]> git.proxmox.com Git - mirror_qemu.git/blob - target/ppc/fpu_helper.c
projects / mirror_qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
target/ppc: overhauled and moved logic of storing fpscr
[mirror_qemu.git] / target / ppc / fpu_helper.c
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 static void do_float_check_status(CPUPPCState *env, uintptr_t raddr)
418 {
419 CPUState *cs = env_cpu(env);
420 int status = get_float_exception_flags(&env->fp_status);
421
422 if (status & float_flag_overflow) {
423 float_overflow_excp(env);
424 } else if (status & float_flag_underflow) {
425 float_underflow_excp(env);
426 }
427 if (status & float_flag_inexact) {
428 float_inexact_excp(env);
429 } else {
430 env->fpscr &= ~FP_FI; /* clear the FPSCR[FI] bit */
431 }
432
433 if (cs->exception_index == POWERPC_EXCP_PROGRAM &&
434 (env->error_code & POWERPC_EXCP_FP)) {
435 /* Deferred floating-point exception after target FPR update */
436 if (fp_exceptions_enabled(env)) {
437 raise_exception_err_ra(env, cs->exception_index,
438 env->error_code, raddr);
439 }
440 }
441 }
442
443 void helper_float_check_status(CPUPPCState *env)
444 {
445 do_float_check_status(env, GETPC());
446 }
447
448 void helper_reset_fpstatus(CPUPPCState *env)
449 {
450 set_float_exception_flags(0, &env->fp_status);
451 }
452
453 static void float_invalid_op_addsub(CPUPPCState *env, bool set_fpcc,
454 uintptr_t retaddr, int classes)
455 {
456 if ((classes & ~is_neg) == is_inf) {
457 /* Magnitude subtraction of infinities */
458 float_invalid_op_vxisi(env, set_fpcc, retaddr);
459 } else if (classes & is_snan) {
460 float_invalid_op_vxsnan(env, retaddr);
461 }
462 }
463
464 /* fadd - fadd. */
465 float64 helper_fadd(CPUPPCState *env, float64 arg1, float64 arg2)
466 {
467 float64 ret = float64_add(arg1, arg2, &env->fp_status);
468 int status = get_float_exception_flags(&env->fp_status);
469
470 if (unlikely(status & float_flag_invalid)) {
471 float_invalid_op_addsub(env, 1, GETPC(),
472 float64_classify(arg1) |
473 float64_classify(arg2));
474 }
475
476 return ret;
477 }
478
479 /* fsub - fsub. */
480 float64 helper_fsub(CPUPPCState *env, float64 arg1, float64 arg2)
481 {
482 float64 ret = float64_sub(arg1, arg2, &env->fp_status);
483 int status = get_float_exception_flags(&env->fp_status);
484
485 if (unlikely(status & float_flag_invalid)) {
486 float_invalid_op_addsub(env, 1, GETPC(),
487 float64_classify(arg1) |
488 float64_classify(arg2));
489 }
490
491 return ret;
492 }
493
494 static void float_invalid_op_mul(CPUPPCState *env, bool set_fprc,
495 uintptr_t retaddr, int classes)
496 {
497 if ((classes & (is_zero | is_inf)) == (is_zero | is_inf)) {
498 /* Multiplication of zero by infinity */
499 float_invalid_op_vximz(env, set_fprc, retaddr);
500 } else if (classes & is_snan) {
501 float_invalid_op_vxsnan(env, retaddr);
502 }
503 }
504
505 /* fmul - fmul. */
506 float64 helper_fmul(CPUPPCState *env, float64 arg1, float64 arg2)
507 {
508 float64 ret = float64_mul(arg1, arg2, &env->fp_status);
509 int status = get_float_exception_flags(&env->fp_status);
510
511 if (unlikely(status & float_flag_invalid)) {
512 float_invalid_op_mul(env, 1, GETPC(),
513 float64_classify(arg1) |
514 float64_classify(arg2));
515 }
516
517 return ret;
518 }
519
520 static void float_invalid_op_div(CPUPPCState *env, bool set_fprc,
521 uintptr_t retaddr, int classes)
522 {
523 classes &= ~is_neg;
524 if (classes == is_inf) {
525 /* Division of infinity by infinity */
526 float_invalid_op_vxidi(env, set_fprc, retaddr);
527 } else if (classes == is_zero) {
528 /* Division of zero by zero */
529 float_invalid_op_vxzdz(env, set_fprc, retaddr);
530 } else if (classes & is_snan) {
531 float_invalid_op_vxsnan(env, retaddr);
532 }
533 }
534
535 /* fdiv - fdiv. */
536 float64 helper_fdiv(CPUPPCState *env, float64 arg1, float64 arg2)
537 {
538 float64 ret = float64_div(arg1, arg2, &env->fp_status);
539 int status = get_float_exception_flags(&env->fp_status);
540
541 if (unlikely(status)) {
542 if (status & float_flag_invalid) {
543 float_invalid_op_div(env, 1, GETPC(),
544 float64_classify(arg1) |
545 float64_classify(arg2));
546 }
547 if (status & float_flag_divbyzero) {
548 float_zero_divide_excp(env, GETPC());
549 }
550 }
551
552 return ret;
553 }
554
555 static void float_invalid_cvt(CPUPPCState *env, bool set_fprc,
556 uintptr_t retaddr, int class1)
557 {
558 float_invalid_op_vxcvi(env, set_fprc, retaddr);
559 if (class1 & is_snan) {
560 float_invalid_op_vxsnan(env, retaddr);
561 }
562 }
563
564 #define FPU_FCTI(op, cvt, nanval) \
565 uint64_t helper_##op(CPUPPCState *env, float64 arg) \
566 { \
567 uint64_t ret = float64_to_##cvt(arg, &env->fp_status); \
568 int status = get_float_exception_flags(&env->fp_status); \
569 \
570 if (unlikely(status)) { \
571 if (status & float_flag_invalid) { \
572 float_invalid_cvt(env, 1, GETPC(), float64_classify(arg)); \
573 ret = nanval; \
574 } \
575 do_float_check_status(env, GETPC()); \
576 } \
577 return ret; \
578 }
579
580 FPU_FCTI(fctiw, int32, 0x80000000U)
581 FPU_FCTI(fctiwz, int32_round_to_zero, 0x80000000U)
582 FPU_FCTI(fctiwu, uint32, 0x00000000U)
583 FPU_FCTI(fctiwuz, uint32_round_to_zero, 0x00000000U)
584 FPU_FCTI(fctid, int64, 0x8000000000000000ULL)
585 FPU_FCTI(fctidz, int64_round_to_zero, 0x8000000000000000ULL)
586 FPU_FCTI(fctidu, uint64, 0x0000000000000000ULL)
587 FPU_FCTI(fctiduz, uint64_round_to_zero, 0x0000000000000000ULL)
588
589 #define FPU_FCFI(op, cvtr, is_single) \
590 uint64_t helper_##op(CPUPPCState *env, uint64_t arg) \
591 { \
592 CPU_DoubleU farg; \
593 \
594 if (is_single) { \
595 float32 tmp = cvtr(arg, &env->fp_status); \
596 farg.d = float32_to_float64(tmp, &env->fp_status); \
597 } else { \
598 farg.d = cvtr(arg, &env->fp_status); \
599 } \
600 do_float_check_status(env, GETPC()); \
601 return farg.ll; \
602 }
603
604 FPU_FCFI(fcfid, int64_to_float64, 0)
605 FPU_FCFI(fcfids, int64_to_float32, 1)
606 FPU_FCFI(fcfidu, uint64_to_float64, 0)
607 FPU_FCFI(fcfidus, uint64_to_float32, 1)
608
609 static inline uint64_t do_fri(CPUPPCState *env, uint64_t arg,
610 int rounding_mode)
611 {
612 CPU_DoubleU farg;
613 FloatRoundMode old_rounding_mode = get_float_rounding_mode(&env->fp_status);
614
615 farg.ll = arg;
616
617 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) {
618 /* sNaN round */
619 float_invalid_op_vxsnan(env, GETPC());
620 farg.ll = arg | 0x0008000000000000ULL;
621 } else {
622 int inexact = get_float_exception_flags(&env->fp_status) &
623 float_flag_inexact;
624 set_float_rounding_mode(rounding_mode, &env->fp_status);
625 farg.ll = float64_round_to_int(farg.d, &env->fp_status);
626 set_float_rounding_mode(old_rounding_mode, &env->fp_status);
627
628 /* fri* does not set FPSCR[XX] */
629 if (!inexact) {
630 env->fp_status.float_exception_flags &= ~float_flag_inexact;
631 }
632 }
633 do_float_check_status(env, GETPC());
634 return farg.ll;
635 }
636
637 uint64_t helper_frin(CPUPPCState *env, uint64_t arg)
638 {
639 return do_fri(env, arg, float_round_ties_away);
640 }
641
642 uint64_t helper_friz(CPUPPCState *env, uint64_t arg)
643 {
644 return do_fri(env, arg, float_round_to_zero);
645 }
646
647 uint64_t helper_frip(CPUPPCState *env, uint64_t arg)
648 {
649 return do_fri(env, arg, float_round_up);
650 }
651
652 uint64_t helper_frim(CPUPPCState *env, uint64_t arg)
653 {
654 return do_fri(env, arg, float_round_down);
655 }
656
657 #define FPU_MADDSUB_UPDATE(NAME, TP) \
658 static void NAME(CPUPPCState *env, TP arg1, TP arg2, TP arg3, \
659 unsigned int madd_flags, uintptr_t retaddr) \
660 { \
661 if (TP##_is_signaling_nan(arg1, &env->fp_status) || \
662 TP##_is_signaling_nan(arg2, &env->fp_status) || \
663 TP##_is_signaling_nan(arg3, &env->fp_status)) { \
664 /* sNaN operation */ \
665 float_invalid_op_vxsnan(env, retaddr); \
666 } \
667 if ((TP##_is_infinity(arg1) && TP##_is_zero(arg2)) || \
668 (TP##_is_zero(arg1) && TP##_is_infinity(arg2))) { \
669 /* Multiplication of zero by infinity */ \
670 float_invalid_op_vximz(env, 1, retaddr); \
671 } \
672 if ((TP##_is_infinity(arg1) || TP##_is_infinity(arg2)) && \
673 TP##_is_infinity(arg3)) { \
674 uint8_t aSign, bSign, cSign; \
675 \
676 aSign = TP##_is_neg(arg1); \
677 bSign = TP##_is_neg(arg2); \
678 cSign = TP##_is_neg(arg3); \
679 if (madd_flags & float_muladd_negate_c) { \
680 cSign ^= 1; \
681 } \
682 if (aSign ^ bSign ^ cSign) { \
683 float_invalid_op_vxisi(env, 1, retaddr); \
684 } \
685 } \
686 }
687 FPU_MADDSUB_UPDATE(float32_maddsub_update_excp, float32)
688 FPU_MADDSUB_UPDATE(float64_maddsub_update_excp, float64)
689
690 #define FPU_FMADD(op, madd_flags) \
691 uint64_t helper_##op(CPUPPCState *env, uint64_t arg1, \
692 uint64_t arg2, uint64_t arg3) \
693 { \
694 uint32_t flags; \
695 float64 ret = float64_muladd(arg1, arg2, arg3, madd_flags, \
696 &env->fp_status); \
697 flags = get_float_exception_flags(&env->fp_status); \
698 if (flags) { \
699 if (flags & float_flag_invalid) { \
700 float64_maddsub_update_excp(env, arg1, arg2, arg3, \
701 madd_flags, GETPC()); \
702 } \
703 do_float_check_status(env, GETPC()); \
704 } \
705 return ret; \
706 }
707
708 #define MADD_FLGS 0
709 #define MSUB_FLGS float_muladd_negate_c
710 #define NMADD_FLGS float_muladd_negate_result
711 #define NMSUB_FLGS (float_muladd_negate_c | float_muladd_negate_result)
712
713 FPU_FMADD(fmadd, MADD_FLGS)
714 FPU_FMADD(fnmadd, NMADD_FLGS)
715 FPU_FMADD(fmsub, MSUB_FLGS)
716 FPU_FMADD(fnmsub, NMSUB_FLGS)
717
718 /* frsp - frsp. */
719 uint64_t helper_frsp(CPUPPCState *env, uint64_t arg)
720 {
721 CPU_DoubleU farg;
722 float32 f32;
723
724 farg.ll = arg;
725
726 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) {
727 float_invalid_op_vxsnan(env, GETPC());
728 }
729 f32 = float64_to_float32(farg.d, &env->fp_status);
730 farg.d = float32_to_float64(f32, &env->fp_status);
731
732 return farg.ll;
733 }
734
735 /* fsqrt - fsqrt. */
736 float64 helper_fsqrt(CPUPPCState *env, float64 arg)
737 {
738 float64 ret = float64_sqrt(arg, &env->fp_status);
739 int status = get_float_exception_flags(&env->fp_status);
740
741 if (unlikely(status & float_flag_invalid)) {
742 if (unlikely(float64_is_any_nan(arg))) {
743 if (unlikely(float64_is_signaling_nan(arg, &env->fp_status))) {
744 /* sNaN square root */
745 float_invalid_op_vxsnan(env, GETPC());
746 }
747 } else {
748 /* Square root of a negative nonzero number */
749 float_invalid_op_vxsqrt(env, 1, GETPC());
750 }
751 }
752
753 return ret;
754 }
755
756 /* fre - fre. */
757 float64 helper_fre(CPUPPCState *env, float64 arg)
758 {
759 /* "Estimate" the reciprocal with actual division. */
760 float64 ret = float64_div(float64_one, arg, &env->fp_status);
761 int status = get_float_exception_flags(&env->fp_status);
762
763 if (unlikely(status)) {
764 if (status & float_flag_invalid) {
765 if (float64_is_signaling_nan(arg, &env->fp_status)) {
766 /* sNaN reciprocal */
767 float_invalid_op_vxsnan(env, GETPC());
768 }
769 }
770 if (status & float_flag_divbyzero) {
771 float_zero_divide_excp(env, GETPC());
