target/ppc: introduce separate generator and helper for xscvqpdp
[qemu/rayw.git] / target / ppc / fpu_helper.c
blob230ee2f07240c6abb1cdbf3143f61d05b9d21c3c
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 {
94 * Zero or Denormal result. If the exponent is in bounds for
95 * a single-precision denormal result, extract the proper
96 * bits. If the input is not zero, and the exponent is out of
97 * bounds, then the result is undefined; this underflows to
98 * zero.
100 ret = extract64(arg, 63, 1) << 31;
101 if (unlikely(exp >= 874)) {
102 /* Denormal result. */
103 ret |= ((1ULL << 52) | extract64(arg, 0, 52)) >> (896 + 30 - exp);
106 return ret;
109 static inline int ppc_float32_get_unbiased_exp(float32 f)
111 return ((f >> 23) & 0xFF) - 127;
114 static inline int ppc_float64_get_unbiased_exp(float64 f)
116 return ((f >> 52) & 0x7FF) - 1023;
119 /* Classify a floating-point number. */
120 enum {
121 is_normal = 1,
122 is_zero = 2,
123 is_denormal = 4,
124 is_inf = 8,
125 is_qnan = 16,
126 is_snan = 32,
127 is_neg = 64,
130 #define COMPUTE_CLASS(tp) \
131 static int tp##_classify(tp arg) \
133 int ret = tp##_is_neg(arg) * is_neg; \
134 if (unlikely(tp##_is_any_nan(arg))) { \
135 float_status dummy = { }; /* snan_bit_is_one = 0 */ \
136 ret |= (tp##_is_signaling_nan(arg, &dummy) \
137 ? is_snan : is_qnan); \
138 } else if (unlikely(tp##_is_infinity(arg))) { \
139 ret |= is_inf; \
140 } else if (tp##_is_zero(arg)) { \
141 ret |= is_zero; \
142 } else if (tp##_is_zero_or_denormal(arg)) { \
143 ret |= is_denormal; \
144 } else { \
145 ret |= is_normal; \
147 return ret; \
150 COMPUTE_CLASS(float16)
151 COMPUTE_CLASS(float32)
152 COMPUTE_CLASS(float64)
153 COMPUTE_CLASS(float128)
155 static void set_fprf_from_class(CPUPPCState *env, int class)
157 static const uint8_t fprf[6][2] = {
158 { 0x04, 0x08 }, /* normalized */
159 { 0x02, 0x12 }, /* zero */
160 { 0x14, 0x18 }, /* denormalized */
161 { 0x05, 0x09 }, /* infinity */
162 { 0x11, 0x11 }, /* qnan */
163 { 0x00, 0x00 }, /* snan -- flags are undefined */
165 bool isneg = class & is_neg;
167 env->fpscr &= ~(0x1F << FPSCR_FPRF);
168 env->fpscr |= fprf[ctz32(class)][isneg] << FPSCR_FPRF;
171 #define COMPUTE_FPRF(tp) \
172 void helper_compute_fprf_##tp(CPUPPCState *env, tp arg) \
174 set_fprf_from_class(env, tp##_classify(arg)); \
177 COMPUTE_FPRF(float16)
178 COMPUTE_FPRF(float32)
179 COMPUTE_FPRF(float64)
180 COMPUTE_FPRF(float128)
182 /* Floating-point invalid operations exception */
183 static void finish_invalid_op_excp(CPUPPCState *env, int op, uintptr_t retaddr)
185 /* Update the floating-point invalid operation summary */
186 env->fpscr |= 1 << FPSCR_VX;
187 /* Update the floating-point exception summary */
188 env->fpscr |= FP_FX;
189 if (fpscr_ve != 0) {
190 /* Update the floating-point enabled exception summary */
191 env->fpscr |= 1 << FPSCR_FEX;
192 if (fp_exceptions_enabled(env)) {
193 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
194 POWERPC_EXCP_FP | op, retaddr);
199 static void finish_invalid_op_arith(CPUPPCState *env, int op,
200 bool set_fpcc, uintptr_t retaddr)
202 env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI));
203 if (fpscr_ve == 0) {
204 if (set_fpcc) {
205 env->fpscr &= ~(0xF << FPSCR_FPCC);
206 env->fpscr |= 0x11 << FPSCR_FPCC;
209 finish_invalid_op_excp(env, op, retaddr);
212 /* Signalling NaN */
213 static void float_invalid_op_vxsnan(CPUPPCState *env, uintptr_t retaddr)
215 env->fpscr |= 1 << FPSCR_VXSNAN;
216 finish_invalid_op_excp(env, POWERPC_EXCP_FP_VXSNAN, retaddr);
219 /* Magnitude subtraction of infinities */
220 static void float_invalid_op_vxisi(CPUPPCState *env, bool set_fpcc,
221 uintptr_t retaddr)
223 env->fpscr |= 1 << FPSCR_VXISI;
224 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXISI, set_fpcc, retaddr);
227 /* Division of infinity by infinity */
228 static void float_invalid_op_vxidi(CPUPPCState *env, bool set_fpcc,
229 uintptr_t retaddr)
231 env->fpscr |= 1 << FPSCR_VXIDI;
232 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXIDI, set_fpcc, retaddr);
235 /* Division of zero by zero */
236 static void float_invalid_op_vxzdz(CPUPPCState *env, bool set_fpcc,
237 uintptr_t retaddr)
239 env->fpscr |= 1 << FPSCR_VXZDZ;
240 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXZDZ, set_fpcc, retaddr);
243 /* Multiplication of zero by infinity */
244 static void float_invalid_op_vximz(CPUPPCState *env, bool set_fpcc,
245 uintptr_t retaddr)
247 env->fpscr |= 1 << FPSCR_VXIMZ;
248 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXIMZ, set_fpcc, retaddr);
251 /* Square root of a negative number */
252 static void float_invalid_op_vxsqrt(CPUPPCState *env, bool set_fpcc,
253 uintptr_t retaddr)
255 env->fpscr |= 1 << FPSCR_VXSQRT;
256 finish_invalid_op_arith(env, POWERPC_EXCP_FP_VXSQRT, set_fpcc, retaddr);
259 /* Ordered comparison of NaN */
260 static void float_invalid_op_vxvc(CPUPPCState *env, bool set_fpcc,
261 uintptr_t retaddr)
263 env->fpscr |= 1 << FPSCR_VXVC;
264 if (set_fpcc) {
265 env->fpscr &= ~(0xF << FPSCR_FPCC);
266 env->fpscr |= 0x11 << FPSCR_FPCC;
268 /* Update the floating-point invalid operation summary */
269 env->fpscr |= 1 << FPSCR_VX;
270 /* Update the floating-point exception summary */
271 env->fpscr |= FP_FX;
272 /* We must update the target FPR before raising the exception */
273 if (fpscr_ve != 0) {
274 CPUState *cs = env_cpu(env);
276 cs->exception_index = POWERPC_EXCP_PROGRAM;
277 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_VXVC;
278 /* Update the floating-point enabled exception summary */
279 env->fpscr |= 1 << FPSCR_FEX;
280 /* Exception is differed */
284 /* Invalid conversion */
285 static void float_invalid_op_vxcvi(CPUPPCState *env, bool set_fpcc,
286 uintptr_t retaddr)
288 env->fpscr |= 1 << FPSCR_VXCVI;
289 env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI));
290 if (fpscr_ve == 0) {
291 if (set_fpcc) {
292 env->fpscr &= ~(0xF << FPSCR_FPCC);
293 env->fpscr |= 0x11 << FPSCR_FPCC;
296 finish_invalid_op_excp(env, POWERPC_EXCP_FP_VXCVI, retaddr);
299 static inline void float_zero_divide_excp(CPUPPCState *env, uintptr_t raddr)
301 env->fpscr |= 1 << FPSCR_ZX;
302 env->fpscr &= ~((1 << FPSCR_FR) | (1 << FPSCR_FI));
303 /* Update the floating-point exception summary */
304 env->fpscr |= FP_FX;
305 if (fpscr_ze != 0) {
306 /* Update the floating-point enabled exception summary */
307 env->fpscr |= 1 << FPSCR_FEX;
308 if (fp_exceptions_enabled(env)) {
309 raise_exception_err_ra(env, POWERPC_EXCP_PROGRAM,
310 POWERPC_EXCP_FP | POWERPC_EXCP_FP_ZX,
311 raddr);
316 static inline void float_overflow_excp(CPUPPCState *env)
318 CPUState *cs = env_cpu(env);
320 env->fpscr |= 1 << FPSCR_OX;
321 /* Update the floating-point exception summary */
322 env->fpscr |= FP_FX;
323 if (fpscr_oe != 0) {
324 /* XXX: should adjust the result */
325 /* Update the floating-point enabled exception summary */
326 env->fpscr |= 1 << FPSCR_FEX;
327 /* We must update the target FPR before raising the exception */
328 cs->exception_index = POWERPC_EXCP_PROGRAM;
329 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_OX;
330 } else {
331 env->fpscr |= 1 << FPSCR_XX;
332 env->fpscr |= 1 << FPSCR_FI;
336 static inline void float_underflow_excp(CPUPPCState *env)
338 CPUState *cs = env_cpu(env);
340 env->fpscr |= 1 << FPSCR_UX;
341 /* Update the floating-point exception summary */
342 env->fpscr |= FP_FX;
343 if (fpscr_ue != 0) {
344 /* XXX: should adjust the result */
345 /* Update the floating-point enabled exception summary */
346 env->fpscr |= 1 << FPSCR_FEX;
347 /* We must update the target FPR before raising the exception */
348 cs->exception_index = POWERPC_EXCP_PROGRAM;
349 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_UX;
353 static inline void float_inexact_excp(CPUPPCState *env)
355 CPUState *cs = env_cpu(env);
357 env->fpscr |= 1 << FPSCR_FI;
358 env->fpscr |= 1 << FPSCR_XX;
359 /* Update the floating-point exception summary */
360 env->fpscr |= FP_FX;
361 if (fpscr_xe != 0) {
362 /* Update the floating-point enabled exception summary */
363 env->fpscr |= 1 << FPSCR_FEX;
364 /* We must update the target FPR before raising the exception */
365 cs->exception_index = POWERPC_EXCP_PROGRAM;
366 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_XX;
370 static inline void fpscr_set_rounding_mode(CPUPPCState *env)
372 int rnd_type;
374 /* Set rounding mode */
375 switch (fpscr_rn) {
376 case 0:
377 /* Best approximation (round to nearest) */
378 rnd_type = float_round_nearest_even;
379 break;
380 case 1:
381 /* Smaller magnitude (round toward zero) */
382 rnd_type = float_round_to_zero;
383 break;
384 case 2:
385 /* Round toward +infinite */
386 rnd_type = float_round_up;
387 break;
388 default:
389 case 3:
390 /* Round toward -infinite */
391 rnd_type = float_round_down;
392 break;
394 set_float_rounding_mode(rnd_type, &env->fp_status);
397 void helper_fpscr_clrbit(CPUPPCState *env, uint32_t bit)
399 int prev;
401 prev = (env->fpscr >> bit) & 1;
402 env->fpscr &= ~(1 << bit);
403 if (prev == 1) {
404 switch (bit) {
405 case FPSCR_RN1:
406 case FPSCR_RN:
407 fpscr_set_rounding_mode(env);
408 break;
409 case FPSCR_VXSNAN:
410 case FPSCR_VXISI:
411 case FPSCR_VXIDI:
412 case FPSCR_VXZDZ:
413 case FPSCR_VXIMZ:
414 case FPSCR_VXVC:
415 case FPSCR_VXSOFT:
416 case FPSCR_VXSQRT:
417 case FPSCR_VXCVI:
418 if (!fpscr_ix) {
419 /* Set VX bit to zero */
420 env->fpscr &= ~(1 << FPSCR_VX);
422 break;
423 case FPSCR_OX:
424 case FPSCR_UX:
425 case FPSCR_ZX:
426 case FPSCR_XX:
427 case FPSCR_VE:
428 case FPSCR_OE:
429 case FPSCR_UE:
430 case FPSCR_ZE:
431 case FPSCR_XE:
432 if (!fpscr_eex) {
433 /* Set the FEX bit */
434 env->fpscr &= ~(1 << FPSCR_FEX);
436 break;
437 default:
438 break;
443 void helper_fpscr_setbit(CPUPPCState *env, uint32_t bit)
445 CPUState *cs = env_cpu(env);
446 int prev;
448 prev = (env->fpscr >> bit) & 1;
449 env->fpscr |= 1 << bit;
450 if (prev == 0) {
451 switch (bit) {
452 case FPSCR_VX:
453 env->fpscr |= FP_FX;
454 if (fpscr_ve) {
455 goto raise_ve;
457 break;
458 case FPSCR_OX:
459 env->fpscr |= FP_FX;
460 if (fpscr_oe) {
461 goto raise_oe;
463 break;
464 case FPSCR_UX:
465 env->fpscr |= FP_FX;
466 if (fpscr_ue) {
467 goto raise_ue;
469 break;
470 case FPSCR_ZX:
471 env->fpscr |= FP_FX;
472 if (fpscr_ze) {
473 goto raise_ze;
475 break;
476 case FPSCR_XX:
477 env->fpscr |= FP_FX;
478 if (fpscr_xe) {
479 goto raise_xe;
481 break;
482 case FPSCR_VXSNAN:
483 case FPSCR_VXISI:
484 case FPSCR_VXIDI:
485 case FPSCR_VXZDZ:
486 case FPSCR_VXIMZ:
487 case FPSCR_VXVC:
488 case FPSCR_VXSOFT:
489 case FPSCR_VXSQRT:
490 case FPSCR_VXCVI:
491 env->fpscr |= 1 << FPSCR_VX;
492 env->fpscr |= FP_FX;
493 if (fpscr_ve != 0) {
494 goto raise_ve;
496 break;
497 case FPSCR_VE:
498 if (fpscr_vx != 0) {
499 raise_ve:
500 env->error_code = POWERPC_EXCP_FP;
501 if (fpscr_vxsnan) {
502 env->error_code |= POWERPC_EXCP_FP_VXSNAN;
504 if (fpscr_vxisi) {
505 env->error_code |= POWERPC_EXCP_FP_VXISI;
507 if (fpscr_vxidi) {
508 env->error_code |= POWERPC_EXCP_FP_VXIDI;
510 if (fpscr_vxzdz) {
511 env->error_code |= POWERPC_EXCP_FP_VXZDZ;
513 if (fpscr_vximz) {
514 env->error_code |= POWERPC_EXCP_FP_VXIMZ;
516 if (fpscr_vxvc) {
517 env->error_code |= POWERPC_EXCP_FP_VXVC;
519 if (fpscr_vxsoft) {
520 env->error_code |= POWERPC_EXCP_FP_VXSOFT;
522 if (fpscr_vxsqrt) {
523 env->error_code |= POWERPC_EXCP_FP_VXSQRT;
525 if (fpscr_vxcvi) {
526 env->error_code |= POWERPC_EXCP_FP_VXCVI;
528 goto raise_excp;
530 break;
531 case FPSCR_OE:
532 if (fpscr_ox != 0) {
533 raise_oe:
534 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_OX;
535 goto raise_excp;
537 break;
538 case FPSCR_UE:
539 if (fpscr_ux != 0) {
540 raise_ue:
541 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_UX;
542 goto raise_excp;
544 break;
545 case FPSCR_ZE:
546 if (fpscr_zx != 0) {
547 raise_ze:
548 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_ZX;
549 goto raise_excp;
551 break;
552 case FPSCR_XE:
553 if (fpscr_xx != 0) {
554 raise_xe:
555 env->error_code = POWERPC_EXCP_FP | POWERPC_EXCP_FP_XX;
556 goto raise_excp;
558 break;
559 case FPSCR_RN1:
560 case FPSCR_RN:
561 fpscr_set_rounding_mode(env);
562 break;
563 default:
564 break;
565 raise_excp:
566 /* Update the floating-point enabled exception summary */
567 env->fpscr |= 1 << FPSCR_FEX;
568 /* We have to update Rc1 before raising the exception */
569 cs->exception_index = POWERPC_EXCP_PROGRAM;
570 break;
575 void helper_store_fpscr(CPUPPCState *env, uint64_t arg, uint32_t mask)
577 CPUState *cs = env_cpu(env);
578 target_ulong prev, new;
579 int i;
581 prev = env->fpscr;
582 new = (target_ulong)arg;
583 new &= ~0x60000000LL;
584 new |= prev & 0x60000000LL;
585 for (i = 0; i < sizeof(target_ulong) * 2; i++) {
586 if (mask & (1 << i)) {
587 env->fpscr &= ~(0xFLL << (4 * i));
588 env->fpscr |= new & (0xFLL << (4 * i));
591 /* Update VX and FEX */
592 if (fpscr_ix != 0) {
593 env->fpscr |= 1 << FPSCR_VX;
594 } else {
595 env->fpscr &= ~(1 << FPSCR_VX);
597 if ((fpscr_ex & fpscr_eex) != 0) {
598 env->fpscr |= 1 << FPSCR_FEX;
599 cs->exception_index = POWERPC_EXCP_PROGRAM;
600 /* XXX: we should compute it properly */
601 env->error_code = POWERPC_EXCP_FP;
602 } else {
603 env->fpscr &= ~(1 << FPSCR_FEX);
605 fpscr_set_rounding_mode(env);
608 void store_fpscr(CPUPPCState *env, uint64_t arg, uint32_t mask)
610 helper_store_fpscr(env, arg, mask);
613 static void do_float_check_status(CPUPPCState *env, uintptr_t raddr)
615 CPUState *cs = env_cpu(env);
616 int status = get_float_exception_flags(&env->fp_status);
617 bool inexact_happened = false;
619 if (status & float_flag_overflow) {
620 float_overflow_excp(env);
621 } else if (status & float_flag_underflow) {
622 float_underflow_excp(env);
623 } else if (status & float_flag_inexact) {
624 float_inexact_excp(env);
625 inexact_happened = true;
628 /* if the inexact flag was not set */
629 if (inexact_happened == false) {
630 env->fpscr &= ~(1 << FPSCR_FI); /* clear the FPSCR[FI] bit */
633 if (cs->exception_index == POWERPC_EXCP_PROGRAM &&
634 (env->error_code & POWERPC_EXCP_FP)) {
635 /* Differred floating-point exception after target FPR update */
636 if (fp_exceptions_enabled(env)) {
637 raise_exception_err_ra(env, cs->exception_index,
638 env->error_code, raddr);
643 void helper_float_check_status(CPUPPCState *env)
645 do_float_check_status(env, GETPC());
648 void helper_reset_fpstatus(CPUPPCState *env)
650 set_float_exception_flags(0, &env->fp_status);
653 static void float_invalid_op_addsub(CPUPPCState *env, bool set_fpcc,
654 uintptr_t retaddr, int classes)
656 if ((classes & ~is_neg) == is_inf) {
657 /* Magnitude subtraction of infinities */
658 float_invalid_op_vxisi(env, set_fpcc, retaddr);
659 } else if (classes & is_snan) {
660 float_invalid_op_vxsnan(env, retaddr);
664 /* fadd - fadd. */
665 float64 helper_fadd(CPUPPCState *env, float64 arg1, float64 arg2)
667 float64 ret = float64_add(arg1, arg2, &env->fp_status);
668 int status = get_float_exception_flags(&env->fp_status);
670 if (unlikely(status & float_flag_invalid)) {
671 float_invalid_op_addsub(env, 1, GETPC(),
672 float64_classify(arg1) |
673 float64_classify(arg2));
676 return ret;
679 /* fsub - fsub. */
680 float64 helper_fsub(CPUPPCState *env, float64 arg1, float64 arg2)
682 float64 ret = float64_sub(arg1, arg2, &env->fp_status);
683 int status = get_float_exception_flags(&env->fp_status);
685 if (unlikely(status & float_flag_invalid)) {
686 float_invalid_op_addsub(env, 1, GETPC(),
687 float64_classify(arg1) |
688 float64_classify(arg2));
691 return ret;
694 static void float_invalid_op_mul(CPUPPCState *env, bool set_fprc,
695 uintptr_t retaddr, int classes)
697 if ((classes & (is_zero | is_inf)) == (is_zero | is_inf)) {
698 /* Multiplication of zero by infinity */
699 float_invalid_op_vximz(env, set_fprc, retaddr);
700 } else if (classes & is_snan) {
701 float_invalid_op_vxsnan(env, retaddr);
705 /* fmul - fmul. */
706 float64 helper_fmul(CPUPPCState *env, float64 arg1, float64 arg2)
708 float64 ret = float64_mul(arg1, arg2, &env->fp_status);
709 int status = get_float_exception_flags(&env->fp_status);
711 if (unlikely(status & float_flag_invalid)) {
712 float_invalid_op_mul(env, 1, GETPC(),
713 float64_classify(arg1) |
714 float64_classify(arg2));
717 return ret;
720 static void float_invalid_op_div(CPUPPCState *env, bool set_fprc,
721 uintptr_t retaddr, int classes)
723 classes &= ~is_neg;
724 if (classes == is_inf) {
725 /* Division of infinity by infinity */
726 float_invalid_op_vxidi(env, set_fprc, retaddr);
727 } else if (classes == is_zero) {
728 /* Division of zero by zero */
729 float_invalid_op_vxzdz(env, set_fprc, retaddr);
730 } else if (classes & is_snan) {
731 float_invalid_op_vxsnan(env, retaddr);
735 /* fdiv - fdiv. */
736 float64 helper_fdiv(CPUPPCState *env, float64 arg1, float64 arg2)
738 float64 ret = float64_div(arg1, arg2, &env->fp_status);
739 int status = get_float_exception_flags(&env->fp_status);
741 if (unlikely(status)) {
742 if (status & float_flag_invalid) {
743 float_invalid_op_div(env, 1, GETPC(),
744 float64_classify(arg1) |
