docker: Use a stable snapshot for Debian Sid
[qemu/ar7.git] / target / ppc / fpu_helper.c
blob2ed4f4227579e07ae24067fc12156d4251bc1f69
1 /*
2 * PowerPC floating point and SPE emulation helpers for QEMU.
4 * Copyright (c) 2003-2007 Jocelyn Mayer
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
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"
26 static inline float128 float128_snan_to_qnan(float128 x)
28 float128 r;
30 r.high = x.high | 0x0000800000000000;
31 r.low = x.low;
32 return r;
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)
39 static inline bool fp_exceptions_enabled(CPUPPCState *env)
41 #ifdef CONFIG_USER_ONLY
42 return true;
43 #else
44 return (env->msr & ((1U << MSR_FE0) | (1U << MSR_FE1))) != 0;
45 #endif
48 /*****************************************************************************/
49 /* Floating point operations helpers */
52 * This is the non-arithmatic conversion that happens e.g. on loads.
53 * In the Power ISA pseudocode, this is called DOUBLE.
55 uint64_t helper_todouble(uint32_t arg)
57 uint32_t abs_arg = arg & 0x7fffffff;
58 uint64_t ret;
60 if (likely(abs_arg >= 0x00800000)) {
61 /* Normalized operand, or Inf, or NaN. */
62 ret = (uint64_t)extract32(arg, 30, 2) << 62;
63 ret |= ((extract32(arg, 30, 1) ^ 1) * (uint64_t)7) << 59;
64 ret |= (uint64_t)extract32(arg, 0, 30) << 29;
65 } else {
66 /* Zero or Denormalized operand. */
67 ret = (uint64_t)extract32(arg, 31, 1) << 63;
68 if (unlikely(abs_arg != 0)) {
69 /* Denormalized operand. */
70 int shift = clz32(abs_arg) - 9;
71 int exp = -126 - shift + 1023;
72 ret |= (uint64_t)exp << 52;
73 ret |= abs_arg << (shift + 29);
76 return ret;
80 * This is the non-arithmatic conversion that happens e.g. on stores.
81 * In the Power ISA pseudocode, this is called SINGLE.
83 uint32_t helper_tosingle(uint64_t arg)
85 int exp = extract64(arg, 52, 11);
86 uint32_t ret;
88 if (likely(exp > 896)) {
89 /* No denormalization required (includes Inf, NaN). */
90 ret = extract64(arg, 62, 2) << 30;
91 ret |= extract64(arg, 29, 30);
92 } else {
93 /* Zero or Denormal result. If the exponent is in bounds for
94 * a single-precision denormal result, extract the proper bits.
95 * If the input is not zero, and the exponent is out of bounds,
96 * then the result is undefined; this underflows to zero.
98 ret = extract64(arg, 63, 1) << 31;
99 if (unlikely(exp >= 874)) {
100 /* Denormal result. */
101 ret |= ((1ULL << 52) | extract64(arg, 0, 52)) >> (896 + 30 - exp);
104 return ret;
107 static inline int ppc_float32_get_unbiased_exp(float32 f)
109 return ((f >> 23) & 0xFF) - 127;
112 static inline int ppc_float64_get_unbiased_exp(float64 f)
114 return ((f >> 52) & 0x7FF) - 1023;
117 /* Classify a floating-point number. */
118 enum {
119 is_normal = 1,
120 is_zero = 2,
121 is_denormal = 4,
122 is_inf = 8,
123 is_qnan = 16,
124 is_snan = 32,
125 is_neg = 64,
128 #define COMPUTE_CLASS(tp) \
129 static int tp##_classify(tp arg) \
131 int ret = tp##_is_neg(arg) * is_neg; \
132 if (unlikely(tp##_is_any_nan(arg))) { \
133 float_status dummy = { }; /* snan_bit_is_one = 0 */ \
134 ret |= (tp##_is_signaling_nan(arg, &dummy) \
135 ? is_snan : is_qnan); \
136 } else if (unlikely(tp##_is_infinity(arg))) { \
137 ret |= is_inf; \
138 } else if (tp##_is_zero(arg)) { \
139 ret |= is_zero; \
140 } else if (tp##_is_zero_or_denormal(arg)) { \
141 ret |= is_denormal; \
142 } else { \
143 ret |= is_normal; \
145 return ret; \
148 COMPUTE_CLASS(float16)
149 COMPUTE_CLASS(float32)
150 COMPUTE_CLASS(float64)
151 COMPUTE_CLASS(float128)
153 static void set_fprf_from_class(CPUPPCState *env, int class)
155 static const uint8_t fprf[6][2] = {
156 { 0x04, 0x08 }, /* normalized */
157 { 0x02, 0x12 }, /* zero */
158 { 0x14, 0x18 }, /* denormalized */
159 { 0x05, 0x09 }, /* infinity */
160 { 0x11, 0x11 }, /* qnan */
161 { 0x00, 0x00 }, /* snan -- flags are undefined */
163 bool isneg = class & is_neg;
165 env->fpscr &= ~(0x1F << FPSCR_FPRF);
166 env->fpscr |= fprf[ctz32(class)][isneg] << FPSCR_FPRF;
169 #define COMPUTE_FPRF(tp) \
170 void helper_compute_fprf_##tp(CPUPPCState *env, tp arg) \
172 set_fprf_from_class(env, tp##_classify(arg)); \
175 COMPUTE_FPRF(float16)
176 COMPUTE_FPRF(float32)
177 COMPUTE_FPRF(float64)
178 COMPUTE_FPRF(float128)
180 /* Floating-point invalid operations exception */
181 static void finish_invalid_op_excp(CPUPPCState *env, int op, uintptr_t retaddr)
183 /* Update the floating-point invalid operation summary */
184 env->fpscr |= 1 << FPSCR_VX;
185 /* Update the floating-point exception summary */
186 env->fpscr |= FP_FX;
187 if (fpscr_ve != 0) {
188 /* Update the floating-point enabled exception summary */
189 env->fpscr |= 1 << FPSCR_FEX;
190 if (fp_exceptions_enabled(env)) {
191 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
192 POWERPC_EXCP_FP | op, retaddr);
197 static void finish_invalid_op_arith(CPUPPCState *env, int op,
198 bool set_fpcc, uintptr_t retaddr)
200 env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI));
201 if (fpscr_ve == 0) {
202 if (set_fpcc) {
203 env->fpscr &= ~(0xF << FPSCR_FPCC);
204 env->fpscr |= 0x11 << FPSCR_FPCC;
207 finish_invalid_op_excp(env, op, retaddr);
210 /* Signalling NaN */
211 static void float_invalid_op_vxsnan(CPUPPCState *env, uintptr_t retaddr)
213 env->fpscr |= 1 << FPSCR_VXSNAN;
214 finish_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, retaddr);
217 /* Magnitude subtraction of infinities */
218 static void float_invalid_op_vxisi(CPUPPCState *env, bool set_fpcc,
219 uintptr_t retaddr)
221 env->fpscr |= 1 << FPSCR_VXISI;
222 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXISI, set_fpcc, retaddr);
225 /* Division of infinity by infinity */
226 static void float_invalid_op_vxidi(CPUPPCState *env, bool set_fpcc,
227 uintptr_t retaddr)
229 env->fpscr |= 1 << FPSCR_VXIDI;
230 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXIDI, set_fpcc, retaddr);
233 /* Division of zero by zero */
234 static void float_invalid_op_vxzdz(CPUPPCState *env, bool set_fpcc,
235 uintptr_t retaddr)
237 env->fpscr |= 1 << FPSCR_VXZDZ;
238 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXZDZ, set_fpcc, retaddr);
241 /* Multiplication of zero by infinity */
242 static void float_invalid_op_vximz(CPUPPCState *env, bool set_fpcc,
243 uintptr_t retaddr)
245 env->fpscr |= 1 << FPSCR_VXIMZ;
246 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXIMZ, set_fpcc, retaddr);
249 /* Square root of a negative number */
250 static void float_invalid_op_vxsqrt(CPUPPCState *env, bool set_fpcc,
251 uintptr_t retaddr)
253 env->fpscr |= 1 << FPSCR_VXSQRT;
254 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXSQRT, set_fpcc, retaddr);
257 /* Ordered comparison of NaN */
258 static void float_invalid_op_vxvc(CPUPPCState *env, bool set_fpcc,
259 uintptr_t retaddr)
261 env->fpscr |= 1 << FPSCR_VXVC;
262 if (set_fpcc) {
263 env->fpscr &= ~(0xF << FPSCR_FPCC);
264 env->fpscr |= 0x11 << FPSCR_FPCC;
266 /* Update the floating-point invalid operation summary */
267 env->fpscr |= 1 << FPSCR_VX;
268 /* Update the floating-point exception summary */
269 env->fpscr |= FP_FX;
270 /* We must update the target FPR before raising the exception */
271 if (fpscr_ve != 0) {
272 CPUState *cs = CPU(ppc_env_get_cpu(env));
274 cs->exception_index = POWERPC_EXCP_PROGRAM;
275 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_VXVC;
276 /* Update the floating-point enabled exception summary */
277 env->fpscr |= 1 << FPSCR_FEX;
278 /* Exception is differed */
282 /* Invalid conversion */
283 static void float_invalid_op_vxcvi(CPUPPCState *env, bool set_fpcc,
284 uintptr_t retaddr)
286 env->fpscr |= 1 << FPSCR_VXCVI;
287 env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI));
288 if (fpscr_ve == 0) {
289 if (set_fpcc) {
290 env->fpscr &= ~(0xF << FPSCR_FPCC);
291 env->fpscr |= 0x11 << FPSCR_FPCC;
294 finish_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, retaddr);
297 static inline void float_zero_divide_excp(CPUPPCState *env, uintptr_t raddr)
299 env->fpscr |= 1 << FPSCR_ZX;
300 env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI));
301 /* Update the floating-point exception summary */
302 env->fpscr |= FP_FX;
303 if (fpscr_ze != 0) {
304 /* Update the floating-point enabled exception summary */
305 env->fpscr |= 1 << FPSCR_FEX;
306 if (fp_exceptions_enabled(env)) {
307 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
308 POWERPC_EXCP_FP | POWERPC_EXCP_FP_ZX,
309 raddr);
314 static inline void float_overflow_excp(CPUPPCState *env)
316 CPUState *cs = CPU(ppc_env_get_cpu(env));
318 env->fpscr |= 1 << FPSCR_OX;
319 /* Update the floating-point exception summary */
320 env->fpscr |= FP_FX;
321 if (fpscr_oe != 0) {
322 /* XXX: should adjust the result */
323 /* Update the floating-point enabled exception summary */
324 env->fpscr |= 1 << FPSCR_FEX;
325 /* We must update the target FPR before raising the exception */
326 cs->exception_index = POWERPC_EXCP_PROGRAM;
327 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_OX;
328 } else {
329 env->fpscr |= 1 << FPSCR_XX;
330 env->fpscr |= 1 << FPSCR_FI;
334 static inline void float_underflow_excp(CPUPPCState *env)
336 CPUState *cs = CPU(ppc_env_get_cpu(env));
338 env->fpscr |= 1 << FPSCR_UX;
339 /* Update the floating-point exception summary */
340 env->fpscr |= FP_FX;
341 if (fpscr_ue != 0) {
342 /* XXX: should adjust the result */
343 /* Update the floating-point enabled exception summary */
344 env->fpscr |= 1 << FPSCR_FEX;
345 /* We must update the target FPR before raising the exception */
346 cs->exception_index = POWERPC_EXCP_PROGRAM;
347 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_UX;
351 static inline void float_inexact_excp(CPUPPCState *env)
353 CPUState *cs = CPU(ppc_env_get_cpu(env));
355 env->fpscr |= 1 << FPSCR_FI;
356 env->fpscr |= 1 << FPSCR_XX;
357 /* Update the floating-point exception summary */
358 env->fpscr |= FP_FX;
359 if (fpscr_xe != 0) {
360 /* Update the floating-point enabled exception summary */
361 env->fpscr |= 1 << FPSCR_FEX;
362 /* We must update the target FPR before raising the exception */
363 cs->exception_index = POWERPC_EXCP_PROGRAM;
364 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_XX;
368 static inline void fpscr_set_rounding_mode(CPUPPCState *env)
370 int rnd_type;
372 /* Set rounding mode */
373 switch (fpscr_rn) {
374 case 0:
375 /* Best approximation (round to nearest) */
376 rnd_type = float_round_nearest_even;
377 break;
378 case 1:
379 /* Smaller magnitude (round toward zero) */
380 rnd_type = float_round_to_zero;
381 break;
382 case 2:
383 /* Round toward +infinite */
384 rnd_type = float_round_up;
385 break;
386 default:
387 case 3:
388 /* Round toward -infinite */
389 rnd_type = float_round_down;
390 break;
392 set_float_rounding_mode(rnd_type, &env->fp_status);
395 void helper_fpscr_clrbit(CPUPPCState *env, uint32_t bit)
397 int prev;
399 prev = (env->fpscr >> bit) & 1;
400 env->fpscr &= ~(1 << bit);
401 if (prev == 1) {
402 switch (bit) {
403 case FPSCR_RN1:
404 case FPSCR_RN:
405 fpscr_set_rounding_mode(env);
406 break;
407 case FPSCR_VXSNAN:
408 case FPSCR_VXISI:
409 case FPSCR_VXIDI:
410 case FPSCR_VXZDZ:
411 case FPSCR_VXIMZ:
412 case FPSCR_VXVC:
413 case FPSCR_VXSOFT:
414 case FPSCR_VXSQRT:
415 case FPSCR_VXCVI:
416 if (!fpscr_ix) {
417 /* Set VX bit to zero */
418 env->fpscr &= ~(1 << FPSCR_VX);
420 break;
421 case FPSCR_OX:
422 case FPSCR_UX:
423 case FPSCR_ZX:
424 case FPSCR_XX:
425 case FPSCR_VE:
426 case FPSCR_OE:
427 case FPSCR_UE:
428 case FPSCR_ZE:
429 case FPSCR_XE:
430 if (!fpscr_eex) {
431 /* Set the FEX bit */
432 env->fpscr &= ~(1 << FPSCR_FEX);
434 break;
435 default:
436 break;
441 void helper_fpscr_setbit(CPUPPCState *env, uint32_t bit)
443 CPUState *cs = CPU(ppc_env_get_cpu(env));
444 int prev;
446 prev = (env->fpscr >> bit) & 1;
447 env->fpscr |= 1 << bit;
448 if (prev == 0) {
449 switch (bit) {
450 case FPSCR_VX:
451 env->fpscr |= FP_FX;
452 if (fpscr_ve) {
453 goto raise_ve;
455 break;
456 case FPSCR_OX:
457 env->fpscr |= FP_FX;
458 if (fpscr_oe) {
459 goto raise_oe;
461 break;
462 case FPSCR_UX:
463 env->fpscr |= FP_FX;
464 if (fpscr_ue) {
465 goto raise_ue;
467 break;
468 case FPSCR_ZX:
469 env->fpscr |= FP_FX;
470 if (fpscr_ze) {
471 goto raise_ze;
473 break;
474 case FPSCR_XX:
475 env->fpscr |= FP_FX;
476 if (fpscr_xe) {
477 goto raise_xe;
479 break;
480 case FPSCR_VXSNAN:
481 case FPSCR_VXISI:
482 case FPSCR_VXIDI:
483 case FPSCR_VXZDZ:
484 case FPSCR_VXIMZ:
485 case FPSCR_VXVC:
486 case FPSCR_VXSOFT:
487 case FPSCR_VXSQRT:
488 case FPSCR_VXCVI:
489 env->fpscr |= 1 << FPSCR_VX;
490 env->fpscr |= FP_FX;
491 if (fpscr_ve != 0) {
492 goto raise_ve;
494 break;
495 case FPSCR_VE:
496 if (fpscr_vx != 0) {
497 raise_ve:
498 env->error_code = POWERPC_EXCP_FP;
499 if (fpscr_vxsnan) {
500 env->error_code |= POWERPC_EXCP_FP_VXSNAN;
502 if (fpscr_vxisi) {
503 env->error_code |= POWERPC_EXCP_FP_VXISI;
505 if (fpscr_vxidi) {
506 env->error_code |= POWERPC_EXCP_FP_VXIDI;
508 if (fpscr_vxzdz) {
509 env->error_code |= POWERPC_EXCP_FP_VXZDZ;
511 if (fpscr_vximz) {
512 env->error_code |= POWERPC_EXCP_FP_VXIMZ;
514 if (fpscr_vxvc) {
515 env->error_code |= POWERPC_EXCP_FP_VXVC;
517 if (fpscr_vxsoft) {
518 env->error_code |= POWERPC_EXCP_FP_VXSOFT;
520 if (fpscr_vxsqrt) {
521 env->error_code |= POWERPC_EXCP_FP_VXSQRT;
523 if (fpscr_vxcvi) {
524 env->error_code |= POWERPC_EXCP_FP_VXCVI;
526 goto raise_excp;
528 break;
529 case FPSCR_OE:
530 if (fpscr_ox != 0) {
531 raise_oe:
532 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_OX;
533 goto raise_excp;
535 break;
536 case FPSCR_UE:
537 if (fpscr_ux != 0) {
538 raise_ue:
539 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_UX;
540 goto raise_excp;
542 break;
543 case FPSCR_ZE:
544 if (fpscr_zx != 0) {
545 raise_ze:
546 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_ZX;
547 goto raise_excp;
549 break;
550 case FPSCR_XE:
551 if (fpscr_xx != 0) {
552 raise_xe:
553 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_XX;
554 goto raise_excp;
556 break;
557 case FPSCR_RN1:
558 case FPSCR_RN:
559 fpscr_set_rounding_mode(env);
560 break;
561 default:
562 break;
563 raise_excp:
564 /* Update the floating-point enabled exception summary */
565 env->fpscr |= 1 << FPSCR_FEX;
566 /* We have to update Rc1 before raising the exception */
567 cs->exception_index = POWERPC_EXCP_PROGRAM;
568 break;
573 void helper_store_fpscr(CPUPPCState *env, uint64_t arg, uint32_t mask)
575 CPUState *cs = CPU(ppc_env_get_cpu(env));
576 target_ulong prev, new;
577 int i;
579 prev = env->fpscr;
580 new = (target_ulong)arg;
581 new &= ~0x60000000LL;
582 new |= prev & 0x60000000LL;
583 for (i = 0; i < sizeof(target_ulong) * 2; i++) {
584 if (mask & (1 << i)) {
585 env->fpscr &= ~(0xFLL << (4 * i));
586 env->fpscr |= new & (0xFLL << (4 * i));
589 /* Update VX and FEX */
590 if (fpscr_ix != 0) {
591 env->fpscr |= 1 << FPSCR_VX;
592 } else {
593 env->fpscr &= ~(1 << FPSCR_VX);
595 if ((fpscr_ex & fpscr_eex) != 0) {
596 env->fpscr |= 1 << FPSCR_FEX;
597 cs->exception_index = POWERPC_EXCP_PROGRAM;
598 /* XXX: we should compute it properly */
599 env->error_code = POWERPC_EXCP_FP;
600 } else {
601 env->fpscr &= ~(1 << FPSCR_FEX);
603 fpscr_set_rounding_mode(env);
606 void store_fpscr(CPUPPCState *env, uint64_t arg, uint32_t mask)
608 helper_store_fpscr(env, arg, mask);
611 static void do_float_check_status(CPUPPCState *env, uintptr_t raddr)
613 CPUState *cs = CPU(ppc_env_get_cpu(env));
614 int status = get_float_exception_flags(&env->fp_status);
615 bool inexact_happened = false;
617 if (status & float_flag_overflow) {
618 float_overflow_excp(env);
619 } else if (status & float_flag_underflow) {
620 float_underflow_excp(env);
621 } else if (status & float_flag_inexact) {
622 float_inexact_excp(env);
623 inexact_happened = true;
626 /* if the inexact flag was not set */
627 if (inexact_happened == false) {
628 env->fpscr &= ~(1 << FPSCR_FI); /* clear the FPSCR[FI] bit */
631 if (cs->exception_index == POWERPC_EXCP_PROGRAM &&
632 (env->error_code & POWERPC_EXCP_FP)) {
633 /* Differred floating-point exception after target FPR update */
634 if (fp_exceptions_enabled(env)) {
635 raise_exception_err_ra(env, cs->exception_index,
636 env->error_code, raddr);
641 void helper_float_check_status(CPUPPCState *env)
643 do_float_check_status(env, GETPC());
646 void helper_reset_fpstatus(CPUPPCState *env)
648 set_float_exception_flags(0, &env->fp_status);
651 static void float_invalid_op_addsub(CPUPPCState *env, bool set_fpcc,
652 uintptr_t retaddr, int classes)
654 if ((classes & ~is_neg) == is_inf) {
655 /* Magnitude subtraction of infinities */
656 float_invalid_op_vxisi(env, set_fpcc, retaddr);
657 } else if (classes & is_snan) {
658 float_invalid_op_vxsnan(env, retaddr);
662 /* fadd - fadd. */
663 float64 helper_fadd(CPUPPCState *env, float64 arg1, float64 arg2)
665 float64 ret = float64_add(arg1, arg2, &env->fp_status);
666 int status = get_float_exception_flags(&env->fp_status);
668 if (unlikely(status & float_flag_invalid)) {
669 float_invalid_op_addsub(env, 1, GETPC(),
670 float64_classify(arg1) |
671 float64_classify(arg2));
674 return ret;
677 /* fsub - fsub. */
678 float64 helper_fsub(CPUPPCState *env, float64 arg1, float64 arg2)
680 float64 ret = float64_sub(arg1, arg2, &env->fp_status);
681 int status = get_float_exception_flags(&env->fp_status);
683 if (unlikely(status & float_flag_invalid)) {
684 float_invalid_op_addsub(env, 1, GETPC(),
685 float64_classify(arg1) |
686 float64_classify(arg2));
689 return ret;
692 static void float_invalid_op_mul(CPUPPCState *env, bool set_fprc,
693 uintptr_t retaddr, int classes)
695 if ((classes & (is_zero | is_inf)) == (is_zero | is_inf)) {
696 /* Multiplication of zero by infinity */
697 float_invalid_op_vximz(env, set_fprc, retaddr);
698 } else if (classes & is_snan) {
699 float_invalid_op_vxsnan(env, retaddr);
703 /* fmul - fmul. */
704 float64 helper_fmul(CPUPPCState *env, float64 arg1, float64 arg2)
706 float64 ret = float64_mul(arg1, arg2, &env->fp_status);
707 int status = get_float_exception_flags(&env->fp_status);
709 if (unlikely(status & float_flag_invalid)) {
710 float_invalid_op_mul(env, 1, GETPC(),
711 float64_classify(arg1) |
712 float64_classify(arg2));
715 return ret;
718 static void float_invalid_op_div(CPUPPCState *env, bool set_fprc,
719 uintptr_t retaddr, int classes)
721 classes &= ~is_neg;
722 if (classes == is_inf) {
723 /* Division of infinity by infinity */
724 float_invalid_op_vxidi(env, set_fprc, retaddr);
725 } else if (classes == is_zero) {
726 /* Division of zero by zero */
727 float_invalid_op_vxzdz(env, set_fprc, retaddr);
728 } else if (classes & is_snan) {
729 float_invalid_op_vxsnan(env, retaddr);
733 /* fdiv - fdiv. */
734 float64 helper_fdiv(CPUPPCState *env, float64 arg1, float64 arg2)
736 float64 ret = float64_div(arg1, arg2, &env->fp_status);
737 int status = get_float_exception_flags(&env->fp_status);
739 if (unlikely(status)) {
740 if (status & float_flag_invalid) {
741 float_invalid_op_div(env, 1, GETPC(),
742 float64_classify(arg1) |
743 float64_classify(arg2));
745 if (status & float_flag_divbyzero) {
746 float_zero_divide_excp(env, GETPC());
750 return ret;
753 static void float_invalid_cvt(CPUPPCState *env, bool set_fprc,
754 uintptr_t retaddr, int class1)
756 float_invalid_op_vxcvi(env, set_fprc, retaddr);
757 if (class1 & is_snan) {
758 float_invalid_op_vxsnan(env, retaddr);
762 #define FPU_FCTI(op, cvt, nanval) \
763 uint64_t helper_##op(CPUPPCState *env, float64 arg) \
765 uint64_t ret = float64_to_##cvt(arg, &env->fp_status); \
766 int status = get_float_exception_flags(&env->fp_status); \
768 if (unlikely(status)) { \
769 if (status & float_flag_invalid) { \
770 float_invalid_cvt(env, 1, GETPC(), float64_classify(arg)); \
771 ret = nanval; \
773 do_float_check_status(env, GETPC()); \
775 return ret; \
778 FPU_FCTI(fctiw, int32, 0x80000000U)
779 FPU_FCTI(fctiwz, int32_round_to_zero, 0x80000000U)
780 FPU_FCTI(fctiwu, uint32, 0x00000000U)
781 FPU_FCTI(fctiwuz, uint32_round_to_zero, 0x00000000U)
782 FPU_FCTI(fctid, int64, 0x8000000000000000ULL)
783 FPU_FCTI(fctidz, int64_round_to_zero, 0x8000000000000000ULL)
784 FPU_FCTI(fctidu, uint64, 0x0000000000000000ULL)
785 FPU_FCTI(fctiduz, uint64_round_to_zero, 0x0000000000000000ULL)
787 #define FPU_FCFI(op, cvtr, is_single) \
788 uint64_t helper_##op(CPUPPCState *env, uint64_t arg) \
790 CPU_DoubleU farg; \
792 if (is_single) { \
793 float32 tmp = cvtr(arg, &env->fp_status); \
794 farg.d = float32_to_float64(tmp, &env->fp_status); \
795 } else { \
796 farg.d = cvtr(arg, &env->fp_status); \
798 do_float_check_status(env, GETPC()); \
799 return farg.ll; \
802 FPU_FCFI(fcfid, int64_to_float64, 0)
803 FPU_FCFI(fcfids, int64_to_float32, 1)
804 FPU_FCFI(fcfidu, uint64_to_float64, 0)
805 FPU_FCFI(fcfidus, uint64_to_float32, 1)
807 static inline uint64_t do_fri(CPUPPCState *env, uint64_t arg,
808 int rounding_mode)
810 CPU_DoubleU farg;
812 farg.ll = arg;
814 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) {
815 /* sNaN round */
816 float_invalid_op_vxsnan(env, GETPC());
817 farg.ll = arg | 0x0008000000000000ULL;
818 } else {
819 int inexact = get_float_exception_flags(&env->fp_status) &
820 float_flag_inexact;
821 set_float_rounding_mode(rounding_mode, &env->fp_status);
822 farg.ll = float64_round_to_int(farg.d, &env->fp_status);
823 /* Restore rounding mode from FPSCR */
824 fpscr_set_rounding_mode(env);
826 /* fri* does not set FPSCR[XX] */
827 if (!inexact) {
828 env->fp_status.float_exception_flags &= ~float_flag_inexact;
831 do_float_check_status(env, GETPC());
832 return farg.ll;
835 uint64_t helper_frin(CPUPPCState *env, uint64_t arg)
837 return do_fri(env, arg, float_round_ties_away);
840 uint64_t helper_friz(CPUPPCState *env, uint64_t arg)
842 return do_fri(env, arg, float_round_to_zero);
845 uint64_t helper_frip(CPUPPCState *env, uint64_t arg)
847 return do_fri(env, arg, float_round_up);
850 uint64_t helper_frim(CPUPPCState *env, uint64_t arg)
852 return do_fri(env, arg, float_round_down);
855 #define FPU_MADDSUB_UPDATE(NAME, TP) \
856 static void NAME(CPUPPCState *env, TP arg1, TP arg2, TP arg3, \
857 unsigned int madd_flags, uintptr_t retaddr) \
859 if (TP##_is_signaling_nan(arg1, &env->fp_status) || \
860 TP##_is_signaling_nan(arg2, &env->fp_status) || \
861 TP##_is_signaling_nan(arg3, &env->fp_status)) { \
862 /* sNaN operation */ \
863 float_invalid_op_vxsnan(env, retaddr); \
865 if ((TP##_is_infinity(arg1) && TP##_is_zero(arg2)) || \
866 (TP##_is_zero(arg1) && TP##_is_infinity(arg2))) { \
867 /* Multiplication of zero by infinity */ \
868 float_invalid_op_vximz(env, 1, retaddr); \
870 if ((TP##_is_infinity(arg1) || TP##_is_infinity(arg2)) && \
871 TP##_is_infinity(arg3)) { \
872 uint8_t aSign, bSign, cSign; \
874 aSign = TP##_is_neg(arg1); \
875 bSign = TP##_is_neg(arg2); \
876 cSign = TP##_is_neg(arg3); \
877 if (madd_flags & float_muladd_negate_c) { \
878 cSign ^= 1; \
880 if (aSign ^ bSign ^ cSign) { \
881 float_invalid_op_vxisi(env, 1, retaddr); \
885 FPU_MADDSUB_UPDATE(float32_maddsub_update_excp, float32)
886 FPU_MADDSUB_UPDATE(float64_maddsub_update_excp, float64)
888 #define FPU_FMADD(op, madd_flags) \
889 uint64_t helper_##op(CPUPPCState *env, uint64_t arg1, \
890 uint64_t arg2, uint64_t arg3) \
892 uint32_t flags; \
893 float64 ret = float64_muladd(arg1, arg2, arg3, madd_flags, \
894 &env->fp_status); \
895 flags = get_float_exception_flags(&env->fp_status); \
896 if (flags) { \
897 if (flags & float_flag_invalid) { \
898 float64_maddsub_update_excp(env, arg1, arg2, arg3, \
899 madd_flags, GETPC()); \
901 do_float_check_status(env, GETPC()); \
903 return ret; \
906 #define MADD_FLGS 0
907 #define MSUB_FLGS float_muladd_negate_c
908 #define NMADD_FLGS float_muladd_negate_result
909 #define NMSUB_FLGS (float_muladd_negate_c | float_muladd_negate_result)
911 FPU_FMADD(fmadd, MADD_FLGS)
912 FPU_FMADD(fnmadd, NMADD_FLGS)
913 FPU_FMADD(fmsub, MSUB_FLGS)
914 FPU_FMADD(fnmsub, NMSUB_FLGS)
916 /* frsp - frsp. */
917 uint64_t helper_frsp(CPUPPCState *env, uint64_t arg)
919 CPU_DoubleU farg;
920 float32 f32;
922 farg.ll = arg;
924 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) {
925 float_invalid_op_vxsnan(env, GETPC());
927 f32 = float64_to_float32(farg.d, &env->fp_status);
928 farg.d = float32_to_float64(f32, &env->fp_status);
930 return farg.ll;
933 /* fsqrt - fsqrt. */
934 float64 helper_fsqrt(CPUPPCState *env, float64 arg)
936 float64 ret = float64_sqrt(arg, &env->fp_status);
937 int status = get_float_exception_flags(&env->fp_status);
939 if (unlikely(status & float_flag_invalid)) {
940 if (unlikely(float64_is_any_nan(arg))) {
941 if (unlikely(float64_is_signaling_nan(arg, &env->fp_status))) {
942 /* sNaN square root */