772 /* For FPSCR.ZE == 0, the result is 1/2. */
773 ret = float64_set_sign(float64_half, float64_is_neg(arg));
774 }
775 }
776
777 return ret;
778 }
779
780 /* fres - fres. */
781 uint64_t helper_fres(CPUPPCState *env, uint64_t arg)
782 {
783 CPU_DoubleU farg;
784 float32 f32;
785
786 farg.ll = arg;
787
788 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) {
789 /* sNaN reciprocal */
790 float_invalid_op_vxsnan(env, GETPC());
791 }
792 farg.d = float64_div(float64_one, farg.d, &env->fp_status);
793 f32 = float64_to_float32(farg.d, &env->fp_status);
794 farg.d = float32_to_float64(f32, &env->fp_status);
795
796 return farg.ll;
797 }
798
799 /* frsqrte - frsqrte. */
800 float64 helper_frsqrte(CPUPPCState *env, float64 arg)
801 {
802 /* "Estimate" the reciprocal with actual division. */
803 float64 rets = float64_sqrt(arg, &env->fp_status);
804 float64 retd = float64_div(float64_one, rets, &env->fp_status);
805 int status = get_float_exception_flags(&env->fp_status);
806
807 if (unlikely(status)) {
808 if (status & float_flag_invalid) {
809 if (float64_is_signaling_nan(arg, &env->fp_status)) {
810 /* sNaN reciprocal */
811 float_invalid_op_vxsnan(env, GETPC());
812 } else {
813 /* Square root of a negative nonzero number */
814 float_invalid_op_vxsqrt(env, 1, GETPC());
815 }
816 }
817 if (status & float_flag_divbyzero) {
818 /* Reciprocal of (square root of) zero. */
819 float_zero_divide_excp(env, GETPC());
820 }
821 }
822
823 return retd;
824 }
825
826 /* fsel - fsel. */
827 uint64_t helper_fsel(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
828 uint64_t arg3)
829 {
830 CPU_DoubleU farg1;
831
832 farg1.ll = arg1;
833
834 if ((!float64_is_neg(farg1.d) || float64_is_zero(farg1.d)) &&
835 !float64_is_any_nan(farg1.d)) {
836 return arg2;
837 } else {
838 return arg3;
839 }
840 }
841
842 uint32_t helper_ftdiv(uint64_t fra, uint64_t frb)
843 {
844 int fe_flag = 0;
845 int fg_flag = 0;
846
847 if (unlikely(float64_is_infinity(fra) ||
848 float64_is_infinity(frb) ||
849 float64_is_zero(frb))) {
850 fe_flag = 1;
851 fg_flag = 1;
852 } else {
853 int e_a = ppc_float64_get_unbiased_exp(fra);
854 int e_b = ppc_float64_get_unbiased_exp(frb);
855
856 if (unlikely(float64_is_any_nan(fra) ||
857 float64_is_any_nan(frb))) {
858 fe_flag = 1;
859 } else if ((e_b <= -1022) || (e_b >= 1021)) {
860 fe_flag = 1;
861 } else if (!float64_is_zero(fra) &&
862 (((e_a - e_b) >= 1023) ||
863 ((e_a - e_b) <= -1021) ||
864 (e_a <= -970))) {
865 fe_flag = 1;
866 }
867
868 if (unlikely(float64_is_zero_or_denormal(frb))) {
869 /* XB is not zero because of the above check and */
870 /* so must be denormalized. */
871 fg_flag = 1;
872 }
873 }
874
875 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
876 }
877
878 uint32_t helper_ftsqrt(uint64_t frb)
879 {
880 int fe_flag = 0;
881 int fg_flag = 0;
882
883 if (unlikely(float64_is_infinity(frb) || float64_is_zero(frb))) {
884 fe_flag = 1;
885 fg_flag = 1;
886 } else {
887 int e_b = ppc_float64_get_unbiased_exp(frb);
888
889 if (unlikely(float64_is_any_nan(frb))) {
890 fe_flag = 1;
891 } else if (unlikely(float64_is_zero(frb))) {
892 fe_flag = 1;
893 } else if (unlikely(float64_is_neg(frb))) {
894 fe_flag = 1;
895 } else if (!float64_is_zero(frb) && (e_b <= (-1022 + 52))) {
896 fe_flag = 1;
897 }
898
899 if (unlikely(float64_is_zero_or_denormal(frb))) {
900 /* XB is not zero because of the above check and */
901 /* therefore must be denormalized. */
902 fg_flag = 1;
903 }
904 }
905
906 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
907 }
908
909 void helper_fcmpu(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
910 uint32_t crfD)
911 {
912 CPU_DoubleU farg1, farg2;
913 uint32_t ret = 0;
914
915 farg1.ll = arg1;
916 farg2.ll = arg2;
917
918 if (unlikely(float64_is_any_nan(farg1.d) ||
919 float64_is_any_nan(farg2.d))) {
920 ret = 0x01UL;
921 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) {
922 ret = 0x08UL;
923 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) {
924 ret = 0x04UL;
925 } else {
926 ret = 0x02UL;
927 }
928
929 env->fpscr &= ~FP_FPCC;
930 env->fpscr |= ret << FPSCR_FPCC;
931 env->crf[crfD] = ret;
932 if (unlikely(ret == 0x01UL
933 && (float64_is_signaling_nan(farg1.d, &env->fp_status) ||
934 float64_is_signaling_nan(farg2.d, &env->fp_status)))) {
935 /* sNaN comparison */
936 float_invalid_op_vxsnan(env, GETPC());
937 }
938 }
939
940 void helper_fcmpo(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
941 uint32_t crfD)
942 {
943 CPU_DoubleU farg1, farg2;
944 uint32_t ret = 0;
945
946 farg1.ll = arg1;
947 farg2.ll = arg2;
948
949 if (unlikely(float64_is_any_nan(farg1.d) ||
950 float64_is_any_nan(farg2.d))) {
951 ret = 0x01UL;
952 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) {
953 ret = 0x08UL;
954 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) {
955 ret = 0x04UL;
956 } else {
957 ret = 0x02UL;
958 }
959
960 env->fpscr &= ~FP_FPCC;
961 env->fpscr |= ret << FPSCR_FPCC;
962 env->crf[crfD] = (uint32_t) ret;
963 if (unlikely(ret == 0x01UL)) {
964 float_invalid_op_vxvc(env, 1, GETPC());
965 if (float64_is_signaling_nan(farg1.d, &env->fp_status) ||
966 float64_is_signaling_nan(farg2.d, &env->fp_status)) {
967 /* sNaN comparison */
968 float_invalid_op_vxsnan(env, GETPC());
969 }
970 }
971 }
972
973 /* Single-precision floating-point conversions */
974 static inline uint32_t efscfsi(CPUPPCState *env, uint32_t val)
975 {
976 CPU_FloatU u;
977
978 u.f = int32_to_float32(val, &env->vec_status);
979
980 return u.l;
981 }
982
983 static inline uint32_t efscfui(CPUPPCState *env, uint32_t val)
984 {
985 CPU_FloatU u;
986
987 u.f = uint32_to_float32(val, &env->vec_status);
988
989 return u.l;
990 }
991
992 static inline int32_t efsctsi(CPUPPCState *env, uint32_t val)
993 {
994 CPU_FloatU u;
995
996 u.l = val;
997 /* NaN are not treated the same way IEEE 754 does */
998 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
999 return 0;
1000 }
1001
1002 return float32_to_int32(u.f, &env->vec_status);
1003 }
1004
1005 static inline uint32_t efsctui(CPUPPCState *env, uint32_t val)
1006 {
1007 CPU_FloatU u;
1008
1009 u.l = val;
1010 /* NaN are not treated the same way IEEE 754 does */
1011 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1012 return 0;
1013 }
1014
1015 return float32_to_uint32(u.f, &env->vec_status);
1016 }
1017
1018 static inline uint32_t efsctsiz(CPUPPCState *env, uint32_t val)
1019 {
1020 CPU_FloatU u;
1021
1022 u.l = val;
1023 /* NaN are not treated the same way IEEE 754 does */
1024 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1025 return 0;
1026 }
1027
1028 return float32_to_int32_round_to_zero(u.f, &env->vec_status);
1029 }
1030
1031 static inline uint32_t efsctuiz(CPUPPCState *env, uint32_t val)
1032 {
1033 CPU_FloatU u;
1034
1035 u.l = val;
1036 /* NaN are not treated the same way IEEE 754 does */
1037 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1038 return 0;
1039 }
1040
1041 return float32_to_uint32_round_to_zero(u.f, &env->vec_status);
1042 }
1043
1044 static inline uint32_t efscfsf(CPUPPCState *env, uint32_t val)
1045 {
1046 CPU_FloatU u;
1047 float32 tmp;
1048
1049 u.f = int32_to_float32(val, &env->vec_status);
1050 tmp = int64_to_float32(1ULL << 32, &env->vec_status);
1051 u.f = float32_div(u.f, tmp, &env->vec_status);
1052
1053 return u.l;
1054 }
1055
1056 static inline uint32_t efscfuf(CPUPPCState *env, uint32_t val)
1057 {
1058 CPU_FloatU u;
1059 float32 tmp;
1060
1061 u.f = uint32_to_float32(val, &env->vec_status);
1062 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1063 u.f = float32_div(u.f, tmp, &env->vec_status);
1064
1065 return u.l;
1066 }
1067
1068 static inline uint32_t efsctsf(CPUPPCState *env, uint32_t val)
1069 {
1070 CPU_FloatU u;
1071 float32 tmp;
1072
1073 u.l = val;
1074 /* NaN are not treated the same way IEEE 754 does */
1075 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1076 return 0;
1077 }
1078 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1079 u.f = float32_mul(u.f, tmp, &env->vec_status);
1080
1081 return float32_to_int32(u.f, &env->vec_status);
1082 }
1083
1084 static inline uint32_t efsctuf(CPUPPCState *env, uint32_t val)
1085 {
1086 CPU_FloatU u;
1087 float32 tmp;
1088
1089 u.l = val;
1090 /* NaN are not treated the same way IEEE 754 does */
1091 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1092 return 0;
1093 }
1094 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1095 u.f = float32_mul(u.f, tmp, &env->vec_status);
1096
1097 return float32_to_uint32(u.f, &env->vec_status);
1098 }
1099
1100 #define HELPER_SPE_SINGLE_CONV(name) \
1101 uint32_t helper_e##name(CPUPPCState *env, uint32_t val) \
1102 { \
1103 return e##name(env, val); \
1104 }
1105 /* efscfsi */
1106 HELPER_SPE_SINGLE_CONV(fscfsi);
1107 /* efscfui */
1108 HELPER_SPE_SINGLE_CONV(fscfui);
1109 /* efscfuf */
1110 HELPER_SPE_SINGLE_CONV(fscfuf);
1111 /* efscfsf */
1112 HELPER_SPE_SINGLE_CONV(fscfsf);
1113 /* efsctsi */
1114 HELPER_SPE_SINGLE_CONV(fsctsi);
1115 /* efsctui */
1116 HELPER_SPE_SINGLE_CONV(fsctui);
1117 /* efsctsiz */
1118 HELPER_SPE_SINGLE_CONV(fsctsiz);
1119 /* efsctuiz */
1120 HELPER_SPE_SINGLE_CONV(fsctuiz);
1121 /* efsctsf */
1122 HELPER_SPE_SINGLE_CONV(fsctsf);
1123 /* efsctuf */
1124 HELPER_SPE_SINGLE_CONV(fsctuf);
1125
1126 #define HELPER_SPE_VECTOR_CONV(name) \
1127 uint64_t helper_ev##name(CPUPPCState *env, uint64_t val) \
1128 { \
1129 return ((uint64_t)e##name(env, val >> 32) << 32) | \
1130 (uint64_t)e##name(env, val); \
1131 }
1132 /* evfscfsi */
1133 HELPER_SPE_VECTOR_CONV(fscfsi);
1134 /* evfscfui */
1135 HELPER_SPE_VECTOR_CONV(fscfui);
1136 /* evfscfuf */
1137 HELPER_SPE_VECTOR_CONV(fscfuf);
1138 /* evfscfsf */
1139 HELPER_SPE_VECTOR_CONV(fscfsf);
1140 /* evfsctsi */
1141 HELPER_SPE_VECTOR_CONV(fsctsi);
1142 /* evfsctui */
1143 HELPER_SPE_VECTOR_CONV(fsctui);
1144 /* evfsctsiz */
1145 HELPER_SPE_VECTOR_CONV(fsctsiz);
1146 /* evfsctuiz */
1147 HELPER_SPE_VECTOR_CONV(fsctuiz);
1148 /* evfsctsf */
1149 HELPER_SPE_VECTOR_CONV(fsctsf);
1150 /* evfsctuf */
1151 HELPER_SPE_VECTOR_CONV(fsctuf);
1152
1153 /* Single-precision floating-point arithmetic */
1154 static inline uint32_t efsadd(CPUPPCState *env, uint32_t op1, uint32_t op2)
1155 {
1156 CPU_FloatU u1, u2;
1157
1158 u1.l = op1;
1159 u2.l = op2;
1160 u1.f = float32_add(u1.f, u2.f, &env->vec_status);
1161 return u1.l;
1162 }
1163
1164 static inline uint32_t efssub(CPUPPCState *env, uint32_t op1, uint32_t op2)
1165 {
1166 CPU_FloatU u1, u2;
1167
1168 u1.l = op1;
1169 u2.l = op2;
1170 u1.f = float32_sub(u1.f, u2.f, &env->vec_status);
1171 return u1.l;
1172 }
1173
1174 static inline uint32_t efsmul(CPUPPCState *env, uint32_t op1, uint32_t op2)
1175 {
1176 CPU_FloatU u1, u2;
1177
1178 u1.l = op1;
1179 u2.l = op2;
1180 u1.f = float32_mul(u1.f, u2.f, &env->vec_status);
1181 return u1.l;
1182 }
1183
1184 static inline uint32_t efsdiv(CPUPPCState *env, uint32_t op1, uint32_t op2)
1185 {
1186 CPU_FloatU u1, u2;
1187
1188 u1.l = op1;
1189 u2.l = op2;