745 float64_classify(arg2));
747 if (status & float_flag_divbyzero) {
748 float_zero_divide_excp(env, GETPC());
752 return ret;
755 static void float_invalid_cvt(CPUPPCState *env, bool set_fprc,
756 uintptr_t retaddr, int class1)
758 float_invalid_op_vxcvi(env, set_fprc, retaddr);
759 if (class1 & is_snan) {
760 float_invalid_op_vxsnan(env, retaddr);
764 #define FPU_FCTI(op, cvt, nanval) \
765 uint64_t helper_##op(CPUPPCState *env, float64 arg) \
767 uint64_t ret = float64_to_##cvt(arg, &env->fp_status); \
768 int status = get_float_exception_flags(&env->fp_status); \
770 if (unlikely(status)) { \
771 if (status & float_flag_invalid) { \
772 float_invalid_cvt(env, 1, GETPC(), float64_classify(arg)); \
773 ret = nanval; \
775 do_float_check_status(env, GETPC()); \
777 return ret; \
780 FPU_FCTI(fctiw, int32, 0x80000000U)
781 FPU_FCTI(fctiwz, int32_round_to_zero, 0x80000000U)
782 FPU_FCTI(fctiwu, uint32, 0x00000000U)
783 FPU_FCTI(fctiwuz, uint32_round_to_zero, 0x00000000U)
784 FPU_FCTI(fctid, int64, 0x8000000000000000ULL)
785 FPU_FCTI(fctidz, int64_round_to_zero, 0x8000000000000000ULL)
786 FPU_FCTI(fctidu, uint64, 0x0000000000000000ULL)
787 FPU_FCTI(fctiduz, uint64_round_to_zero, 0x0000000000000000ULL)
789 #define FPU_FCFI(op, cvtr, is_single) \
790 uint64_t helper_##op(CPUPPCState *env, uint64_t arg) \
792 CPU_DoubleU farg; \
794 if (is_single) { \
795 float32 tmp = cvtr(arg, &env->fp_status); \
796 farg.d = float32_to_float64(tmp, &env->fp_status); \
797 } else { \
798 farg.d = cvtr(arg, &env->fp_status); \
800 do_float_check_status(env, GETPC()); \
801 return farg.ll; \
804 FPU_FCFI(fcfid, int64_to_float64, 0)
805 FPU_FCFI(fcfids, int64_to_float32, 1)
806 FPU_FCFI(fcfidu, uint64_to_float64, 0)
807 FPU_FCFI(fcfidus, uint64_to_float32, 1)
809 static inline uint64_t do_fri(CPUPPCState *env, uint64_t arg,
810 int rounding_mode)
812 CPU_DoubleU farg;
814 farg.ll = arg;
816 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) {
817 /* sNaN round */
818 float_invalid_op_vxsnan(env, GETPC());
819 farg.ll = arg | 0x0008000000000000ULL;
820 } else {
821 int inexact = get_float_exception_flags(&env->fp_status) &
822 float_flag_inexact;
823 set_float_rounding_mode(rounding_mode, &env->fp_status);
824 farg.ll = float64_round_to_int(farg.d, &env->fp_status);
825 /* Restore rounding mode from FPSCR */
826 fpscr_set_rounding_mode(env);
828 /* fri* does not set FPSCR[XX] */
829 if (!inexact) {
830 env->fp_status.float_exception_flags &= ~float_flag_inexact;
833 do_float_check_status(env, GETPC());
834 return farg.ll;
837 uint64_t helper_frin(CPUPPCState *env, uint64_t arg)
839 return do_fri(env, arg, float_round_ties_away);
842 uint64_t helper_friz(CPUPPCState *env, uint64_t arg)
844 return do_fri(env, arg, float_round_to_zero);
847 uint64_t helper_frip(CPUPPCState *env, uint64_t arg)
849 return do_fri(env, arg, float_round_up);
852 uint64_t helper_frim(CPUPPCState *env, uint64_t arg)
854 return do_fri(env, arg, float_round_down);
857 #define FPU_MADDSUB_UPDATE(NAME, TP) \
858 static void NAME(CPUPPCState *env, TP arg1, TP arg2, TP arg3, \
859 unsigned int madd_flags, uintptr_t retaddr) \
861 if (TP##_is_signaling_nan(arg1, &env->fp_status) || \
862 TP##_is_signaling_nan(arg2, &env->fp_status) || \
863 TP##_is_signaling_nan(arg3, &env->fp_status)) { \
864 /* sNaN operation */ \
865 float_invalid_op_vxsnan(env, retaddr); \
867 if ((TP##_is_infinity(arg1) && TP##_is_zero(arg2)) || \
868 (TP##_is_zero(arg1) && TP##_is_infinity(arg2))) { \
869 /* Multiplication of zero by infinity */ \
870 float_invalid_op_vximz(env, 1, retaddr); \
872 if ((TP##_is_infinity(arg1) || TP##_is_infinity(arg2)) && \
873 TP##_is_infinity(arg3)) { \
874 uint8_t aSign, bSign, cSign; \
876 aSign = TP##_is_neg(arg1); \
877 bSign = TP##_is_neg(arg2); \
878 cSign = TP##_is_neg(arg3); \
879 if (madd_flags & float_muladd_negate_c) { \
880 cSign ^= 1; \
882 if (aSign ^ bSign ^ cSign) { \
883 float_invalid_op_vxisi(env, 1, retaddr); \
887 FPU_MADDSUB_UPDATE(float32_maddsub_update_excp, float32)
888 FPU_MADDSUB_UPDATE(float64_maddsub_update_excp, float64)
890 #define FPU_FMADD(op, madd_flags) \
891 uint64_t helper_##op(CPUPPCState *env, uint64_t arg1, \
892 uint64_t arg2, uint64_t arg3) \
894 uint32_t flags; \
895 float64 ret = float64_muladd(arg1, arg2, arg3, madd_flags, \
896 &env->fp_status); \
897 flags = get_float_exception_flags(&env->fp_status); \
898 if (flags) { \
899 if (flags & float_flag_invalid) { \
900 float64_maddsub_update_excp(env, arg1, arg2, arg3, \
901 madd_flags, GETPC()); \
903 do_float_check_status(env, GETPC()); \
905 return ret; \
908 #define MADD_FLGS 0
909 #define MSUB_FLGS float_muladd_negate_c
910 #define NMADD_FLGS float_muladd_negate_result
911 #define NMSUB_FLGS (float_muladd_negate_c | float_muladd_negate_result)
913 FPU_FMADD(fmadd, MADD_FLGS)
914 FPU_FMADD(fnmadd, NMADD_FLGS)
915 FPU_FMADD(fmsub, MSUB_FLGS)
916 FPU_FMADD(fnmsub, NMSUB_FLGS)
918 /* frsp - frsp. */
919 uint64_t helper_frsp(CPUPPCState *env, uint64_t arg)
921 CPU_DoubleU farg;
922 float32 f32;
924 farg.ll = arg;
926 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) {
927 float_invalid_op_vxsnan(env, GETPC());
929 f32 = float64_to_float32(farg.d, &env->fp_status);
930 farg.d = float32_to_float64(f32, &env->fp_status);
932 return farg.ll;
935 /* fsqrt - fsqrt. */
936 float64 helper_fsqrt(CPUPPCState *env, float64 arg)
938 float64 ret = float64_sqrt(arg, &env->fp_status);
939 int status = get_float_exception_flags(&env->fp_status);
941 if (unlikely(status & float_flag_invalid)) {
942 if (unlikely(float64_is_any_nan(arg))) {
943 if (unlikely(float64_is_signaling_nan(arg, &env->fp_status))) {
944 /* sNaN square root */
945 float_invalid_op_vxsnan(env, GETPC());
947 } else {
948 /* Square root of a negative nonzero number */
949 float_invalid_op_vxsqrt(env, 1, GETPC());
953 return ret;
956 /* fre - fre. */
957 float64 helper_fre(CPUPPCState *env, float64 arg)
959 /* "Estimate" the reciprocal with actual division. */
960 float64 ret = float64_div(float64_one, arg, &env->fp_status);
961 int status = get_float_exception_flags(&env->fp_status);
963 if (unlikely(status)) {
964 if (status & float_flag_invalid) {
965 if (float64_is_signaling_nan(arg, &env->fp_status)) {
966 /* sNaN reciprocal */
967 float_invalid_op_vxsnan(env, GETPC());
970 if (status & float_flag_divbyzero) {
971 float_zero_divide_excp(env, GETPC());
972 /* For FPSCR.ZE == 0, the result is 1/2. */
973 ret = float64_set_sign(float64_half, float64_is_neg(arg));
977 return ret;
980 /* fres - fres. */
981 uint64_t helper_fres(CPUPPCState *env, uint64_t arg)
983 CPU_DoubleU farg;
984 float32 f32;
986 farg.ll = arg;
988 if (unlikely(float64_is_signaling_nan(farg.d, &env->fp_status))) {
989 /* sNaN reciprocal */
990 float_invalid_op_vxsnan(env, GETPC());
992 farg.d = float64_div(float64_one, farg.d, &env->fp_status);
993 f32 = float64_to_float32(farg.d, &env->fp_status);
994 farg.d = float32_to_float64(f32, &env->fp_status);
996 return farg.ll;
999 /* frsqrte - frsqrte. */
1000 float64 helper_frsqrte(CPUPPCState *env, float64 arg)
1002 /* "Estimate" the reciprocal with actual division. */
1003 float64 rets = float64_sqrt(arg, &env->fp_status);
1004 float64 retd = float64_div(float64_one, rets, &env->fp_status);
1005 int status = get_float_exception_flags(&env->fp_status);
1007 if (unlikely(status)) {
1008 if (status & float_flag_invalid) {
1009 if (float64_is_signaling_nan(arg, &env->fp_status)) {
1010 /* sNaN reciprocal */
1011 float_invalid_op_vxsnan(env, GETPC());
1012 } else {
1013 /* Square root of a negative nonzero number */
1014 float_invalid_op_vxsqrt(env, 1, GETPC());
1017 if (status & float_flag_divbyzero) {
1018 /* Reciprocal of (square root of) zero. */
1019 float_zero_divide_excp(env, GETPC());
1023 return retd;
1026 /* fsel - fsel. */
1027 uint64_t helper_fsel(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
1028 uint64_t arg3)
1030 CPU_DoubleU farg1;
1032 farg1.ll = arg1;
1034 if ((!float64_is_neg(farg1.d) || float64_is_zero(farg1.d)) &&
1035 !float64_is_any_nan(farg1.d)) {
1036 return arg2;
1037 } else {
1038 return arg3;
1042 uint32_t helper_ftdiv(uint64_t fra, uint64_t frb)
1044 int fe_flag = 0;
1045 int fg_flag = 0;
1047 if (unlikely(float64_is_infinity(fra) ||
1048 float64_is_infinity(frb) ||
1049 float64_is_zero(frb))) {
1050 fe_flag = 1;
1051 fg_flag = 1;
1052 } else {
1053 int e_a = ppc_float64_get_unbiased_exp(fra);
1054 int e_b = ppc_float64_get_unbiased_exp(frb);
1056 if (unlikely(float64_is_any_nan(fra) ||