943 float_invalid_op_vxsnan(env, GETPC());
945 } else {
946 /* Square root of a negative nonzero number */
947 float_invalid_op_vxsqrt(env, 1, GETPC());
951 return ret;
954 /* fre - fre. */
955 float64 helper_fre(CPUPPCState *env, float64 arg)
957 /* "Estimate" the reciprocal with actual division. */
958 float64 ret = float64_div(float64_one, arg, &env->fp_status);
959 int status = get_float_exception_flags(&env->fp_status);
961 if (unlikely(status)) {
962 if (status & float_flag_invalid) {
963 if (float64_is_signaling_nan(arg, &env->fp_status)) {
964 /* sNaN reciprocal */
965 float_invalid_op_vxsnan(env, GETPC());
968 if (status & float_flag_divbyzero) {
969 float_zero_divide_excp(env, GETPC());
970 /* For FPSCR.ZE == 0, the result is 1/2. */
971 ret = float64_set_sign(float64_half, float64_is_neg(arg));
975 return ret;
978 /* fres - fres. */
979 uint64_t helper_fres(CPUPPCState *env, uint64_t arg)
981 CPU_DoubleU farg;
982 float32 f32;
984 farg.ll = arg;
986 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) {
987 /* sNaN reciprocal */
988 float_invalid_op_vxsnan(env, GETPC());
990 farg.d = float64_div(float64_one, farg.d, &env->fp_status);
991 f32 = float64_to_float32(farg.d, &env->fp_status);
992 farg.d = float32_to_float64(f32, &env->fp_status);
994 return farg.ll;
997 /* frsqrte - frsqrte. */
998 float64 helper_frsqrte(CPUPPCState *env, float64 arg)
1000 /* "Estimate" the reciprocal with actual division. */
1001 float64 rets = float64_sqrt(arg, &env->fp_status);
1002 float64 retd = float64_div(float64_one, rets, &env->fp_status);
1003 int status = get_float_exception_flags(&env->fp_status);
1005 if (unlikely(status)) {
1006 if (status & float_flag_invalid) {
1007 if (float64_is_signaling_nan(arg, &env->fp_status)) {
1008 /* sNaN reciprocal */
1009 float_invalid_op_vxsnan(env, GETPC());
1010 } else {
1011 /* Square root of a negative nonzero number */
1012 float_invalid_op_vxsqrt(env, 1, GETPC());
1015 if (status & float_flag_divbyzero) {
1016 /* Reciprocal of (square root of) zero. */
1017 float_zero_divide_excp(env, GETPC());
1021 return retd;
1024 /* fsel - fsel. */
1025 uint64_t helper_fsel(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
1026 uint64_t arg3)
1028 CPU_DoubleU farg1;
1030 farg1.ll = arg1;
1032 if ((!float64_is_neg(farg1.d) || float64_is_zero(farg1.d)) &&
1033 !float64_is_any_nan(farg1.d)) {
1034 return arg2;
1035 } else {
1036 return arg3;
1040 uint32_t helper_ftdiv(uint64_t fra, uint64_t frb)
1042 int fe_flag = 0;
1043 int fg_flag = 0;
1045 if (unlikely(float64_is_infinity(fra) ||
1046 float64_is_infinity(frb) ||
1047 float64_is_zero(frb))) {
1048 fe_flag = 1;
1049 fg_flag = 1;
1050 } else {
1051 int e_a = ppc_float64_get_unbiased_exp(fra);
1052 int e_b = ppc_float64_get_unbiased_exp(frb);
1054 if (unlikely(float64_is_any_nan(fra) ||
1055 float64_is_any_nan(frb))) {
1056 fe_flag = 1;
1057 } else if ((e_b <= -1022) || (e_b >= 1021)) {
1058 fe_flag = 1;
1059 } else if (!float64_is_zero(fra) &&
1060 (((e_a - e_b) >= 1023) ||
1061 ((e_a - e_b) <= -1021) ||
1062 (e_a <= -970))) {
1063 fe_flag = 1;
1066 if (unlikely(float64_is_zero_or_denormal(frb))) {
1067 /* XB is not zero because of the above check and */
1068 /* so must be denormalized. */
1069 fg_flag = 1;
1073 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
1076 uint32_t helper_ftsqrt(uint64_t frb)
1078 int fe_flag = 0;
1079 int fg_flag = 0;
1081 if (unlikely(float64_is_infinity(frb) || float64_is_zero(frb))) {
1082 fe_flag = 1;
1083 fg_flag = 1;
1084 } else {
1085 int e_b = ppc_float64_get_unbiased_exp(frb);
1087 if (unlikely(float64_is_any_nan(frb))) {
1088 fe_flag = 1;
1089 } else if (unlikely(float64_is_zero(frb))) {
1090 fe_flag = 1;
1091 } else if (unlikely(float64_is_neg(frb))) {
1092 fe_flag = 1;
1093 } else if (!float64_is_zero(frb) && (e_b <= (-1022+52))) {
1094 fe_flag = 1;
1097 if (unlikely(float64_is_zero_or_denormal(frb))) {
1098 /* XB is not zero because of the above check and */
1099 /* therefore must be denormalized. */
1100 fg_flag = 1;
1104 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
1107 void helper_fcmpu(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
1108 uint32_t crfD)
1110 CPU_DoubleU farg1, farg2;
1111 uint32_t ret = 0;
1113 farg1.ll = arg1;
1114 farg2.ll = arg2;
1116 if (unlikely(float64_is_any_nan(farg1.d) ||
1117 float64_is_any_nan(farg2.d))) {
1118 ret = 0x01UL;
1119 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) {
1120 ret = 0x08UL;
1121 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) {
1122 ret = 0x04UL;
1123 } else {
1124 ret = 0x02UL;
1127 env->fpscr &= ~(0x0F << FPSCR_FPRF);
1128 env->fpscr |= ret << FPSCR_FPRF;
1129 env->crf[crfD] = ret;
1130 if (unlikely(ret == 0x01UL
1131 && (float64_is_signaling_nan(farg1.d, &env->fp_status) ||
1132 float64_is_signaling_nan(farg2.d, &env->fp_status)))) {
1133 /* sNaN comparison */
1134 float_invalid_op_vxsnan(env, GETPC());
1138 void helper_fcmpo(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
1139 uint32_t crfD)
1141 CPU_DoubleU farg1, farg2;
1142 uint32_t ret = 0;
1144 farg1.ll = arg1;
1145 farg2.ll = arg2;
1147 if (unlikely(float64_is_any_nan(farg1.d) ||
1148 float64_is_any_nan(farg2.d))) {
1149 ret = 0x01UL;
1150 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) {
1151 ret = 0x08UL;
1152 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) {
1153 ret = 0x04UL;
1154 } else {
1155 ret = 0x02UL;
1158 env->fpscr &= ~(0x0F << FPSCR_FPRF);
1159 env->fpscr |= ret << FPSCR_FPRF;
1160 env->crf[crfD] = ret;
1161 if (unlikely(ret == 0x01UL)) {
1162 float_invalid_op_vxvc(env, 1, GETPC());
1163 if (float64_is_signaling_nan(farg1.d, &env->fp_status) ||
1164 float64_is_signaling_nan(farg2.d, &env->fp_status)) {
1165 /* sNaN comparison */
1166 float_invalid_op_vxsnan(env, GETPC());
1171 /* Single-precision floating-point conversions */
1172 static inline uint32_t efscfsi(CPUPPCState *env, uint32_t val)
1174 CPU_FloatU u;
1176 u.f = int32_to_float32(val, &env->vec_status);
1178 return u.l;
1181 static inline uint32_t efscfui(CPUPPCState *env, uint32_t val)
1183 CPU_FloatU u;
1185 u.f = uint32_to_float32(val, &env->vec_status);
1187 return u.l;
1190 static inline int32_t efsctsi(CPUPPCState *env, uint32_t val)
1192 CPU_FloatU u;
1194 u.l = val;
1195 /* NaN are not treated the same way IEEE 754 does */
1196 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1197 return 0;
1200 return float32_to_int32(u.f, &env->vec_status);
1203 static inline uint32_t efsctui(CPUPPCState *env, uint32_t val)
1205 CPU_FloatU u;
1207 u.l = val;
1208 /* NaN are not treated the same way IEEE 754 does */
1209 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1210 return 0;
1213 return float32_to_uint32(u.f, &env->vec_status);
1216 static inline uint32_t efsctsiz(CPUPPCState *env, uint32_t val)
1218 CPU_FloatU u;
1220 u.l = val;
1221 /* NaN are not treated the same way IEEE 754 does */
1222 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1223 return 0;
1226 return float32_to_int32_round_to_zero(u.f, &env->vec_status);
1229 static inline uint32_t efsctuiz(CPUPPCState *env, uint32_t val)
1231 CPU_FloatU u;
1233 u.l = val;
1234 /* NaN are not treated the same way IEEE 754 does */
1235 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1236 return 0;
1239 return float32_to_uint32_round_to_zero(u.f, &env->vec_status);
1242 static inline uint32_t efscfsf(CPUPPCState *env, uint32_t val)
1244 CPU_FloatU u;
1245 float32 tmp;
1247 u.f = int32_to_float32(val, &env->vec_status);
1248 tmp = int64_to_float32(1ULL << 32, &env->vec_status);
1249 u.f = float32_div(u.f, tmp, &env->vec_status);
1251 return u.l;
1254 static inline uint32_t efscfuf(CPUPPCState *env, uint32_t val)
1256 CPU_FloatU u;
1257 float32 tmp;
1259 u.f = uint32_to_float32(val, &env->vec_status);
1260 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1261 u.f = float32_div(u.f, tmp, &env->vec_status);
1263 return u.l;
1266 static inline uint32_t efsctsf(CPUPPCState *env, uint32_t val)
1268 CPU_FloatU u;
1269 float32 tmp;
1271 u.l = val;
1272 /* NaN are not treated the same way IEEE 754 does */
1273 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1274 return 0;
1276 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1277 u.f = float32_mul(u.f, tmp, &env->vec_status);
1279 return float32_to_int32(u.f, &env->vec_status);
1282 static inline uint32_t efsctuf(CPUPPCState *env, uint32_t val)
1284 CPU_FloatU u;
1285 float32 tmp;
1287 u.l = val;
1288 /* NaN are not treated the same way IEEE 754 does */
1289 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1290 return 0;
1292 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1293 u.f = float32_mul(u.f, tmp, &env->vec_status);
1295 return float32_to_uint32(u.f, &env->vec_status);
1298 #define HELPER_SPE_SINGLE_CONV(name) \
1299 uint32_t helper_e##name(CPUPPCState *env, uint32_t val) \
1301 return e##name(env, val); \
1303 /* efscfsi */
1304 HELPER_SPE_SINGLE_CONV(fscfsi);
1305 /* efscfui */
1306 HELPER_SPE_SINGLE_CONV(fscfui);
1307 /* efscfuf */
1308 HELPER_SPE_SINGLE_CONV(fscfuf);
1309 /* efscfsf */
1310 HELPER_SPE_SINGLE_CONV(fscfsf);
1311 /* efsctsi */
1312 HELPER_SPE_SINGLE_CONV(fsctsi);
1313 /* efsctui */
1314 HELPER_SPE_SINGLE_CONV(fsctui);
1315 /* efsctsiz */
1316 HELPER_SPE_SINGLE_CONV(fsctsiz);
1317 /* efsctuiz */
1318 HELPER_SPE_SINGLE_CONV(fsctuiz);
1319 /* efsctsf */
1320 HELPER_SPE_SINGLE_CONV(fsctsf);
1321 /* efsctuf */
1322 HELPER_SPE_SINGLE_CONV(fsctuf);
1324 #define HELPER_SPE_VECTOR_CONV(name) \
1325 uint64_t helper_ev##name(CPUPPCState *env, uint64_t val) \
1327 return ((uint64_t)e##name(env, val >> 32) << 32) | \
1328 (uint64_t)e##name(env, val); \
1330 /* evfscfsi */
1331 HELPER_SPE_VECTOR_CONV(fscfsi);
1332 /* evfscfui */
1333 HELPER_SPE_VECTOR_CONV(fscfui);
1334 /* evfscfuf */
1335 HELPER_SPE_VECTOR_CONV(fscfuf);
1336 /* evfscfsf */
1337 HELPER_SPE_VECTOR_CONV(fscfsf);
1338 /* evfsctsi */
1339 HELPER_SPE_VECTOR_CONV(fsctsi);
1340 /* evfsctui */
1341 HELPER_SPE_VECTOR_CONV(fsctui);
1342 /* evfsctsiz */
1343 HELPER_SPE_VECTOR_CONV(fsctsiz);
1344 /* evfsctuiz */
1345 HELPER_SPE_VECTOR_CONV(fsctuiz);
1346 /* evfsctsf */
1347 HELPER_SPE_VECTOR_CONV(fsctsf);
1348 /* evfsctuf */
1349 HELPER_SPE_VECTOR_CONV(fsctuf);
1351 /* Single-precision floating-point arithmetic */
1352 static inline uint32_t efsadd(CPUPPCState *env, uint32_t op1, uint32_t op2)
1354 CPU_FloatU u1, u2;
1356 u1.l = op1;
1357 u2.l = op2;
1358 u1.f = float32_add(u1.f, u2.f, &env->vec_status);
1359 return u1.l;
1362 static inline uint32_t efssub(CPUPPCState *env, uint32_t op1, uint32_t op2)
1364 CPU_FloatU u1, u2;
1366 u1.l = op1;
1367 u2.l = op2;
1368 u1.f = float32_sub(u1.f, u2.f, &env->vec_status);
1369 return u1.l;
1372 static inline uint32_t efsmul(CPUPPCState *env, uint32_t op1, uint32_t op2)
1374 CPU_FloatU u1, u2;
1376 u1.l = op1;
1377 u2.l = op2;
1378 u1.f = float32_mul(u1.f, u2.f, &env->vec_status);
1379 return u1.l;
1382 static inline uint32_t efsdiv(CPUPPCState *env, uint32_t op1, uint32_t op2)
1384 CPU_FloatU u1, u2;
1386 u1.l = op1;
1387 u2.l = op2;
1388 u1.f = float32_div(u1.f, u2.f, &env->vec_status);
1389 return u1.l;
1392 #define HELPER_SPE_SINGLE_ARITH(name) \
1393 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \
1395 return e##name(env, op1, op2); \
1397 /* efsadd */
1398 HELPER_SPE_SINGLE_ARITH(fsadd);
1399 /* efssub */
1400 HELPER_SPE_SINGLE_ARITH(fssub);
1401 /* efsmul */
1402 HELPER_SPE_SINGLE_ARITH(fsmul);
1403 /* efsdiv */
1404 HELPER_SPE_SINGLE_ARITH(fsdiv);
1406 #define HELPER_SPE_VECTOR_ARITH(name) \
1407 uint64_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \
1409 return ((uint64_t)e##name(env, op1 >> 32, op2 >> 32) << 32) | \
1410 (uint64_t)e##name(env, op1, op2); \
1412 /* evfsadd */
1413 HELPER_SPE_VECTOR_ARITH(fsadd);
1414 /* evfssub */
1415 HELPER_SPE_VECTOR_ARITH(fssub);
1416 /* evfsmul */
1417 HELPER_SPE_VECTOR_ARITH(fsmul);
1418 /* evfsdiv */
1419 HELPER_SPE_VECTOR_ARITH(fsdiv);
1421 /* Single-precision floating-point comparisons */
1422 static inline uint32_t efscmplt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1424 CPU_FloatU u1, u2;
1426 u1.l = op1;
1427 u2.l = op2;
1428 return float32_lt(u1.f, u2.f, &env->vec_status) ? 4 : 0;
1431 static inline uint32_t efscmpgt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1433 CPU_FloatU u1, u2;
1435 u1.l = op1;
1436 u2.l = op2;
1437 return float32_le(u1.f, u2.f, &env->vec_status) ? 0 : 4;