1190 u1.f = float32_div(u1.f, u2.f, &env->vec_status);
1191 return u1.l;
1192 }
1193
1194 #define HELPER_SPE_SINGLE_ARITH(name) \
1195 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \
1196 { \
1197 return e##name(env, op1, op2); \
1198 }
1199 /* efsadd */
1200 HELPER_SPE_SINGLE_ARITH(fsadd);
1201 /* efssub */
1202 HELPER_SPE_SINGLE_ARITH(fssub);
1203 /* efsmul */
1204 HELPER_SPE_SINGLE_ARITH(fsmul);
1205 /* efsdiv */
1206 HELPER_SPE_SINGLE_ARITH(fsdiv);
1207
1208 #define HELPER_SPE_VECTOR_ARITH(name) \
1209 uint64_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \
1210 { \
1211 return ((uint64_t)e##name(env, op1 >> 32, op2 >> 32) << 32) | \
1212 (uint64_t)e##name(env, op1, op2); \
1213 }
1214 /* evfsadd */
1215 HELPER_SPE_VECTOR_ARITH(fsadd);
1216 /* evfssub */
1217 HELPER_SPE_VECTOR_ARITH(fssub);
1218 /* evfsmul */
1219 HELPER_SPE_VECTOR_ARITH(fsmul);
1220 /* evfsdiv */
1221 HELPER_SPE_VECTOR_ARITH(fsdiv);
1222
1223 /* Single-precision floating-point comparisons */
1224 static inline uint32_t efscmplt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1225 {
1226 CPU_FloatU u1, u2;
1227
1228 u1.l = op1;
1229 u2.l = op2;
1230 return float32_lt(u1.f, u2.f, &env->vec_status) ? 4 : 0;
1231 }
1232
1233 static inline uint32_t efscmpgt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1234 {
1235 CPU_FloatU u1, u2;
1236
1237 u1.l = op1;
1238 u2.l = op2;
1239 return float32_le(u1.f, u2.f, &env->vec_status) ? 0 : 4;
1240 }
1241
1242 static inline uint32_t efscmpeq(CPUPPCState *env, uint32_t op1, uint32_t op2)
1243 {
1244 CPU_FloatU u1, u2;
1245
1246 u1.l = op1;
1247 u2.l = op2;
1248 return float32_eq(u1.f, u2.f, &env->vec_status) ? 4 : 0;
1249 }
1250
1251 static inline uint32_t efststlt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1252 {
1253 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1254 return efscmplt(env, op1, op2);
1255 }
1256
1257 static inline uint32_t efststgt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1258 {
1259 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1260 return efscmpgt(env, op1, op2);
1261 }
1262
1263 static inline uint32_t efststeq(CPUPPCState *env, uint32_t op1, uint32_t op2)
1264 {
1265 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1266 return efscmpeq(env, op1, op2);
1267 }
1268
1269 #define HELPER_SINGLE_SPE_CMP(name) \
1270 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \
1271 { \
1272 return e##name(env, op1, op2); \
1273 }
1274 /* efststlt */
1275 HELPER_SINGLE_SPE_CMP(fststlt);
1276 /* efststgt */
1277 HELPER_SINGLE_SPE_CMP(fststgt);
1278 /* efststeq */
1279 HELPER_SINGLE_SPE_CMP(fststeq);
1280 /* efscmplt */
1281 HELPER_SINGLE_SPE_CMP(fscmplt);
1282 /* efscmpgt */
1283 HELPER_SINGLE_SPE_CMP(fscmpgt);
1284 /* efscmpeq */
1285 HELPER_SINGLE_SPE_CMP(fscmpeq);
1286
1287 static inline uint32_t evcmp_merge(int t0, int t1)
1288 {
1289 return (t0 << 3) | (t1 << 2) | ((t0 | t1) << 1) | (t0 & t1);
1290 }
1291
1292 #define HELPER_VECTOR_SPE_CMP(name) \
1293 uint32_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \
1294 { \
1295 return evcmp_merge(e##name(env, op1 >> 32, op2 >> 32), \
1296 e##name(env, op1, op2)); \
1297 }
1298 /* evfststlt */
1299 HELPER_VECTOR_SPE_CMP(fststlt);
1300 /* evfststgt */
1301 HELPER_VECTOR_SPE_CMP(fststgt);
1302 /* evfststeq */
1303 HELPER_VECTOR_SPE_CMP(fststeq);
1304 /* evfscmplt */
1305 HELPER_VECTOR_SPE_CMP(fscmplt);
1306 /* evfscmpgt */
1307 HELPER_VECTOR_SPE_CMP(fscmpgt);
1308 /* evfscmpeq */
1309 HELPER_VECTOR_SPE_CMP(fscmpeq);
1310
1311 /* Double-precision floating-point conversion */
1312 uint64_t helper_efdcfsi(CPUPPCState *env, uint32_t val)
1313 {
1314 CPU_DoubleU u;
1315
1316 u.d = int32_to_float64(val, &env->vec_status);
1317
1318 return u.ll;
1319 }
1320
1321 uint64_t helper_efdcfsid(CPUPPCState *env, uint64_t val)
1322 {
1323 CPU_DoubleU u;
1324
1325 u.d = int64_to_float64(val, &env->vec_status);
1326
1327 return u.ll;
1328 }
1329
1330 uint64_t helper_efdcfui(CPUPPCState *env, uint32_t val)
1331 {
1332 CPU_DoubleU u;
1333
1334 u.d = uint32_to_float64(val, &env->vec_status);
1335
1336 return u.ll;
1337 }
1338
1339 uint64_t helper_efdcfuid(CPUPPCState *env, uint64_t val)
1340 {
1341 CPU_DoubleU u;
1342
1343 u.d = uint64_to_float64(val, &env->vec_status);
1344
1345 return u.ll;
1346 }
1347
1348 uint32_t helper_efdctsi(CPUPPCState *env, uint64_t val)
1349 {
1350 CPU_DoubleU u;
1351
1352 u.ll = val;
1353 /* NaN are not treated the same way IEEE 754 does */
1354 if (unlikely(float64_is_any_nan(u.d))) {
1355 return 0;
1356 }
1357
1358 return float64_to_int32(u.d, &env->vec_status);
1359 }
1360
1361 uint32_t helper_efdctui(CPUPPCState *env, uint64_t val)
1362 {
1363 CPU_DoubleU u;
1364
1365 u.ll = val;
1366 /* NaN are not treated the same way IEEE 754 does */
1367 if (unlikely(float64_is_any_nan(u.d))) {
1368 return 0;
1369 }
1370
1371 return float64_to_uint32(u.d, &env->vec_status);
1372 }
1373
1374 uint32_t helper_efdctsiz(CPUPPCState *env, uint64_t val)
1375 {
1376 CPU_DoubleU u;
1377
1378 u.ll = val;
1379 /* NaN are not treated the same way IEEE 754 does */
1380 if (unlikely(float64_is_any_nan(u.d))) {
1381 return 0;
1382 }
1383
1384 return float64_to_int32_round_to_zero(u.d, &env->vec_status);
1385 }
1386
1387 uint64_t helper_efdctsidz(CPUPPCState *env, uint64_t val)
1388 {
1389 CPU_DoubleU u;
1390
1391 u.ll = val;
1392 /* NaN are not treated the same way IEEE 754 does */
1393 if (unlikely(float64_is_any_nan(u.d))) {
1394 return 0;
1395 }
1396
1397 return float64_to_int64_round_to_zero(u.d, &env->vec_status);
1398 }
1399
1400 uint32_t helper_efdctuiz(CPUPPCState *env, uint64_t val)
1401 {
1402 CPU_DoubleU u;
1403
1404 u.ll = val;
1405 /* NaN are not treated the same way IEEE 754 does */
1406 if (unlikely(float64_is_any_nan(u.d))) {
1407 return 0;
1408 }
1409
1410 return float64_to_uint32_round_to_zero(u.d, &env->vec_status);
1411 }
1412
1413 uint64_t helper_efdctuidz(CPUPPCState *env, uint64_t val)
1414 {
1415 CPU_DoubleU u;
1416
1417 u.ll = val;
1418 /* NaN are not treated the same way IEEE 754 does */
1419 if (unlikely(float64_is_any_nan(u.d))) {
1420 return 0;
1421 }
1422
1423 return float64_to_uint64_round_to_zero(u.d, &env->vec_status);
1424 }
1425
1426 uint64_t helper_efdcfsf(CPUPPCState *env, uint32_t val)
1427 {
1428 CPU_DoubleU u;
1429 float64 tmp;
1430
1431 u.d = int32_to_float64(val, &env->vec_status);
1432 tmp = int64_to_float64(1ULL << 32, &env->vec_status);
1433 u.d = float64_div(u.d, tmp, &env->vec_status);
1434
1435 return u.ll;
1436 }
1437
1438 uint64_t helper_efdcfuf(CPUPPCState *env, uint32_t val)
1439 {
1440 CPU_DoubleU u;
1441 float64 tmp;
1442
1443 u.d = uint32_to_float64(val, &env->vec_status);
1444 tmp = int64_to_float64(1ULL << 32, &env->vec_status);
1445 u.d = float64_div(u.d, tmp, &env->vec_status);
1446
1447 return u.ll;
1448 }
1449
1450 uint32_t helper_efdctsf(CPUPPCState *env, uint64_t val)
1451 {
1452 CPU_DoubleU u;
1453 float64 tmp;
1454
1455 u.ll = val;
1456 /* NaN are not treated the same way IEEE 754 does */
1457 if (unlikely(float64_is_any_nan(u.d))) {
1458 return 0;
1459 }
1460 tmp = uint64_to_float64(1ULL << 32, &env->vec_status);
1461 u.d = float64_mul(u.d, tmp, &env->vec_status);
1462
1463 return float64_to_int32(u.d, &env->vec_status);
1464 }
1465
1466 uint32_t helper_efdctuf(CPUPPCState *env, uint64_t val)
1467 {
1468 CPU_DoubleU u;
1469 float64 tmp;
1470
1471 u.ll = val;
1472 /* NaN are not treated the same way IEEE 754 does */
1473 if (unlikely(float64_is_any_nan(u.d))) {
1474 return 0;
1475 }
1476 tmp = uint64_to_float64(1ULL << 32, &env->vec_status);
1477 u.d = float64_mul(u.d, tmp, &env->vec_status);
1478
1479 return float64_to_uint32(u.d, &env->vec_status);
1480 }
1481
1482 uint32_t helper_efscfd(CPUPPCState *env, uint64_t val)
1483 {
1484 CPU_DoubleU u1;
1485 CPU_FloatU u2;
1486
1487 u1.ll = val;
1488 u2.f = float64_to_float32(u1.d, &env->vec_status);
1489
1490 return u2.l;
1491 }
1492
1493 uint64_t helper_efdcfs(CPUPPCState *env, uint32_t val)
1494 {
1495 CPU_DoubleU u2;
1496 CPU_FloatU u1;
1497
1498 u1.l = val;
1499 u2.d = float32_to_float64(u1.f, &env->vec_status);
1500
1501 return u2.ll;
1502 }
1503
1504 /* Double precision fixed-point arithmetic */
1505 uint64_t helper_efdadd(CPUPPCState *env, uint64_t op1, uint64_t op2)
1506 {
1507 CPU_DoubleU u1, u2;
1508
1509 u1.ll = op1;
1510 u2.ll = op2;
1511 u1.d = float64_add(u1.d, u2.d, &env->vec_status);
1512 return u1.ll;
1513 }
1514
1515 uint64_t helper_efdsub(CPUPPCState *env, uint64_t op1, uint64_t op2)
1516 {
1517 CPU_DoubleU u1, u2;
1518
1519 u1.ll = op1;
1520 u2.ll = op2;
1521 u1.d = float64_sub(u1.d, u2.d, &env->vec_status);
1522 return u1.ll;
1523 }
1524
1525 uint64_t helper_efdmul(CPUPPCState *env, uint64_t op1, uint64_t op2)
1526 {
1527 CPU_DoubleU u1, u2;
1528
1529 u1.ll = op1;
1530 u2.ll = op2;
1531 u1.d = float64_mul(u1.d, u2.d, &env->vec_status);
1532 return u1.ll;
1533 }
1534
1535 uint64_t helper_efddiv(CPUPPCState *env, uint64_t op1, uint64_t op2)
1536 {
1537 CPU_DoubleU u1, u2;
1538
1539 u1.ll = op1;
1540 u2.ll = op2;
1541 u1.d = float64_div(u1.d, u2.d, &env->vec_status);
1542 return u1.ll;
1543 }
1544
1545 /* Double precision floating point helpers */
1546 uint32_t helper_efdtstlt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1547 {
1548 CPU_DoubleU u1, u2;
1549
1550 u1.ll = op1;
1551 u2.ll = op2;
1552 return float64_lt(u1.d, u2.d, &env->vec_status) ? 4 : 0;
1553 }
1554
1555 uint32_t helper_efdtstgt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1556 {
1557 CPU_DoubleU u1, u2;
1558
1559 u1.ll = op1;
1560 u2.ll = op2;
1561 return float64_le(u1.d, u2.d, &env->vec_status) ? 0 : 4;
1562 }
1563
1564 uint32_t helper_efdtsteq(CPUPPCState *env, uint64_t op1, uint64_t op2)
1565 {
1566 CPU_DoubleU u1, u2;
1567
1568 u1.ll = op1;
1569 u2.ll = op2;
1570 return float64_eq_quiet(u1.d, u2.d, &env->vec_status) ? 4 : 0;
1571 }
1572
1573 uint32_t helper_efdcmplt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1574 {
1575 /* XXX: TODO: test special values (NaN, infinites, ...) */
1576 return helper_efdtstlt(env, op1, op2);
1577 }
1578
1579 uint32_t helper_efdcmpgt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1580 {
1581 /* XXX: TODO: test special values (NaN, infinites, ...) */
1582 return helper_efdtstgt(env, op1, op2);
1583 }
1584
1585 uint32_t helper_efdcmpeq(CPUPPCState *env, uint64_t op1, uint64_t op2)
1586 {
1587 /* XXX: TODO: test special values (NaN, infinites, ...) */
1588 return helper_efdtsteq(env, op1, op2);
1589 }
1590
1591 #define float64_to_float64(x, env) x
1592
1593
1594 /*
1595 * VSX_ADD_SUB - VSX floating point add/subtract
1596 * name - instruction mnemonic
1597 * op - operation (add or sub)
1598 * nels - number of elements (1, 2 or 4)
1599 * tp - type (float32 or float64)
1600 * fld - vsr_t field (VsrD(*) or VsrW(*))
1601 * sfprf - set FPRF
1602 */
1603 #define VSX_ADD_SUB(name, op, nels, tp, fld, sfprf, r2sp) \
1604 void helper_##name(CPUPPCState *env, ppc_vsr_t *xt, \
1605 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1606 { \
