, VsrD(0))
2491 VSX_MAX_MIN(xvmaxdp, maxnum, 2, float64, VsrD(i))
2492 VSX_MAX_MIN(xvmaxsp, maxnum, 4, float32, VsrW(i))
2493 VSX_MAX_MIN(xsmindp, minnum, 1, float64, VsrD(0))
2494 VSX_MAX_MIN(xvmindp, minnum, 2, float64, VsrD(i))
2495 VSX_MAX_MIN(xvminsp, minnum, 4, float32, VsrW(i))
2496
2497 #define VSX_MAX_MINC(name, max) \
2498 void helper_##name(CPUPPCState *env, \
2499 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) \
2500 { \
2501 ppc_vsr_t t = *xt; \
2502 bool vxsnan_flag = false, vex_flag = false; \
2503 \
2504 if (unlikely(float64_is_any_nan(xa->VsrD(0)) || \
2505 float64_is_any_nan(xb->VsrD(0)))) { \
2506 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \
2507 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2508 vxsnan_flag = true; \
2509 } \
2510 t.VsrD(0) = xb->VsrD(0); \
2511 } else if ((max && \
2512 !float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) || \
2513 (!max && \
2514 float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status))) { \
2515 t.VsrD(0) = xa->VsrD(0); \
2516 } else { \
2517 t.VsrD(0) = xb->VsrD(0); \
2518 } \
2519 \
2520 vex_flag = fpscr_ve & vxsnan_flag; \
2521 if (vxsnan_flag) { \
2522 float_invalid_op_vxsnan(env, GETPC()); \
2523 } \
2524 if (!vex_flag) { \
2525 *xt = t; \
2526 } \
2527 } \
2528
2529 VSX_MAX_MINC(xsmaxcdp, 1);
2530 VSX_MAX_MINC(xsmincdp, 0);
2531
2532 #define VSX_MAX_MINJ(name, max) \
2533 void helper_##name(CPUPPCState *env, \
2534 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) \
2535 { \
2536 ppc_vsr_t t = *xt; \
2537 bool vxsnan_flag = false, vex_flag = false; \
2538 \
2539 if (unlikely(float64_is_any_nan(xa->VsrD(0)))) { \
2540 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status)) { \
2541 vxsnan_flag = true; \
2542 } \
2543 t.VsrD(0) = xa->VsrD(0); \
2544 } else if (unlikely(float64_is_any_nan(xb->VsrD(0)))) { \
2545 if (float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2546 vxsnan_flag = true; \
2547 } \
2548 t.VsrD(0) = xb->VsrD(0); \
2549 } else if (float64_is_zero(xa->VsrD(0)) && \
2550 float64_is_zero(xb->VsrD(0))) { \
2551 if (max) { \
2552 if (!float64_is_neg(xa->VsrD(0)) || \
2553 !float64_is_neg(xb->VsrD(0))) { \
2554 t.VsrD(0) = 0ULL; \
2555 } else { \
2556 t.VsrD(0) = 0x8000000000000000ULL; \
2557 } \
2558 } else { \
2559 if (float64_is_neg(xa->VsrD(0)) || \
2560 float64_is_neg(xb->VsrD(0))) { \
2561 t.VsrD(0) = 0x8000000000000000ULL; \
2562 } else { \
2563 t.VsrD(0) = 0ULL; \
2564 } \
2565 } \
2566 } else if ((max && \
2567 !float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) || \
2568 (!max && \
2569 float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status))) { \
2570 t.VsrD(0) = xa->VsrD(0); \
2571 } else { \
2572 t.VsrD(0) = xb->VsrD(0); \
2573 } \
2574 \
2575 vex_flag = fpscr_ve & vxsnan_flag; \
2576 if (vxsnan_flag) { \
2577 float_invalid_op_vxsnan(env, GETPC()); \
2578 } \
2579 if (!vex_flag) { \
2580 *xt = t; \
2581 } \
2582 } \
2583
2584 VSX_MAX_MINJ(xsmaxjdp, 1);
2585 VSX_MAX_MINJ(xsminjdp, 0);
2586
2587 /*
2588 * VSX_CMP - VSX floating point compare
2589 * op - instruction mnemonic
2590 * nels - number of elements (1, 2 or 4)
2591 * tp - type (float32 or float64)
2592 * fld - vsr_t field (VsrD(*) or VsrW(*))
2593 * cmp - comparison operation
2594 * svxvc - set VXVC bit
2595 * exp - expected result of comparison
2596 */
2597 #define VSX_CMP(op, nels, tp, fld, cmp, svxvc, exp) \