1057 float64_is_any_nan(frb))) {
1058 fe_flag = 1;
1059 } else if ((e_b <= -1022) || (e_b >= 1021)) {
1060 fe_flag = 1;
1061 } else if (!float64_is_zero(fra) &&
1062 (((e_a - e_b) >= 1023) ||
1063 ((e_a - e_b) <= -1021) ||
1064 (e_a <= -970))) {
1065 fe_flag = 1;
1068 if (unlikely(float64_is_zero_or_denormal(frb))) {
1069 /* XB is not zero because of the above check and */
1070 /* so must be denormalized. */
1071 fg_flag = 1;
1075 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
1078 uint32_t helper_ftsqrt(uint64_t frb)
1080 int fe_flag = 0;
1081 int fg_flag = 0;
1083 if (unlikely(float64_is_infinity(frb) || float64_is_zero(frb))) {
1084 fe_flag = 1;
1085 fg_flag = 1;
1086 } else {
1087 int e_b = ppc_float64_get_unbiased_exp(frb);
1089 if (unlikely(float64_is_any_nan(frb))) {
1090 fe_flag = 1;
1091 } else if (unlikely(float64_is_zero(frb))) {
1092 fe_flag = 1;
1093 } else if (unlikely(float64_is_neg(frb))) {
1094 fe_flag = 1;
1095 } else if (!float64_is_zero(frb) && (e_b <= (-1022 + 52))) {
1096 fe_flag = 1;
1099 if (unlikely(float64_is_zero_or_denormal(frb))) {
1100 /* XB is not zero because of the above check and */
1101 /* therefore must be denormalized. */
1102 fg_flag = 1;
1106 return 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0);
1109 void helper_fcmpu(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
1110 uint32_t crfD)
1112 CPU_DoubleU farg1, farg2;
1113 uint32_t ret = 0;
1115 farg1.ll = arg1;
1116 farg2.ll = arg2;
1118 if (unlikely(float64_is_any_nan(farg1.d) ||
1119 float64_is_any_nan(farg2.d))) {
1120 ret = 0x01UL;
1121 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) {
1122 ret = 0x08UL;
1123 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) {
1124 ret = 0x04UL;
1125 } else {
1126 ret = 0x02UL;
1129 env->fpscr &= ~(0x0F << FPSCR_FPRF);
1130 env->fpscr |= ret << FPSCR_FPRF;
1131 env->crf[crfD] = ret;
1132 if (unlikely(ret == 0x01UL
1133 && (float64_is_signaling_nan(farg1.d, &env->fp_status) ||
1134 float64_is_signaling_nan(farg2.d, &env->fp_status)))) {
1135 /* sNaN comparison */
1136 float_invalid_op_vxsnan(env, GETPC());
1140 void helper_fcmpo(CPUPPCState *env, uint64_t arg1, uint64_t arg2,
1141 uint32_t crfD)
1143 CPU_DoubleU farg1, farg2;
1144 uint32_t ret = 0;
1146 farg1.ll = arg1;
1147 farg2.ll = arg2;
1149 if (unlikely(float64_is_any_nan(farg1.d) ||
1150 float64_is_any_nan(farg2.d))) {
1151 ret = 0x01UL;
1152 } else if (float64_lt(farg1.d, farg2.d, &env->fp_status)) {
1153 ret = 0x08UL;
1154 } else if (!float64_le(farg1.d, farg2.d, &env->fp_status)) {
1155 ret = 0x04UL;
1156 } else {
1157 ret = 0x02UL;
1160 env->fpscr &= ~(0x0F << FPSCR_FPRF);
1161 env->fpscr |= ret << FPSCR_FPRF;
1162 env->crf[crfD] = ret;
1163 if (unlikely(ret == 0x01UL)) {
1164 float_invalid_op_vxvc(env, 1, GETPC());
1165 if (float64_is_signaling_nan(farg1.d, &env->fp_status) ||
1166 float64_is_signaling_nan(farg2.d, &env->fp_status)) {
1167 /* sNaN comparison */
1168 float_invalid_op_vxsnan(env, GETPC());
1173 /* Single-precision floating-point conversions */
1174 static inline uint32_t efscfsi(CPUPPCState *env, uint32_t val)
1176 CPU_FloatU u;
1178 u.f = int32_to_float32(val, &env->vec_status);
1180 return u.l;
1183 static inline uint32_t efscfui(CPUPPCState *env, uint32_t val)
1185 CPU_FloatU u;
1187 u.f = uint32_to_float32(val, &env->vec_status);
1189 return u.l;
1192 static inline int32_t efsctsi(CPUPPCState *env, uint32_t val)
1194 CPU_FloatU u;
1196 u.l = val;
1197 /* NaN are not treated the same way IEEE 754 does */
1198 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1199 return 0;
1202 return float32_to_int32(u.f, &env->vec_status);
1205 static inline uint32_t efsctui(CPUPPCState *env, uint32_t val)
1207 CPU_FloatU u;
1209 u.l = val;
1210 /* NaN are not treated the same way IEEE 754 does */
1211 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1212 return 0;
1215 return float32_to_uint32(u.f, &env->vec_status);
1218 static inline uint32_t efsctsiz(CPUPPCState *env, uint32_t val)
1220 CPU_FloatU u;
1222 u.l = val;
1223 /* NaN are not treated the same way IEEE 754 does */
1224 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1225 return 0;
1228 return float32_to_int32_round_to_zero(u.f, &env->vec_status);
1231 static inline uint32_t efsctuiz(CPUPPCState *env, uint32_t val)
1233 CPU_FloatU u;
1235 u.l = val;
1236 /* NaN are not treated the same way IEEE 754 does */
1237 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1238 return 0;
1241 return float32_to_uint32_round_to_zero(u.f, &env->vec_status);
1244 static inline uint32_t efscfsf(CPUPPCState *env, uint32_t val)
1246 CPU_FloatU u;
1247 float32 tmp;
1249 u.f = int32_to_float32(val, &env->vec_status);
1250 tmp = int64_to_float32(1ULL << 32, &env->vec_status);
1251 u.f = float32_div(u.f, tmp, &env->vec_status);
1253 return u.l;
1256 static inline uint32_t efscfuf(CPUPPCState *env, uint32_t val)
1258 CPU_FloatU u;
1259 float32 tmp;
1261 u.f = uint32_to_float32(val, &env->vec_status);
1262 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1263 u.f = float32_div(u.f, tmp, &env->vec_status);
1265 return u.l;
1268 static inline uint32_t efsctsf(CPUPPCState *env, uint32_t val)
1270 CPU_FloatU u;
1271 float32 tmp;
1273 u.l = val;
1274 /* NaN are not treated the same way IEEE 754 does */
1275 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1276 return 0;
1278 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1279 u.f = float32_mul(u.f, tmp, &env->vec_status);
1281 return float32_to_int32(u.f, &env->vec_status);
1284 static inline uint32_t efsctuf(CPUPPCState *env, uint32_t val)
1286 CPU_FloatU u;
1287 float32 tmp;
1289 u.l = val;
1290 /* NaN are not treated the same way IEEE 754 does */
1291 if (unlikely(float32_is_quiet_nan(u.f, &env->vec_status))) {
1292 return 0;
1294 tmp = uint64_to_float32(1ULL << 32, &env->vec_status);
1295 u.f = float32_mul(u.f, tmp, &env->vec_status);
1297 return float32_to_uint32(u.f, &env->vec_status);
1300 #define HELPER_SPE_SINGLE_CONV(name) \
1301 uint32_t helper_e##name(CPUPPCState *env, uint32_t val) \
1303 return e##name(env, val); \
1305 /* efscfsi */
1306 HELPER_SPE_SINGLE_CONV(fscfsi);
1307 /* efscfui */
1308 HELPER_SPE_SINGLE_CONV(fscfui);
1309 /* efscfuf */
1310 HELPER_SPE_SINGLE_CONV(fscfuf);
1311 /* efscfsf */
1312 HELPER_SPE_SINGLE_CONV(fscfsf);
1313 /* efsctsi */
1314 HELPER_SPE_SINGLE_CONV(fsctsi);
1315 /* efsctui */
1316 HELPER_SPE_SINGLE_CONV(fsctui);
1317 /* efsctsiz */
1318 HELPER_SPE_SINGLE_CONV(fsctsiz);
1319 /* efsctuiz */
1320 HELPER_SPE_SINGLE_CONV(fsctuiz);
1321 /* efsctsf */
1322 HELPER_SPE_SINGLE_CONV(fsctsf);
1323 /* efsctuf */
1324 HELPER_SPE_SINGLE_CONV(fsctuf);
1326 #define HELPER_SPE_VECTOR_CONV(name) \
1327 uint64_t helper_ev##name(CPUPPCState *env, uint64_t val) \
1329 return ((uint64_t)e##name(env, val >> 32) << 32) | \
1330 (uint64_t)e##name(env, val); \
1332 /* evfscfsi */
1333 HELPER_SPE_VECTOR_CONV(fscfsi);
1334 /* evfscfui */
1335 HELPER_SPE_VECTOR_CONV(fscfui);
1336 /* evfscfuf */
1337 HELPER_SPE_VECTOR_CONV(fscfuf);
1338 /* evfscfsf */
1339 HELPER_SPE_VECTOR_CONV(fscfsf);
1340 /* evfsctsi */
1341 HELPER_SPE_VECTOR_CONV(fsctsi);
1342 /* evfsctui */
1343 HELPER_SPE_VECTOR_CONV(fsctui);
1344 /* evfsctsiz */
1345 HELPER_SPE_VECTOR_CONV(fsctsiz);
1346 /* evfsctuiz */
1347 HELPER_SPE_VECTOR_CONV(fsctuiz);
1348 /* evfsctsf */
1349 HELPER_SPE_VECTOR_CONV(fsctsf);
1350 /* evfsctuf */
1351 HELPER_SPE_VECTOR_CONV(fsctuf);
1353 /* Single-precision floating-point arithmetic */
1354 static inline uint32_t efsadd(CPUPPCState *env, uint32_t op1, uint32_t op2)
1356 CPU_FloatU u1, u2;
1358 u1.l = op1;
1359 u2.l = op2;
1360 u1.f = float32_add(u1.f, u2.f, &env->vec_status);
1361 return u1.l;
1364 static inline uint32_t efssub(CPUPPCState *env, uint32_t op1, uint32_t op2)
1366 CPU_FloatU u1, u2;
1368 u1.l = op1;
1369 u2.l = op2;
1370 u1.f = float32_sub(u1.f, u2.f, &env->vec_status);
1371 return u1.l;
1374 static inline uint32_t efsmul(CPUPPCState *env, uint32_t op1, uint32_t op2)
1376 CPU_FloatU u1, u2;
1378 u1.l = op1;
1379 u2.l = op2;
1380 u1.f = float32_mul(u1.f, u2.f, &env->vec_status);
1381 return u1.l;
1384 static inline uint32_t efsdiv(CPUPPCState *env, uint32_t op1, uint32_t op2)
1386 CPU_FloatU u1, u2;
1388 u1.l = op1;
1389 u2.l = op2;
1390 u1.f = float32_div(u1.f, u2.f, &env->vec_status);
1391 return u1.l;
1394 #define HELPER_SPE_SINGLE_ARITH(name) \
1395 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \
1397 return e##name(env, op1, op2); \
1399 /* efsadd */
1400 HELPER_SPE_SINGLE_ARITH(fsadd);
1401 /* efssub */