1607 ppc_vsr_t t = *xt; \
1608 int i; \
1609 \
1610 helper_reset_fpstatus(env); \
1611 \
1612 for (i = 0; i < nels; i++) { \
1613 float_status tstat = env->fp_status; \
1614 set_float_exception_flags(0, &tstat); \
1615 t.fld = tp##_##op(xa->fld, xb->fld, &tstat); \
1616 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1617 \
1618 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1619 float_invalid_op_addsub(env, sfprf, GETPC(), \
1620 tp##_classify(xa->fld) | \
1621 tp##_classify(xb->fld)); \
1622 } \
1623 \
1624 if (r2sp) { \
1625 t.fld = helper_frsp(env, t.fld); \
1626 } \
1627 \
1628 if (sfprf) { \
1629 helper_compute_fprf_float64(env, t.fld); \
1630 } \
1631 } \
1632 *xt = t; \
1633 do_float_check_status(env, GETPC()); \
1634 }
1635
1636 VSX_ADD_SUB(xsadddp, add, 1, float64, VsrD(0), 1, 0)
1637 VSX_ADD_SUB(xsaddsp, add, 1, float64, VsrD(0), 1, 1)
1638 VSX_ADD_SUB(xvadddp, add, 2, float64, VsrD(i), 0, 0)
1639 VSX_ADD_SUB(xvaddsp, add, 4, float32, VsrW(i), 0, 0)
1640 VSX_ADD_SUB(xssubdp, sub, 1, float64, VsrD(0), 1, 0)
1641 VSX_ADD_SUB(xssubsp, sub, 1, float64, VsrD(0), 1, 1)
1642 VSX_ADD_SUB(xvsubdp, sub, 2, float64, VsrD(i), 0, 0)
1643 VSX_ADD_SUB(xvsubsp, sub, 4, float32, VsrW(i), 0, 0)
1644
1645 void helper_xsaddqp(CPUPPCState *env, uint32_t opcode,
1646 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb)
1647 {
1648 ppc_vsr_t t = *xt;
1649 float_status tstat;
1650
1651 helper_reset_fpstatus(env);
1652
1653 tstat = env->fp_status;
1654 if (unlikely(Rc(opcode) != 0)) {
1655 tstat.float_rounding_mode = float_round_to_odd;
1656 }
1657
1658 set_float_exception_flags(0, &tstat);
1659 t.f128 = float128_add(xa->f128, xb->f128, &tstat);
1660 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1661
1662 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1663 float_invalid_op_addsub(env, 1, GETPC(),
1664 float128_classify(xa->f128) |
1665 float128_classify(xb->f128));
1666 }
1667
1668 helper_compute_fprf_float128(env, t.f128);
1669
1670 *xt = t;
1671 do_float_check_status(env, GETPC());
1672 }
1673
1674 /*
1675 * VSX_MUL - VSX floating point multiply
1676 * op - instruction mnemonic
1677 * nels - number of elements (1, 2 or 4)
1678 * tp - type (float32 or float64)
1679 * fld - vsr_t field (VsrD(*) or VsrW(*))
1680 * sfprf - set FPRF
1681 */
1682 #define VSX_MUL(op, nels, tp, fld, sfprf, r2sp) \
1683 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
1684 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1685 { \
1686 ppc_vsr_t t = *xt; \
1687 int i; \
1688 \
1689 helper_reset_fpstatus(env); \
1690 \
1691 for (i = 0; i < nels; i++) { \
1692 float_status tstat = env->fp_status; \
1693 set_float_exception_flags(0, &tstat); \
1694 t.fld = tp##_mul(xa->fld, xb->fld, &tstat); \
1695 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1696 \
1697 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1698 float_invalid_op_mul(env, sfprf, GETPC(), \
1699 tp##_classify(xa->fld) | \
1700 tp##_classify(xb->fld)); \
1701 } \
1702 \
1703 if (r2sp) { \
1704 t.fld = helper_frsp(env, t.fld); \
1705 } \
1706 \
1707 if (sfprf) { \
1708 helper_compute_fprf_float64(env, t.fld); \
1709 } \
1710 } \
1711 \
1712 *xt = t; \
1713 do_float_check_status(env, GETPC()); \
1714 }
1715
1716 VSX_MUL(xsmuldp, 1, float64, VsrD(0), 1, 0)
1717 VSX_MUL(xsmulsp, 1, float64, VsrD(0), 1, 1)
1718 VSX_MUL(xvmuldp, 2, float64, VsrD(i), 0, 0)
1719 VSX_MUL(xvmulsp, 4, float32, VsrW(i), 0, 0)
1720
1721 void helper_xsmulqp(CPUPPCState *env, uint32_t opcode,
1722 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb)
1723 {
1724 ppc_vsr_t t = *xt;
1725 float_status tstat;
1726
1727 helper_reset_fpstatus(env);
1728 tstat = env->fp_status;
1729 if (unlikely(Rc(opcode) != 0)) {
1730 tstat.float_rounding_mode = float_round_to_odd;
1731 }
1732
1733 set_float_exception_flags(0, &tstat);
1734 t.f128 = float128_mul(xa->f128, xb->f128, &tstat);
1735 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1736
1737 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1738 float_invalid_op_mul(env, 1, GETPC(),
1739 float128_classify(xa->f128) |
1740 float128_classify(xb->f128));
1741 }
1742 helper_compute_fprf_float128(env, t.f128);
1743
1744 *xt = t;
1745 do_float_check_status(env, GETPC());
1746 }
1747
1748 /*
1749 * VSX_DIV - VSX floating point divide
1750 * op - instruction mnemonic
1751 * nels - number of elements (1, 2 or 4)
1752 * tp - type (float32 or float64)
1753 * fld - vsr_t field (VsrD(*) or VsrW(*))
1754 * sfprf - set FPRF
1755 */
1756 #define VSX_DIV(op, nels, tp, fld, sfprf, r2sp) \
1757 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
1758 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1759 { \
1760 ppc_vsr_t t = *xt; \
1761 int i; \
1762 \
1763 helper_reset_fpstatus(env); \
1764 \
1765 for (i = 0; i < nels; i++) { \
1766 float_status tstat = env->fp_status; \
1767 set_float_exception_flags(0, &tstat); \
1768 t.fld = tp##_div(xa->fld, xb->fld, &tstat); \
1769 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1770 \
1771 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1772 float_invalid_op_div(env, sfprf, GETPC(), \
1773 tp##_classify(xa->fld) | \
1774 tp##_classify(xb->fld)); \
1775 } \
1776 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) { \
1777 float_zero_divide_excp(env, GETPC()); \
1778 } \
1779 \
1780 if (r2sp) { \
1781 t.fld = helper_frsp(env, t.fld); \
1782 } \
1783 \
1784 if (sfprf) { \
1785 helper_compute_fprf_float64(env, t.fld); \
1786 } \
1787 } \
1788 \
1789 *xt = t; \
1790 do_float_check_status(env, GETPC()); \
1791 }
1792
1793 VSX_DIV(xsdivdp, 1, float64, VsrD(0), 1, 0)
1794 VSX_DIV(xsdivsp, 1, float64, VsrD(0), 1, 1)
1795 VSX_DIV(xvdivdp, 2, float64, VsrD(i), 0, 0)
1796 VSX_DIV(xvdivsp, 4, float32, VsrW(i), 0, 0)
1797
1798 void helper_xsdivqp(CPUPPCState *env, uint32_t opcode,
1799 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb)
1800 {
1801 ppc_vsr_t t = *xt;
1802 float_status tstat;
1803
1804 helper_reset_fpstatus(env);
1805 tstat = env->fp_status;
1806 if (unlikely(Rc(opcode) != 0)) {
1807 tstat.float_rounding_mode = float_round_to_odd;
1808 }
1809
1810 set_float_exception_flags(0, &tstat);
1811 t.f128 = float128_div(xa->f128, xb->f128, &tstat);
1812 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1813
1814 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1815 float_invalid_op_div(env, 1, GETPC(),
1816 float128_classify(xa->f128) |
1817 float128_classify(xb->f128));
1818 }
1819 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) {
1820 float_zero_divide_excp(env, GETPC());
1821 }
1822
1823 helper_compute_fprf_float128(env, t.f128);
1824 *xt = t;
1825 do_float_check_status(env, GETPC());
1826 }
1827
1828 /*
1829 * VSX_RE - VSX floating point reciprocal estimate
1830 * op - instruction mnemonic
1831 * nels - number of elements (1, 2 or 4)
1832 * tp - type (float32 or float64)
1833 * fld - vsr_t field (VsrD(*) or VsrW(*))
1834 * sfprf - set FPRF
1835 */
1836 #define VSX_RE(op, nels, tp, fld, sfprf, r2sp) \
1837 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
1838 { \
1839 ppc_vsr_t t = *xt; \
1840 int i; \
1841 \
1842 helper_reset_fpstatus(env); \
1843 \
1844 for (i = 0; i < nels; i++) { \
1845 if (unlikely(tp##_is_signaling_nan(xb->fld, &env->fp_status))) { \
1846 float_invalid_op_vxsnan(env, GETPC()); \
1847 } \
1848 t.fld = tp##_div(tp##_one, xb->fld, &env->fp_status); \
1849 \
1850 if (r2sp) { \
1851 t.fld = helper_frsp(env, t.fld); \
1852 } \
1853 \
1854 if (sfprf) { \
1855 helper_compute_fprf_float64(env, t.fld); \
1856 } \
1857 } \
1858 \
1859 *xt = t; \
1860 do_float_check_status(env, GETPC()); \
1861 }
1862
1863 VSX_RE(xsredp, 1, float64, VsrD(0), 1, 0)
1864 VSX_RE(xsresp, 1, float64, VsrD(0), 1, 1)
1865 VSX_RE(xvredp, 2, float64, VsrD(i), 0, 0)
1866 VSX_RE(xvresp, 4, float32, VsrW(i), 0, 0)
1867
1868 /*
1869 * VSX_SQRT - VSX floating point square root
1870 * op - instruction mnemonic
1871 * nels - number of elements (1, 2 or 4)
1872 * tp - type (float32 or float64)
1873 * fld - vsr_t field (VsrD(*) or VsrW(*))
1874 * sfprf - set FPRF
1875 */
1876 #define VSX_SQRT(op, nels, tp, fld, sfprf, r2sp) \
1877 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
1878 { \
1879 ppc_vsr_t t = *xt; \
1880 int i; \
1881 \
1882 helper_reset_fpstatus(env); \
1883 \
1884 for (i = 0; i < nels; i++) { \
1885 float_status tstat = env->fp_status; \
1886 set_float_exception_flags(0, &tstat); \
1887 t.fld = tp##_sqrt(xb->fld, &tstat); \
1888 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1889 \
1890 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1891 if (tp##_is_neg(xb->fld) && !tp##_is_zero(xb->fld)) { \
1892 float_invalid_op_vxsqrt(env, sfprf, GETPC()); \
1893 } else if (tp##_is_signaling_nan(xb->fld, &tstat)) { \
1894 float_invalid_op_vxsnan(env, GETPC()); \
1895 } \
1896 } \
1897 \
1898 if (r2sp) { \
1899 t.fld = helper_frsp(env, t.fld); \
1900 } \
1901 \
1902 if (sfprf) { \
1903 helper_compute_fprf_float64(env, t.fld); \
1904 } \
1905 } \
1906 \
1907 *xt = t; \
1908 do_float_check_status(env, GETPC()); \
1909 }
1910
1911 VSX_SQRT(xssqrtdp, 1, float64, VsrD(0), 1, 0)
1912 VSX_SQRT(xssqrtsp, 1, float64, VsrD(0), 1, 1)
1913 VSX_SQRT(xvsqrtdp, 2, float64, VsrD(i), 0, 0)
1914 VSX_SQRT(xvsqrtsp, 4, float32, VsrW(i), 0, 0)
1915
1916 /*
1917 *VSX_RSQRTE - VSX floating point reciprocal square root estimate
1918 * op - instruction mnemonic
1919 * nels - number of elements (1, 2 or 4)
1920 * tp - type (float32 or float64)
1921 * fld - vsr_t field (VsrD(*) or VsrW(*))
1922 * sfprf - set FPRF
1923 */
1924 #define VSX_RSQRTE(op, nels, tp, fld, sfprf, r2sp) \
1925 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
1926 { \
1927 ppc_vsr_t t = *xt; \
1928 int i; \
1929 \
1930 helper_reset_fpstatus(env); \
1931 \
1932 for (i = 0; i < nels; i++) { \
1933 float_status tstat = env->fp_status; \
1934 set_float_exception_flags(0, &tstat); \
1935 t.fld = tp##_sqrt(xb->fld, &tstat); \
1936 t.fld = tp##_div(tp##_one, t.fld, &tstat); \
1937 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1938 \
1939 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1940 if (tp##_is_neg(xb->fld) && !tp##_is_zero(xb->fld)) { \
1941 float_invalid_op_vxsqrt(env, sfprf, GETPC()); \
1942 } else if (tp##_is_signaling_nan(xb->fld, &tstat)) { \
1943 float_invalid_op_vxsnan(env, GETPC()); \
1944 } \
1945 } \
1946 \
1947 if (r2sp) { \
1948 t.fld = helper_frsp(env, t.fld); \
1949 } \
1950 \
1951 if (sfprf) { \
1952 helper_compute_fprf_float64(env, t.fld); \
1953 } \
1954 } \
1955 \
1956 *xt = t; \
1957 do_float_check_status(env, GETPC()); \
1958 }
1959
1960 VSX_RSQRTE(xsrsqrtedp, 1, float64, VsrD(0), 1, 0)
1961 VSX_RSQRTE(xsrsqrtesp, 1, float64, VsrD(0), 1, 1)
1962 VSX_RSQRTE(xvrsqrtedp, 2, float64, VsrD(i), 0, 0)
1963 VSX_RSQRTE(xvrsqrtesp, 4, float32, VsrW(i), 0, 0)
1964
1965 /*
1966 * VSX_TDIV - VSX floating point test for divide
1967 * op - instruction mnemonic
1968 * nels - number of elements (1, 2 or 4)
1969 * tp - type (float32 or float64)