2598 uint32_t helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2599 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2600 { \
2601 ppc_vsr_t t = *xt; \
2602 uint32_t crf6 = 0; \
2603 int i; \
2604 int all_true = 1; \
2605 int all_false = 1; \
2606 \
2607 for (i = 0; i < nels; i++) { \
2608 if (unlikely(tp##_is_any_nan(xa->fld) || \
2609 tp##_is_any_nan(xb->fld))) { \
2610 if (tp##_is_signaling_nan(xa->fld, &env->fp_status) || \
2611 tp##_is_signaling_nan(xb->fld, &env->fp_status)) { \
2612 float_invalid_op_vxsnan(env, GETPC()); \
2613 } \
2614 if (svxvc) { \
2615 float_invalid_op_vxvc(env, 0, GETPC()); \
2616 } \
2617 t.fld = 0; \
2618 all_true = 0; \
2619 } else { \
2620 if (tp##_##cmp(xb->fld, xa->fld, &env->fp_status) == exp) { \
2621 t.fld = -1; \
2622 all_false = 0; \
2623 } else { \
2624 t.fld = 0; \
2625 all_true = 0; \
2626 } \
2627 } \
2628 } \
2629 \
2630 *xt = t; \
2631 crf6 = (all_true ? 0x8 : 0) | (all_false ? 0x2 : 0); \
2632 return crf6; \
2633 }
2634
2635 VSX_CMP(xvcmpeqdp, 2, float64, VsrD(i), eq, 0, 1)
2636 VSX_CMP(xvcmpgedp, 2, float64, VsrD(i), le, 1, 1)
2637 VSX_CMP(xvcmpgtdp, 2, float64, VsrD(i), lt, 1, 1)
2638 VSX_CMP(xvcmpnedp, 2, float64, VsrD(i), eq, 0, 0)
2639 VSX_CMP(xvcmpeqsp, 4, float32, VsrW(i), eq, 0, 1)
2640 VSX_CMP(xvcmpgesp, 4, float32, VsrW(i), le, 1, 1)
2641 VSX_CMP(xvcmpgtsp, 4, float32, VsrW(i), lt, 1, 1)
2642 VSX_CMP(xvcmpnesp, 4, float32, VsrW(i), eq, 0, 0)
2643
2644 /*
2645 * VSX_CVT_FP_TO_FP - VSX floating point/floating point conversion
2646 * op - instruction mnemonic
2647 * nels - number of elements (1, 2 or 4)
2648 * stp - source type (float32 or float64)
2649 * ttp - target type (float32 or float64)
2650 * sfld - source vsr_t field
2651 * tfld - target vsr_t field (f32 or f64)
2652 * sfprf - set FPRF
2653 */
2654 #define VSX_CVT_FP_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf) \
2655 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2656 { \
2657 ppc_vsr_t t = *xt; \
2658 int i; \
2659 \
2660 for (i = 0; i < nels; i++) { \
2661 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2662 if (unlikely(stp##_is_signaling_nan(xb->sfld, \
2663 &env->fp_status))) { \
2664 float_invalid_op_vxsnan(env, GETPC()); \
2665 t.tfld = ttp##_snan_to_qnan(t.tfld); \
2666 } \
2667 if (sfprf) { \
2668 helper_compute_fprf_##ttp(env, t.tfld); \
2669 } \
2670 } \
2671 \
2672 *xt = t; \
2673 do_float_check_status(env, GETPC()); \
2674 }
2675
2676 VSX_CVT_FP_TO_FP(xscvdpsp, 1, float64, float32, VsrD(0), VsrW(0), 1)
2677 VSX_CVT_FP_TO_FP(xscvspdp, 1, float32, float64, VsrW(0), VsrD(0), 1)
2678 VSX_CVT_FP_TO_FP(xvcvdpsp, 2, float64, float32, VsrD(i), VsrW(2 * i), 0)
2679 VSX_CVT_FP_TO_FP(xvcvspdp, 2, float32, float64, VsrW(2 * i), VsrD(i), 0)
2680
2681 /*
2682 * VSX_CVT_FP_TO_FP_VECTOR - VSX floating point/floating point conversion
2683 * op - instruction mnemonic
2684 * nels - number of elements (1, 2 or 4)
2685 * stp - source type (float32 or float64)
2686 * ttp - target type (float32 or float64)
2687 * sfld - source vsr_t field
2688 * tfld - target vsr_t field (f32 or f64)
2689 * sfprf - set FPRF
2690 */
2691 #define VSX_CVT_FP_TO_FP_VECTOR(op, nels, stp, ttp, sfld, tfld, sfprf) \
2692 void helper_##op(CPUPPCState *env, uint32_t opcode, \
2693 ppc_vsr_t *xt, ppc_vsr_t *xb) \
2694 { \
2695 ppc_vsr_t t = *xt; \
2696 int i; \
2697 \