1440 static inline uint32_t efscmpeq(CPUPPCState *env, uint32_t op1, uint32_t op2)
1442 CPU_FloatU u1, u2;
1444 u1.l = op1;
1445 u2.l = op2;
1446 return float32_eq(u1.f, u2.f, &env->vec_status) ? 4 : 0;
1449 static inline uint32_t efststlt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1451 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1452 return efscmplt(env, op1, op2);
1455 static inline uint32_t efststgt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1457 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1458 return efscmpgt(env, op1, op2);
1461 static inline uint32_t efststeq(CPUPPCState *env, uint32_t op1, uint32_t op2)
1463 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1464 return efscmpeq(env, op1, op2);
1467 #define HELPER_SINGLE_SPE_CMP(name) \
1468 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \
1470 return e##name(env, op1, op2); \
1472 /* efststlt */
1473 HELPER_SINGLE_SPE_CMP(fststlt);
1474 /* efststgt */
1475 HELPER_SINGLE_SPE_CMP(fststgt);
1476 /* efststeq */
1477 HELPER_SINGLE_SPE_CMP(fststeq);
1478 /* efscmplt */
1479 HELPER_SINGLE_SPE_CMP(fscmplt);
1480 /* efscmpgt */
1481 HELPER_SINGLE_SPE_CMP(fscmpgt);
1482 /* efscmpeq */
1483 HELPER_SINGLE_SPE_CMP(fscmpeq);
1485 static inline uint32_t evcmp_merge(int t0, int t1)
1487 return (t0 << 3) | (t1 << 2) | ((t0 | t1) << 1) | (t0 & t1);
1490 #define HELPER_VECTOR_SPE_CMP(name) \
1491 uint32_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \
1493 return evcmp_merge(e##name(env, op1 >> 32, op2 >> 32), \
1494 e##name(env, op1, op2)); \
1496 /* evfststlt */
1497 HELPER_VECTOR_SPE_CMP(fststlt);
1498 /* evfststgt */
1499 HELPER_VECTOR_SPE_CMP(fststgt);
1500 /* evfststeq */
1501 HELPER_VECTOR_SPE_CMP(fststeq);
1502 /* evfscmplt */
1503 HELPER_VECTOR_SPE_CMP(fscmplt);
1504 /* evfscmpgt */
1505 HELPER_VECTOR_SPE_CMP(fscmpgt);
1506 /* evfscmpeq */
1507 HELPER_VECTOR_SPE_CMP(fscmpeq);
1509 /* Double-precision floating-point conversion */
1510 uint64_t helper_efdcfsi(CPUPPCState *env, uint32_t val)
1512 CPU_DoubleU u;
1514 u.d = int32_to_float64(val, &env->vec_status);
1516 return u.ll;
1519 uint64_t helper_efdcfsid(CPUPPCState *env, uint64_t val)
1521 CPU_DoubleU u;
1523 u.d = int64_to_float64(val, &env->vec_status);
1525 return u.ll;
1528 uint64_t helper_efdcfui(CPUPPCState *env, uint32_t val)
1530 CPU_DoubleU u;
1532 u.d = uint32_to_float64(val, &env->vec_status);
1534 return u.ll;
1537 uint64_t helper_efdcfuid(CPUPPCState *env, uint64_t val)
1539 CPU_DoubleU u;
1541 u.d = uint64_to_float64(val, &env->vec_status);
1543 return u.ll;
1546 uint32_t helper_efdctsi(CPUPPCState *env, uint64_t val)
1548 CPU_DoubleU u;
1550 u.ll = val;
1551 /* NaN are not treated the same way IEEE 754 does */
1552 if (unlikely(float64_is_any_nan(u.d))) {
1553 return 0;
1556 return float64_to_int32(u.d, &env->vec_status);
1559 uint32_t helper_efdctui(CPUPPCState *env, uint64_t val)
1561 CPU_DoubleU u;
1563 u.ll = val;
1564 /* NaN are not treated the same way IEEE 754 does */
1565 if (unlikely(float64_is_any_nan(u.d))) {
1566 return 0;
1569 return float64_to_uint32(u.d, &env->vec_status);
1572 uint32_t helper_efdctsiz(CPUPPCState *env, uint64_t val)
1574 CPU_DoubleU u;
1576 u.ll = val;
1577 /* NaN are not treated the same way IEEE 754 does */
1578 if (unlikely(float64_is_any_nan(u.d))) {
1579 return 0;
1582 return float64_to_int32_round_to_zero(u.d, &env->vec_status);
1585 uint64_t helper_efdctsidz(CPUPPCState *env, uint64_t val)
1587 CPU_DoubleU u;
1589 u.ll = val;
1590 /* NaN are not treated the same way IEEE 754 does */
1591 if (unlikely(float64_is_any_nan(u.d))) {
1592 return 0;
1595 return float64_to_int64_round_to_zero(u.d, &env->vec_status);
1598 uint32_t helper_efdctuiz(CPUPPCState *env, uint64_t val)
1600 CPU_DoubleU u;
1602 u.ll = val;
1603 /* NaN are not treated the same way IEEE 754 does */
1604 if (unlikely(float64_is_any_nan(u.d))) {
1605 return 0;
1608 return float64_to_uint32_round_to_zero(u.d, &env->vec_status);
1611 uint64_t helper_efdctuidz(CPUPPCState *env, uint64_t val)
1613 CPU_DoubleU u;
1615 u.ll = val;
1616 /* NaN are not treated the same way IEEE 754 does */
1617 if (unlikely(float64_is_any_nan(u.d))) {
1618 return 0;
1621 return float64_to_uint64_round_to_zero(u.d, &env->vec_status);
1624 uint64_t helper_efdcfsf(CPUPPCState *env, uint32_t val)
1626 CPU_DoubleU u;
1627 float64 tmp;
1629 u.d = int32_to_float64(val, &env->vec_status);
1630 tmp = int64_to_float64(1ULL << 32, &env->vec_status);
1631 u.d = float64_div(u.d, tmp, &env->vec_status);
1633 return u.ll;
1636 uint64_t helper_efdcfuf(CPUPPCState *env, uint32_t val)
1638 CPU_DoubleU u;
1639 float64 tmp;
1641 u.d = uint32_to_float64(val, &env->vec_status);
1642 tmp = int64_to_float64(1ULL << 32, &env->vec_status);
1643 u.d = float64_div(u.d, tmp, &env->vec_status);
1645 return u.ll;
1648 uint32_t helper_efdctsf(CPUPPCState *env, uint64_t val)
1650 CPU_DoubleU u;
1651 float64 tmp;
1653 u.ll = val;
1654 /* NaN are not treated the same way IEEE 754 does */
1655 if (unlikely(float64_is_any_nan(u.d))) {
1656 return 0;
1658 tmp = uint64_to_float64(1ULL << 32, &env->vec_status);
1659 u.d = float64_mul(u.d, tmp, &env->vec_status);
1661 return float64_to_int32(u.d, &env->vec_status);
1664 uint32_t helper_efdctuf(CPUPPCState *env, uint64_t val)
1666 CPU_DoubleU u;
1667 float64 tmp;
1669 u.ll = val;
1670 /* NaN are not treated the same way IEEE 754 does */
1671 if (unlikely(float64_is_any_nan(u.d))) {
1672 return 0;
1674 tmp = uint64_to_float64(1ULL << 32, &env->vec_status);
1675 u.d = float64_mul(u.d, tmp, &env->vec_status);
1677 return float64_to_uint32(u.d, &env->vec_status);
1680 uint32_t helper_efscfd(CPUPPCState *env, uint64_t val)
1682 CPU_DoubleU u1;
1683 CPU_FloatU u2;
1685 u1.ll = val;
1686 u2.f = float64_to_float32(u1.d, &env->vec_status);
1688 return u2.l;
1691 uint64_t helper_efdcfs(CPUPPCState *env, uint32_t val)
1693 CPU_DoubleU u2;
1694 CPU_FloatU u1;
1696 u1.l = val;
1697 u2.d = float32_to_float64(u1.f, &env->vec_status);
1699 return u2.ll;
1702 /* Double precision fixed-point arithmetic */
1703 uint64_t helper_efdadd(CPUPPCState *env, uint64_t op1, uint64_t op2)
1705 CPU_DoubleU u1, u2;
1707 u1.ll = op1;
1708 u2.ll = op2;
1709 u1.d = float64_add(u1.d, u2.d, &env->vec_status);
1710 return u1.ll;
1713 uint64_t helper_efdsub(CPUPPCState *env, uint64_t op1, uint64_t op2)
1715 CPU_DoubleU u1, u2;
1717 u1.ll = op1;
1718 u2.ll = op2;
1719 u1.d = float64_sub(u1.d, u2.d, &env->vec_status);
1720 return u1.ll;
1723 uint64_t helper_efdmul(CPUPPCState *env, uint64_t op1, uint64_t op2)
1725 CPU_DoubleU u1, u2;
1727 u1.ll = op1;
1728 u2.ll = op2;
1729 u1.d = float64_mul(u1.d, u2.d, &env->vec_status);
1730 return u1.ll;
1733 uint64_t helper_efddiv(CPUPPCState *env, uint64_t op1, uint64_t op2)
1735 CPU_DoubleU u1, u2;
1737 u1.ll = op1;
1738 u2.ll = op2;
1739 u1.d = float64_div(u1.d, u2.d, &env->vec_status);
1740 return u1.ll;
1743 /* Double precision floating point helpers */
1744 uint32_t helper_efdtstlt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1746 CPU_DoubleU u1, u2;
1748 u1.ll = op1;
1749 u2.ll = op2;
1750 return float64_lt(u1.d, u2.d, &env->vec_status) ? 4 : 0;
1753 uint32_t helper_efdtstgt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1755 CPU_DoubleU u1, u2;
1757 u1.ll = op1;
1758 u2.ll = op2;
1759 return float64_le(u1.d, u2.d, &env->vec_status) ? 0 : 4;
1762 uint32_t helper_efdtsteq(CPUPPCState *env, uint64_t op1, uint64_t op2)
1764 CPU_DoubleU u1, u2;
1766 u1.ll = op1;
1767 u2.ll = op2;
1768 return float64_eq_quiet(u1.d, u2.d, &env->vec_status) ? 4 : 0;
1771 uint32_t helper_efdcmplt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1773 /* XXX: TODO: test special values (NaN, infinites, ...) */
1774 return helper_efdtstlt(env, op1, op2);
1777 uint32_t helper_efdcmpgt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1779 /* XXX: TODO: test special values (NaN, infinites, ...) */
1780 return helper_efdtstgt(env, op1, op2);
1783 uint32_t helper_efdcmpeq(CPUPPCState *env, uint64_t op1, uint64_t op2)
1785 /* XXX: TODO: test special values (NaN, infinites, ...) */
1786 return helper_efdtsteq(env, op1, op2);
1789 #define float64_to_float64(x, env) x
1792 /* VSX_ADD_SUB - VSX floating point add/subract
1793 * name - instruction mnemonic
1794 * op - operation (add or sub)
1795 * nels - number of elements (1, 2 or 4)
1796 * tp - type (float32 or float64)
1797 * fld - vsr_t field (VsrD(*) or VsrW(*))
1798 * sfprf - set FPRF
1800 #define VSX_ADD_SUB(name, op, nels, tp, fld, sfprf, r2sp) \
1801 void helper_##name(CPUPPCState *env, uint32_t opcode) \
1803 ppc_vsr_t xt, xa, xb; \
1804 int i; \
1806 getVSR(xA(opcode), &xa, env); \
1807 getVSR(xB(opcode), &xb, env); \
1808 getVSR(xT(opcode), &xt, env); \
1809 helper_reset_fpstatus(env); \
1811 for (i = 0; i < nels; i++) { \
1812 float_status tstat = env->fp_status; \
1813 set_float_exception_flags(0, &tstat); \
1814 xt.fld = tp##_##op(xa.fld, xb.fld, &tstat); \
1815 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1817 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1818 float_invalid_op_addsub(env, sfprf, GETPC(), \
1819 tp##_classify(xa.fld) | \
1820 tp##_classify(xb.fld)); \
1823 if (r2sp) { \
1824 xt.fld = helper_frsp(env, xt.fld); \
1827 if (sfprf) { \
1828 helper_compute_fprf_float64(env, xt.fld); \
1831 putVSR(xT(opcode), &xt, env); \
1832 do_float_check_status(env, GETPC()); \
1835 VSX_ADD_SUB(xsadddp, add, 1, float64, VsrD(0), 1, 0)
1836 VSX_ADD_SUB(xsaddsp, add, 1, float64, VsrD(0), 1, 1)
1837 VSX_ADD_SUB(xvadddp, add, 2, float64, VsrD(i), 0, 0)
1838 VSX_ADD_SUB(xvaddsp, add, 4, float32, VsrW(i), 0, 0)
1839 VSX_ADD_SUB(xssubdp, sub, 1, float64, VsrD(0), 1, 0)
1840 VSX_ADD_SUB(xssubsp, sub, 1, float64, VsrD(0), 1, 1)
1841 VSX_ADD_SUB(xvsubdp, sub, 2, float64, VsrD(i), 0, 0)
1842 VSX_ADD_SUB(xvsubsp, sub, 4, float32, VsrW(i), 0, 0)
1844 void helper_xsaddqp(CPUPPCState *env, uint32_t opcode)
1846 ppc_vsr_t xt, xa, xb;
1847 float_status tstat;
1849 getVSR(rA(opcode) + 32, &xa, env);
1850 getVSR(rB(opcode) + 32, &xb, env);
1851 getVSR(rD(opcode) + 32, &xt, env);
1852 helper_reset_fpstatus(env);
1854 tstat = env->fp_status;
1855 if (unlikely(Rc(opcode) != 0)) {
1856 tstat.float_rounding_mode = float_round_to_odd;
1859 set_float_exception_flags(0, &tstat);
1860 xt.f128 = float128_add(xa.f128, xb.f128, &tstat);
1861 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1863 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1864 float_invalid_op_addsub(env, 1, GETPC(),
1865 float128_classify(xa.f128) |
1866 float128_classify(xb.f128));
1869 helper_compute_fprf_float128(env, xt.f128);
1871 putVSR(rD(opcode) + 32, &xt, env);
1872 do_float_check_status(env, GETPC());
1875 /* VSX_MUL - VSX floating point multiply
1876 * op - instruction mnemonic
1877 * nels - number of elements (1, 2 or 4)
1878 * tp - type (float32 or float64)
1879 * fld - vsr_t field (VsrD(*) or VsrW(*))
1880 * sfprf - set FPRF
1882 #define VSX_MUL(op, nels, tp, fld, sfprf, r2sp) \
1883 void helper_##op(CPUPPCState *env, uint32_t opcode) \
1885 ppc_vsr_t xt, xa, xb; \
1886 int i; \
1888 getVSR(xA(opcode), &xa, env); \
1889 getVSR(xB(opcode), &xb, env); \
1890 getVSR(xT(opcode), &xt, env); \
1891 helper_reset_fpstatus(env); \
1893 for (i = 0; i < nels; i++) { \
1894 float_status tstat = env->fp_status; \
1895 set_float_exception_flags(0, &tstat); \
1896 xt.fld = tp##_mul(xa.fld, xb.fld, &tstat); \
1897 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1899 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1900 float_invalid_op_mul(env, sfprf, GETPC(), \
1901 tp##_classify(xa.fld) | \
1902 tp##_classify(xb.fld)); \
1905 if (r2sp) { \
1906 xt.fld = helper_frsp(env, xt.fld); \
1909 if (sfprf) { \
1910 helper_compute_fprf_float64(env, xt.fld); \
1914 putVSR(xT(opcode), &xt, env); \
1915 do_float_check_status(env, GETPC()); \
1918 VSX_MUL(xsmuldp, 1, float64, VsrD(0), 1, 0)
1919 VSX_MUL(xsmulsp, 1, float64, VsrD(0), 1, 1)
1920 VSX_MUL(xvmuldp, 2, float64, VsrD(i), 0, 0)
1921 VSX_MUL(xvmulsp, 4, float32, VsrW(i), 0, 0)
1923 void helper_xsmulqp(CPUPPCState *env, uint32_t opcode)
1925 ppc_vsr_t xt, xa, xb;
1926 float_status tstat;
1928 getVSR(rA(opcode) + 32, &xa, env);
1929 getVSR(rB(opcode) + 32, &xb, env);
1930 getVSR(rD(opcode) + 32, &xt, env);
1932 helper_reset_fpstatus(env);
1933 tstat = env->fp_status;