1970 * fld - vsr_t field (VsrD(*) or VsrW(*))
1971 * emin - minimum unbiased exponent
1972 * emax - maximum unbiased exponent
1973 * nbits - number of fraction bits
1974 */
1975 #define VSX_TDIV(op, nels, tp, fld, emin, emax, nbits) \
1976 void helper_##op(CPUPPCState *env, uint32_t opcode, \
1977 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1978 { \
1979 int i; \
1980 int fe_flag = 0; \
1981 int fg_flag = 0; \
1982 \
1983 for (i = 0; i < nels; i++) { \
1984 if (unlikely(tp##_is_infinity(xa->fld) || \
1985 tp##_is_infinity(xb->fld) || \
1986 tp##_is_zero(xb->fld))) { \
1987 fe_flag = 1; \
1988 fg_flag = 1; \
1989 } else { \
1990 int e_a = ppc_##tp##_get_unbiased_exp(xa->fld); \
1991 int e_b = ppc_##tp##_get_unbiased_exp(xb->fld); \
1992 \
1993 if (unlikely(tp##_is_any_nan(xa->fld) || \
1994 tp##_is_any_nan(xb->fld))) { \
1995 fe_flag = 1; \
1996 } else if ((e_b <= emin) || (e_b >= (emax - 2))) { \
1997 fe_flag = 1; \
1998 } else if (!tp##_is_zero(xa->fld) && \
1999 (((e_a - e_b) >= emax) || \
2000 ((e_a - e_b) <= (emin + 1)) || \
2001 (e_a <= (emin + nbits)))) { \
2002 fe_flag = 1; \
2003 } \
2004 \
2005 if (unlikely(tp##_is_zero_or_denormal(xb->fld))) { \
2006 /* \
2007 * XB is not zero because of the above check and so \
2008 * must be denormalized. \
2009 */ \
2010 fg_flag = 1; \
2011 } \
2012 } \
2013 } \
2014 \
2015 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
2016 }
2017
2018 VSX_TDIV(xstdivdp, 1, float64, VsrD(0), -1022, 1023, 52)
2019 VSX_TDIV(xvtdivdp, 2, float64, VsrD(i), -1022, 1023, 52)
2020 VSX_TDIV(xvtdivsp, 4, float32, VsrW(i), -126, 127, 23)
2021
2022 /*
2023 * VSX_TSQRT - VSX floating point test for square root
2024 * op - instruction mnemonic
2025 * nels - number of elements (1, 2 or 4)
2026 * tp - type (float32 or float64)
2027 * fld - vsr_t field (VsrD(*) or VsrW(*))
2028 * emin - minimum unbiased exponent
2029 * emax - maximum unbiased exponent
2030 * nbits - number of fraction bits
2031 */
2032 #define VSX_TSQRT(op, nels, tp, fld, emin, nbits) \
2033 void helper_##op(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xb) \
2034 { \
2035 int i; \
2036 int fe_flag = 0; \
2037 int fg_flag = 0; \
2038 \
2039 for (i = 0; i < nels; i++) { \
2040 if (unlikely(tp##_is_infinity(xb->fld) || \
2041 tp##_is_zero(xb->fld))) { \
2042 fe_flag = 1; \
2043 fg_flag = 1; \
2044 } else { \
2045 int e_b = ppc_##tp##_get_unbiased_exp(xb->fld); \
2046 \
2047 if (unlikely(tp##_is_any_nan(xb->fld))) { \
2048 fe_flag = 1; \
2049 } else if (unlikely(tp##_is_zero(xb->fld))) { \
2050 fe_flag = 1; \
2051 } else if (unlikely(tp##_is_neg(xb->fld))) { \
2052 fe_flag = 1; \
2053 } else if (!tp##_is_zero(xb->fld) && \
2054 (e_b <= (emin + nbits))) { \
2055 fe_flag = 1; \
2056 } \
2057 \
2058 if (unlikely(tp##_is_zero_or_denormal(xb->fld))) { \
2059 /* \
2060 * XB is not zero because of the above check and \
2061 * therefore must be denormalized. \
2062 */ \
2063 fg_flag = 1; \
2064 } \
2065 } \
2066 } \
2067 \
2068 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
2069 }
2070
2071 VSX_TSQRT(xstsqrtdp, 1, float64, VsrD(0), -1022, 52)
2072 VSX_TSQRT(xvtsqrtdp, 2, float64, VsrD(i), -1022, 52)
2073 VSX_TSQRT(xvtsqrtsp, 4, float32, VsrW(i), -126, 23)
2074
2075 /*
2076 * VSX_MADD - VSX floating point muliply/add variations
2077 * op - instruction mnemonic
2078 * nels - number of elements (1, 2 or 4)
2079 * tp - type (float32 or float64)
2080 * fld - vsr_t field (VsrD(*) or VsrW(*))
2081 * maddflgs - flags for the float*muladd routine that control the
2082 * various forms (madd, msub, nmadd, nmsub)
2083 * sfprf - set FPRF
2084 */
2085 #define VSX_MADD(op, nels, tp, fld, maddflgs, sfprf, r2sp) \
2086 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2087 ppc_vsr_t *xa, ppc_vsr_t *b, ppc_vsr_t *c) \
2088 { \
2089 ppc_vsr_t t = *xt; \
2090 int i; \
2091 \
2092 helper_reset_fpstatus(env); \
2093 \
2094 for (i = 0; i < nels; i++) { \
2095 float_status tstat = env->fp_status; \
2096 set_float_exception_flags(0, &tstat); \
2097 if (r2sp && (tstat.float_rounding_mode == float_round_nearest_even)) {\
2098 /* \
2099 * Avoid double rounding errors by rounding the intermediate \
2100 * result to odd. \
2101 */ \
2102 set_float_rounding_mode(float_round_to_zero, &tstat); \
2103 t.fld = tp##_muladd(xa->fld, b->fld, c->fld, \
2104 maddflgs, &tstat); \
2105 t.fld |= (get_float_exception_flags(&tstat) & \
2106 float_flag_inexact) != 0; \
2107 } else { \
2108 t.fld = tp##_muladd(xa->fld, b->fld, c->fld, \
2109 maddflgs, &tstat); \
2110 } \
2111 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
2112 \
2113 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
2114 tp##_maddsub_update_excp(env, xa->fld, b->fld, \
2115 c->fld, maddflgs, GETPC()); \
2116 } \
2117 \
2118 if (r2sp) { \
2119 t.fld = helper_frsp(env, t.fld); \
2120 } \
2121 \
2122 if (sfprf) { \
2123 helper_compute_fprf_float64(env, t.fld); \
2124 } \
2125 } \
2126 *xt = t; \
2127 do_float_check_status(env, GETPC()); \
2128 }
2129
2130 VSX_MADD(xsmadddp, 1, float64, VsrD(0), MADD_FLGS, 1, 0)
2131 VSX_MADD(xsmsubdp, 1, float64, VsrD(0), MSUB_FLGS, 1, 0)
2132 VSX_MADD(xsnmadddp, 1, float64, VsrD(0), NMADD_FLGS, 1, 0)
2133 VSX_MADD(xsnmsubdp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 0)
2134 VSX_MADD(xsmaddsp, 1, float64, VsrD(0), MADD_FLGS, 1, 1)
2135 VSX_MADD(xsmsubsp, 1, float64, VsrD(0), MSUB_FLGS, 1, 1)
2136 VSX_MADD(xsnmaddsp, 1, float64, VsrD(0), NMADD_FLGS, 1, 1)
2137 VSX_MADD(xsnmsubsp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 1)
2138
2139 VSX_MADD(xvmadddp, 2, float64, VsrD(i), MADD_FLGS, 0, 0)
2140 VSX_MADD(xvmsubdp, 2, float64, VsrD(i), MSUB_FLGS, 0, 0)
2141 VSX_MADD(xvnmadddp, 2, float64, VsrD(i), NMADD_FLGS, 0, 0)
2142 VSX_MADD(xvnmsubdp, 2, float64, VsrD(i), NMSUB_FLGS, 0, 0)
2143
2144 VSX_MADD(xvmaddsp, 4, float32, VsrW(i), MADD_FLGS, 0, 0)
2145 VSX_MADD(xvmsubsp, 4, float32, VsrW(i), MSUB_FLGS, 0, 0)
2146 VSX_MADD(xvnmaddsp, 4, float32, VsrW(i), NMADD_FLGS, 0, 0)
2147 VSX_MADD(xvnmsubsp, 4, float32, VsrW(i), NMSUB_FLGS, 0, 0)
2148
2149 /*
2150 * VSX_SCALAR_CMP_DP - VSX scalar floating point compare double precision
2151 * op - instruction mnemonic
2152 * cmp - comparison operation
2153 * exp - expected result of comparison
2154 * svxvc - set VXVC bit
2155 */
2156 #define VSX_SCALAR_CMP_DP(op, cmp, exp, svxvc) \
2157 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2158 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2159 { \
2160 ppc_vsr_t t = *xt; \
2161 bool vxsnan_flag = false, vxvc_flag = false, vex_flag = false; \
2162 \
2163 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \
2164 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2165 vxsnan_flag = true; \
2166 if (fpscr_ve == 0 && svxvc) { \
2167 vxvc_flag = true; \
2168 } \
2169 } else if (svxvc) { \
2170 vxvc_flag = float64_is_quiet_nan(xa->VsrD(0), &env->fp_status) || \
2171 float64_is_quiet_nan(xb->VsrD(0), &env->fp_status); \
2172 } \
2173 if (vxsnan_flag) { \
2174 float_invalid_op_vxsnan(env, GETPC()); \
2175 } \
2176 if (vxvc_flag) { \
2177 float_invalid_op_vxvc(env, 0, GETPC()); \
2178 } \
2179 vex_flag = fpscr_ve && (vxvc_flag || vxsnan_flag); \
2180 \
2181 if (!vex_flag) { \
2182 if (float64_##cmp(xb->VsrD(0), xa->VsrD(0), \
2183 &env->fp_status) == exp) { \
2184 t.VsrD(0) = -1; \
2185 t.VsrD(1) = 0; \
2186 } else { \
2187 t.VsrD(0) = 0; \
2188 t.VsrD(1) = 0; \
2189 } \
2190 } \
2191 *xt = t; \
2192 do_float_check_status(env, GETPC()); \
2193 }
2194
2195 VSX_SCALAR_CMP_DP(xscmpeqdp, eq, 1, 0)
2196 VSX_SCALAR_CMP_DP(xscmpgedp, le, 1, 1)
2197 VSX_SCALAR_CMP_DP(xscmpgtdp, lt, 1, 1)
2198 VSX_SCALAR_CMP_DP(xscmpnedp, eq, 0, 0)
2199
2200 void helper_xscmpexpdp(CPUPPCState *env, uint32_t opcode,
2201 ppc_vsr_t *xa, ppc_vsr_t *xb)
2202 {
2203 int64_t exp_a, exp_b;
2204 uint32_t cc;
2205
2206 exp_a = extract64(xa->VsrD(0), 52, 11);
2207 exp_b = extract64(xb->VsrD(0), 52, 11);
2208
2209 if (unlikely(float64_is_any_nan(xa->VsrD(0)) ||
2210 float64_is_any_nan(xb->VsrD(0)))) {
2211 cc = CRF_SO;
2212 } else {
2213 if (exp_a < exp_b) {
2214 cc = CRF_LT;
2215 } else if (exp_a > exp_b) {
2216 cc = CRF_GT;
2217 } else {
2218 cc = CRF_EQ;
2219 }
2220 }
2221
2222 env->fpscr &= ~FP_FPCC;
2223 env->fpscr |= cc << FPSCR_FPCC;
2224 env->crf[BF(opcode)] = cc;
2225
2226 do_float_check_status(env, GETPC());
2227 }
2228
2229 void helper_xscmpexpqp(CPUPPCState *env, uint32_t opcode,
2230 ppc_vsr_t *xa, ppc_vsr_t *xb)
2231 {
2232 int64_t exp_a, exp_b;
2233 uint32_t cc;
2234
2235 exp_a = extract64(xa->VsrD(0), 48, 15);
2236 exp_b = extract64(xb->VsrD(0), 48, 15);
2237
2238 if (unlikely(float128_is_any_nan(xa->f128) ||
2239 float128_is_any_nan(xb->f128))) {
2240 cc = CRF_SO;
2241 } else {
2242 if (exp_a < exp_b) {
2243 cc = CRF_LT;
2244 } else if (exp_a > exp_b) {
2245 cc = CRF_GT;
2246 } else {
2247 cc = CRF_EQ;
2248 }
2249 }
2250
2251 env->fpscr &= ~FP_FPCC;
2252 env->fpscr |= cc << FPSCR_FPCC;
2253 env->crf[BF(opcode)] = cc;
2254
2255 do_float_check_status(env, GETPC());
2256 }
2257
2258 static inline void do_scalar_cmp(CPUPPCState *env, ppc_vsr_t *xa, ppc_vsr_t *xb,
2259 int crf_idx, bool ordered)
2260 {
2261 uint32_t cc;
2262 bool vxsnan_flag = false, vxvc_flag = false;
2263
2264 helper_reset_fpstatus(env);
2265
2266 switch (float64_compare(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) {
2267 case float_relation_less:
2268 cc = CRF_LT;
2269 break;
2270 case float_relation_equal:
2271 cc = CRF_EQ;
2272 break;
2273 case float_relation_greater:
2274 cc = CRF_GT;
2275 break;
2276 case float_relation_unordered:
2277 cc = CRF_SO;
2278
2279 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) ||
2280 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) {
2281 vxsnan_flag = true;
2282 if (fpscr_ve == 0 && ordered) {
2283 vxvc_flag = true;
2284 }
2285 } else if (float64_is_quiet_nan(xa->VsrD(0), &env->fp_status) ||
2286 float64_is_quiet_nan(xb->VsrD(0), &env->fp_status)) {
2287 if (ordered) {
2288 vxvc_flag = true;
2289 }
2290 }
2291
2292 break;
2293 default:
2294 g_assert_not_reached();
2295 }
2296
2297 env->fpscr &= ~FP_FPCC;
2298 env->fpscr |= cc << FPSCR_FPCC;
2299 env->crf[crf_idx] = cc;
2300
2301 if (vxsnan_flag) {
2302 float_invalid_op_vxsnan(env, GETPC());
2303 }
2304 if (vxvc_flag) {
2305 float_invalid_op_vxvc(env, 0, GETPC());
2306 }
2307
2308 do_float_check_status(env, GETPC());
2309 }
2310
2311 void helper_xscmpodp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xa,
2312 ppc_vsr_t *xb)
2313 {
2314 do_scalar_cmp(env, xa, xb, BF(opcode), true);
2315 }
2316
2317 void helper_xscmpudp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xa,
2318 ppc_vsr_t *xb)
2319 {
2320 do_scalar_cmp(env, xa, xb, BF(opcode), false);
2321 }
2322
2323 static inline void do_scalar_cmpq(CPUPPCState *env, ppc_vsr_t *xa,
2324 ppc_vsr_t *xb, int crf_idx, bool ordered)
2325 {
2326 uint32_t cc;
2327 bool vxsnan_flag = false, vxvc_flag = false;
2328
2329 helper_reset_fpstatus(env);
2330
2331 switch (float128_compare(xa->f128, xb->f128, &env->fp_status)) {