2698 for (i = 0; i < nels; i++) { \
2699 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2700 if (unlikely(stp##_is_signaling_nan(xb->sfld, \
2701 &env->fp_status))) { \
2702 float_invalid_op_vxsnan(env, GETPC()); \
2703 t.tfld = ttp##_snan_to_qnan(t.tfld); \
2704 } \
2705 if (sfprf) { \
2706 helper_compute_fprf_##ttp(env, t.tfld); \
2707 } \
2708 } \
2709 \
2710 *xt = t; \
2711 do_float_check_status(env, GETPC()); \
2712 }
2713
2714 VSX_CVT_FP_TO_FP_VECTOR(xscvdpqp, 1, float64, float128, VsrD(0), f128, 1)
2715
2716 /*
2717 * VSX_CVT_FP_TO_FP_HP - VSX floating point/floating point conversion
2718 * involving one half precision value
2719 * op - instruction mnemonic
2720 * nels - number of elements (1, 2 or 4)
2721 * stp - source type
2722 * ttp - target type
2723 * sfld - source vsr_t field
2724 * tfld - target vsr_t field
2725 * sfprf - set FPRF
2726 */
2727 #define VSX_CVT_FP_TO_FP_HP(op, nels, stp, ttp, sfld, tfld, sfprf) \
2728 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2729 { \
2730 ppc_vsr_t t = { }; \
2731 int i; \
2732 \
2733 for (i = 0; i < nels; i++) { \
2734 t.tfld = stp##_to_##ttp(xb->sfld, 1, &env->fp_status); \
2735 if (unlikely(stp##_is_signaling_nan(xb->sfld, \
2736 &env->fp_status))) { \
2737 float_invalid_op_vxsnan(env, GETPC()); \
2738 t.tfld = ttp##_snan_to_qnan(t.tfld); \
2739 } \
2740 if (sfprf) { \
2741 helper_compute_fprf_##ttp(env, t.tfld); \
2742 } \
2743 } \
2744 \
2745 *xt = t; \
2746 do_float_check_status(env, GETPC()); \
2747 }
2748
2749 VSX_CVT_FP_TO_FP_HP(xscvdphp, 1, float64, float16, VsrD(0), VsrH(3), 1)
2750 VSX_CVT_FP_TO_FP_HP(xscvhpdp, 1, float16, float64, VsrH(3), VsrD(0), 1)
2751 VSX_CVT_FP_TO_FP_HP(xvcvsphp, 4, float32, float16, VsrW(i), VsrH(2 * i + 1), 0)
2752 VSX_CVT_FP_TO_FP_HP(xvcvhpsp, 4, float16, float32, VsrH(2 * i + 1), VsrW(i), 0)
2753
2754 void helper_XSCVQPDP(CPUPPCState *env, uint32_t ro, ppc_vsr_t *xt,
2755 ppc_vsr_t *xb)
2756 {
2757 ppc_vsr_t t = { };
2758 float_status tstat;
2759
2760 tstat = env->fp_status;
2761 if (ro != 0) {
2762 tstat.float_rounding_mode = float_round_to_odd;
2763 }
2764
2765 t.VsrD(0) = float128_to_float64(xb->f128, &tstat);
2766 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
2767 if (unlikely(float128_is_signaling_nan(xb->f128, &tstat))) {
2768 float_invalid_op_vxsnan(env, GETPC());
2769 t.VsrD(0) = float64_snan_to_qnan(t.VsrD(0));
2770 }
2771 helper_compute_fprf_float64(env, t.VsrD(0));
2772
2773 *xt = t;
2774 do_float_check_status(env, GETPC());
2775 }
2776
2777 uint64_t helper_xscvdpspn(CPUPPCState *env, uint64_t xb)
2778 {
2779 uint64_t result, sign, exp, frac;
2780
2781 float_status tstat = env->fp_status;
2782 set_float_exception_flags(0, &tstat);
2783
2784 sign = extract64(xb, 63, 1);
2785 exp = extract64(xb, 52, 11);
2786 frac = extract64(xb, 0, 52) | 0x10000000000000ULL;
2787
2788 if (unlikely(exp == 0 && extract64(frac, 0, 52) != 0)) {
2789 /* DP denormal operand. */
2790 /* Exponent override to DP min exp. */
2791 exp = 1;
2792 /* Implicit bit override to 0. */
2793 frac = deposit64(frac, 53, 1, 0);
2794 }
2795
2796 if (unlikely(exp < 897 && frac != 0)) {
2797 /* SP tiny operand. */
2798 if (897 - exp > 63) {
2799 frac = 0;
2800 } else {
2801 /* Denormalize until exp = SP min exp. */
2802 frac >>= (897 - exp);
2803 }
2804 /* Exponent override to SP min exp - 1. */
2805 exp = 896;
2806 }
2807