1402 HELPER_SPE_SINGLE_ARITH(fssub);
1403 /* efsmul */
1404 HELPER_SPE_SINGLE_ARITH(fsmul);
1405 /* efsdiv */
1406 HELPER_SPE_SINGLE_ARITH(fsdiv);
1408 #define HELPER_SPE_VECTOR_ARITH(name) \
1409 uint64_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \
1411 return ((uint64_t)e##name(env, op1 >> 32, op2 >> 32) << 32) | \
1412 (uint64_t)e##name(env, op1, op2); \
1414 /* evfsadd */
1415 HELPER_SPE_VECTOR_ARITH(fsadd);
1416 /* evfssub */
1417 HELPER_SPE_VECTOR_ARITH(fssub);
1418 /* evfsmul */
1419 HELPER_SPE_VECTOR_ARITH(fsmul);
1420 /* evfsdiv */
1421 HELPER_SPE_VECTOR_ARITH(fsdiv);
1423 /* Single-precision floating-point comparisons */
1424 static inline uint32_t efscmplt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1426 CPU_FloatU u1, u2;
1428 u1.l = op1;
1429 u2.l = op2;
1430 return float32_lt(u1.f, u2.f, &env->vec_status) ? 4 : 0;
1433 static inline uint32_t efscmpgt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1435 CPU_FloatU u1, u2;
1437 u1.l = op1;
1438 u2.l = op2;
1439 return float32_le(u1.f, u2.f, &env->vec_status) ? 0 : 4;
1442 static inline uint32_t efscmpeq(CPUPPCState *env, uint32_t op1, uint32_t op2)
1444 CPU_FloatU u1, u2;
1446 u1.l = op1;
1447 u2.l = op2;
1448 return float32_eq(u1.f, u2.f, &env->vec_status) ? 4 : 0;
1451 static inline uint32_t efststlt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1453 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1454 return efscmplt(env, op1, op2);
1457 static inline uint32_t efststgt(CPUPPCState *env, uint32_t op1, uint32_t op2)
1459 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1460 return efscmpgt(env, op1, op2);
1463 static inline uint32_t efststeq(CPUPPCState *env, uint32_t op1, uint32_t op2)
1465 /* XXX: TODO: ignore special values (NaN, infinites, ...) */
1466 return efscmpeq(env, op1, op2);
1469 #define HELPER_SINGLE_SPE_CMP(name) \
1470 uint32_t helper_e##name(CPUPPCState *env, uint32_t op1, uint32_t op2) \
1472 return e##name(env, op1, op2); \
1474 /* efststlt */
1475 HELPER_SINGLE_SPE_CMP(fststlt);
1476 /* efststgt */
1477 HELPER_SINGLE_SPE_CMP(fststgt);
1478 /* efststeq */
1479 HELPER_SINGLE_SPE_CMP(fststeq);
1480 /* efscmplt */
1481 HELPER_SINGLE_SPE_CMP(fscmplt);
1482 /* efscmpgt */
1483 HELPER_SINGLE_SPE_CMP(fscmpgt);
1484 /* efscmpeq */
1485 HELPER_SINGLE_SPE_CMP(fscmpeq);
1487 static inline uint32_t evcmp_merge(int t0, int t1)
1489 return (t0 << 3) | (t1 << 2) | ((t0 | t1) << 1) | (t0 & t1);
1492 #define HELPER_VECTOR_SPE_CMP(name) \
1493 uint32_t helper_ev##name(CPUPPCState *env, uint64_t op1, uint64_t op2) \
1495 return evcmp_merge(e##name(env, op1 >> 32, op2 >> 32), \
1496 e##name(env, op1, op2)); \
1498 /* evfststlt */
1499 HELPER_VECTOR_SPE_CMP(fststlt);
1500 /* evfststgt */
1501 HELPER_VECTOR_SPE_CMP(fststgt);
1502 /* evfststeq */
1503 HELPER_VECTOR_SPE_CMP(fststeq);
1504 /* evfscmplt */
1505 HELPER_VECTOR_SPE_CMP(fscmplt);
1506 /* evfscmpgt */
1507 HELPER_VECTOR_SPE_CMP(fscmpgt);
1508 /* evfscmpeq */
1509 HELPER_VECTOR_SPE_CMP(fscmpeq);
1511 /* Double-precision floating-point conversion */
1512 uint64_t helper_efdcfsi(CPUPPCState *env, uint32_t val)
1514 CPU_DoubleU u;
1516 u.d = int32_to_float64(val, &env->vec_status);
1518 return u.ll;
1521 uint64_t helper_efdcfsid(CPUPPCState *env, uint64_t val)
1523 CPU_DoubleU u;
1525 u.d = int64_to_float64(val, &env->vec_status);
1527 return u.ll;
1530 uint64_t helper_efdcfui(CPUPPCState *env, uint32_t val)
1532 CPU_DoubleU u;
1534 u.d = uint32_to_float64(val, &env->vec_status);
1536 return u.ll;
1539 uint64_t helper_efdcfuid(CPUPPCState *env, uint64_t val)
1541 CPU_DoubleU u;
1543 u.d = uint64_to_float64(val, &env->vec_status);
1545 return u.ll;
1548 uint32_t helper_efdctsi(CPUPPCState *env, uint64_t val)
1550 CPU_DoubleU u;
1552 u.ll = val;
1553 /* NaN are not treated the same way IEEE 754 does */
1554 if (unlikely(float64_is_any_nan(u.d))) {
1555 return 0;
1558 return float64_to_int32(u.d, &env->vec_status);
1561 uint32_t helper_efdctui(CPUPPCState *env, uint64_t val)
1563 CPU_DoubleU u;
1565 u.ll = val;
1566 /* NaN are not treated the same way IEEE 754 does */
1567 if (unlikely(float64_is_any_nan(u.d))) {
1568 return 0;
1571 return float64_to_uint32(u.d, &env->vec_status);
1574 uint32_t helper_efdctsiz(CPUPPCState *env, uint64_t val)
1576 CPU_DoubleU u;
1578 u.ll = val;
1579 /* NaN are not treated the same way IEEE 754 does */
1580 if (unlikely(float64_is_any_nan(u.d))) {
1581 return 0;
1584 return float64_to_int32_round_to_zero(u.d, &env->vec_status);
1587 uint64_t helper_efdctsidz(CPUPPCState *env, uint64_t val)
1589 CPU_DoubleU u;
1591 u.ll = val;
1592 /* NaN are not treated the same way IEEE 754 does */
1593 if (unlikely(float64_is_any_nan(u.d))) {
1594 return 0;
1597 return float64_to_int64_round_to_zero(u.d, &env->vec_status);
1600 uint32_t helper_efdctuiz(CPUPPCState *env, uint64_t val)
1602 CPU_DoubleU u;
1604 u.ll = val;
1605 /* NaN are not treated the same way IEEE 754 does */
1606 if (unlikely(float64_is_any_nan(u.d))) {
1607 return 0;
1610 return float64_to_uint32_round_to_zero(u.d, &env->vec_status);
1613 uint64_t helper_efdctuidz(CPUPPCState *env, uint64_t val)
1615 CPU_DoubleU u;
1617 u.ll = val;
1618 /* NaN are not treated the same way IEEE 754 does */
1619 if (unlikely(float64_is_any_nan(u.d))) {
1620 return 0;
1623 return float64_to_uint64_round_to_zero(u.d, &env->vec_status);
1626 uint64_t helper_efdcfsf(CPUPPCState *env, uint32_t val)
1628 CPU_DoubleU u;
1629 float64 tmp;
1631 u.d = int32_to_float64(val, &env->vec_status);
1632 tmp = int64_to_float64(1ULL << 32, &env->vec_status);
1633 u.d = float64_div(u.d, tmp, &env->vec_status);
1635 return u.ll;
1638 uint64_t helper_efdcfuf(CPUPPCState *env, uint32_t val)
1640 CPU_DoubleU u;
1641 float64 tmp;
1643 u.d = uint32_to_float64(val, &env->vec_status);
1644 tmp = int64_to_float64(1ULL << 32, &env->vec_status);
1645 u.d = float64_div(u.d, tmp, &env->vec_status);
1647 return u.ll;
1650 uint32_t helper_efdctsf(CPUPPCState *env, uint64_t val)
1652 CPU_DoubleU u;
1653 float64 tmp;
1655 u.ll = val;
1656 /* NaN are not treated the same way IEEE 754 does */
1657 if (unlikely(float64_is_any_nan(u.d))) {
1658 return 0;
1660 tmp = uint64_to_float64(1ULL << 32, &env->vec_status);
1661 u.d = float64_mul(u.d, tmp, &env->vec_status);
1663 return float64_to_int32(u.d, &env->vec_status);
1666 uint32_t helper_efdctuf(CPUPPCState *env, uint64_t val)
1668 CPU_DoubleU u;
1669 float64 tmp;
1671 u.ll = val;
1672 /* NaN are not treated the same way IEEE 754 does */
1673 if (unlikely(float64_is_any_nan(u.d))) {
1674 return 0;
1676 tmp = uint64_to_float64(1ULL << 32, &env->vec_status);
1677 u.d = float64_mul(u.d, tmp, &env->vec_status);
1679 return float64_to_uint32(u.d, &env->vec_status);
1682 uint32_t helper_efscfd(CPUPPCState *env, uint64_t val)
1684 CPU_DoubleU u1;
1685 CPU_FloatU u2;
1687 u1.ll = val;
1688 u2.f = float64_to_float32(u1.d, &env->vec_status);
1690 return u2.l;
1693 uint64_t helper_efdcfs(CPUPPCState *env, uint32_t val)
1695 CPU_DoubleU u2;
1696 CPU_FloatU u1;
1698 u1.l = val;
1699 u2.d = float32_to_float64(u1.f, &env->vec_status);
1701 return u2.ll;
1704 /* Double precision fixed-point arithmetic */
1705 uint64_t helper_efdadd(CPUPPCState *env, uint64_t op1, uint64_t op2)
1707 CPU_DoubleU u1, u2;
1709 u1.ll = op1;
1710 u2.ll = op2;
1711 u1.d = float64_add(u1.d, u2.d, &env->vec_status);
1712 return u1.ll;
1715 uint64_t helper_efdsub(CPUPPCState *env, uint64_t op1, uint64_t op2)
1717 CPU_DoubleU u1, u2;
1719 u1.ll = op1;
1720 u2.ll = op2;
1721 u1.d = float64_sub(u1.d, u2.d, &env->vec_status);
1722 return u1.ll;
1725 uint64_t helper_efdmul(CPUPPCState *env, uint64_t op1, uint64_t op2)
1727 CPU_DoubleU u1, u2;
1729 u1.ll = op1;
1730 u2.ll = op2;
1731 u1.d = float64_mul(u1.d, u2.d, &env->vec_status);
1732 return u1.ll;
1735 uint64_t helper_efddiv(CPUPPCState *env, uint64_t op1, uint64_t op2)
1737 CPU_DoubleU u1, u2;
1739 u1.ll = op1;
1740 u2.ll = op2;
1741 u1.d = float64_div(u1.d, u2.d, &env->vec_status);
1742 return u1.ll;
1745 /* Double precision floating point helpers */
1746 uint32_t helper_efdtstlt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1748 CPU_DoubleU u1, u2;
1750 u1.ll = op1;
1751 u2.ll = op2;
1752 return float64_lt(u1.d, u2.d, &env->vec_status) ? 4 : 0;
1755 uint32_t helper_efdtstgt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1757 CPU_DoubleU u1, u2;
1759 u1.ll = op1;
1760 u2.ll = op2;
1761 return float64_le(u1.d, u2.d, &env->vec_status) ? 0 : 4;
1764 uint32_t helper_efdtsteq(CPUPPCState *env, uint64_t op1, uint64_t op2)