2332 case float_relation_less:
2333 cc = CRF_LT;
2334 break;
2335 case float_relation_equal:
2336 cc = CRF_EQ;
2337 break;
2338 case float_relation_greater:
2339 cc = CRF_GT;
2340 break;
2341 case float_relation_unordered:
2342 cc = CRF_SO;
2343
2344 if (float128_is_signaling_nan(xa->f128, &env->fp_status) ||
2345 float128_is_signaling_nan(xb->f128, &env->fp_status)) {
2346 vxsnan_flag = true;
2347 if (fpscr_ve == 0 && ordered) {
2348 vxvc_flag = true;
2349 }
2350 } else if (float128_is_quiet_nan(xa->f128, &env->fp_status) ||
2351 float128_is_quiet_nan(xb->f128, &env->fp_status)) {
2352 if (ordered) {
2353 vxvc_flag = true;
2354 }
2355 }
2356
2357 break;
2358 default:
2359 g_assert_not_reached();
2360 }
2361
2362 env->fpscr &= ~FP_FPCC;
2363 env->fpscr |= cc << FPSCR_FPCC;
2364 env->crf[crf_idx] = cc;
2365
2366 if (vxsnan_flag) {
2367 float_invalid_op_vxsnan(env, GETPC());
2368 }
2369 if (vxvc_flag) {
2370 float_invalid_op_vxvc(env, 0, GETPC());
2371 }
2372
2373 do_float_check_status(env, GETPC());
2374 }
2375
2376 void helper_xscmpoqp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xa,
2377 ppc_vsr_t *xb)
2378 {
2379 do_scalar_cmpq(env, xa, xb, BF(opcode), true);
2380 }
2381
2382 void helper_xscmpuqp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xa,
2383 ppc_vsr_t *xb)
2384 {
2385 do_scalar_cmpq(env, xa, xb, BF(opcode), false);
2386 }
2387
2388 /*
2389 * VSX_MAX_MIN - VSX floating point maximum/minimum
2390 * name - instruction mnemonic
2391 * op - operation (max or min)
2392 * nels - number of elements (1, 2 or 4)
2393 * tp - type (float32 or float64)
2394 * fld - vsr_t field (VsrD(*) or VsrW(*))
2395 */
2396 #define VSX_MAX_MIN(name, op, nels, tp, fld) \
2397 void helper_##name(CPUPPCState *env, ppc_vsr_t *xt, \
2398 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2399 { \
2400 ppc_vsr_t t = *xt; \
2401 int i; \
2402 \
2403 for (i = 0; i < nels; i++) { \
2404 t.fld = tp##_##op(xa->fld, xb->fld, &env->fp_status); \
2405 if (unlikely(tp##_is_signaling_nan(xa->fld, &env->fp_status) || \
2406 tp##_is_signaling_nan(xb->fld, &env->fp_status))) { \
2407 float_invalid_op_vxsnan(env, GETPC()); \
2408 } \
2409 } \
2410 \
2411 *xt = t; \
2412 do_float_check_status(env, GETPC()); \
2413 }
2414
2415 VSX_MAX_MIN(xsmaxdp, maxnum, 1, float64, VsrD(0))
2416 VSX_MAX_MIN(xvmaxdp, maxnum, 2, float64, VsrD(i))
2417 VSX_MAX_MIN(xvmaxsp, maxnum, 4, float32, VsrW(i))
2418 VSX_MAX_MIN(xsmindp, minnum, 1, float64, VsrD(0))
2419 VSX_MAX_MIN(xvmindp, minnum, 2, float64, VsrD(i))
2420 VSX_MAX_MIN(xvminsp, minnum, 4, float32, VsrW(i))
2421
2422 #define VSX_MAX_MINC(name, max) \
2423 void helper_##name(CPUPPCState *env, uint32_t opcode, \
2424 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) \
2425 { \
2426 ppc_vsr_t t = *xt; \
2427 bool vxsnan_flag = false, vex_flag = false; \
2428 \
2429 if (unlikely(float64_is_any_nan(xa->VsrD(0)) || \
2430 float64_is_any_nan(xb->VsrD(0)))) { \
2431 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \
2432 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2433 vxsnan_flag = true; \
2434 } \
2435 t.VsrD(0) = xb->VsrD(0); \
2436 } else if ((max && \
2437 !float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) || \
2438 (!max && \
2439 float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status))) { \
2440 t.VsrD(0) = xa->VsrD(0); \
2441 } else { \
2442 t.VsrD(0) = xb->VsrD(0); \
2443 } \
2444 \
2445 vex_flag = fpscr_ve & vxsnan_flag; \
2446 if (vxsnan_flag) { \
2447 float_invalid_op_vxsnan(env, GETPC()); \
2448 } \
2449 if (!vex_flag) { \
2450 *xt = t; \
2451 } \
2452 } \
2453
2454 VSX_MAX_MINC(xsmaxcdp, 1);
2455 VSX_MAX_MINC(xsmincdp, 0);
2456
2457 #define VSX_MAX_MINJ(name, max) \
2458 void helper_##name(CPUPPCState *env, uint32_t opcode, \
2459 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb) \
2460 { \
2461 ppc_vsr_t t = *xt; \
2462 bool vxsnan_flag = false, vex_flag = false; \
2463 \
2464 if (unlikely(float64_is_any_nan(xa->VsrD(0)))) { \
2465 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status)) { \
2466 vxsnan_flag = true; \
2467 } \
2468 t.VsrD(0) = xa->VsrD(0); \
2469 } else if (unlikely(float64_is_any_nan(xb->VsrD(0)))) { \
2470 if (float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2471 vxsnan_flag = true; \
2472 } \
2473 t.VsrD(0) = xb->VsrD(0); \
2474 } else if (float64_is_zero(xa->VsrD(0)) && \
2475 float64_is_zero(xb->VsrD(0))) { \
2476 if (max) { \
2477 if (!float64_is_neg(xa->VsrD(0)) || \
2478 !float64_is_neg(xb->VsrD(0))) { \
2479 t.VsrD(0) = 0ULL; \
2480 } else { \
2481 t.VsrD(0) = 0x8000000000000000ULL; \
2482 } \
2483 } else { \
2484 if (float64_is_neg(xa->VsrD(0)) || \
2485 float64_is_neg(xb->VsrD(0))) { \
2486 t.VsrD(0) = 0x8000000000000000ULL; \
2487 } else { \
2488 t.VsrD(0) = 0ULL; \
2489 } \
2490 } \
2491 } else if ((max && \
2492 !float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) || \
2493 (!max && \
2494 float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status))) { \
2495 t.VsrD(0) = xa->VsrD(0); \
2496 } else { \
2497 t.VsrD(0) = xb->VsrD(0); \
2498 } \
2499 \
2500 vex_flag = fpscr_ve & vxsnan_flag; \
2501 if (vxsnan_flag) { \
2502 float_invalid_op_vxsnan(env, GETPC()); \
2503 } \
2504 if (!vex_flag) { \
2505 *xt = t; \
2506 } \
2507 } \
2508
2509 VSX_MAX_MINJ(xsmaxjdp, 1);
2510 VSX_MAX_MINJ(xsminjdp, 0);
2511
2512 /*
2513 * VSX_CMP - VSX floating point compare
2514 * op - instruction mnemonic
2515 * nels - number of elements (1, 2 or 4)
2516 * tp - type (float32 or float64)
2517 * fld - vsr_t field (VsrD(*) or VsrW(*))
2518 * cmp - comparison operation
2519 * svxvc - set VXVC bit
2520 * exp - expected result of comparison
2521 */
2522 #define VSX_CMP(op, nels, tp, fld, cmp, svxvc, exp) \
2523 uint32_t helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2524 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2525 { \
2526 ppc_vsr_t t = *xt; \
2527 uint32_t crf6 = 0; \
2528 int i; \
2529 int all_true = 1; \
2530 int all_false = 1; \
2531 \
2532 for (i = 0; i < nels; i++) { \
2533 if (unlikely(tp##_is_any_nan(xa->fld) || \
2534 tp##_is_any_nan(xb->fld))) { \
2535 if (tp##_is_signaling_nan(xa->fld, &env->fp_status) || \
2536 tp##_is_signaling_nan(xb->fld, &env->fp_status)) { \
2537 float_invalid_op_vxsnan(env, GETPC()); \
2538 } \
2539 if (svxvc) { \
2540 float_invalid_op_vxvc(env, 0, GETPC()); \
2541 } \
2542 t.fld = 0; \
2543 all_true = 0; \
2544 } else { \
2545 if (tp##_##cmp(xb->fld, xa->fld, &env->fp_status) == exp) { \
2546 t.fld = -1; \
2547 all_false = 0; \
2548 } else { \
2549 t.fld = 0; \
2550 all_true = 0; \
2551 } \
2552 } \
2553 } \
2554 \
2555 *xt = t; \
2556 crf6 = (all_true ? 0x8 : 0) | (all_false ? 0x2 : 0); \
2557 return crf6; \
2558 }
2559
2560 VSX_CMP(xvcmpeqdp, 2, float64, VsrD(i), eq, 0, 1)
2561 VSX_CMP(xvcmpgedp, 2, float64, VsrD(i), le, 1, 1)
2562 VSX_CMP(xvcmpgtdp, 2, float64, VsrD(i), lt, 1, 1)
2563 VSX_CMP(xvcmpnedp, 2, float64, VsrD(i), eq, 0, 0)
2564 VSX_CMP(xvcmpeqsp, 4, float32, VsrW(i), eq, 0, 1)
2565 VSX_CMP(xvcmpgesp, 4, float32, VsrW(i), le, 1, 1)
2566 VSX_CMP(xvcmpgtsp, 4, float32, VsrW(i), lt, 1, 1)
2567 VSX_CMP(xvcmpnesp, 4, float32, VsrW(i), eq, 0, 0)
2568
2569 /*
2570 * VSX_CVT_FP_TO_FP - VSX floating point/floating point conversion
2571 * op - instruction mnemonic
2572 * nels - number of elements (1, 2 or 4)
2573 * stp - source type (float32 or float64)
2574 * ttp - target type (float32 or float64)
2575 * sfld - source vsr_t field
2576 * tfld - target vsr_t field (f32 or f64)
2577 * sfprf - set FPRF
2578 */
2579 #define VSX_CVT_FP_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf) \
2580 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2581 { \
2582 ppc_vsr_t t = *xt; \
2583 int i; \
2584 \
2585 for (i = 0; i < nels; i++) { \
2586 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2587 if (unlikely(stp##_is_signaling_nan(xb->sfld, \
2588 &env->fp_status))) { \
2589 float_invalid_op_vxsnan(env, GETPC()); \
2590 t.tfld = ttp##_snan_to_qnan(t.tfld); \
2591 } \
2592 if (sfprf) { \
2593 helper_compute_fprf_##ttp(env, t.tfld); \
2594 } \
2595 } \
2596 \
2597 *xt = t; \
2598 do_float_check_status(env, GETPC()); \
2599 }
2600
2601 VSX_CVT_FP_TO_FP(xscvdpsp, 1, float64, float32, VsrD(0), VsrW(0), 1)
2602 VSX_CVT_FP_TO_FP(xscvspdp, 1, float32, float64, VsrW(0), VsrD(0), 1)
2603 VSX_CVT_FP_TO_FP(xvcvdpsp, 2, float64, float32, VsrD(i), VsrW(2 * i), 0)
2604 VSX_CVT_FP_TO_FP(xvcvspdp, 2, float32, float64, VsrW(2 * i), VsrD(i), 0)
2605
2606 /*
2607 * VSX_CVT_FP_TO_FP_VECTOR - VSX floating point/floating point conversion
2608 * op - instruction mnemonic
2609 * nels - number of elements (1, 2 or 4)
2610 * stp - source type (float32 or float64)
2611 * ttp - target type (float32 or float64)
2612 * sfld - source vsr_t field
2613 * tfld - target vsr_t field (f32 or f64)
2614 * sfprf - set FPRF
2615 */
2616 #define VSX_CVT_FP_TO_FP_VECTOR(op, nels, stp, ttp, sfld, tfld, sfprf) \
2617 void helper_##op(CPUPPCState *env, uint32_t opcode, \
2618 ppc_vsr_t *xt, ppc_vsr_t *xb) \
2619 { \
2620 ppc_vsr_t t = *xt; \
2621 int i; \
2622 \
2623 for (i = 0; i < nels; i++) { \
2624 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2625 if (unlikely(stp##_is_signaling_nan(xb->sfld, \
2626 &env->fp_status))) { \
2627 float_invalid_op_vxsnan(env, GETPC()); \
2628 t.tfld = ttp##_snan_to_qnan(t.tfld); \
2629 } \
2630 if (sfprf) { \
2631 helper_compute_fprf_##ttp(env, t.tfld); \
2632 } \
2633 } \
2634 \
2635 *xt = t; \
2636 do_float_check_status(env, GETPC()); \
2637 }
2638
2639 VSX_CVT_FP_TO_FP_VECTOR(xscvdpqp, 1, float64, float128, VsrD(0), f128, 1)
2640
2641 /*
2642 * VSX_CVT_FP_TO_FP_HP - VSX floating point/floating point conversion
2643 * involving one half precision value
2644 * op - instruction mnemonic
2645 * nels - number of elements (1, 2 or 4)
2646 * stp - source type
2647 * ttp - target type
2648 * sfld - source vsr_t field
2649 * tfld - target vsr_t field
2650 * sfprf - set FPRF
2651 */
2652 #define VSX_CVT_FP_TO_FP_HP(op, nels, stp, ttp, sfld, tfld, sfprf) \
2653 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2654 { \
2655 ppc_vsr_t t = { }; \
2656 int i; \
2657 \
2658 for (i = 0; i < nels; i++) { \
2659 t.tfld = stp##_to_##ttp(xb->sfld, 1, &env->fp_status); \
2660 if (unlikely(stp##_is_signaling_nan(xb->sfld, \
2661 &env->fp_status))) { \
2662 float_invalid_op_vxsnan(env, GETPC()); \
2663 t.tfld = ttp##_snan_to_qnan(t.tfld); \
2664 } \
2665 if (sfprf) { \
2666 helper_compute_fprf_##ttp(env, t.tfld); \
2667 } \
2668 } \
2669 \
2670 *xt = t; \
2671 do_float_check_status(env, GETPC()); \
2672 }
2673
2674 VSX_CVT_FP_TO_FP_HP(xscvdphp, 1, float64, float16, VsrD(0), VsrH(3), 1)
2675 VSX_CVT_FP_TO_FP_HP(xscvhpdp, 1, float16, float64, VsrH(3), VsrD(0), 1)
2676 VSX_CVT_FP_TO_FP_HP(xvcvsphp, 4, float32, float16, VsrW(i), VsrH(2 * i + 1), 0)
2677 VSX_CVT_FP_TO_FP_HP(xvcvhpsp, 4, float16, float32, VsrH(2 * i + 1), VsrW(i), 0)
2678
2679 /*
2680 * xscvqpdp isn't using VSX_CVT_FP_TO_FP() because xscvqpdpo will be
2681 * added to this later.