1934 if (unlikely(Rc(opcode) != 0)) {
1935 tstat.float_rounding_mode = float_round_to_odd;
1938 set_float_exception_flags(0, &tstat);
1939 xt.f128 = float128_mul(xa.f128, xb.f128, &tstat);
1940 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1942 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1943 float_invalid_op_mul(env, 1, GETPC(),
1944 float128_classify(xa.f128) |
1945 float128_classify(xb.f128));
1947 helper_compute_fprf_float128(env, xt.f128);
1949 putVSR(rD(opcode) + 32, &xt, env);
1950 do_float_check_status(env, GETPC());
1953 /* VSX_DIV - VSX floating point divide
1954 * op - instruction mnemonic
1955 * nels - number of elements (1, 2 or 4)
1956 * tp - type (float32 or float64)
1957 * fld - vsr_t field (VsrD(*) or VsrW(*))
1958 * sfprf - set FPRF
1960 #define VSX_DIV(op, nels, tp, fld, sfprf, r2sp) \
1961 void helper_##op(CPUPPCState *env, uint32_t opcode) \
1963 ppc_vsr_t xt, xa, xb; \
1964 int i; \
1966 getVSR(xA(opcode), &xa, env); \
1967 getVSR(xB(opcode), &xb, env); \
1968 getVSR(xT(opcode), &xt, env); \
1969 helper_reset_fpstatus(env); \
1971 for (i = 0; i < nels; i++) { \
1972 float_status tstat = env->fp_status; \
1973 set_float_exception_flags(0, &tstat); \
1974 xt.fld = tp##_div(xa.fld, xb.fld, &tstat); \
1975 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1977 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1978 float_invalid_op_div(env, sfprf, GETPC(), \
1979 tp##_classify(xa.fld) | \
1980 tp##_classify(xb.fld)); \
1982 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) { \
1983 float_zero_divide_excp(env, GETPC()); \
1986 if (r2sp) { \
1987 xt.fld = helper_frsp(env, xt.fld); \
1990 if (sfprf) { \
1991 helper_compute_fprf_float64(env, xt.fld); \
1995 putVSR(xT(opcode), &xt, env); \
1996 do_float_check_status(env, GETPC()); \
1999 VSX_DIV(xsdivdp, 1, float64, VsrD(0), 1, 0)
2000 VSX_DIV(xsdivsp, 1, float64, VsrD(0), 1, 1)
2001 VSX_DIV(xvdivdp, 2, float64, VsrD(i), 0, 0)
2002 VSX_DIV(xvdivsp, 4, float32, VsrW(i), 0, 0)
2004 void helper_xsdivqp(CPUPPCState *env, uint32_t opcode)
2006 ppc_vsr_t xt, xa, xb;
2007 float_status tstat;
2009 getVSR(rA(opcode) + 32, &xa, env);
2010 getVSR(rB(opcode) + 32, &xb, env);
2011 getVSR(rD(opcode) + 32, &xt, env);
2013 helper_reset_fpstatus(env);
2014 tstat = env->fp_status;
2015 if (unlikely(Rc(opcode) != 0)) {
2016 tstat.float_rounding_mode = float_round_to_odd;
2019 set_float_exception_flags(0, &tstat);
2020 xt.f128 = float128_div(xa.f128, xb.f128, &tstat);
2021 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
2023 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
2024 float_invalid_op_div(env, 1, GETPC(),
2025 float128_classify(xa.f128) |
2026 float128_classify(xb.f128));
2028 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) {
2029 float_zero_divide_excp(env, GETPC());
2032 helper_compute_fprf_float128(env, xt.f128);
2033 putVSR(rD(opcode) + 32, &xt, env);
2034 do_float_check_status(env, GETPC());
2037 /* VSX_RE - VSX floating point reciprocal estimate
2038 * op - instruction mnemonic
2039 * nels - number of elements (1, 2 or 4)
2040 * tp - type (float32 or float64)
2041 * fld - vsr_t field (VsrD(*) or VsrW(*))
2042 * sfprf - set FPRF
2044 #define VSX_RE(op, nels, tp, fld, sfprf, r2sp) \
2045 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2047 ppc_vsr_t xt, xb; \
2048 int i; \
2050 getVSR(xB(opcode), &xb, env); \
2051 getVSR(xT(opcode), &xt, env); \
2052 helper_reset_fpstatus(env); \
2054 for (i = 0; i < nels; i++) { \
2055 if (unlikely(tp##_is_signaling_nan(xb.fld, &env->fp_status))) { \
2056 float_invalid_op_vxsnan(env, GETPC()); \
2058 xt.fld = tp##_div(tp##_one, xb.fld, &env->fp_status); \
2060 if (r2sp) { \
2061 xt.fld = helper_frsp(env, xt.fld); \
2064 if (sfprf) { \
2065 helper_compute_fprf_float64(env, xt.fld); \
2069 putVSR(xT(opcode), &xt, env); \
2070 do_float_check_status(env, GETPC()); \
2073 VSX_RE(xsredp, 1, float64, VsrD(0), 1, 0)
2074 VSX_RE(xsresp, 1, float64, VsrD(0), 1, 1)
2075 VSX_RE(xvredp, 2, float64, VsrD(i), 0, 0)
2076 VSX_RE(xvresp, 4, float32, VsrW(i), 0, 0)
2078 /* VSX_SQRT - VSX floating point square root
2079 * op - instruction mnemonic
2080 * nels - number of elements (1, 2 or 4)
2081 * tp - type (float32 or float64)
2082 * fld - vsr_t field (VsrD(*) or VsrW(*))
2083 * sfprf - set FPRF
2085 #define VSX_SQRT(op, nels, tp, fld, sfprf, r2sp) \
2086 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2088 ppc_vsr_t xt, xb; \
2089 int i; \
2091 getVSR(xB(opcode), &xb, env); \
2092 getVSR(xT(opcode), &xt, env); \
2093 helper_reset_fpstatus(env); \
2095 for (i = 0; i < nels; i++) { \
2096 float_status tstat = env->fp_status; \
2097 set_float_exception_flags(0, &tstat); \
2098 xt.fld = tp##_sqrt(xb.fld, &tstat); \
2099 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
2101 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
2102 if (tp##_is_neg(xb.fld) && !tp##_is_zero(xb.fld)) { \
2103 float_invalid_op_vxsqrt(env, sfprf, GETPC()); \
2104 } else if (tp##_is_signaling_nan(xb.fld, &tstat)) { \
2105 float_invalid_op_vxsnan(env, GETPC()); \
2109 if (r2sp) { \
2110 xt.fld = helper_frsp(env, xt.fld); \
2113 if (sfprf) { \
2114 helper_compute_fprf_float64(env, xt.fld); \
2118 putVSR(xT(opcode), &xt, env); \
2119 do_float_check_status(env, GETPC()); \
2122 VSX_SQRT(xssqrtdp, 1, float64, VsrD(0), 1, 0)
2123 VSX_SQRT(xssqrtsp, 1, float64, VsrD(0), 1, 1)
2124 VSX_SQRT(xvsqrtdp, 2, float64, VsrD(i), 0, 0)
2125 VSX_SQRT(xvsqrtsp, 4, float32, VsrW(i), 0, 0)
2127 /* VSX_RSQRTE - VSX floating point reciprocal square root estimate
2128 * op - instruction mnemonic
2129 * nels - number of elements (1, 2 or 4)
2130 * tp - type (float32 or float64)
2131 * fld - vsr_t field (VsrD(*) or VsrW(*))
2132 * sfprf - set FPRF
2134 #define VSX_RSQRTE(op, nels, tp, fld, sfprf, r2sp) \
2135 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2137 ppc_vsr_t xt, xb; \
2138 int i; \
2140 getVSR(xB(opcode), &xb, env); \
2141 getVSR(xT(opcode), &xt, env); \
2142 helper_reset_fpstatus(env); \
2144 for (i = 0; i < nels; i++) { \
2145 float_status tstat = env->fp_status; \
2146 set_float_exception_flags(0, &tstat); \
2147 xt.fld = tp##_sqrt(xb.fld, &tstat); \
2148 xt.fld = tp##_div(tp##_one, xt.fld, &tstat); \
2149 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
2151 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
2152 if (tp##_is_neg(xb.fld) && !tp##_is_zero(xb.fld)) { \
2153 float_invalid_op_vxsqrt(env, sfprf, GETPC()); \
2154 } else if (tp##_is_signaling_nan(xb.fld, &tstat)) { \
2155 float_invalid_op_vxsnan(env, GETPC()); \
2159 if (r2sp) { \
2160 xt.fld = helper_frsp(env, xt.fld); \
2163 if (sfprf) { \
2164 helper_compute_fprf_float64(env, xt.fld); \
2168 putVSR(xT(opcode), &xt, env); \
2169 do_float_check_status(env, GETPC()); \
2172 VSX_RSQRTE(xsrsqrtedp, 1, float64, VsrD(0), 1, 0)
2173 VSX_RSQRTE(xsrsqrtesp, 1, float64, VsrD(0), 1, 1)
2174 VSX_RSQRTE(xvrsqrtedp, 2, float64, VsrD(i), 0, 0)
2175 VSX_RSQRTE(xvrsqrtesp, 4, float32, VsrW(i), 0, 0)
2177 /* VSX_TDIV - VSX floating point test for divide
2178 * op - instruction mnemonic
2179 * nels - number of elements (1, 2 or 4)
2180 * tp - type (float32 or float64)
2181 * fld - vsr_t field (VsrD(*) or VsrW(*))
2182 * emin - minimum unbiased exponent
2183 * emax - maximum unbiased exponent
2184 * nbits - number of fraction bits
2186 #define VSX_TDIV(op, nels, tp, fld, emin, emax, nbits) \
2187 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2189 ppc_vsr_t xa, xb; \
2190 int i; \
2191 int fe_flag = 0; \
2192 int fg_flag = 0; \
2194 getVSR(xA(opcode), &xa, env); \
2195 getVSR(xB(opcode), &xb, env); \
2197 for (i = 0; i < nels; i++) { \
2198 if (unlikely(tp##_is_infinity(xa.fld) || \
2199 tp##_is_infinity(xb.fld) || \
2200 tp##_is_zero(xb.fld))) { \
2201 fe_flag = 1; \
2202 fg_flag = 1; \
2203 } else { \
2204 int e_a = ppc_##tp##_get_unbiased_exp(xa.fld); \
2205 int e_b = ppc_##tp##_get_unbiased_exp(xb.fld); \
2207 if (unlikely(tp##_is_any_nan(xa.fld) || \
2208 tp##_is_any_nan(xb.fld))) { \
2209 fe_flag = 1; \
2210 } else if ((e_b <= emin) || (e_b >= (emax-2))) { \
2211 fe_flag = 1; \
2212 } else if (!tp##_is_zero(xa.fld) && \
2213 (((e_a - e_b) >= emax) || \
2214 ((e_a - e_b) <= (emin+1)) || \
2215 (e_a <= (emin+nbits)))) { \
2216 fe_flag = 1; \
2219 if (unlikely(tp##_is_zero_or_denormal(xb.fld))) { \
2220 /* XB is not zero because of the above check and */ \
2221 /* so must be denormalized. */ \
2222 fg_flag = 1; \
2227 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
2230 VSX_TDIV(xstdivdp, 1, float64, VsrD(0), -1022, 1023, 52)
2231 VSX_TDIV(xvtdivdp, 2, float64, VsrD(i), -1022, 1023, 52)
2232 VSX_TDIV(xvtdivsp, 4, float32, VsrW(i), -126, 127, 23)
2234 /* VSX_TSQRT - VSX floating point test for square root
2235 * op - instruction mnemonic
2236 * nels - number of elements (1, 2 or 4)
2237 * tp - type (float32 or float64)
2238 * fld - vsr_t field (VsrD(*) or VsrW(*))
2239 * emin - minimum unbiased exponent
2240 * emax - maximum unbiased exponent
2241 * nbits - number of fraction bits
2243 #define VSX_TSQRT(op, nels, tp, fld, emin, nbits) \
2244 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2246 ppc_vsr_t xa, xb; \
2247 int i; \
2248 int fe_flag = 0; \
2249 int fg_flag = 0; \
2251 getVSR(xA(opcode), &xa, env); \
2252 getVSR(xB(opcode), &xb, env); \
2254 for (i = 0; i < nels; i++) { \
2255 if (unlikely(tp##_is_infinity(xb.fld) || \
2256 tp##_is_zero(xb.fld))) { \
2257 fe_flag = 1; \
2258 fg_flag = 1; \
2259 } else { \
2260 int e_b = ppc_##tp##_get_unbiased_exp(xb.fld); \
2262 if (unlikely(tp##_is_any_nan(xb.fld))) { \
2263 fe_flag = 1; \
2264 } else if (unlikely(tp##_is_zero(xb.fld))) { \
2265 fe_flag = 1; \
2266 } else if (unlikely(tp##_is_neg(xb.fld))) { \
2267 fe_flag = 1; \
2268 } else if (!tp##_is_zero(xb.fld) && \
2269 (e_b <= (emin+nbits))) { \
2270 fe_flag = 1; \
2273 if (unlikely(tp##_is_zero_or_denormal(xb.fld))) { \
2274 /* XB is not zero because of the above check and */ \
2275 /* therefore must be denormalized. */ \
2276 fg_flag = 1; \
2281 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
2284 VSX_TSQRT(xstsqrtdp, 1, float64, VsrD(0), -1022, 52)
2285 VSX_TSQRT(xvtsqrtdp, 2, float64, VsrD(i), -1022, 52)
2286 VSX_TSQRT(xvtsqrtsp, 4, float32, VsrW(i), -126, 23)
2288 /* VSX_MADD - VSX floating point muliply/add variations
2289 * op - instruction mnemonic
2290 * nels - number of elements (1, 2 or 4)
2291 * tp - type (float32 or float64)
2292 * fld - vsr_t field (VsrD(*) or VsrW(*))
2293 * maddflgs - flags for the float*muladd routine that control the
2294 * various forms (madd, msub, nmadd, nmsub)
2295 * afrm - A form (1=A, 0=M)
2296 * sfprf - set FPRF
2298 #define VSX_MADD(op, nels, tp, fld, maddflgs, afrm, sfprf, r2sp) \
2299 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2301 ppc_vsr_t xt_in, xa, xb, xt_out; \
2302 ppc_vsr_t *b, *c; \
2303 int i; \
2305 if (afrm) { /* AxB + T */ \
2306 b = &xb; \
2307 c = &xt_in; \
2308 } else { /* AxT + B */ \
2309 b = &xt_in; \
2310 c = &xb; \
2313 getVSR(xA(opcode), &xa, env); \
2314 getVSR(xB(opcode), &xb, env); \
2315 getVSR(xT(opcode), &xt_in, env); \
2317 xt_out = xt_in; \
2319 helper_reset_fpstatus(env); \
2321 for (i = 0; i < nels; i++) { \
2322 float_status tstat = env->fp_status; \
2323 set_float_exception_flags(0, &tstat); \
2324 if (r2sp && (tstat.float_rounding_mode == float_round_nearest_even)) {\
2325 /* Avoid double rounding errors by rounding the intermediate */ \
2326 /* result to odd. */ \
2327 set_float_rounding_mode(float_round_to_zero, &tstat); \
2328 xt_out.fld = tp##_muladd(xa.fld, b->fld, c->fld, \
2329 maddflgs, &tstat); \
2330 xt_out.fld |= (get_float_exception_flags(&tstat) & \
2331 float_flag_inexact) != 0; \
2332 } else { \
2333 xt_out.fld = tp##_muladd(xa.fld, b->fld, c->fld, \
2334 maddflgs, &tstat); \
2336 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
2338 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
2339 tp##_maddsub_update_excp(env, xa.fld, b->fld, \
2340 c->fld, maddflgs, GETPC()); \
2343 if (r2sp) { \
2344 xt_out.fld = helper_frsp(env, xt_out.fld); \
2347 if (sfprf) { \
2348 helper_compute_fprf_float64(env, xt_out.fld); \
2351 putVSR(xT(opcode), &xt_out, env); \
2352 do_float_check_status(env, GETPC()); \
2355 VSX_MADD(xsmaddadp, 1, float64, VsrD(0), MADD_FLGS, 1, 1, 0)
2356 VSX_MADD(xsmaddmdp, 1, float64, VsrD(0), MADD_FLGS, 0, 1, 0)