1766 CPU_DoubleU u1, u2;
1768 u1.ll = op1;
1769 u2.ll = op2;
1770 return float64_eq_quiet(u1.d, u2.d, &env->vec_status) ? 4 : 0;
1773 uint32_t helper_efdcmplt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1775 /* XXX: TODO: test special values (NaN, infinites, ...) */
1776 return helper_efdtstlt(env, op1, op2);
1779 uint32_t helper_efdcmpgt(CPUPPCState *env, uint64_t op1, uint64_t op2)
1781 /* XXX: TODO: test special values (NaN, infinites, ...) */
1782 return helper_efdtstgt(env, op1, op2);
1785 uint32_t helper_efdcmpeq(CPUPPCState *env, uint64_t op1, uint64_t op2)
1787 /* XXX: TODO: test special values (NaN, infinites, ...) */
1788 return helper_efdtsteq(env, op1, op2);
1791 #define float64_to_float64(x, env) x
1795 * VSX_ADD_SUB - VSX floating point add/subract
1796 * name - instruction mnemonic
1797 * op - operation (add or sub)
1798 * nels - number of elements (1, 2 or 4)
1799 * tp - type (float32 or float64)
1800 * fld - vsr_t field (VsrD(*) or VsrW(*))
1801 * sfprf - set FPRF
1803 #define VSX_ADD_SUB(name, op, nels, tp, fld, sfprf, r2sp) \
1804 void helper_##name(CPUPPCState *env, ppc_vsr_t *xt, \
1805 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1807 ppc_vsr_t t = *xt; \
1808 int i; \
1810 helper_reset_fpstatus(env); \
1812 for (i = 0; i < nels; i++) { \
1813 float_status tstat = env->fp_status; \
1814 set_float_exception_flags(0, &tstat); \
1815 t.fld = tp##_##op(xa->fld, xb->fld, &tstat); \
1816 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1818 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1819 float_invalid_op_addsub(env, sfprf, GETPC(), \
1820 tp##_classify(xa->fld) | \
1821 tp##_classify(xb->fld)); \
1824 if (r2sp) { \
1825 t.fld = helper_frsp(env, t.fld); \
1828 if (sfprf) { \
1829 helper_compute_fprf_float64(env, t.fld); \
1832 *xt = t; \
1833 do_float_check_status(env, GETPC()); \
1836 VSX_ADD_SUB(xsadddp, add, 1, float64, VsrD(0), 1, 0)
1837 VSX_ADD_SUB(xsaddsp, add, 1, float64, VsrD(0), 1, 1)
1838 VSX_ADD_SUB(xvadddp, add, 2, float64, VsrD(i), 0, 0)
1839 VSX_ADD_SUB(xvaddsp, add, 4, float32, VsrW(i), 0, 0)
1840 VSX_ADD_SUB(xssubdp, sub, 1, float64, VsrD(0), 1, 0)
1841 VSX_ADD_SUB(xssubsp, sub, 1, float64, VsrD(0), 1, 1)
1842 VSX_ADD_SUB(xvsubdp, sub, 2, float64, VsrD(i), 0, 0)
1843 VSX_ADD_SUB(xvsubsp, sub, 4, float32, VsrW(i), 0, 0)
1845 void helper_xsaddqp(CPUPPCState *env, uint32_t opcode)
1847 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32];
1848 ppc_vsr_t *xa = &env->vsr[rA(opcode) + 32];
1849 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32];
1850 ppc_vsr_t t = *xt;
1851 float_status tstat;
1853 helper_reset_fpstatus(env);
1855 tstat = env->fp_status;
1856 if (unlikely(Rc(opcode) != 0)) {
1857 tstat.float_rounding_mode = float_round_to_odd;
1860 set_float_exception_flags(0, &tstat);
1861 t.f128 = float128_add(xa->f128, xb->f128, &tstat);
1862 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1864 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1865 float_invalid_op_addsub(env, 1, GETPC(),
1866 float128_classify(xa->f128) |
1867 float128_classify(xb->f128));
1870 helper_compute_fprf_float128(env, t.f128);
1872 *xt = t;
1873 do_float_check_status(env, GETPC());
1877 * VSX_MUL - VSX floating point multiply
1878 * op - instruction mnemonic
1879 * nels - number of elements (1, 2 or 4)
1880 * tp - type (float32 or float64)
1881 * fld - vsr_t field (VsrD(*) or VsrW(*))
1882 * sfprf - set FPRF
1884 #define VSX_MUL(op, nels, tp, fld, sfprf, r2sp) \
1885 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
1886 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1888 ppc_vsr_t t = *xt; \
1889 int i; \
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 t.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 t.fld = helper_frsp(env, t.fld); \
1909 if (sfprf) { \
1910 helper_compute_fprf_float64(env, t.fld); \
1914 *xt = t; \
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 = &env->vsr[rD(opcode) + 32];
1926 ppc_vsr_t *xa = &env->vsr[rA(opcode) + 32];
1927 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32];
1928 ppc_vsr_t t = *xt;
1929 float_status tstat;
1931 helper_reset_fpstatus(env);
1932 tstat = env->fp_status;
1933 if (unlikely(Rc(opcode) != 0)) {
1934 tstat.float_rounding_mode = float_round_to_odd;
1937 set_float_exception_flags(0, &tstat);
1938 t.f128 = float128_mul(xa->f128, xb->f128, &tstat);
1939 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
1941 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
1942 float_invalid_op_mul(env, 1, GETPC(),
1943 float128_classify(xa->f128) |
1944 float128_classify(xb->f128));
1946 helper_compute_fprf_float128(env, t.f128);
1948 *xt = t;
1949 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, ppc_vsr_t *xt, \
1962 ppc_vsr_t *xa, ppc_vsr_t *xb) \
1964 ppc_vsr_t t = *xt; \
1965 int i; \
1967 helper_reset_fpstatus(env); \
1969 for (i = 0; i < nels; i++) { \
1970 float_status tstat = env->fp_status; \
1971 set_float_exception_flags(0, &tstat); \
1972 t.fld = tp##_div(xa->fld, xb->fld, &tstat); \
1973 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
1975 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
1976 float_invalid_op_div(env, sfprf, GETPC(), \
1977 tp##_classify(xa->fld) | \
1978 tp##_classify(xb->fld)); \
1980 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) { \
1981 float_zero_divide_excp(env, GETPC()); \
1984 if (r2sp) { \
1985 t.fld = helper_frsp(env, t.fld); \
1988 if (sfprf) { \
1989 helper_compute_fprf_float64(env, t.fld); \
1993 *xt = t; \
1994 do_float_check_status(env, GETPC()); \
1997 VSX_DIV(xsdivdp, 1, float64, VsrD(0), 1, 0)
1998 VSX_DIV(xsdivsp, 1, float64, VsrD(0), 1, 1)
1999 VSX_DIV(xvdivdp, 2, float64, VsrD(i), 0, 0)
2000 VSX_DIV(xvdivsp, 4, float32, VsrW(i), 0, 0)
2002 void helper_xsdivqp(CPUPPCState *env, uint32_t opcode)
2004 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32];
2005 ppc_vsr_t *xa = &env->vsr[rA(opcode) + 32];
2006 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32];
2007 ppc_vsr_t t = *xt;
2008 float_status tstat;
2010 helper_reset_fpstatus(env);
2011 tstat = env->fp_status;
2012 if (unlikely(Rc(opcode) != 0)) {
2013 tstat.float_rounding_mode = float_round_to_odd;
2016 set_float_exception_flags(0, &tstat);
2017 t.f128 = float128_div(xa->f128, xb->f128, &tstat);
2018 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
2020 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
2021 float_invalid_op_div(env, 1, GETPC(),
2022 float128_classify(xa->f128) |
2023 float128_classify(xb->f128));
2025 if (unlikely(tstat.float_exception_flags & float_flag_divbyzero)) {
2026 float_zero_divide_excp(env, GETPC());
2029 helper_compute_fprf_float128(env, t.f128);
2030 *xt = t;
2031 do_float_check_status(env, GETPC());
2035 * VSX_RE - VSX floating point reciprocal estimate
2036 * op - instruction mnemonic
2037 * nels - number of elements (1, 2 or 4)
2038 * tp - type (float32 or float64)
2039 * fld - vsr_t field (VsrD(*) or VsrW(*))
2040 * sfprf - set FPRF
2042 #define VSX_RE(op, nels, tp, fld, sfprf, r2sp) \
2043 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2045 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
2046 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
2047 ppc_vsr_t t = *xt; \
2048 int i; \
2050 helper_reset_fpstatus(env); \
2052 for (i = 0; i < nels; i++) { \
2053 if (unlikely(tp##_is_signaling_nan(xb->fld, &env->fp_status))) { \
2054 float_invalid_op_vxsnan(env, GETPC()); \
2056 t.fld = tp##_div(tp##_one, xb->fld, &env->fp_status); \
2058 if (r2sp) { \
2059 t.fld = helper_frsp(env, t.fld); \
2062 if (sfprf) { \
2063 helper_compute_fprf_float64(env, t.fld); \
2067 *xt = t; \
2068 do_float_check_status(env, GETPC()); \
2071 VSX_RE(xsredp, 1, float64, VsrD(0), 1, 0)
2072 VSX_RE(xsresp, 1, float64, VsrD(0), 1, 1)
2073 VSX_RE(xvredp, 2, float64, VsrD(i), 0, 0)
2074 VSX_RE(xvresp, 4, float32, VsrW(i), 0, 0)
2077 * VSX_SQRT - VSX floating point square root
2078 * op - instruction mnemonic
2079 * nels - number of elements (1, 2 or 4)
2080 * tp - type (float32 or float64)
2081 * fld - vsr_t field (VsrD(*) or VsrW(*))
2082 * sfprf - set FPRF
2084 #define VSX_SQRT(op, nels, tp, fld, sfprf, r2sp) \
2085 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2087 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
2088 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
2089 ppc_vsr_t t = *xt; \
2090 int i; \
2092 helper_reset_fpstatus(env); \
2094 for (i = 0; i < nels; i++) { \
2095 float_status tstat = env->fp_status; \
2096 set_float_exception_flags(0, &tstat); \