2808 result = sign << 31;
2809 result |= extract64(exp, 10, 1) << 30;
2810 result |= extract64(exp, 0, 7) << 23;
2811 result |= extract64(frac, 29, 23);
2812
2813 /* hardware replicates result to both words of the doubleword result. */
2814 return (result << 32) | result;
2815 }
2816
2817 uint64_t helper_xscvspdpn(CPUPPCState *env, uint64_t xb)
2818 {
2819 return helper_todouble(xb >> 32);
2820 }
2821
2822 /*
2823 * VSX_CVT_FP_TO_INT - VSX floating point to integer conversion
2824 * op - instruction mnemonic
2825 * nels - number of elements (1, 2 or 4)
2826 * stp - source type (float32 or float64)
2827 * ttp - target type (int32, uint32, int64 or uint64)
2828 * sfld - source vsr_t field
2829 * tfld - target vsr_t field
2830 * rnan - resulting NaN
2831 */
2832 #define VSX_CVT_FP_TO_INT(op, nels, stp, ttp, sfld, tfld, rnan) \
2833 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2834 { \
2835 int all_flags = env->fp_status.float_exception_flags, flags; \
2836 ppc_vsr_t t = *xt; \
2837 int i; \
2838 \
2839 for (i = 0; i < nels; i++) { \
2840 env->fp_status.float_exception_flags = 0; \
2841 t.tfld = stp##_to_##ttp##_round_to_zero(xb->sfld, &env->fp_status); \
2842 flags = env->fp_status.float_exception_flags; \
2843 if (unlikely(flags & float_flag_invalid)) { \
2844 t.tfld = float_invalid_cvt(env, flags, t.tfld, rnan, 0, GETPC());\
2845 } \
2846 all_flags |= flags; \
2847 } \
2848 \
2849 *xt = t; \
2850 env->fp_status.float_exception_flags = all_flags; \
2851 do_float_check_status(env, GETPC()); \
2852 }
2853
2854 VSX_CVT_FP_TO_INT(xscvdpsxds, 1, float64, int64, VsrD(0), VsrD(0), \
2855 0x8000000000000000ULL)
2856 VSX_CVT_FP_TO_INT(xscvdpsxws, 1, float64, int32, VsrD(0), VsrW(1), \
2857 0x80000000U)
2858 VSX_CVT_FP_TO_INT(xscvdpuxds, 1, float64, uint64, VsrD(0), VsrD(0), 0ULL)
2859 VSX_CVT_FP_TO_INT(xscvdpuxws, 1, float64, uint32, VsrD(0), VsrW(1), 0U)
2860 VSX_CVT_FP_TO_INT(xvcvdpsxds, 2, float64, int64, VsrD(i), VsrD(i), \
2861 0x8000000000000000ULL)
2862 VSX_CVT_FP_TO_INT(xvcvdpsxws, 2, float64, int32, VsrD(i), VsrW(2 * i), \
2863 0x80000000U)
2864 VSX_CVT_FP_TO_INT(xvcvdpuxds, 2, float64, uint64, VsrD(i), VsrD(i), 0ULL)
2865 VSX_CVT_FP_TO_INT(xvcvdpuxws, 2, float64, uint32, VsrD(i), VsrW(2 * i), 0U)
2866 VSX_CVT_FP_TO_INT(xvcvspsxds, 2, float32, int64, VsrW(2 * i), VsrD(i), \
2867 0x8000000000000000ULL)
2868 VSX_CVT_FP_TO_INT(xvcvspsxws, 4, float32, int32, VsrW(i), VsrW(i), 0x80000000U)
2869 VSX_CVT_FP_TO_INT(xvcvspuxds, 2, float32, uint64, VsrW(2 * i), VsrD(i), 0ULL)
2870 VSX_CVT_FP_TO_INT(xvcvspuxws, 4, float32, uint32, VsrW(i), VsrW(i), 0U)
2871
2872 /*
2873 * VSX_CVT_FP_TO_INT_VECTOR - VSX floating point to integer conversion
2874 * op - instruction mnemonic
2875 * stp - source type (float32 or float64)
2876 * ttp - target type (int32, uint32, int64 or uint64)
2877 * sfld - source vsr_t field
2878 * tfld - target vsr_t field
2879 * rnan - resulting NaN
2880 */
2881 #define VSX_CVT_FP_TO_INT_VECTOR(op, stp, ttp, sfld, tfld, rnan) \
2882 void helper_##op(CPUPPCState *env, uint32_t opcode, \
2883 ppc_vsr_t *xt, ppc_vsr_t *xb) \
2884 { \
2885 ppc_vsr_t t = { }; \
2886 int flags; \
2887 \
2888 t.tfld = stp##_to_##ttp##_round_to_zero(xb->sfld, &env->fp_status); \
2889 flags = get_float_exception_flags(&env->fp_status); \
2890 if (flags & float_flag_invalid) { \
2891 t.tfld = float_invalid_cvt(env, flags, t.tfld, rnan, 0, GETPC()); \