2682 */
2683 void helper_xscvqpdp(CPUPPCState *env, uint32_t opcode,
2684 ppc_vsr_t *xt, ppc_vsr_t *xb)
2685 {
2686 ppc_vsr_t t = { };
2687 float_status tstat;
2688
2689 tstat = env->fp_status;
2690 if (unlikely(Rc(opcode) != 0)) {
2691 tstat.float_rounding_mode = float_round_to_odd;
2692 }
2693
2694 t.VsrD(0) = float128_to_float64(xb->f128, &tstat);
2695 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
2696 if (unlikely(float128_is_signaling_nan(xb->f128, &tstat))) {
2697 float_invalid_op_vxsnan(env, GETPC());
2698 t.VsrD(0) = float64_snan_to_qnan(t.VsrD(0));
2699 }
2700 helper_compute_fprf_float64(env, t.VsrD(0));
2701
2702 *xt = t;
2703 do_float_check_status(env, GETPC());
2704 }
2705
2706 uint64_t helper_xscvdpspn(CPUPPCState *env, uint64_t xb)
2707 {
2708 uint64_t result, sign, exp, frac;
2709
2710 float_status tstat = env->fp_status;
2711 set_float_exception_flags(0, &tstat);
2712
2713 sign = extract64(xb, 63, 1);
2714 exp = extract64(xb, 52, 11);
2715 frac = extract64(xb, 0, 52) | 0x10000000000000ULL;
2716
2717 if (unlikely(exp == 0 && extract64(frac, 0, 52) != 0)) {
2718 /* DP denormal operand. */
2719 /* Exponent override to DP min exp. */
2720 exp = 1;
2721 /* Implicit bit override to 0. */
2722 frac = deposit64(frac, 53, 1, 0);
2723 }
2724
2725 if (unlikely(exp < 897 && frac != 0)) {
2726 /* SP tiny operand. */
2727 if (897 - exp > 63) {
2728 frac = 0;
2729 } else {
2730 /* Denormalize until exp = SP min exp. */
2731 frac >>= (897 - exp);
2732 }
2733 /* Exponent override to SP min exp - 1. */
2734 exp = 896;
2735 }
2736
2737 result = sign << 31;
2738 result |= extract64(exp, 10, 1) << 30;
2739 result |= extract64(exp, 0, 7) << 23;
2740 result |= extract64(frac, 29, 23);
2741
2742 /* hardware replicates result to both words of the doubleword result. */
2743 return (result << 32) | result;
2744 }
2745
2746 uint64_t helper_xscvspdpn(CPUPPCState *env, uint64_t xb)
2747 {
2748 float_status tstat = env->fp_status;
2749 set_float_exception_flags(0, &tstat);
2750
2751 return float32_to_float64(xb >> 32, &tstat);
2752 }
2753
2754 /*
2755 * VSX_CVT_FP_TO_INT - VSX floating point to integer conversion
2756 * op - instruction mnemonic
2757 * nels - number of elements (1, 2 or 4)
2758 * stp - source type (float32 or float64)
2759 * ttp - target type (int32, uint32, int64 or uint64)
2760 * sfld - source vsr_t field
2761 * tfld - target vsr_t field
2762 * rnan - resulting NaN
2763 */
2764 #define VSX_CVT_FP_TO_INT(op, nels, stp, ttp, sfld, tfld, rnan) \
2765 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2766 { \
2767 int all_flags = env->fp_status.float_exception_flags, flags; \
2768 ppc_vsr_t t = *xt; \
2769 int i; \
2770 \
2771 for (i = 0; i < nels; i++) { \
2772 env->fp_status.float_exception_flags = 0; \
2773 t.tfld = stp##_to_##ttp##_round_to_zero(xb->sfld, &env->fp_status); \
2774 flags = env->fp_status.float_exception_flags; \
2775 if (unlikely(flags & float_flag_invalid)) { \
2776 float_invalid_cvt(env, 0, GETPC(), stp##_classify(xb->sfld)); \
2777 t.tfld = rnan; \
2778 } \
2779 all_flags |= flags; \
2780 } \
2781 \
2782 *xt = t; \
2783 env->fp_status.float_exception_flags = all_flags; \
2784 do_float_check_status(env, GETPC()); \
2785 }
2786
2787 VSX_CVT_FP_TO_INT(xscvdpsxds, 1, float64, int64, VsrD(0), VsrD(0), \
2788 0x8000000000000000ULL)
2789 VSX_CVT_FP_TO_INT(xscvdpsxws, 1, float64, int32, VsrD(0), VsrW(1), \
2790 0x80000000U)
2791 VSX_CVT_FP_TO_INT(xscvdpuxds, 1, float64, uint64, VsrD(0), VsrD(0), 0ULL)
2792 VSX_CVT_FP_TO_INT(xscvdpuxws, 1, float64, uint32, VsrD(0), VsrW(1), 0U)
2793 VSX_CVT_FP_TO_INT(xvcvdpsxds, 2, float64, int64, VsrD(i), VsrD(i), \
2794 0x8000000000000000ULL)
2795 VSX_CVT_FP_TO_INT(xvcvdpsxws, 2, float64, int32, VsrD(i), VsrW(2 * i), \
2796 0x80000000U)
2797 VSX_CVT_FP_TO_INT(xvcvdpuxds, 2, float64, uint64, VsrD(i), VsrD(i), 0ULL)
2798 VSX_CVT_FP_TO_INT(xvcvdpuxws, 2, float64, uint32, VsrD(i), VsrW(2 * i), 0U)
2799 VSX_CVT_FP_TO_INT(xvcvspsxds, 2, float32, int64, VsrW(2 * i), VsrD(i), \
2800 0x8000000000000000ULL)
2801 VSX_CVT_FP_TO_INT(xvcvspsxws, 4, float32, int32, VsrW(i), VsrW(i), 0x80000000U)
2802 VSX_CVT_FP_TO_INT(xvcvspuxds, 2, float32, uint64, VsrW(2 * i), VsrD(i), 0ULL)
2803 VSX_CVT_FP_TO_INT(xvcvspuxws, 4, float32, uint32, VsrW(i), VsrW(i), 0U)
2804
2805 /*
2806 * VSX_CVT_FP_TO_INT_VECTOR - VSX floating point to integer conversion
2807 * op - instruction mnemonic
2808 * stp - source type (float32 or float64)
2809 * ttp - target type (int32, uint32, int64 or uint64)
2810 * sfld - source vsr_t field
2811 * tfld - target vsr_t field
2812 * rnan - resulting NaN
2813 */
2814 #define VSX_CVT_FP_TO_INT_VECTOR(op, stp, ttp, sfld, tfld, rnan) \
2815 void helper_##op(CPUPPCState *env, uint32_t opcode, \
2816 ppc_vsr_t *xt, ppc_vsr_t *xb) \
2817 { \
2818 ppc_vsr_t t = { }; \
2819 \
2820 t.tfld = stp##_to_##ttp##_round_to_zero(xb->sfld, &env->fp_status); \
2821 if (env->fp_status.float_exception_flags & float_flag_invalid) { \
2822 float_invalid_cvt(env, 0, GETPC(), stp##_classify(xb->sfld)); \
2823 t.tfld = rnan; \
2824 } \
2825 \
2826 *xt = t; \
2827 do_float_check_status(env, GETPC()); \
2828 }
2829
2830 VSX_CVT_FP_TO_INT_VECTOR(xscvqpsdz, float128, int64, f128, VsrD(0), \
2831 0x8000000000000000ULL)
2832
2833 VSX_CVT_FP_TO_INT_VECTOR(xscvqpswz, float128, int32, f128, VsrD(0), \
2834 0xffffffff80000000ULL)
2835 VSX_CVT_FP_TO_INT_VECTOR(xscvqpudz, float128, uint64, f128, VsrD(0), 0x0ULL)
2836 VSX_CVT_FP_TO_INT_VECTOR(xscvqpuwz, float128, uint32, f128, VsrD(0), 0x0ULL)
2837
2838 /*
2839 * VSX_CVT_INT_TO_FP - VSX integer to floating point conversion
2840 * op - instruction mnemonic
2841 * nels - number of elements (1, 2 or 4)
2842 * stp - source type (int32, uint32, int64 or uint64)
2843 * ttp - target type (float32 or float64)
2844 * sfld - source vsr_t field
2845 * tfld - target vsr_t field
2846 * jdef - definition of the j index (i or 2*i)
2847 * sfprf - set FPRF
2848 */
2849 #define VSX_CVT_INT_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf, r2sp) \
2850 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2851 { \
2852 ppc_vsr_t t = *xt; \
2853 int i; \
2854 \
2855 for (i = 0; i < nels; i++) { \
2856 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2857 if (r2sp) { \
2858 t.tfld = helper_frsp(env, t.tfld); \
2859 } \
2860 if (sfprf) { \
2861 helper_compute_fprf_float64(env, t.tfld); \
2862 } \
2863 } \
2864 \
2865 *xt = t; \
2866 do_float_check_status(env, GETPC()); \
2867 }
2868
2869 VSX_CVT_INT_TO_FP(xscvsxddp, 1, int64, float64, VsrD(0), VsrD(0), 1, 0)
2870 VSX_CVT_INT_TO_FP(xscvuxddp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 0)
2871 VSX_CVT_INT_TO_FP(xscvsxdsp, 1, int64, float64, VsrD(0), VsrD(0), 1, 1)
2872 VSX_CVT_INT_TO_FP(xscvuxdsp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 1)
2873 VSX_CVT_INT_TO_FP(xvcvsxddp, 2, int64, float64, VsrD(i), VsrD(i), 0, 0)
2874 VSX_CVT_INT_TO_FP(xvcvuxddp, 2, uint64, float64, VsrD(i), VsrD(i), 0, 0)
2875 VSX_CVT_INT_TO_FP(xvcvsxwdp, 2, int32, float64, VsrW(2 * i), VsrD(i), 0, 0)
2876 VSX_CVT_INT_TO_FP(xvcvuxwdp, 2, uint64, float64, VsrW(2 * i), VsrD(i), 0, 0)
2877 VSX_CVT_INT_TO_FP(xvcvsxdsp, 2, int64, float32, VsrD(i), VsrW(2 * i), 0, 0)
2878 VSX_CVT_INT_TO_FP(xvcvuxdsp, 2, uint64, float32, VsrD(i), VsrW(2 * i), 0, 0)
2879 VSX_CVT_INT_TO_FP(xvcvsxwsp, 4, int32, float32, VsrW(i), VsrW(i), 0, 0)
2880 VSX_CVT_INT_TO_FP(xvcvuxwsp, 4, uint32, float32, VsrW(i), VsrW(i), 0, 0)
2881
2882 /*
2883 * VSX_CVT_INT_TO_FP_VECTOR - VSX integer to floating point conversion
2884 * op - instruction mnemonic
2885 * stp - source type (int32, uint32, int64 or uint64)
2886 * ttp - target type (float32 or float64)
2887 * sfld - source vsr_t field
2888 * tfld - target vsr_t field
2889 */
2890 #define VSX_CVT_INT_TO_FP_VECTOR(op, stp, ttp, sfld, tfld) \
2891 void helper_##op(CPUPPCState *env, uint32_t opcode, \
2892 ppc_vsr_t *xt, ppc_vsr_t *xb) \
2893 { \
2894 ppc_vsr_t t = *xt; \
2895 \
2896 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2897 helper_compute_fprf_##ttp(env, t.tfld); \
2898 \
2899 *xt = t; \
2900 do_float_check_status(env, GETPC()); \
2901 }
2902
2903 VSX_CVT_INT_TO_FP_VECTOR(xscvsdqp, int64, float128, VsrD(0), f128)
2904 VSX_CVT_INT_TO_FP_VECTOR(xscvudqp, uint64, float128, VsrD(0), f128)
2905
2906 /*
2907 * For "use current rounding mode", define a value that will not be
2908 * one of the existing rounding model enums.