2357 VSX_MADD(xsmsubadp, 1, float64, VsrD(0), MSUB_FLGS, 1, 1, 0)
2358 VSX_MADD(xsmsubmdp, 1, float64, VsrD(0), MSUB_FLGS, 0, 1, 0)
2359 VSX_MADD(xsnmaddadp, 1, float64, VsrD(0), NMADD_FLGS, 1, 1, 0)
2360 VSX_MADD(xsnmaddmdp, 1, float64, VsrD(0), NMADD_FLGS, 0, 1, 0)
2361 VSX_MADD(xsnmsubadp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 1, 0)
2362 VSX_MADD(xsnmsubmdp, 1, float64, VsrD(0), NMSUB_FLGS, 0, 1, 0)
2364 VSX_MADD(xsmaddasp, 1, float64, VsrD(0), MADD_FLGS, 1, 1, 1)
2365 VSX_MADD(xsmaddmsp, 1, float64, VsrD(0), MADD_FLGS, 0, 1, 1)
2366 VSX_MADD(xsmsubasp, 1, float64, VsrD(0), MSUB_FLGS, 1, 1, 1)
2367 VSX_MADD(xsmsubmsp, 1, float64, VsrD(0), MSUB_FLGS, 0, 1, 1)
2368 VSX_MADD(xsnmaddasp, 1, float64, VsrD(0), NMADD_FLGS, 1, 1, 1)
2369 VSX_MADD(xsnmaddmsp, 1, float64, VsrD(0), NMADD_FLGS, 0, 1, 1)
2370 VSX_MADD(xsnmsubasp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 1, 1)
2371 VSX_MADD(xsnmsubmsp, 1, float64, VsrD(0), NMSUB_FLGS, 0, 1, 1)
2373 VSX_MADD(xvmaddadp, 2, float64, VsrD(i), MADD_FLGS, 1, 0, 0)
2374 VSX_MADD(xvmaddmdp, 2, float64, VsrD(i), MADD_FLGS, 0, 0, 0)
2375 VSX_MADD(xvmsubadp, 2, float64, VsrD(i), MSUB_FLGS, 1, 0, 0)
2376 VSX_MADD(xvmsubmdp, 2, float64, VsrD(i), MSUB_FLGS, 0, 0, 0)
2377 VSX_MADD(xvnmaddadp, 2, float64, VsrD(i), NMADD_FLGS, 1, 0, 0)
2378 VSX_MADD(xvnmaddmdp, 2, float64, VsrD(i), NMADD_FLGS, 0, 0, 0)
2379 VSX_MADD(xvnmsubadp, 2, float64, VsrD(i), NMSUB_FLGS, 1, 0, 0)
2380 VSX_MADD(xvnmsubmdp, 2, float64, VsrD(i), NMSUB_FLGS, 0, 0, 0)
2382 VSX_MADD(xvmaddasp, 4, float32, VsrW(i), MADD_FLGS, 1, 0, 0)
2383 VSX_MADD(xvmaddmsp, 4, float32, VsrW(i), MADD_FLGS, 0, 0, 0)
2384 VSX_MADD(xvmsubasp, 4, float32, VsrW(i), MSUB_FLGS, 1, 0, 0)
2385 VSX_MADD(xvmsubmsp, 4, float32, VsrW(i), MSUB_FLGS, 0, 0, 0)
2386 VSX_MADD(xvnmaddasp, 4, float32, VsrW(i), NMADD_FLGS, 1, 0, 0)
2387 VSX_MADD(xvnmaddmsp, 4, float32, VsrW(i), NMADD_FLGS, 0, 0, 0)
2388 VSX_MADD(xvnmsubasp, 4, float32, VsrW(i), NMSUB_FLGS, 1, 0, 0)
2389 VSX_MADD(xvnmsubmsp, 4, float32, VsrW(i), NMSUB_FLGS, 0, 0, 0)
2391 /* VSX_SCALAR_CMP_DP - VSX scalar floating point compare double precision
2392 * op - instruction mnemonic
2393 * cmp - comparison operation
2394 * exp - expected result of comparison
2395 * svxvc - set VXVC bit
2397 #define VSX_SCALAR_CMP_DP(op, cmp, exp, svxvc) \
2398 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2400 ppc_vsr_t xt, xa, xb; \
2401 bool vxsnan_flag = false, vxvc_flag = false, vex_flag = false; \
2403 getVSR(xA(opcode), &xa, env); \
2404 getVSR(xB(opcode), &xb, env); \
2405 getVSR(xT(opcode), &xt, env); \
2407 if (float64_is_signaling_nan(xa.VsrD(0), &env->fp_status) || \
2408 float64_is_signaling_nan(xb.VsrD(0), &env->fp_status)) { \
2409 vxsnan_flag = true; \
2410 if (fpscr_ve == 0 && svxvc) { \
2411 vxvc_flag = true; \
2413 } else if (svxvc) { \
2414 vxvc_flag = float64_is_quiet_nan(xa.VsrD(0), &env->fp_status) || \
2415 float64_is_quiet_nan(xb.VsrD(0), &env->fp_status); \
2417 if (vxsnan_flag) { \
2418 float_invalid_op_vxsnan(env, GETPC()); \
2420 if (vxvc_flag) { \
2421 float_invalid_op_vxvc(env, 0, GETPC()); \
2423 vex_flag = fpscr_ve && (vxvc_flag || vxsnan_flag); \
2425 if (!vex_flag) { \
2426 if (float64_##cmp(xb.VsrD(0), xa.VsrD(0), &env->fp_status) == exp) { \
2427 xt.VsrD(0) = -1; \
2428 xt.VsrD(1) = 0; \
2429 } else { \
2430 xt.VsrD(0) = 0; \
2431 xt.VsrD(1) = 0; \
2434 putVSR(xT(opcode), &xt, env); \
2435 do_float_check_status(env, GETPC()); \
2438 VSX_SCALAR_CMP_DP(xscmpeqdp, eq, 1, 0)
2439 VSX_SCALAR_CMP_DP(xscmpgedp, le, 1, 1)
2440 VSX_SCALAR_CMP_DP(xscmpgtdp, lt, 1, 1)
2441 VSX_SCALAR_CMP_DP(xscmpnedp, eq, 0, 0)
2443 void helper_xscmpexpdp(CPUPPCState *env, uint32_t opcode)
2445 ppc_vsr_t xa, xb;
2446 int64_t exp_a, exp_b;
2447 uint32_t cc;
2449 getVSR(xA(opcode), &xa, env);
2450 getVSR(xB(opcode), &xb, env);
2452 exp_a = extract64(xa.VsrD(0), 52, 11);
2453 exp_b = extract64(xb.VsrD(0), 52, 11);
2455 if (unlikely(float64_is_any_nan(xa.VsrD(0)) ||
2456 float64_is_any_nan(xb.VsrD(0)))) {
2457 cc = CRF_SO;
2458 } else {
2459 if (exp_a < exp_b) {
2460 cc = CRF_LT;
2461 } else if (exp_a > exp_b) {
2462 cc = CRF_GT;
2463 } else {
2464 cc = CRF_EQ;
2468 env->fpscr &= ~(0x0F << FPSCR_FPRF);
2469 env->fpscr |= cc << FPSCR_FPRF;
2470 env->crf[BF(opcode)] = cc;
2472 do_float_check_status(env, GETPC());
2475 void helper_xscmpexpqp(CPUPPCState *env, uint32_t opcode)
2477 ppc_vsr_t xa, xb;
2478 int64_t exp_a, exp_b;
2479 uint32_t cc;
2481 getVSR(rA(opcode) + 32, &xa, env);
2482 getVSR(rB(opcode) + 32, &xb, env);
2484 exp_a = extract64(xa.VsrD(0), 48, 15);
2485 exp_b = extract64(xb.VsrD(0), 48, 15);
2487 if (unlikely(float128_is_any_nan(xa.f128) ||
2488 float128_is_any_nan(xb.f128))) {
2489 cc = CRF_SO;
2490 } else {
2491 if (exp_a < exp_b) {
2492 cc = CRF_LT;
2493 } else if (exp_a > exp_b) {
2494 cc = CRF_GT;
2495 } else {
2496 cc = CRF_EQ;
2500 env->fpscr &= ~(0x0F << FPSCR_FPRF);
2501 env->fpscr |= cc << FPSCR_FPRF;
2502 env->crf[BF(opcode)] = cc;
2504 do_float_check_status(env, GETPC());
2507 #define VSX_SCALAR_CMP(op, ordered) \
2508 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2510 ppc_vsr_t xa, xb; \
2511 uint32_t cc = 0; \
2512 bool vxsnan_flag = false, vxvc_flag = false; \
2514 helper_reset_fpstatus(env); \
2515 getVSR(xA(opcode), &xa, env); \
2516 getVSR(xB(opcode), &xb, env); \
2518 if (float64_is_signaling_nan(xa.VsrD(0), &env->fp_status) || \
2519 float64_is_signaling_nan(xb.VsrD(0), &env->fp_status)) { \
2520 vxsnan_flag = true; \
2521 cc = CRF_SO; \
2522 if (fpscr_ve == 0 && ordered) { \
2523 vxvc_flag = true; \
2525 } else if (float64_is_quiet_nan(xa.VsrD(0), &env->fp_status) || \
2526 float64_is_quiet_nan(xb.VsrD(0), &env->fp_status)) { \
2527 cc = CRF_SO; \
2528 if (ordered) { \
2529 vxvc_flag = true; \
2532 if (vxsnan_flag) { \
2533 float_invalid_op_vxsnan(env, GETPC()); \
2535 if (vxvc_flag) { \
2536 float_invalid_op_vxvc(env, 0, GETPC()); \
2539 if (float64_lt(xa.VsrD(0), xb.VsrD(0), &env->fp_status)) { \
2540 cc |= CRF_LT; \
2541 } else if (!float64_le(xa.VsrD(0), xb.VsrD(0), &env->fp_status)) { \
2542 cc |= CRF_GT; \
2543 } else { \
2544 cc |= CRF_EQ; \
2547 env->fpscr &= ~(0x0F << FPSCR_FPRF); \
2548 env->fpscr |= cc << FPSCR_FPRF; \
2549 env->crf[BF(opcode)] = cc; \
2551 do_float_check_status(env, GETPC()); \
2554 VSX_SCALAR_CMP(xscmpodp, 1)
2555 VSX_SCALAR_CMP(xscmpudp, 0)
2557 #define VSX_SCALAR_CMPQ(op, ordered) \
2558 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2560 ppc_vsr_t xa, xb; \
2561 uint32_t cc = 0; \
2562 bool vxsnan_flag = false, vxvc_flag = false; \
2564 helper_reset_fpstatus(env); \
2565 getVSR(rA(opcode) + 32, &xa, env); \
2566 getVSR(rB(opcode) + 32, &xb, env); \
2568 if (float128_is_signaling_nan(xa.f128, &env->fp_status) || \
2569 float128_is_signaling_nan(xb.f128, &env->fp_status)) { \
2570 vxsnan_flag = true; \
2571 cc = CRF_SO; \
2572 if (fpscr_ve == 0 && ordered) { \
2573 vxvc_flag = true; \
2575 } else if (float128_is_quiet_nan(xa.f128, &env->fp_status) || \
2576 float128_is_quiet_nan(xb.f128, &env->fp_status)) { \
2577 cc = CRF_SO; \
2578 if (ordered) { \
2579 vxvc_flag = true; \
2582 if (vxsnan_flag) { \
2583 float_invalid_op_vxsnan(env, GETPC()); \
2585 if (vxvc_flag) { \
2586 float_invalid_op_vxvc(env, 0, GETPC()); \
2589 if (float128_lt(xa.f128, xb.f128, &env->fp_status)) { \
2590 cc |= CRF_LT; \
2591 } else if (!float128_le(xa.f128, xb.f128, &env->fp_status)) { \
2592 cc |= CRF_GT; \
2593 } else { \
2594 cc |= CRF_EQ; \
2597 env->fpscr &= ~(0x0F << FPSCR_FPRF); \
2598 env->fpscr |= cc << FPSCR_FPRF; \
2599 env->crf[BF(opcode)] = cc; \
2601 do_float_check_status(env, GETPC()); \
2604 VSX_SCALAR_CMPQ(xscmpoqp, 1)
2605 VSX_SCALAR_CMPQ(xscmpuqp, 0)
2607 /* VSX_MAX_MIN - VSX floating point maximum/minimum
2608 * name - instruction mnemonic
2609 * op - operation (max or min)
2610 * nels - number of elements (1, 2 or 4)
2611 * tp - type (float32 or float64)
2612 * fld - vsr_t field (VsrD(*) or VsrW(*))
2614 #define VSX_MAX_MIN(name, op, nels, tp, fld) \
2615 void helper_##name(CPUPPCState *env, uint32_t opcode) \
2617 ppc_vsr_t xt, xa, xb; \
2618 int i; \
2620 getVSR(xA(opcode), &xa, env); \
2621 getVSR(xB(opcode), &xb, env); \
2622 getVSR(xT(opcode), &xt, env); \
2624 for (i = 0; i < nels; i++) { \
2625 xt.fld = tp##_##op(xa.fld, xb.fld, &env->fp_status); \
2626 if (unlikely(tp##_is_signaling_nan(xa.fld, &env->fp_status) || \
2627 tp##_is_signaling_nan(xb.fld, &env->fp_status))) { \
2628 float_invalid_op_vxsnan(env, GETPC()); \
2632 putVSR(xT(opcode), &xt, env); \
2633 do_float_check_status(env, GETPC()); \
2636 VSX_MAX_MIN(xsmaxdp, maxnum, 1, float64, VsrD(0))
2637 VSX_MAX_MIN(xvmaxdp, maxnum, 2, float64, VsrD(i))
2638 VSX_MAX_MIN(xvmaxsp, maxnum, 4, float32, VsrW(i))
2639 VSX_MAX_MIN(xsmindp, minnum, 1, float64, VsrD(0))
2640 VSX_MAX_MIN(xvmindp, minnum, 2, float64, VsrD(i))
2641 VSX_MAX_MIN(xvminsp, minnum, 4, float32, VsrW(i))
2643 #define VSX_MAX_MINC(name, max) \
2644 void helper_##name(CPUPPCState *env, uint32_t opcode) \
2646 ppc_vsr_t xt, xa, xb; \
2647 bool vxsnan_flag = false, vex_flag = false; \
2649 getVSR(rA(opcode) + 32, &xa, env); \
2650 getVSR(rB(opcode) + 32, &xb, env); \
2651 getVSR(rD(opcode) + 32, &xt, env); \
2653 if (unlikely(float64_is_any_nan(xa.VsrD(0)) || \
2654 float64_is_any_nan(xb.VsrD(0)))) { \
2655 if (float64_is_signaling_nan(xa.VsrD(0), &env->fp_status) || \
2656 float64_is_signaling_nan(xb.VsrD(0), &env->fp_status)) { \
2657 vxsnan_flag = true; \
2659 xt.VsrD(0) = xb.VsrD(0); \
2660 } else if ((max && \
2661 !float64_lt(xa.VsrD(0), xb.VsrD(0), &env->fp_status)) || \
2662 (!max && \
2663 float64_lt(xa.VsrD(0), xb.VsrD(0), &env->fp_status))) { \
2664 xt.VsrD(0) = xa.VsrD(0); \
2665 } else { \
2666 xt.VsrD(0) = xb.VsrD(0); \
2669 vex_flag = fpscr_ve & vxsnan_flag; \
2670 if (vxsnan_flag) { \
2671 float_invalid_op_vxsnan(env, GETPC()); \
2673 if (!vex_flag) { \
2674 putVSR(rD(opcode) + 32, &xt, env); \
2678 VSX_MAX_MINC(xsmaxcdp, 1);
2679 VSX_MAX_MINC(xsmincdp, 0);
2681 #define VSX_MAX_MINJ(name, max) \
2682 void helper_##name(CPUPPCState *env, uint32_t opcode) \
2684 ppc_vsr_t xt, xa, xb; \
2685 bool vxsnan_flag = false, vex_flag = false; \
2687 getVSR(rA(opcode) + 32, &xa, env); \
2688 getVSR(rB(opcode) + 32, &xb, env); \
2689 getVSR(rD(opcode) + 32, &xt, env); \
2691 if (unlikely(float64_is_any_nan(xa.VsrD(0)))) { \
2692 if (float64_is_signaling_nan(xa.VsrD(0), &env->fp_status)) { \
2693 vxsnan_flag = true; \
2695 xt.VsrD(0) = xa.VsrD(0); \
2696 } else if (unlikely(float64_is_any_nan(xb.VsrD(0)))) { \
2697 if (float64_is_signaling_nan(xb.VsrD(0), &env->fp_status)) { \
2698 vxsnan_flag = true; \
2700 xt.VsrD(0) = xb.VsrD(0); \
2701 } else if (float64_is_zero(xa.VsrD(0)) && float64_is_zero(xb.VsrD(0))) { \
2702 if (max) { \
2703 if (!float64_is_neg(xa.VsrD(0)) || !float64_is_neg(xb.VsrD(0))) { \
2704 xt.VsrD(0) = 0ULL; \
2705 } else { \
2706 xt.VsrD(0) = 0x8000000000000000ULL; \
2708 } else { \
2709 if (float64_is_neg(xa.VsrD(0)) || float64_is_neg(xb.VsrD(0))) { \
2710 xt.VsrD(0) = 0x8000000000000000ULL; \
2711 } else { \
2712 xt.VsrD(0) = 0ULL; \
2715 } else if ((max && \
2716 !float64_lt(xa.VsrD(0), xb.VsrD(0), &env->fp_status)) || \
2717 (!max && \
2718 float64_lt(xa.VsrD(0), xb.VsrD(0), &env->fp_status))) { \
2719 xt.VsrD(0) = xa.VsrD(0); \
2720 } else { \
2721 xt.VsrD(0) = xb.VsrD(0); \
2724 vex_flag = fpscr_ve & vxsnan_flag; \
2725 if (vxsnan_flag) { \
2726 float_invalid_op_vxsnan(env, GETPC()); \
2728 if (!vex_flag) { \
2729 putVSR(rD(opcode) + 32, &xt, env); \
2733 VSX_MAX_MINJ(xsmaxjdp, 1);
2734 VSX_MAX_MINJ(xsminjdp, 0);
2736 /* VSX_CMP - VSX floating point compare
2737 * op - instruction mnemonic
2738 * nels - number of elements (1, 2 or 4)
2739 * tp - type (float32 or float64)
2740 * fld - vsr_t field (VsrD(*) or VsrW(*))
2741 * cmp - comparison operation
2742 * svxvc - set VXVC bit
2743 * exp - expected result of comparison
2745 #define VSX_CMP(op, nels, tp, fld, cmp, svxvc, exp) \
2746 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2748 ppc_vsr_t xt, xa, xb; \
2749 int i; \
2750 int all_true = 1; \
2751 int all_false = 1; \
2753 getVSR(xA(opcode), &xa, env); \
2754 getVSR(xB(opcode), &xb, env); \
2755 getVSR(xT(opcode), &xt, env); \
2757 for (i = 0; i < nels; i++) { \
2758 if (unlikely(tp##_is_any_nan(xa.fld) || \
2759 tp##_is_any_nan(xb.fld))) { \
2760 if (tp##_is_signaling_nan(xa.fld, &env->fp_status) || \