2097 t.fld = tp##_sqrt(xb->fld, &tstat); \
2098 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
2100 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
2101 if (tp##_is_neg(xb->fld) && !tp##_is_zero(xb->fld)) { \
2102 float_invalid_op_vxsqrt(env, sfprf, GETPC()); \
2103 } else if (tp##_is_signaling_nan(xb->fld, &tstat)) { \
2104 float_invalid_op_vxsnan(env, GETPC()); \
2108 if (r2sp) { \
2109 t.fld = helper_frsp(env, t.fld); \
2112 if (sfprf) { \
2113 helper_compute_fprf_float64(env, t.fld); \
2117 *xt = t; \
2118 do_float_check_status(env, GETPC()); \
2121 VSX_SQRT(xssqrtdp, 1, float64, VsrD(0), 1, 0)
2122 VSX_SQRT(xssqrtsp, 1, float64, VsrD(0), 1, 1)
2123 VSX_SQRT(xvsqrtdp, 2, float64, VsrD(i), 0, 0)
2124 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 = &env->vsr[xT(opcode)]; \
2138 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
2139 ppc_vsr_t t = *xt; \
2140 int i; \
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 t.fld = tp##_sqrt(xb->fld, &tstat); \
2148 t.fld = tp##_div(tp##_one, t.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 t.fld = helper_frsp(env, t.fld); \
2163 if (sfprf) { \
2164 helper_compute_fprf_float64(env, t.fld); \
2168 *xt = t; \
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)
2178 * VSX_TDIV - VSX floating point test for divide
2179 * op - instruction mnemonic
2180 * nels - number of elements (1, 2 or 4)
2181 * tp - type (float32 or float64)
2182 * fld - vsr_t field (VsrD(*) or VsrW(*))
2183 * emin - minimum unbiased exponent
2184 * emax - maximum unbiased exponent
2185 * nbits - number of fraction bits
2187 #define VSX_TDIV(op, nels, tp, fld, emin, emax, nbits) \
2188 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2190 ppc_vsr_t *xa = &env->vsr[xA(opcode)]; \
2191 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
2192 int i; \
2193 int fe_flag = 0; \
2194 int fg_flag = 0; \
2196 for (i = 0; i < nels; i++) { \
2197 if (unlikely(tp##_is_infinity(xa->fld) || \
2198 tp##_is_infinity(xb->fld) || \
2199 tp##_is_zero(xb->fld))) { \
2200 fe_flag = 1; \
2201 fg_flag = 1; \
2202 } else { \
2203 int e_a = ppc_##tp##_get_unbiased_exp(xa->fld); \
2204 int e_b = ppc_##tp##_get_unbiased_exp(xb->fld); \
2206 if (unlikely(tp##_is_any_nan(xa->fld) || \
2207 tp##_is_any_nan(xb->fld))) { \
2208 fe_flag = 1; \
2209 } else if ((e_b <= emin) || (e_b >= (emax - 2))) { \
2210 fe_flag = 1; \
2211 } else if (!tp##_is_zero(xa->fld) && \
2212 (((e_a - e_b) >= emax) || \
2213 ((e_a - e_b) <= (emin + 1)) || \
2214 (e_a <= (emin + nbits)))) { \
2215 fe_flag = 1; \
2218 if (unlikely(tp##_is_zero_or_denormal(xb->fld))) { \
2219 /* \
2220 * XB is not zero because of the above check and so \
2221 * must be denormalized. \
2222 */ \
2223 fg_flag = 1; \
2228 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
2231 VSX_TDIV(xstdivdp, 1, float64, VsrD(0), -1022, 1023, 52)
2232 VSX_TDIV(xvtdivdp, 2, float64, VsrD(i), -1022, 1023, 52)
2233 VSX_TDIV(xvtdivsp, 4, float32, VsrW(i), -126, 127, 23)
2236 * VSX_TSQRT - VSX floating point test for square root
2237 * op - instruction mnemonic
2238 * nels - number of elements (1, 2 or 4)
2239 * tp - type (float32 or float64)
2240 * fld - vsr_t field (VsrD(*) or VsrW(*))
2241 * emin - minimum unbiased exponent
2242 * emax - maximum unbiased exponent
2243 * nbits - number of fraction bits
2245 #define VSX_TSQRT(op, nels, tp, fld, emin, nbits) \
2246 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2248 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
2249 int i; \
2250 int fe_flag = 0; \
2251 int fg_flag = 0; \
2253 for (i = 0; i < nels; i++) { \
2254 if (unlikely(tp##_is_infinity(xb->fld) || \
2255 tp##_is_zero(xb->fld))) { \
2256 fe_flag = 1; \
2257 fg_flag = 1; \
2258 } else { \
2259 int e_b = ppc_##tp##_get_unbiased_exp(xb->fld); \
2261 if (unlikely(tp##_is_any_nan(xb->fld))) { \
2262 fe_flag = 1; \
2263 } else if (unlikely(tp##_is_zero(xb->fld))) { \
2264 fe_flag = 1; \
2265 } else if (unlikely(tp##_is_neg(xb->fld))) { \
2266 fe_flag = 1; \
2267 } else if (!tp##_is_zero(xb->fld) && \
2268 (e_b <= (emin + nbits))) { \
2269 fe_flag = 1; \
2272 if (unlikely(tp##_is_zero_or_denormal(xb->fld))) { \
2273 /* \
2274 * XB is not zero because of the above check and \
2275 * therefore must be denormalized. \
2276 */ \
2277 fg_flag = 1; \
2282 env->crf[BF(opcode)] = 0x8 | (fg_flag ? 4 : 0) | (fe_flag ? 2 : 0); \
2285 VSX_TSQRT(xstsqrtdp, 1, float64, VsrD(0), -1022, 52)
2286 VSX_TSQRT(xvtsqrtdp, 2, float64, VsrD(i), -1022, 52)
2287 VSX_TSQRT(xvtsqrtsp, 4, float32, VsrW(i), -126, 23)
2290 * VSX_MADD - VSX floating point muliply/add variations
2291 * op - instruction mnemonic
2292 * nels - number of elements (1, 2 or 4)
2293 * tp - type (float32 or float64)
2294 * fld - vsr_t field (VsrD(*) or VsrW(*))
2295 * maddflgs - flags for the float*muladd routine that control the
2296 * various forms (madd, msub, nmadd, nmsub)
2297 * afrm - A form (1=A, 0=M)
2298 * sfprf - set FPRF
2300 #define VSX_MADD(op, nels, tp, fld, maddflgs, afrm, sfprf, r2sp) \
2301 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2302 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2304 ppc_vsr_t t = *xt, *b, *c; \
2305 int i; \
2307 if (afrm) { /* AxB + T */ \
2308 b = xb; \
2309 c = xt; \
2310 } else { /* AxT + B */ \
2311 b = xt; \
2312 c = xb; \
2315 helper_reset_fpstatus(env); \
2317 for (i = 0; i < nels; i++) { \
2318 float_status tstat = env->fp_status; \
2319 set_float_exception_flags(0, &tstat); \
2320 if (r2sp && (tstat.float_rounding_mode == float_round_nearest_even)) {\
2321 /* \
2322 * Avoid double rounding errors by rounding the intermediate \
2323 * result to odd. \
2324 */ \
2325 set_float_rounding_mode(float_round_to_zero, &tstat); \
2326 t.fld = tp##_muladd(xa->fld, b->fld, c->fld, \
2327 maddflgs, &tstat); \
2328 t.fld |= (get_float_exception_flags(&tstat) & \
2329 float_flag_inexact) != 0; \
2330 } else { \
2331 t.fld = tp##_muladd(xa->fld, b->fld, c->fld, \
2332 maddflgs, &tstat); \
2334 env->fp_status.float_exception_flags |= tstat.float_exception_flags; \
2336 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) { \
2337 tp##_maddsub_update_excp(env, xa->fld, b->fld, \
2338 c->fld, maddflgs, GETPC()); \
2341 if (r2sp) { \
2342 t.fld = helper_frsp(env, t.fld); \
2345 if (sfprf) { \
2346 helper_compute_fprf_float64(env, t.fld); \
2349 *xt = t; \
2350 do_float_check_status(env, GETPC()); \
2353 VSX_MADD(xsmaddadp, 1, float64, VsrD(0), MADD_FLGS, 1, 1, 0)
2354 VSX_MADD(xsmaddmdp, 1, float64, VsrD(0), MADD_FLGS, 0, 1, 0)
2355 VSX_MADD(xsmsubadp, 1, float64, VsrD(0), MSUB_FLGS, 1, 1, 0)
2356 VSX_MADD(xsmsubmdp, 1, float64, VsrD(0), MSUB_FLGS, 0, 1, 0)
2357 VSX_MADD(xsnmaddadp, 1, float64, VsrD(0), NMADD_FLGS, 1, 1, 0)
2358 VSX_MADD(xsnmaddmdp, 1, float64, VsrD(0), NMADD_FLGS, 0, 1, 0)
2359 VSX_MADD(xsnmsubadp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 1, 0)
2360 VSX_MADD(xsnmsubmdp, 1, float64, VsrD(0), NMSUB_FLGS, 0, 1, 0)
2362 VSX_MADD(xsmaddasp, 1, float64, VsrD(0), MADD_FLGS, 1, 1, 1)
2363 VSX_MADD(xsmaddmsp, 1, float64, VsrD(0), MADD_FLGS, 0, 1, 1)
2364 VSX_MADD(xsmsubasp, 1, float64, VsrD(0), MSUB_FLGS, 1, 1, 1)
2365 VSX_MADD(xsmsubmsp, 1, float64, VsrD(0), MSUB_FLGS, 0, 1, 1)
2366 VSX_MADD(xsnmaddasp, 1, float64, VsrD(0), NMADD_FLGS, 1, 1, 1)
2367 VSX_MADD(xsnmaddmsp, 1, float64, VsrD(0), NMADD_FLGS, 0, 1, 1)
2368 VSX_MADD(xsnmsubasp, 1, float64, VsrD(0), NMSUB_FLGS, 1, 1, 1)
2369 VSX_MADD(xsnmsubmsp, 1, float64, VsrD(0), NMSUB_FLGS, 0, 1, 1)
2371 VSX_MADD(xvmaddadp, 2, float64, VsrD(i), MADD_FLGS, 1, 0, 0)
2372 VSX_MADD(xvmaddmdp, 2, float64, VsrD(i), MADD_FLGS, 0, 0, 0)
2373 VSX_MADD(xvmsubadp, 2, float64, VsrD(i), MSUB_FLGS, 1, 0, 0)
2374 VSX_MADD(xvmsubmdp, 2, float64, VsrD(i), MSUB_FLGS, 0, 0, 0)
2375 VSX_MADD(xvnmaddadp, 2, float64, VsrD(i), NMADD_FLGS, 1, 0, 0)
2376 VSX_MADD(xvnmaddmdp, 2, float64, VsrD(i), NMADD_FLGS, 0, 0, 0)
2377 VSX_MADD(xvnmsubadp, 2, float64, VsrD(i), NMSUB_FLGS, 1, 0, 0)
2378 VSX_MADD(xvnmsubmdp, 2, float64, VsrD(i), NMSUB_FLGS, 0, 0, 0)
2380 VSX_MADD(xvmaddasp, 4, float32, VsrW(i), MADD_FLGS, 1, 0, 0)
2381 VSX_MADD(xvmaddmsp, 4, float32, VsrW(i), MADD_FLGS, 0, 0, 0)
2382 VSX_MADD(xvmsubasp, 4, float32, VsrW(i), MSUB_FLGS, 1, 0, 0)
2383 VSX_MADD(xvmsubmsp, 4, float32, VsrW(i), MSUB_FLGS, 0, 0, 0)
2384 VSX_MADD(xvnmaddasp, 4, float32, VsrW(i), NMADD_FLGS, 1, 0, 0)
2385 VSX_MADD(xvnmaddmsp, 4, float32, VsrW(i), NMADD_FLGS, 0, 0, 0)
2386 VSX_MADD(xvnmsubasp, 4, float32, VsrW(i), NMSUB_FLGS, 1, 0, 0)
2387 VSX_MADD(xvnmsubmsp, 4, float32, VsrW(i), NMSUB_FLGS, 0, 0, 0)