2909 */
2910 #define FLOAT_ROUND_CURRENT (float_round_nearest_even + float_round_down + \
2911 float_round_up + float_round_to_zero)
2912
2913 /*
2914 * VSX_ROUND - VSX floating point round
2915 * op - instruction mnemonic
2916 * nels - number of elements (1, 2 or 4)
2917 * tp - type (float32 or float64)
2918 * fld - vsr_t field (VsrD(*) or VsrW(*))
2919 * rmode - rounding mode
2920 * sfprf - set FPRF
2921 */
2922 #define VSX_ROUND(op, nels, tp, fld, rmode, sfprf) \
2923 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb) \
2924 { \
2925 ppc_vsr_t t = *xt; \
2926 int i; \
2927 FloatRoundMode curr_rounding_mode; \
2928 \
2929 if (rmode != FLOAT_ROUND_CURRENT) { \
2930 curr_rounding_mode = get_float_rounding_mode(&env->fp_status); \
2931 set_float_rounding_mode(rmode, &env->fp_status); \
2932 } \
2933 \
2934 for (i = 0; i < nels; i++) { \
2935 if (unlikely(tp##_is_signaling_nan(xb->fld, \
2936 &env->fp_status))) { \
2937 float_invalid_op_vxsnan(env, GETPC()); \
2938 t.fld = tp##_snan_to_qnan(xb->fld); \
2939 } else { \
2940 t.fld = tp##_round_to_int(xb->fld, &env->fp_status); \
2941 } \
2942 if (sfprf) { \
2943 helper_compute_fprf_float64(env, t.fld); \
2944 } \
2945 } \
2946 \
2947 /* \
2948 * If this is not a "use current rounding mode" instruction, \
2949 * then inhibit setting of the XX bit and restore rounding \
2950 * mode from FPSCR \
2951 */ \
2952 if (rmode != FLOAT_ROUND_CURRENT) { \
2953 set_float_rounding_mode(curr_rounding_mode, &env->fp_status); \
2954 env->fp_status.float_exception_flags &= ~float_flag_inexact; \
2955 } \
2956 \
2957 *xt = t; \
2958 do_float_check_status(env, GETPC()); \
2959 }
2960
2961 VSX_ROUND(xsrdpi, 1, float64, VsrD(0), float_round_ties_away, 1)
2962 VSX_ROUND(xsrdpic, 1, float64, VsrD(0), FLOAT_ROUND_CURRENT, 1)
2963 VSX_ROUND(xsrdpim, 1, float64, VsrD(0), float_round_down, 1)
2964 VSX_ROUND(xsrdpip, 1, float64, VsrD(0), float_round_up, 1)
2965 VSX_ROUND(xsrdpiz, 1, float64, VsrD(0), float_round_to_zero, 1)
2966
2967 VSX_ROUND(xvrdpi, 2, float64, VsrD(i), float_round_ties_away, 0)
2968 VSX_ROUND(xvrdpic, 2, float64, VsrD(i), FLOAT_ROUND_CURRENT, 0)
2969 VSX_ROUND(xvrdpim, 2, float64, VsrD(i), float_round_down, 0)
2970 VSX_ROUND(xvrdpip, 2, float64, VsrD(i), float_round_up, 0)
2971 VSX_ROUND(xvrdpiz, 2, float64, VsrD(i), float_round_to_zero, 0)
2972
2973 VSX_ROUND(xvrspi, 4, float32, VsrW(i), float_round_ties_away, 0)
2974 VSX_ROUND(xvrspic, 4, float32, VsrW(i), FLOAT_ROUND_CURRENT, 0)
2975 VSX_ROUND(xvrspim, 4, float32, VsrW(i), float_round_down, 0)
2976 VSX_ROUND(xvrspip, 4, float32, VsrW(i), float_round_up, 0)
2977 VSX_ROUND(xvrspiz, 4, float32, VsrW(i), float_round_to_zero, 0)
2978
2979 uint64_t helper_xsrsp(CPUPPCState *env, uint64_t xb)
2980 {
2981 helper_reset_fpstatus(env);
2982
2983 uint64_t xt = helper_frsp(env, xb);
2984
2985 helper_compute_fprf_float64(env, xt);
2986 do_float_check_status(env, GETPC());
2987 return xt;
2988 }
2989
2990 #define VSX_XXPERM(op, indexed) \
2991 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2992 ppc_vsr_t *xa, ppc_vsr_t *pcv) \
2993 { \
2994 ppc_vsr_t t = *xt; \
2995 int i, idx; \
2996 \
2997 for (i = 0; i < 16; i++) { \
2998 idx = pcv->VsrB(i) & 0x1F; \
2999 if (indexed) { \
3000 idx = 31 - idx; \
3001 } \
3002 t.VsrB(i) = (idx <= 15) ? xa->VsrB(idx) \
3003 : xt->VsrB(idx - 16); \
3004 } \
3005 *xt = t; \
3006 }
3007
3008 VSX_XXPERM(xxperm, 0)
3009 VSX_XXPERM(xxpermr, 1)
3010
3011 void helper_xvxsigsp(CPUPPCState *env, ppc_vsr_t *xt, ppc_vsr_t *xb)
3012 {
3013 ppc_vsr_t t = { };
3014 uint32_t exp, i, fraction;
3015
3016 for (i = 0; i < 4; i++) {
3017 exp = (xb->VsrW(i) >> 23) & 0xFF;
3018 fraction = xb->VsrW(i) & 0x7FFFFF;
3019 if (exp != 0 && exp != 255) {
3020 t.VsrW(i) = fraction | 0x00800000;
3021 } else {
3022 t.VsrW(i) = fraction;
3023 }
3024 }
3025 *xt = t;
3026 }
3027
3028 /*
3029 * VSX_TEST_DC - VSX floating point test data class
3030 * op - instruction mnemonic
3031 * nels - number of elements (1, 2 or 4)
3032 * xbn - VSR register number
3033 * tp - type (float32 or float64)
3034 * fld - vsr_t field (VsrD(*) or VsrW(*))
3035 * tfld - target vsr_t field (VsrD(*) or VsrW(*))
3036 * fld_max - target field max
3037 * scrf - set result in CR and FPCC
3038 */
3039 #define VSX_TEST_DC(op, nels, xbn, tp, fld, tfld, fld_max, scrf) \
3040 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3041 { \
3042 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
3043 ppc_vsr_t *xb = &env->vsr[xbn]; \
3044 ppc_vsr_t t = { }; \
3045 uint32_t i, sign, dcmx; \
3046 uint32_t cc, match = 0; \
3047 \
3048 if (!scrf) { \
3049 dcmx = DCMX_XV(opcode); \
3050 } else { \
3051 t = *xt; \
3052 dcmx = DCMX(opcode); \
3053 } \
3054 \
3055 for (i = 0; i < nels; i++) { \
3056 sign = tp##_is_neg(xb->fld); \
3057 if (tp##_is_any_nan(xb->fld)) { \
3058 match = extract32(dcmx, 6, 1); \
3059 } else if (tp##_is_infinity(xb->fld)) { \
3060 match = extract32(dcmx, 4 + !sign, 1); \
3061 } else if (tp##_is_zero(xb->fld)) { \
3062 match = extract32(dcmx, 2 + !sign, 1); \
3063 } else if (tp##_is_zero_or_denormal(xb->fld)) { \
3064 match = extract32(dcmx, 0 + !sign, 1); \
3065 } \
3066 \
3067 if (scrf) { \
3068 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT; \
3069 env->fpscr &= ~FP_FPCC; \
3070 env->fpscr |= cc << FPSCR_FPCC; \
3071 env->crf[BF(opcode)] = cc; \
3072 } else { \
3073 t.tfld = match ? fld_max : 0; \
3074 } \
3075 match = 0; \
3076 } \
3077 if (!scrf) { \
3078 *xt = t; \
3079 } \
3080 }
3081
3082 VSX_TEST_DC(xvtstdcdp, 2, xB(opcode), float64, VsrD(i), VsrD(i), UINT64_MAX, 0)
3083 VSX_TEST_DC(xvtstdcsp, 4, xB(opcode), float32, VsrW(i), VsrW(i), UINT32_MAX, 0)
3084 VSX_TEST_DC(xststdcdp, 1, xB(opcode), float64, VsrD(0), VsrD(0), 0, 1)
3085 VSX_TEST_DC(xststdcqp, 1, (rB(opcode) + 32), float128, f128, VsrD(0), 0, 1)
3086
3087 void helper_xststdcsp(CPUPPCState *env, uint32_t opcode, ppc_vsr_t *xb)
3088 {
3089 uint32_t dcmx, sign, exp;
3090 uint32_t cc, match = 0, not_sp = 0;
3091
3092 dcmx = DCMX(opcode);
3093 exp = (xb->VsrD(0) >> 52) & 0x7FF;
3094
3095 sign = float64_is_neg(xb->VsrD(0));
3096 if (float64_is_any_nan(xb->VsrD(0))) {
3097 match = extract32(dcmx, 6, 1);
3098 } else if (float64_is_infinity(xb->VsrD(0))) {
3099 match = extract32(dcmx, 4 + !sign, 1);
3100 } else if (float64_is_zero(xb->VsrD(0))) {
3101 match = extract32(dcmx, 2 + !sign, 1);
3102 } else if (float64_is_zero_or_denormal(xb->VsrD(0)) ||
3103 (exp > 0 && exp < 0x381)) {
3104 match = extract32(dcmx, 0 + !sign, 1);
3105 }
3106
3107 not_sp = !float64_eq(xb->VsrD(0),
3108 float32_to_float64(
3109 float64_to_float32(xb->VsrD(0), &env->fp_status),
3110 &env->fp_status), &env->fp_status);
3111
3112 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT | not_sp << CRF_SO_BIT;
3113 env->fpscr &= ~FP_FPCC;
3114 env->fpscr |= cc << FPSCR_FPCC;
3115 env->crf[BF(opcode)] = cc;
3116 }
3117
3118 void helper_xsrqpi(CPUPPCState *env, uint32_t opcode,
3119 ppc_vsr_t *xt, ppc_vsr_t *xb)
3120 {
3121 ppc_vsr_t t = { };
3122 uint8_t r = Rrm(opcode);
3123 uint8_t ex = Rc(opcode);
3124 uint8_t rmc = RMC(opcode);
3125 uint8_t rmode = 0;
3126 float_status tstat;
3127
3128 helper_reset_fpstatus(env);
3129
3130 if (r == 0 && rmc == 0) {
3131 rmode = float_round_ties_away;
3132 } else if (r == 0 && rmc == 0x3) {
3133 rmode = fpscr_rn;
3134 } else if (r == 1) {
3135 switch (rmc) {
3136 case 0:
3137 rmode = float_round_nearest_even;
3138 break;
3139 case 1:
3140 rmode = float_round_to_zero;
3141 break;
3142 case 2:
3143 rmode = float_round_up;
3144 break;
3145 case 3:
3146 rmode = float_round_down;
3147 break;
3148 default:
3149 abort();
3150 }
3151 }
3152
3153 tstat = env->fp_status;
3154 set_float_exception_flags(0, &tstat);
3155 set_float_rounding_mode(rmode, &tstat);
3156 t.f128 = float128_round_to_int(xb->f128, &tstat);
3157 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3158
3159 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3160 if (float128_is_signaling_nan(xb->f128, &tstat)) {
3161 float_invalid_op_vxsnan(env, GETPC());
3162 t.f128 = float128_snan_to_qnan(t.f128);
3163 }
3164 }
3165
3166 if (ex == 0 && (tstat.float_exception_flags & float_flag_inexact)) {
3167 env->fp_status.float_exception_flags &= ~float_flag_inexact;
3168 }
3169
3170 helper_compute_fprf_float128(env, t.f128);
3171 do_float_check_status(env, GETPC());
3172 *xt = t;
3173 }
3174
3175 void helper_xsrqpxp(CPUPPCState *env, uint32_t opcode,
3176 ppc_vsr_t *xt, ppc_vsr_t *xb)
3177 {
3178 ppc_vsr_t t = { };
3179 uint8_t r = Rrm(opcode);
3180 uint8_t rmc = RMC(opcode);
3181 uint8_t rmode = 0;
3182 floatx80 round_res;
3183 float_status tstat;
3184
3185 helper_reset_fpstatus(env);
3186
3187 if (r == 0 && rmc == 0) {
3188 rmode = float_round_ties_away;
3189 } else if (r == 0 && rmc == 0x3) {
3190 rmode = fpscr_rn;
3191 } else if (r == 1) {
3192 switch (rmc) {
3193 case 0:
3194 rmode = float_round_nearest_even;
3195 break;
3196 case 1:
3197 rmode = float_round_to_zero;
3198 break;
3199 case 2:
3200 rmode = float_round_up;
3201 break;
3202 case 3:
3203 rmode = float_round_down;
3204 break;
3205 default:
3206 abort();
3207 }
3208 }
3209
3210 tstat = env->fp_status;
3211 set_float_exception_flags(0, &tstat);
3212 set_float_rounding_mode(rmode, &tstat);
3213 round_res = float128_to_floatx80(xb->f128, &tstat);
3214 t.f128 = floatx80_to_float128(round_res, &tstat);
3215 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3216
3217 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3218 if (float128_is_signaling_nan(xb->f128, &tstat)) {
3219 float_invalid_op_vxsnan(env, GETPC());
3220 t.f128 = float128_snan_to_qnan(t.f128);
3221 }
3222 }
3223
3224 helper_compute_fprf_float128(env, t.f128);
3225 *xt = t;
3226 do_float_check_status(env, GETPC());
3227 }
3228
3229 void helper_xssqrtqp(CPUPPCState *env, uint32_t opcode,
3230 ppc_vsr_t *xt, ppc_vsr_t *xb)
3231 {
3232 ppc_vsr_t t = { };
3233 float_status tstat;
3234
3235 helper_reset_fpstatus(env);
3236
3237 tstat = env->fp_status;
3238 if (unlikely(Rc(opcode) != 0)) {
3239 tstat.float_rounding_mode = float_round_to_odd;
3240 }
3241
3242 set_float_exception_flags(0, &tstat);
3243 t.f128 = float128_sqrt(xb->f128, &tstat);
3244 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3245
3246 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3247 if (float128_is_signaling_nan(xb->f128, &tstat)) {
3248 float_invalid_op_vxsnan(env, GETPC());
3249 t.f128 = float128_snan_to_qnan(xb->f128);
3250 } else if (float128_is_quiet_nan(xb->f128, &tstat)) {
3251 t.f128 = xb->f128;
3252 } else if (float128_is_neg(xb->f128) && !float128_is_zero(xb->f128)) {
3253 float_invalid_op_vxsqrt(env, 1, GETPC());
3254 t.f128 = float128_default_nan(&env->fp_status);
3255 }
3256 }
3257
3258 helper_compute_fprf_float128(env, t.f128);
3259 *xt = t;
3260 do_float_check_status(env, GETPC());
3261 }
3262
3263 void helper_xssubqp(CPUPPCState *env, uint32_t opcode,
3264 ppc_vsr_t *xt, ppc_vsr_t *xa, ppc_vsr_t *xb)
3265 {
3266 ppc_vsr_t t = *xt;
3267 float_status tstat;
3268
3269 helper_reset_fpstatus(env);
3270
3271 tstat = env->fp_status;
3272 if (unlikely(Rc(opcode) != 0)) {
3273 tstat.float_rounding_mode = float_round_to_odd;
3274 }
3275
3276 set_float_exception_flags(0, &tstat);
3277 t.f128 = float128_sub(xa->f128, xb->f128, &tstat);
3278 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3279
3280 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3281 float_invalid_op_addsub(env, 1, GETPC(),
3282 float128_classify(xa->f128) |
3283 float128_classify(xb->f128));
3284 }
3285
3286 helper_compute_fprf_float128(env, t.f128);
3287 *xt = t;
3288 do_float_check_status(env, GETPC());
3289 }
QEMU mirror