2761 tp##_is_signaling_nan(xb.fld, &env->fp_status)) { \
2762 float_invalid_op_vxsnan(env, GETPC()); \
2764 if (svxvc) { \
2765 float_invalid_op_vxvc(env, 0, GETPC()); \
2767 xt.fld = 0; \
2768 all_true = 0; \
2769 } else { \
2770 if (tp##_##cmp(xb.fld, xa.fld, &env->fp_status) == exp) { \
2771 xt.fld = -1; \
2772 all_false = 0; \
2773 } else { \
2774 xt.fld = 0; \
2775 all_true = 0; \
2780 putVSR(xT(opcode), &xt, env); \
2781 if ((opcode >> (31-21)) & 1) { \
2782 env->crf[6] = (all_true ? 0x8 : 0) | (all_false ? 0x2 : 0); \
2784 do_float_check_status(env, GETPC()); \
2787 VSX_CMP(xvcmpeqdp, 2, float64, VsrD(i), eq, 0, 1)
2788 VSX_CMP(xvcmpgedp, 2, float64, VsrD(i), le, 1, 1)
2789 VSX_CMP(xvcmpgtdp, 2, float64, VsrD(i), lt, 1, 1)
2790 VSX_CMP(xvcmpnedp, 2, float64, VsrD(i), eq, 0, 0)
2791 VSX_CMP(xvcmpeqsp, 4, float32, VsrW(i), eq, 0, 1)
2792 VSX_CMP(xvcmpgesp, 4, float32, VsrW(i), le, 1, 1)
2793 VSX_CMP(xvcmpgtsp, 4, float32, VsrW(i), lt, 1, 1)
2794 VSX_CMP(xvcmpnesp, 4, float32, VsrW(i), eq, 0, 0)
2796 /* VSX_CVT_FP_TO_FP - VSX floating point/floating point conversion
2797 * op - instruction mnemonic
2798 * nels - number of elements (1, 2 or 4)
2799 * stp - source type (float32 or float64)
2800 * ttp - target type (float32 or float64)
2801 * sfld - source vsr_t field
2802 * tfld - target vsr_t field (f32 or f64)
2803 * sfprf - set FPRF
2805 #define VSX_CVT_FP_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf) \
2806 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2808 ppc_vsr_t xt, xb; \
2809 int i; \
2811 getVSR(xB(opcode), &xb, env); \
2812 getVSR(xT(opcode), &xt, env); \
2814 for (i = 0; i < nels; i++) { \
2815 xt.tfld = stp##_to_##ttp(xb.sfld, &env->fp_status); \
2816 if (unlikely(stp##_is_signaling_nan(xb.sfld, \
2817 &env->fp_status))) { \
2818 float_invalid_op_vxsnan(env, GETPC()); \
2819 xt.tfld = ttp##_snan_to_qnan(xt.tfld); \
2821 if (sfprf) { \
2822 helper_compute_fprf_##ttp(env, xt.tfld); \
2826 putVSR(xT(opcode), &xt, env); \
2827 do_float_check_status(env, GETPC()); \
2830 VSX_CVT_FP_TO_FP(xscvdpsp, 1, float64, float32, VsrD(0), VsrW(0), 1)
2831 VSX_CVT_FP_TO_FP(xscvspdp, 1, float32, float64, VsrW(0), VsrD(0), 1)
2832 VSX_CVT_FP_TO_FP(xvcvdpsp, 2, float64, float32, VsrD(i), VsrW(2*i), 0)
2833 VSX_CVT_FP_TO_FP(xvcvspdp, 2, float32, float64, VsrW(2*i), VsrD(i), 0)
2835 /* VSX_CVT_FP_TO_FP_VECTOR - VSX floating point/floating point conversion
2836 * op - instruction mnemonic
2837 * nels - number of elements (1, 2 or 4)
2838 * stp - source type (float32 or float64)
2839 * ttp - target type (float32 or float64)
2840 * sfld - source vsr_t field
2841 * tfld - target vsr_t field (f32 or f64)
2842 * sfprf - set FPRF
2844 #define VSX_CVT_FP_TO_FP_VECTOR(op, nels, stp, ttp, sfld, tfld, sfprf) \
2845 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2847 ppc_vsr_t xt, xb; \
2848 int i; \
2850 getVSR(rB(opcode) + 32, &xb, env); \
2851 getVSR(rD(opcode) + 32, &xt, env); \
2853 for (i = 0; i < nels; i++) { \
2854 xt.tfld = stp##_to_##ttp(xb.sfld, &env->fp_status); \
2855 if (unlikely(stp##_is_signaling_nan(xb.sfld, \
2856 &env->fp_status))) { \
2857 float_invalid_op_vxsnan(env, GETPC()); \
2858 xt.tfld = ttp##_snan_to_qnan(xt.tfld); \
2860 if (sfprf) { \
2861 helper_compute_fprf_##ttp(env, xt.tfld); \
2865 putVSR(rD(opcode) + 32, &xt, env); \
2866 do_float_check_status(env, GETPC()); \
2869 VSX_CVT_FP_TO_FP_VECTOR(xscvdpqp, 1, float64, float128, VsrD(0), f128, 1)
2871 /* VSX_CVT_FP_TO_FP_HP - VSX floating point/floating point conversion
2872 * involving one half precision value
2873 * op - instruction mnemonic
2874 * nels - number of elements (1, 2 or 4)
2875 * stp - source type
2876 * ttp - target type
2877 * sfld - source vsr_t field
2878 * tfld - target vsr_t field
2879 * sfprf - set FPRF
2881 #define VSX_CVT_FP_TO_FP_HP(op, nels, stp, ttp, sfld, tfld, sfprf) \
2882 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2884 ppc_vsr_t xt, xb; \
2885 int i; \
2887 getVSR(xB(opcode), &xb, env); \
2888 memset(&xt, 0, sizeof(xt)); \
2890 for (i = 0; i < nels; i++) { \
2891 xt.tfld = stp##_to_##ttp(xb.sfld, 1, &env->fp_status); \
2892 if (unlikely(stp##_is_signaling_nan(xb.sfld, \
2893 &env->fp_status))) { \
2894 float_invalid_op_vxsnan(env, GETPC()); \
2895 xt.tfld = ttp##_snan_to_qnan(xt.tfld); \
2897 if (sfprf) { \
2898 helper_compute_fprf_##ttp(env, xt.tfld); \
2902 putVSR(xT(opcode), &xt, env); \
2903 do_float_check_status(env, GETPC()); \
2906 VSX_CVT_FP_TO_FP_HP(xscvdphp, 1, float64, float16, VsrD(0), VsrH(3), 1)
2907 VSX_CVT_FP_TO_FP_HP(xscvhpdp, 1, float16, float64, VsrH(3), VsrD(0), 1)
2908 VSX_CVT_FP_TO_FP_HP(xvcvsphp, 4, float32, float16, VsrW(i), VsrH(2 * i + 1), 0)
2909 VSX_CVT_FP_TO_FP_HP(xvcvhpsp, 4, float16, float32, VsrH(2 * i + 1), VsrW(i), 0)
2912 * xscvqpdp isn't using VSX_CVT_FP_TO_FP() because xscvqpdpo will be
2913 * added to this later.
2915 void helper_xscvqpdp(CPUPPCState *env, uint32_t opcode)
2917 ppc_vsr_t xt, xb;
2918 float_status tstat;
2920 getVSR(rB(opcode) + 32, &xb, env);
2921 memset(&xt, 0, sizeof(xt));
2923 tstat = env->fp_status;
2924 if (unlikely(Rc(opcode) != 0)) {
2925 tstat.float_rounding_mode = float_round_to_odd;
2928 xt.VsrD(0) = float128_to_float64(xb.f128, &tstat);
2929 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
2930 if (unlikely(float128_is_signaling_nan(xb.f128, &tstat))) {
2931 float_invalid_op_vxsnan(env, GETPC());
2932 xt.VsrD(0) = float64_snan_to_qnan(xt.VsrD(0));
2934 helper_compute_fprf_float64(env, xt.VsrD(0));
2936 putVSR(rD(opcode) + 32, &xt, env);
2937 do_float_check_status(env, GETPC());
2940 uint64_t helper_xscvdpspn(CPUPPCState *env, uint64_t xb)
2942 float_status tstat = env->fp_status;
2943 set_float_exception_flags(0, &tstat);
2945 return (uint64_t)float64_to_float32(xb, &tstat) << 32;
2948 uint64_t helper_xscvspdpn(CPUPPCState *env, uint64_t xb)
2950 float_status tstat = env->fp_status;
2951 set_float_exception_flags(0, &tstat);
2953 return float32_to_float64(xb >> 32, &tstat);
2956 /* VSX_CVT_FP_TO_INT - VSX floating point to integer conversion
2957 * op - instruction mnemonic
2958 * nels - number of elements (1, 2 or 4)
2959 * stp - source type (float32 or float64)
2960 * ttp - target type (int32, uint32, int64 or uint64)
2961 * sfld - source vsr_t field
2962 * tfld - target vsr_t field
2963 * rnan - resulting NaN
2965 #define VSX_CVT_FP_TO_INT(op, nels, stp, ttp, sfld, tfld, rnan) \
2966 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2968 int all_flags = env->fp_status.float_exception_flags, flags; \
2969 ppc_vsr_t xt, xb; \
2970 int i; \
2972 getVSR(xB(opcode), &xb, env); \
2973 getVSR(xT(opcode), &xt, env); \
2975 for (i = 0; i < nels; i++) { \
2976 env->fp_status.float_exception_flags = 0; \
2977 xt.tfld = stp##_to_##ttp##_round_to_zero(xb.sfld, &env->fp_status); \
2978 flags = env->fp_status.float_exception_flags; \
2979 if (unlikely(flags & float_flag_invalid)) { \
2980 float_invalid_cvt(env, 0, GETPC(), stp##_classify(xb.sfld)); \
2981 xt.tfld = rnan; \
2983 all_flags |= flags; \
2986 putVSR(xT(opcode), &xt, env); \
2987 env->fp_status.float_exception_flags = all_flags; \
2988 do_float_check_status(env, GETPC()); \
2991 VSX_CVT_FP_TO_INT(xscvdpsxds, 1, float64, int64, VsrD(0), VsrD(0), \
2992 0x8000000000000000ULL)
2993 VSX_CVT_FP_TO_INT(xscvdpsxws, 1, float64, int32, VsrD(0), VsrW(1), \
2994 0x80000000U)
2995 VSX_CVT_FP_TO_INT(xscvdpuxds, 1, float64, uint64, VsrD(0), VsrD(0), 0ULL)
2996 VSX_CVT_FP_TO_INT(xscvdpuxws, 1, float64, uint32, VsrD(0), VsrW(1), 0U)
2997 VSX_CVT_FP_TO_INT(xvcvdpsxds, 2, float64, int64, VsrD(i), VsrD(i), \
2998 0x8000000000000000ULL)
2999 VSX_CVT_FP_TO_INT(xvcvdpsxws, 2, float64, int32, VsrD(i), VsrW(2*i), \
3000 0x80000000U)
3001 VSX_CVT_FP_TO_INT(xvcvdpuxds, 2, float64, uint64, VsrD(i), VsrD(i), 0ULL)
3002 VSX_CVT_FP_TO_INT(xvcvdpuxws, 2, float64, uint32, VsrD(i), VsrW(2*i), 0U)
3003 VSX_CVT_FP_TO_INT(xvcvspsxds, 2, float32, int64, VsrW(2*i), VsrD(i), \
3004 0x8000000000000000ULL)
3005 VSX_CVT_FP_TO_INT(xvcvspsxws, 4, float32, int32, VsrW(i), VsrW(i), 0x80000000U)
3006 VSX_CVT_FP_TO_INT(xvcvspuxds, 2, float32, uint64, VsrW(2*i), VsrD(i), 0ULL)
3007 VSX_CVT_FP_TO_INT(xvcvspuxws, 4, float32, uint32, VsrW(i), VsrW(i), 0U)
3009 /* VSX_CVT_FP_TO_INT_VECTOR - VSX floating point to integer conversion
3010 * op - instruction mnemonic
3011 * stp - source type (float32 or float64)
3012 * ttp - target type (int32, uint32, int64 or uint64)
3013 * sfld - source vsr_t field
3014 * tfld - target vsr_t field
3015 * rnan - resulting NaN
3017 #define VSX_CVT_FP_TO_INT_VECTOR(op, stp, ttp, sfld, tfld, rnan) \
3018 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3020 ppc_vsr_t xt, xb; \
3022 getVSR(rB(opcode) + 32, &xb, env); \
3023 memset(&xt, 0, sizeof(xt)); \
3025 xt.tfld = stp##_to_##ttp##_round_to_zero(xb.sfld, &env->fp_status); \
3026 if (env->fp_status.float_exception_flags & float_flag_invalid) { \
3027 float_invalid_cvt(env, 0, GETPC(), stp##_classify(xb.sfld)); \
3028 xt.tfld = rnan; \
3031 putVSR(rD(opcode) + 32, &xt, env); \
3032 do_float_check_status(env, GETPC()); \
3035 VSX_CVT_FP_TO_INT_VECTOR(xscvqpsdz, float128, int64, f128, VsrD(0), \
3036 0x8000000000000000ULL)
3038 VSX_CVT_FP_TO_INT_VECTOR(xscvqpswz, float128, int32, f128, VsrD(0), \
3039 0xffffffff80000000ULL)
3040 VSX_CVT_FP_TO_INT_VECTOR(xscvqpudz, float128, uint64, f128, VsrD(0), 0x0ULL)
3041 VSX_CVT_FP_TO_INT_VECTOR(xscvqpuwz, float128, uint32, f128, VsrD(0), 0x0ULL)
3043 /* VSX_CVT_INT_TO_FP - VSX integer to floating point conversion
3044 * op - instruction mnemonic
3045 * nels - number of elements (1, 2 or 4)
3046 * stp - source type (int32, uint32, int64 or uint64)
3047 * ttp - target type (float32 or float64)
3048 * sfld - source vsr_t field
3049 * tfld - target vsr_t field
3050 * jdef - definition of the j index (i or 2*i)
3051 * sfprf - set FPRF
3053 #define VSX_CVT_INT_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf, r2sp) \
3054 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3056 ppc_vsr_t xt, xb; \
3057 int i; \
3059 getVSR(xB(opcode), &xb, env); \
3060 getVSR(xT(opcode), &xt, env); \
3062 for (i = 0; i < nels; i++) { \
3063 xt.tfld = stp##_to_##ttp(xb.sfld, &env->fp_status); \
3064 if (r2sp) { \
3065 xt.tfld = helper_frsp(env, xt.tfld); \
3067 if (sfprf) { \
3068 helper_compute_fprf_float64(env, xt.tfld); \
3072 putVSR(xT(opcode), &xt, env); \
3073 do_float_check_status(env, GETPC()); \
3076 VSX_CVT_INT_TO_FP(xscvsxddp, 1, int64, float64, VsrD(0), VsrD(0), 1, 0)
3077 VSX_CVT_INT_TO_FP(xscvuxddp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 0)
3078 VSX_CVT_INT_TO_FP(xscvsxdsp, 1, int64, float64, VsrD(0), VsrD(0), 1, 1)
3079 VSX_CVT_INT_TO_FP(xscvuxdsp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 1)
3080 VSX_CVT_INT_TO_FP(xvcvsxddp, 2, int64, float64, VsrD(i), VsrD(i), 0, 0)
3081 VSX_CVT_INT_TO_FP(xvcvuxddp, 2, uint64, float64, VsrD(i), VsrD(i), 0, 0)
3082 VSX_CVT_INT_TO_FP(xvcvsxwdp, 2, int32, float64, VsrW(2*i), VsrD(i), 0, 0)
3083 VSX_CVT_INT_TO_FP(xvcvuxwdp, 2, uint64, float64, VsrW(2*i), VsrD(i), 0, 0)
3084 VSX_CVT_INT_TO_FP(xvcvsxdsp, 2, int64, float32, VsrD(i), VsrW(2*i), 0, 0)
3085 VSX_CVT_INT_TO_FP(xvcvuxdsp, 2, uint64, float32, VsrD(i), VsrW(2*i), 0, 0)
3086 VSX_CVT_INT_TO_FP(xvcvsxwsp, 4, int32, float32, VsrW(i), VsrW(i), 0, 0)
3087 VSX_CVT_INT_TO_FP(xvcvuxwsp, 4, uint32, float32, VsrW(i), VsrW(i), 0, 0)
3089 /* VSX_CVT_INT_TO_FP_VECTOR - VSX integer to floating point conversion
3090 * op - instruction mnemonic
3091 * stp - source type (int32, uint32, int64 or uint64)
3092 * ttp - target type (float32 or float64)
3093 * sfld - source vsr_t field
3094 * tfld - target vsr_t field
3096 #define VSX_CVT_INT_TO_FP_VECTOR(op, stp, ttp, sfld, tfld) \
3097 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3099 ppc_vsr_t xt, xb; \
3101 getVSR(rB(opcode) + 32, &xb, env); \
3102 getVSR(rD(opcode) + 32, &xt, env); \
3104 xt.tfld = stp##_to_##ttp(xb.sfld, &env->fp_status); \
3105 helper_compute_fprf_##ttp(env, xt.tfld); \
3107 putVSR(xT(opcode) + 32, &xt, env); \
3108 do_float_check_status(env, GETPC()); \
3111 VSX_CVT_INT_TO_FP_VECTOR(xscvsdqp, int64, float128, VsrD(0), f128)
3112 VSX_CVT_INT_TO_FP_VECTOR(xscvudqp, uint64, float128, VsrD(0), f128)
3114 /* For "use current rounding mode", define a value that will not be one of
3115 * the existing rounding model enums.