2892 } \
2893 \
2894 *xt = t; \
2895 do_float_check_status(env, GETPC()); \
2896 }
2897
2898 VSX_CVT_FP_TO_INT_VECTOR(xscvqpsdz, float128, int64, f128, VsrD(0), \
2899 0x8000000000000000ULL)
2900
2901 VSX_CVT_FP_TO_INT_VECTOR(xscvqpswz, float128, int32, f128, VsrD(0), \
2902 0xffffffff80000000ULL)
2903 VSX_CVT_FP_TO_INT_VECTOR(xscvqpudz, float128, uint64, f128, VsrD(0), 0x0ULL)
2904 VSX_CVT_FP_TO_INT_VECTOR(xscvqpuwz, float128, uint32, f128, VsrD(0), 0x0ULL)
2905
2906 /*
2907 * VSX_CVT_INT_TO_FP - VSX integer to floating point conversion
2908 * op - instruction mnemonic
2909 * nels - number of elements (1, 2 or 4)
2910 * stp - source type (int32, uint32, int64 or uint64)
2911 * ttp - target type (float32 or float64)
2912 * sfld - source vsr_t field
2913 * tfld - target vsr_t field
2914 * jdef - definition of the j index (i or 2*i)
2915 * sfprf - set FPRF
2916 */
2917 #define VSX_CVT_INT_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf, r2sp) \
2918 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2919 { \
2920 ppc_vsr_t t = *xt; \
2921 int i; \
2922 \
2923 for (i = 0; i < nels; i++) { \
2924 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2925 if (r2sp) { \
2926 t.tfld = do_frsp(env, t.tfld, GETPC()); \
2927 } \
2928 if (sfprf) { \
2929 helper_compute_fprf_float64(env, t.tfld); \
2930 } \
2931 } \
2932 \
2933 *xt = t; \
2934 do_float_check_status(env, GETPC()); \
2935 }
2936
2937 VSX_CVT_INT_TO_FP(xscvsxddp, 1, int64, float64, VsrD(0), VsrD(0), 1, 0)
2938 VSX_CVT_INT_TO_FP(xscvuxddp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 0)
2939 VSX_CVT_INT_TO_FP(xscvsxdsp, 1, int64, float64, VsrD(0), VsrD(0), 1, 1)
2940 VSX_CVT_INT_TO_FP(xscvuxdsp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 1)
2941 VSX_CVT_INT_TO_FP(xvcvsxddp, 2, int64, float64, VsrD(i), VsrD(i), 0, 0)
2942 VSX_CVT_INT_TO_FP(xvcvuxddp, 2, uint64, float64, VsrD(i), VsrD(i), 0, 0)
2943 VSX_CVT_INT_TO_FP(xvcvsxwdp, 2, int32, float64, VsrW(2 * i), VsrD(i), 0, 0)
2944 VSX_CVT_INT_TO_FP(xvcvuxwdp, 2, uint64, float64, VsrW(2 * i), VsrD(i), 0, 0)
2945 VSX_CVT_INT_TO_FP(xvcvsxdsp, 2, int64, float32, VsrD(i), VsrW(2 * i), 0, 0)
2946 VSX_CVT_INT_TO_FP(xvcvuxdsp, 2, uint64, float32, VsrD(i), VsrW(2 * i), 0, 0)
2947 VSX_CVT_INT_TO_FP(xvcvsxwsp, 4, int32, float32, VsrW(i), VsrW(i), 0, 0)
2948 VSX_CVT_INT_TO_FP(xvcvuxwsp, 4, uint32, float32, VsrW(i), VsrW(i), 0, 0)
2949
2950 /*
2951 * VSX_CVT_INT_TO_FP_VECTOR - VSX integer to floating point conversion
2952 * op - instruction mnemonic
2953 * stp - source type (int32, uint32, int64 or uint64)
2954 * ttp - target type (float32 or float64)
2955 * sfld - source vsr_t field
2956 * tfld - target vsr_t field
2957 */
2958 #define VSX_CVT_INT_TO_FP_VECTOR(op, stp, ttp, sfld, tfld) \
2959 void helper_##op(CPUPPCState *env, uint32_t opcode, \
2960 ppc_vsr_t *xt, ppc_vsr_t *xb) \
2961 { \
2962 ppc_vsr_t t = *xt; \
2963 \
2964 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2965 helper_compute_fprf_##ttp(env, t.tfld); \
2966 \
2967 *xt = t; \
2968 do_float_check_status(env, GETPC()); \
2969 }
2970
2971 VSX_CVT_INT_TO_FP_VECTOR(xscvsdqp, int64, float128, VsrD(0), f128)
2972 VSX_CVT_INT_TO_FP_VECTOR(xscvudqp, uint64, float128, VsrD(0), f128)
2973
2974 /*
2975 * For "use current rounding mode", define a value that will not be
2976 * one of the existing rounding model enums.
2977 */
2978 #define FLOAT_ROUND_CURRENT (float_round_nearest_even + float_round_down + \
2979 float_round_up + float_round_to_zero)
2980
2981 /*
2982 * VSX_ROUND - VSX floating point round
2983 * op - instruction mnemonic
2984 * nels - number of elements (1, 2 or 4)
2985 * tp - type (float32 or float64)
2986 * fld - vsr_t field (VsrD(*) or VsrW(*))
2987 * rmode - rounding mode
2988 * sfprf - set FPRF
2989 */
2990 #define VSX_ROUND(op, nels, tp, fld, rmode, sfprf) \
2991 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2992 { \
2993 ppc_vsr_t t = *xt; \
2994 int i; \
2995 FloatRoundMode curr_rounding_mode; \
2996 \
2997 if (rmode != FLOAT_ROUND_CURRENT) { \
2998 curr_rounding_mode = get_float_rounding_mode(&env->fp_status); \
2999 set_float_rounding_mode(rmode, &env->fp_status); \
3000 } \
3001 \
3002 for (i = 0; i < nels; i++) { \
3003 if (unlikely(tp##_is_signaling_nan(xb->fld, \
3004 &env->fp_status))) { \
3005 float_invalid_op_vxsnan(env, GETPC()); \
3006 t.fld = tp##_snan_to_qnan(xb->fld); \
3007 } else { \
3008 t.fld = tp##_round_to_int(xb->fld, &env->fp_status); \
3009 } \
3010 if (sfprf) { \
3011 helper_compute_fprf_float64(env, t.fld); \
3012 } \
3013 } \
3014 \
3015 /* \
3016 * If this is not a "use current rounding mode" instruction, \
3017 * then inhibit setting of the XX bit and restore rounding \
3018 * mode from FPSCR \
3019 */ \
3020 if (rmode != FLOAT_ROUND_CURRENT) { \
3021 set_float_rounding_mode(curr_rounding_mode, &env->fp_status); \
3022 env->fp_status.float_exception_flags &= ~float_flag_inexact; \
3023 } \
3024 \
3025 *xt = t; \
3026 do_float_check_status(env, GETPC()); \
3027 }
3028
3029 VSX_ROUND(xsrdpi, 1, float64, VsrD(0), float_round_ties_away, 1)
3030 VSX_ROUND(xsrdpic, 1, float64, VsrD(0), FLOAT_ROUND_CURRENT, 1)
3031 VSX_ROUND(xsrdpim, 1, float64, VsrD(0), float_round_down, 1)
3032 VSX_ROUND(xsrdpip, 1, float64, VsrD(0), float_round_up, 1)
3033 VSX_ROUND(xsrdpiz, 1, float64, VsrD(0), float_round_to_zero, 1)
3034
3035 VSX_ROUND(xvrdpi, 2, float64, VsrD(i), float_round_ties_away, 0)
3036 VSX_ROUND(xvrdpic, 2, float64, VsrD(i), FLOAT_ROUND_CURRENT, 0)
3037 VSX_ROUND(xvrdpim, 2, float64, VsrD(i), float_round_down, 0)
3038 VSX_ROUND(xvrdpip, 2, float64, VsrD(i), float_round_up, 0)
3039 VSX_ROUND(xvrdpiz, 2, float64, VsrD(i), float_round_to_zero, 0)
3040
3041 VSX_ROUND(xvrspi, 4, float32, VsrW(i), float_round_ties_away, 0)
3042 VSX_ROUND(xvrspic, 4, float32, VsrW(i), FLOAT_ROUND_CURRENT, 0)
3043 VSX_ROUND(xvrspim, 4, float32, VsrW(i), float_round_down, 0)
3044 VSX_ROUND(xvrspip, 4, float32, VsrW(i), float_round_up, 0)
3045 VSX_ROUND(xvrspiz, 4, float32, VsrW(i), float_round_to_zero, 0)
3046
3047 uint64_t helper_xsrsp(CPUPPCState *env, uint64_t xb)
3048 {
3049 helper_reset_fpstatus(env);
3050
3051 uint64_t xt = do_frsp(env, xb, GETPC());
3052
3053 helper_compute_fprf_float64(env, xt);
3054 do_float_check_status(env, GETPC());
3055 return xt;
3056 }
3057
3058 #define VSX_XXPERM(op, indexed) \
3059 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
3060 ppc_vsr_t *xa, ppc_vsr_t *pcv) \
3061 { \
3062 ppc_vsr_t t = *xt; \
3063 int i, idx; \
3064 \
3065 for (i = 0; i < 16; i++) { \
3066 idx = pcv->VsrB(i) & 0x1F; \
3067 if (indexed) { \
3068 idx = 31 - idx; \
3069 } \
3070 t.VsrB(i) = (idx <= 15) ? xa->VsrB(idx) \
3071 : xt->VsrB(idx - 16); \
3072 } \
3073 *xt = t; \
3074 }
3075
3076 VSX_XXPERM(xxperm, 0)
3077 VSX_XXPERM(xxpermr, 1)
3078
3079 void helper_xvxsigsp(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb)
3080 {
3081 ppc_vsr_t t = { };
3082 uint32_t exp, i, fraction;
3083
3084 for (i = 0; i < 4; i++) {
3085 exp = (xb->VsrW(i) >> 23) & 0xFF;
3086 fraction = xb->VsrW(i) & 0x7FFFFF;