3117 #define FLOAT_ROUND_CURRENT (float_round_nearest_even + float_round_down + \
3118 float_round_up + float_round_to_zero)
3120 /* VSX_ROUND - VSX floating point round
3121 * op - instruction mnemonic
3122 * nels - number of elements (1, 2 or 4)
3123 * tp - type (float32 or float64)
3124 * fld - vsr_t field (VsrD(*) or VsrW(*))
3125 * rmode - rounding mode
3126 * sfprf - set FPRF
3128 #define VSX_ROUND(op, nels, tp, fld, rmode, sfprf) \
3129 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3131 ppc_vsr_t xt, xb; \
3132 int i; \
3133 getVSR(xB(opcode), &xb, env); \
3134 getVSR(xT(opcode), &xt, env); \
3136 if (rmode != FLOAT_ROUND_CURRENT) { \
3137 set_float_rounding_mode(rmode, &env->fp_status); \
3140 for (i = 0; i < nels; i++) { \
3141 if (unlikely(tp##_is_signaling_nan(xb.fld, \
3142 &env->fp_status))) { \
3143 float_invalid_op_vxsnan(env, GETPC()); \
3144 xt.fld = tp##_snan_to_qnan(xb.fld); \
3145 } else { \
3146 xt.fld = tp##_round_to_int(xb.fld, &env->fp_status); \
3148 if (sfprf) { \
3149 helper_compute_fprf_float64(env, xt.fld); \
3153 /* If this is not a "use current rounding mode" instruction, \
3154 * then inhibit setting of the XX bit and restore rounding \
3155 * mode from FPSCR */ \
3156 if (rmode != FLOAT_ROUND_CURRENT) { \
3157 fpscr_set_rounding_mode(env); \
3158 env->fp_status.float_exception_flags &= ~float_flag_inexact; \
3161 putVSR(xT(opcode), &xt, env); \
3162 do_float_check_status(env, GETPC()); \
3165 VSX_ROUND(xsrdpi, 1, float64, VsrD(0), float_round_ties_away, 1)
3166 VSX_ROUND(xsrdpic, 1, float64, VsrD(0), FLOAT_ROUND_CURRENT, 1)
3167 VSX_ROUND(xsrdpim, 1, float64, VsrD(0), float_round_down, 1)
3168 VSX_ROUND(xsrdpip, 1, float64, VsrD(0), float_round_up, 1)
3169 VSX_ROUND(xsrdpiz, 1, float64, VsrD(0), float_round_to_zero, 1)
3171 VSX_ROUND(xvrdpi, 2, float64, VsrD(i), float_round_ties_away, 0)
3172 VSX_ROUND(xvrdpic, 2, float64, VsrD(i), FLOAT_ROUND_CURRENT, 0)
3173 VSX_ROUND(xvrdpim, 2, float64, VsrD(i), float_round_down, 0)
3174 VSX_ROUND(xvrdpip, 2, float64, VsrD(i), float_round_up, 0)
3175 VSX_ROUND(xvrdpiz, 2, float64, VsrD(i), float_round_to_zero, 0)
3177 VSX_ROUND(xvrspi, 4, float32, VsrW(i), float_round_ties_away, 0)
3178 VSX_ROUND(xvrspic, 4, float32, VsrW(i), FLOAT_ROUND_CURRENT, 0)
3179 VSX_ROUND(xvrspim, 4, float32, VsrW(i), float_round_down, 0)
3180 VSX_ROUND(xvrspip, 4, float32, VsrW(i), float_round_up, 0)
3181 VSX_ROUND(xvrspiz, 4, float32, VsrW(i), float_round_to_zero, 0)
3183 uint64_t helper_xsrsp(CPUPPCState *env, uint64_t xb)
3185 helper_reset_fpstatus(env);
3187 uint64_t xt = helper_frsp(env, xb);
3189 helper_compute_fprf_float64(env, xt);
3190 do_float_check_status(env, GETPC());
3191 return xt;
3194 #define VSX_XXPERM(op, indexed) \
3195 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3197 ppc_vsr_t xt, xa, pcv, xto; \
3198 int i, idx; \
3200 getVSR(xA(opcode), &xa, env); \
3201 getVSR(xT(opcode), &xt, env); \
3202 getVSR(xB(opcode), &pcv, env); \
3204 for (i = 0; i < 16; i++) { \
3205 idx = pcv.VsrB(i) & 0x1F; \
3206 if (indexed) { \
3207 idx = 31 - idx; \
3209 xto.VsrB(i) = (idx <= 15) ? xa.VsrB(idx) : xt.VsrB(idx - 16); \
3211 putVSR(xT(opcode), &xto, env); \
3214 VSX_XXPERM(xxperm, 0)
3215 VSX_XXPERM(xxpermr, 1)
3217 void helper_xvxsigsp(CPUPPCState *env, uint32_t opcode)
3219 ppc_vsr_t xt, xb;
3220 uint32_t exp, i, fraction;
3222 getVSR(xB(opcode), &xb, env);
3223 memset(&xt, 0, sizeof(xt));
3225 for (i = 0; i < 4; i++) {
3226 exp = (xb.VsrW(i) >> 23) & 0xFF;
3227 fraction = xb.VsrW(i) & 0x7FFFFF;
3228 if (exp != 0 && exp != 255) {
3229 xt.VsrW(i) = fraction | 0x00800000;
3230 } else {
3231 xt.VsrW(i) = fraction;
3234 putVSR(xT(opcode), &xt, env);
3237 /* VSX_TEST_DC - VSX floating point test data class
3238 * op - instruction mnemonic
3239 * nels - number of elements (1, 2 or 4)
3240 * xbn - VSR register number
3241 * tp - type (float32 or float64)
3242 * fld - vsr_t field (VsrD(*) or VsrW(*))
3243 * tfld - target vsr_t field (VsrD(*) or VsrW(*))
3244 * fld_max - target field max
3245 * scrf - set result in CR and FPCC
3247 #define VSX_TEST_DC(op, nels, xbn, tp, fld, tfld, fld_max, scrf) \
3248 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3250 ppc_vsr_t xt, xb; \
3251 uint32_t i, sign, dcmx; \
3252 uint32_t cc, match = 0; \
3254 getVSR(xbn, &xb, env); \
3255 if (!scrf) { \
3256 memset(&xt, 0, sizeof(xt)); \
3257 dcmx = DCMX_XV(opcode); \
3258 } else { \
3259 dcmx = DCMX(opcode); \
3262 for (i = 0; i < nels; i++) { \
3263 sign = tp##_is_neg(xb.fld); \
3264 if (tp##_is_any_nan(xb.fld)) { \
3265 match = extract32(dcmx, 6, 1); \
3266 } else if (tp##_is_infinity(xb.fld)) { \
3267 match = extract32(dcmx, 4 + !sign, 1); \
3268 } else if (tp##_is_zero(xb.fld)) { \
3269 match = extract32(dcmx, 2 + !sign, 1); \
3270 } else if (tp##_is_zero_or_denormal(xb.fld)) { \
3271 match = extract32(dcmx, 0 + !sign, 1); \
3274 if (scrf) { \
3275 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT; \
3276 env->fpscr &= ~(0x0F << FPSCR_FPRF); \
3277 env->fpscr |= cc << FPSCR_FPRF; \
3278 env->crf[BF(opcode)] = cc; \
3279 } else { \
3280 xt.tfld = match ? fld_max : 0; \
3282 match = 0; \
3284 if (!scrf) { \
3285 putVSR(xT(opcode), &xt, env); \
3289 VSX_TEST_DC(xvtstdcdp, 2, xB(opcode), float64, VsrD(i), VsrD(i), UINT64_MAX, 0)
3290 VSX_TEST_DC(xvtstdcsp, 4, xB(opcode), float32, VsrW(i), VsrW(i), UINT32_MAX, 0)
3291 VSX_TEST_DC(xststdcdp, 1, xB(opcode), float64, VsrD(0), VsrD(0), 0, 1)
3292 VSX_TEST_DC(xststdcqp, 1, (rB(opcode) + 32), float128, f128, VsrD(0), 0, 1)
3294 void helper_xststdcsp(CPUPPCState *env, uint32_t opcode)
3296 ppc_vsr_t xb;
3297 uint32_t dcmx, sign, exp;
3298 uint32_t cc, match = 0, not_sp = 0;
3300 getVSR(xB(opcode), &xb, env);
3301 dcmx = DCMX(opcode);
3302 exp = (xb.VsrD(0) >> 52) & 0x7FF;
3304 sign = float64_is_neg(xb.VsrD(0));
3305 if (float64_is_any_nan(xb.VsrD(0))) {
3306 match = extract32(dcmx, 6, 1);
3307 } else if (float64_is_infinity(xb.VsrD(0))) {
3308 match = extract32(dcmx, 4 + !sign, 1);
3309 } else if (float64_is_zero(xb.VsrD(0))) {
3310 match = extract32(dcmx, 2 + !sign, 1);
3311 } else if (float64_is_zero_or_denormal(xb.VsrD(0)) ||
3312 (exp > 0 && exp < 0x381)) {
3313 match = extract32(dcmx, 0 + !sign, 1);
3316 not_sp = !float64_eq(xb.VsrD(0),
3317 float32_to_float64(
3318 float64_to_float32(xb.VsrD(0), &env->fp_status),
3319 &env->fp_status), &env->fp_status);
3321 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT | not_sp << CRF_SO_BIT;
3322 env->fpscr &= ~(0x0F << FPSCR_FPRF);
3323 env->fpscr |= cc << FPSCR_FPRF;
3324 env->crf[BF(opcode)] = cc;
3327 void helper_xsrqpi(CPUPPCState *env, uint32_t opcode)
3329 ppc_vsr_t xb;
3330 ppc_vsr_t xt;
3331 uint8_t r = Rrm(opcode);
3332 uint8_t ex = Rc(opcode);
3333 uint8_t rmc = RMC(opcode);
3334 uint8_t rmode = 0;
3335 float_status tstat;
3337 getVSR(rB(opcode) + 32, &xb, env);
3338 memset(&xt, 0, sizeof(xt));
3339 helper_reset_fpstatus(env);
3341 if (r == 0 && rmc == 0) {
3342 rmode = float_round_ties_away;
3343 } else if (r == 0 && rmc == 0x3) {
3344 rmode = fpscr_rn;
3345 } else if (r == 1) {
3346 switch (rmc) {
3347 case 0:
3348 rmode = float_round_nearest_even;
3349 break;
3350 case 1:
3351 rmode = float_round_to_zero;
3352 break;
3353 case 2:
3354 rmode = float_round_up;
3355 break;
3356 case 3:
3357 rmode = float_round_down;
3358 break;
3359 default:
3360 abort();
3364 tstat = env->fp_status;
3365 set_float_exception_flags(0, &tstat);
3366 set_float_rounding_mode(rmode, &tstat);
3367 xt.f128 = float128_round_to_int(xb.f128, &tstat);
3368 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3370 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3371 if (float128_is_signaling_nan(xb.f128, &tstat)) {
3372 float_invalid_op_vxsnan(env, GETPC());
3373 xt.f128 = float128_snan_to_qnan(xt.f128);
3377 if (ex == 0 && (tstat.float_exception_flags & float_flag_inexact)) {
3378 env->fp_status.float_exception_flags &= ~float_flag_inexact;
3381 helper_compute_fprf_float128(env, xt.f128);
3382 do_float_check_status(env, GETPC());
3383 putVSR(rD(opcode) + 32, &xt, env);
3386 void helper_xsrqpxp(CPUPPCState *env, uint32_t opcode)
3388 ppc_vsr_t xb;
3389 ppc_vsr_t xt;
3390 uint8_t r = Rrm(opcode);
3391 uint8_t rmc = RMC(opcode);
3392 uint8_t rmode = 0;
3393 floatx80 round_res;
3394 float_status tstat;
3396 getVSR(rB(opcode) + 32, &xb, env);
3397 memset(&xt, 0, sizeof(xt));
3398 helper_reset_fpstatus(env);
3400 if (r == 0 && rmc == 0) {
3401 rmode = float_round_ties_away;
3402 } else if (r == 0 && rmc == 0x3) {
3403 rmode = fpscr_rn;
3404 } else if (r == 1) {
3405 switch (rmc) {
3406 case 0:
3407 rmode = float_round_nearest_even;
3408 break;
3409 case 1:
3410 rmode = float_round_to_zero;
3411 break;
3412 case 2:
3413 rmode = float_round_up;
3414 break;
3415 case 3:
3416 rmode = float_round_down;
3417 break;
3418 default:
3419 abort();
3423 tstat = env->fp_status;
3424 set_float_exception_flags(0, &tstat);
3425 set_float_rounding_mode(rmode, &tstat);
3426 round_res = float128_to_floatx80(xb.f128, &tstat);
3427 xt.f128 = floatx80_to_float128(round_res, &tstat);
3428 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3430 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3431 if (float128_is_signaling_nan(xb.f128, &tstat)) {
3432 float_invalid_op_vxsnan(env, GETPC());
3433 xt.f128 = float128_snan_to_qnan(xt.f128);
3437 helper_compute_fprf_float128(env, xt.f128);
3438 putVSR(rD(opcode) + 32, &xt, env);
3439 do_float_check_status(env, GETPC());
3442 void helper_xssqrtqp(CPUPPCState *env, uint32_t opcode)
3444 ppc_vsr_t xb;
3445 ppc_vsr_t xt;
3446 float_status tstat;
3448 getVSR(rB(opcode) + 32, &xb, env);
3449 memset(&xt, 0, sizeof(xt));
3450 helper_reset_fpstatus(env);
3452 tstat = env->fp_status;
3453 if (unlikely(Rc(opcode) != 0)) {
3454 tstat.float_rounding_mode = float_round_to_odd;
3457 set_float_exception_flags(0, &tstat);
3458 xt.f128 = float128_sqrt(xb.f128, &tstat);
3459 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3461 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3462 if (float128_is_signaling_nan(xb.f128, &tstat)) {
3463 float_invalid_op_vxsnan(env, GETPC());
3464 xt.f128 = float128_snan_to_qnan(xb.f128);
3465 } else if (float128_is_quiet_nan(xb.f128, &tstat)) {
3466 xt.f128 = xb.f128;
3467 } else if (float128_is_neg(xb.f128) && !float128_is_zero(xb.f128)) {
3468 float_invalid_op_vxsqrt(env, 1, GETPC());
3469 xt.f128 = float128_default_nan(&env->fp_status);
3473 helper_compute_fprf_float128(env, xt.f128);
3474 putVSR(rD(opcode) + 32, &xt, env);
3475 do_float_check_status(env, GETPC());
3478 void helper_xssubqp(CPUPPCState *env, uint32_t opcode)
3480 ppc_vsr_t xt, xa, xb;
3481 float_status tstat;
3483 getVSR(rA(opcode) + 32, &xa, env);
3484 getVSR(rB(opcode) + 32, &xb, env);
3485 getVSR(rD(opcode) + 32, &xt, env);
3486 helper_reset_fpstatus(env);
3488 tstat = env->fp_status;
3489 if (unlikely(Rc(opcode) != 0)) {
3490 tstat.float_rounding_mode = float_round_to_odd;
3493 set_float_exception_flags(0, &tstat);
3494 xt.f128 = float128_sub(xa.f128, xb.f128, &tstat);
3495 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3497 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3498 float_invalid_op_addsub(env, 1, GETPC(),
3499 float128_classify(xa.f128) |
3500 float128_classify(xb.f128));
3503 helper_compute_fprf_float128(env, xt.f128);
3504 putVSR(rD(opcode) + 32, &xt, env);
3505 do_float_check_status(env, GETPC());