3087 if (exp != 0 && exp != 255) {
3088 t.VsrW(i) = fraction | 0x00800000;
3089 } else {
3090 t.VsrW(i) = fraction;
3091 }
3092 }
3093 *xt = t;
3094 }
3095
3096 /*
3097 * VSX_TEST_DC - VSX floating point test data class
3098 * op - instruction mnemonic
3099 * nels - number of elements (1, 2 or 4)
3100 * xbn - VSR register number
3101 * tp - type (float32 or float64)
3102 * fld - vsr_t field (VsrD(*) or VsrW(*))
3103 * tfld - target vsr_t field (VsrD(*) or VsrW(*))
3104 * fld_max - target field max
3105 * scrf - set result in CR and FPCC
3106 */
3107 #define VSX_TEST_DC(op, nels, xbn, tp, fld, tfld, fld_max, scrf) \
3108 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3109 { \
3110 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
3111 ppc_vsr_t *xb = &env->vsr[xbn]; \
3112 ppc_vsr_t t = { }; \
3113 uint32_t i, sign, dcmx; \
3114 uint32_t cc, match = 0; \
3115 \
3116 if (!scrf) { \
3117 dcmx = DCMX_XV(opcode); \
3118 } else { \
3119 t = *xt; \
3120 dcmx = DCMX(opcode); \
3121 } \
3122 \
3123 for (i = 0; i < nels; i++) { \
3124 sign = tp##_is_neg(xb->fld); \
3125 if (tp##_is_any_nan(xb->fld)) { \
3126 match = extract32(dcmx, 6, 1); \
3127 } else if (tp##_is_infinity(xb->fld)) { \
3128 match = extract32(dcmx, 4 + !sign, 1); \
3129 } else if (tp##_is_zero(xb->fld)) { \
3130 match = extract32(dcmx, 2 + !sign, 1); \
3131 } else if (tp##_is_zero_or_denormal(xb->fld)) { \
3132 match = extract32(dcmx, 0 + !sign, 1); \
3133 } \
3134 \
3135 if (scrf) { \
3136 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT; \
3137 env->fpscr &= ~FP_FPCC; \
3138 env->fpscr |= cc << FPSCR_FPCC; \
3139 env->crf[BF(opcode)] = cc; \
3140 } else { \
3141 t.tfld = match ? fld_max : 0; \
3142 } \
3143 match = 0; \
3144 } \
3145 if (!scrf) { \
3146 *xt = t; \
3147 } \
3148 }
3149
3150 VSX_TEST_DC(xvtstdcdp, 2, xB(opcode), float64, VsrD(i), VsrD(i), UINT64_MAX, 0)
3151 VSX_TEST_DC(xvtstdcsp, 4, xB(opcode), float32, VsrW(i), VsrW(i), UINT32_MAX, 0)
3152 VSX_TEST_DC(xststdcdp, 1, xB(opcode), float64, VsrD(0), VsrD(0), 0, 1)
3153 VSX_TEST_DC(xststdcqp, 1, (rB(opcode) + 32), float128, f128, VsrD(0), 0, 1)
3154
3155 void helper_xststdcsp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xb)
3156 {
3157 uint32_t dcmx, sign, exp;
3158 uint32_t cc, match = 0, not_sp = 0;
3159 float64 arg = xb->VsrD(0);
3160 float64 arg_sp;
3161
3162 dcmx = DCMX(opcode);
3163 exp = (arg >> 52) & 0x7FF;
3164 sign = float64_is_neg(arg);
3165
3166 if (float64_is_any_nan(arg)) {
3167 match = extract32(dcmx, 6, 1);
3168 } else if (float64_is_infinity(arg)) {
3169 match = extract32(dcmx, 4 + !sign, 1);
3170 } else if (float64_is_zero(arg)) {
3171 match = extract32(dcmx, 2 + !sign, 1);
3172 } else if (float64_is_zero_or_denormal(arg) || (exp > 0 && exp < 0x381)) {
3173 match = extract32(dcmx, 0 + !sign, 1);
3174 }
3175
3176 arg_sp = helper_todouble(helper_tosingle(arg));
3177 not_sp = arg != arg_sp;
3178
3179 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT | not_sp << CRF_SO_BIT;
3180 env->fpscr &= ~FP_FPCC;
3181 env->fpscr |= cc << FPSCR_FPCC;
3182 env->crf[BF(opcode)] = cc;
3183 }
3184
3185 void helper_xsrqpi(CPUPPCState *env, uint32_t opcode,
3186 ppc_vsr_t *xt, ppc_vsr_t *xb)
3187 {
3188 ppc_vsr_t t = { };
3189 uint8_t r = Rrm(opcode);
3190 uint8_t ex = Rc(opcode);
3191 uint8_t rmc = RMC(opcode);
3192 uint8_t rmode = 0;
3193 float_status tstat;
3194
3195 helper_reset_fpstatus(env);
3196
3197 if (r == 0 && rmc == 0) {
3198 rmode = float_round_ties_away;
3199 } else if (r == 0 && rmc == 0x3) {
3200 rmode = fpscr_rn;
3201 } else if (r == 1) {
3202 switch (rmc) {
3203 case 0:
3204 rmode = float_round_nearest_even;
3205 break;
3206 case 1:
3207 rmode = float_round_to_zero;
3208 break;
3209 case 2:
3210 rmode = float_round_up;
3211 break;
3212 case 3:
3213 rmode = float_round_down;
3214 break;
3215 default:
3216 abort();
3217 }
3218 }
3219
3220 tstat = env->fp_status;
3221 set_float_exception_flags(0, &tstat);
3222 set_float_rounding_mode(rmode, &tstat);
3223 t.f128 = float128_round_to_int(xb->f128, &tstat);
3224 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3225
3226 if (unlikely(tstat.float_exception_flags & float_flag_invalid_snan)) {
3227 float_invalid_op_vxsnan(env, GETPC());
3228 }
3229
3230 if (ex == 0 && (tstat.float_exception_flags & float_flag_inexact)) {
3231 env->fp_status.float_exception_flags &= ~float_flag_inexact;
3232 }
3233
3234 helper_compute_fprf_float128(env, t.f128);
3235 do_float_check_status(env, GETPC());
3236 *xt = t;
3237 }
3238
3239 void helper_xsrqpxp(CPUPPCState *env, uint32_t opcode,
3240 ppc_vsr_t *xt, ppc_vsr_t *xb)
3241 {
3242 ppc_vsr_t t = { };
3243 uint8_t r = Rrm(opcode);
3244 uint8_t rmc = RMC(opcode);
3245 uint8_t rmode = 0;
3246 floatx80 round_res;
3247 float_status tstat;
3248
3249 helper_reset_fpstatus(env);
3250
3251 if (r == 0 && rmc == 0) {
3252 rmode = float_round_ties_away;
3253 } else if (r == 0 && rmc == 0x3) {
3254 rmode = fpscr_rn;
3255 } else if (r == 1) {
3256 switch (rmc) {
3257 case 0:
3258 rmode = float_round_nearest_even;
3259 break;
3260 case 1:
3261 rmode = float_round_to_zero;
3262 break;
3263 case 2:
3264 rmode = float_round_up;
3265 break;
3266 case 3:
3267 rmode = float_round_down;
3268 break;
3269 default:
3270 abort();
3271 }
3272 }
3273
3274 tstat = env->fp_status;
3275 set_float_exception_flags(0, &tstat);
3276 set_float_rounding_mode(rmode, &tstat);
3277 round_res = float128_to_floatx80(xb->f128, &tstat);
3278 t.f128 = floatx80_to_float128(round_res, &tstat);
3279 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3280
3281 if (unlikely(tstat.float_exception_flags & float_flag_invalid_snan)) {
3282 float_invalid_op_vxsnan(env, GETPC());
3283 t.f128 = float128_snan_to_qnan(t.f128);
3284 }
3285
3286 helper_compute_fprf_float128(env, t.f128);
3287 *xt = t;
3288 do_float_check_status(env, GETPC());
3289 }
3290
3291 void helper_xssqrtqp(CPUPPCState *env, uint32_t opcode,
3292 ppc_vsr_t *xt, ppc_vsr_t *xb)
3293 {
3294 ppc_vsr_t t = { };
3295 float_status tstat;
3296
3297 helper_reset_fpstatus(env);
3298
3299 tstat = env->fp_status;
3300 if (unlikely(Rc(opcode) != 0)) {
3301 tstat.float_rounding_mode = float_round_to_odd;
3302 }
3303
3304 set_float_exception_flags(0, &tstat);
3305 t.f128 = float128_sqrt(xb->f128, &tstat);
3306 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3307
3308 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3309 float_invalid_op_sqrt(env, tstat.float_exception_flags, 1, GETPC());
3310 }
3311
3312 helper_compute_fprf_float128(env, t.f128);
3313 *xt = t;
3314 do_float_check_status(env, GETPC());
3315 }
3316
3317 void helper_xssubqp(CPUPPCState *env, uint32_t opcode,
3318 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb)
3319 {
3320 ppc_vsr_t t = *xt;
3321 float_status tstat;
3322
3323 helper_reset_fpstatus(env);
3324
3325 tstat = env->fp_status;
3326 if (unlikely(Rc(opcode) != 0)) {
3327 tstat.float_rounding_mode = float_round_to_odd;
3328 }
3329
3330 set_float_exception_flags(0, &tstat);
3331 t.f128 = float128_sub(xa->f128, xb->f128, &tstat);
3332 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3333
3334 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3335 float_invalid_op_addsub(env, tstat.float_exception_flags, 1, GETPC());
3336 }
3337
3338 helper_compute_fprf_float128(env, t.f128);
3339 *xt = t;
3340 do_float_check_status(env, GETPC());
3341 }
QEmu