2390 * VSX_SCALAR_CMP_DP - VSX scalar floating point compare double precision
2391 * op - instruction mnemonic
2392 * cmp - comparison operation
2393 * exp - expected result of comparison
2394 * svxvc - set VXVC bit
2396 #define VSX_SCALAR_CMP_DP(op, cmp, exp, svxvc) \
2397 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2398 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2400 ppc_vsr_t t = *xt; \
2401 bool vxsnan_flag = false, vxvc_flag = false, vex_flag = false; \
2403 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \
2404 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2405 vxsnan_flag = true; \
2406 if (fpscr_ve == 0 && svxvc) { \
2407 vxvc_flag = true; \
2409 } else if (svxvc) { \
2410 vxvc_flag = float64_is_quiet_nan(xa->VsrD(0), &env->fp_status) || \
2411 float64_is_quiet_nan(xb->VsrD(0), &env->fp_status); \
2413 if (vxsnan_flag) { \
2414 float_invalid_op_vxsnan(env, GETPC()); \
2416 if (vxvc_flag) { \
2417 float_invalid_op_vxvc(env, 0, GETPC()); \
2419 vex_flag = fpscr_ve && (vxvc_flag || vxsnan_flag); \
2421 if (!vex_flag) { \
2422 if (float64_##cmp(xb->VsrD(0), xa->VsrD(0), \
2423 &env->fp_status) == exp) { \
2424 t.VsrD(0) = -1; \
2425 t.VsrD(1) = 0; \
2426 } else { \
2427 t.VsrD(0) = 0; \
2428 t.VsrD(1) = 0; \
2431 *xt = t; \
2432 do_float_check_status(env, GETPC()); \
2435 VSX_SCALAR_CMP_DP(xscmpeqdp, eq, 1, 0)
2436 VSX_SCALAR_CMP_DP(xscmpgedp, le, 1, 1)
2437 VSX_SCALAR_CMP_DP(xscmpgtdp, lt, 1, 1)
2438 VSX_SCALAR_CMP_DP(xscmpnedp, eq, 0, 0)
2440 void helper_xscmpexpdp(CPUPPCState *env, uint32_t opcode)
2442 ppc_vsr_t *xa = &env->vsr[xA(opcode)];
2443 ppc_vsr_t *xb = &env->vsr[xB(opcode)];
2444 int64_t exp_a, exp_b;
2445 uint32_t cc;
2447 exp_a = extract64(xa->VsrD(0), 52, 11);
2448 exp_b = extract64(xb->VsrD(0), 52, 11);
2450 if (unlikely(float64_is_any_nan(xa->VsrD(0)) ||
2451 float64_is_any_nan(xb->VsrD(0)))) {
2452 cc = CRF_SO;
2453 } else {
2454 if (exp_a < exp_b) {
2455 cc = CRF_LT;
2456 } else if (exp_a > exp_b) {
2457 cc = CRF_GT;
2458 } else {
2459 cc = CRF_EQ;
2463 env->fpscr &= ~(0x0F << FPSCR_FPRF);
2464 env->fpscr |= cc << FPSCR_FPRF;
2465 env->crf[BF(opcode)] = cc;
2467 do_float_check_status(env, GETPC());
2470 void helper_xscmpexpqp(CPUPPCState *env, uint32_t opcode)
2472 ppc_vsr_t *xa = &env->vsr[rA(opcode) + 32];
2473 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32];
2474 int64_t exp_a, exp_b;
2475 uint32_t cc;
2477 exp_a = extract64(xa->VsrD(0), 48, 15);
2478 exp_b = extract64(xb->VsrD(0), 48, 15);
2480 if (unlikely(float128_is_any_nan(xa->f128) ||
2481 float128_is_any_nan(xb->f128))) {
2482 cc = CRF_SO;
2483 } else {
2484 if (exp_a < exp_b) {
2485 cc = CRF_LT;
2486 } else if (exp_a > exp_b) {
2487 cc = CRF_GT;
2488 } else {
2489 cc = CRF_EQ;
2493 env->fpscr &= ~(0x0F << FPSCR_FPRF);
2494 env->fpscr |= cc << FPSCR_FPRF;
2495 env->crf[BF(opcode)] = cc;
2497 do_float_check_status(env, GETPC());
2500 #define VSX_SCALAR_CMP(op, ordered) \
2501 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2503 ppc_vsr_t *xa = &env->vsr[xA(opcode)]; \
2504 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
2505 uint32_t cc = 0; \
2506 bool vxsnan_flag = false, vxvc_flag = false; \
2508 helper_reset_fpstatus(env); \
2510 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \
2511 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2512 vxsnan_flag = true; \
2513 cc = CRF_SO; \
2514 if (fpscr_ve == 0 && ordered) { \
2515 vxvc_flag = true; \
2517 } else if (float64_is_quiet_nan(xa->VsrD(0), &env->fp_status) || \
2518 float64_is_quiet_nan(xb->VsrD(0), &env->fp_status)) { \
2519 cc = CRF_SO; \
2520 if (ordered) { \
2521 vxvc_flag = true; \
2524 if (vxsnan_flag) { \
2525 float_invalid_op_vxsnan(env, GETPC()); \
2527 if (vxvc_flag) { \
2528 float_invalid_op_vxvc(env, 0, GETPC()); \
2531 if (float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) { \
2532 cc |= CRF_LT; \
2533 } else if (!float64_le(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) { \
2534 cc |= CRF_GT; \
2535 } else { \
2536 cc |= CRF_EQ; \
2539 env->fpscr &= ~(0x0F << FPSCR_FPRF); \
2540 env->fpscr |= cc << FPSCR_FPRF; \
2541 env->crf[BF(opcode)] = cc; \
2543 do_float_check_status(env, GETPC()); \
2546 VSX_SCALAR_CMP(xscmpodp, 1)
2547 VSX_SCALAR_CMP(xscmpudp, 0)
2549 #define VSX_SCALAR_CMPQ(op, ordered) \
2550 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2552 ppc_vsr_t *xa = &env->vsr[rA(opcode) + 32]; \
2553 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32]; \
2554 uint32_t cc = 0; \
2555 bool vxsnan_flag = false, vxvc_flag = false; \
2557 helper_reset_fpstatus(env); \
2559 if (float128_is_signaling_nan(xa->f128, &env->fp_status) || \
2560 float128_is_signaling_nan(xb->f128, &env->fp_status)) { \
2561 vxsnan_flag = true; \
2562 cc = CRF_SO; \
2563 if (fpscr_ve == 0 && ordered) { \
2564 vxvc_flag = true; \
2566 } else if (float128_is_quiet_nan(xa->f128, &env->fp_status) || \
2567 float128_is_quiet_nan(xb->f128, &env->fp_status)) { \
2568 cc = CRF_SO; \
2569 if (ordered) { \
2570 vxvc_flag = true; \
2573 if (vxsnan_flag) { \
2574 float_invalid_op_vxsnan(env, GETPC()); \
2576 if (vxvc_flag) { \
2577 float_invalid_op_vxvc(env, 0, GETPC()); \
2580 if (float128_lt(xa->f128, xb->f128, &env->fp_status)) { \
2581 cc |= CRF_LT; \
2582 } else if (!float128_le(xa->f128, xb->f128, &env->fp_status)) { \
2583 cc |= CRF_GT; \
2584 } else { \
2585 cc |= CRF_EQ; \
2588 env->fpscr &= ~(0x0F << FPSCR_FPRF); \
2589 env->fpscr |= cc << FPSCR_FPRF; \
2590 env->crf[BF(opcode)] = cc; \
2592 do_float_check_status(env, GETPC()); \
2595 VSX_SCALAR_CMPQ(xscmpoqp, 1)
2596 VSX_SCALAR_CMPQ(xscmpuqp, 0)
2599 * VSX_MAX_MIN - VSX floating point maximum/minimum
2600 * name - instruction mnemonic
2601 * op - operation (max or min)
2602 * nels - number of elements (1, 2 or 4)
2603 * tp - type (float32 or float64)
2604 * fld - vsr_t field (VsrD(*) or VsrW(*))
2606 #define VSX_MAX_MIN(name, op, nels, tp, fld) \
2607 void helper_##name(CPUPPCState *env, ppc_vsr_t *xt, \
2608 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2610 ppc_vsr_t t = *xt; \
2611 int i; \
2613 for (i = 0; i < nels; i++) { \
2614 t.fld = tp##_##op(xa->fld, xb->fld, &env->fp_status); \
2615 if (unlikely(tp##_is_signaling_nan(xa->fld, &env->fp_status) || \
2616 tp##_is_signaling_nan(xb->fld, &env->fp_status))) { \
2617 float_invalid_op_vxsnan(env, GETPC()); \
2621 *xt = t; \
2622 do_float_check_status(env, GETPC()); \
2625 VSX_MAX_MIN(xsmaxdp, maxnum, 1, float64, VsrD(0))
2626 VSX_MAX_MIN(xvmaxdp, maxnum, 2, float64, VsrD(i))
2627 VSX_MAX_MIN(xvmaxsp, maxnum, 4, float32, VsrW(i))
2628 VSX_MAX_MIN(xsmindp, minnum, 1, float64, VsrD(0))
2629 VSX_MAX_MIN(xvmindp, minnum, 2, float64, VsrD(i))
2630 VSX_MAX_MIN(xvminsp, minnum, 4, float32, VsrW(i))
2632 #define VSX_MAX_MINC(name, max) \
2633 void helper_##name(CPUPPCState *env, uint32_t opcode) \
2635 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32]; \
2636 ppc_vsr_t *xa = &env->vsr[rA(opcode) + 32]; \
2637 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32]; \
2638 ppc_vsr_t t = *xt; \
2639 bool vxsnan_flag = false, vex_flag = false; \
2641 if (unlikely(float64_is_any_nan(xa->VsrD(0)) || \
2642 float64_is_any_nan(xb->VsrD(0)))) { \
2643 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status) || \
2644 float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2645 vxsnan_flag = true; \
2647 t.VsrD(0) = xb->VsrD(0); \
2648 } else if ((max && \
2649 !float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) || \
2650 (!max && \
2651 float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status))) { \
2652 t.VsrD(0) = xa->VsrD(0); \
2653 } else { \
2654 t.VsrD(0) = xb->VsrD(0); \
2657 vex_flag = fpscr_ve & vxsnan_flag; \
2658 if (vxsnan_flag) { \
2659 float_invalid_op_vxsnan(env, GETPC()); \
2661 if (!vex_flag) { \
2662 *xt = t; \
2666 VSX_MAX_MINC(xsmaxcdp, 1);
2667 VSX_MAX_MINC(xsmincdp, 0);
2669 #define VSX_MAX_MINJ(name, max) \
2670 void helper_##name(CPUPPCState *env, uint32_t opcode) \
2672 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32]; \
2673 ppc_vsr_t *xa = &env->vsr[rA(opcode) + 32]; \
2674 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32]; \
2675 ppc_vsr_t t = *xt; \
2676 bool vxsnan_flag = false, vex_flag = false; \
2678 if (unlikely(float64_is_any_nan(xa->VsrD(0)))) { \
2679 if (float64_is_signaling_nan(xa->VsrD(0), &env->fp_status)) { \
2680 vxsnan_flag = true; \
2682 t.VsrD(0) = xa->VsrD(0); \
2683 } else if (unlikely(float64_is_any_nan(xb->VsrD(0)))) { \
2684 if (float64_is_signaling_nan(xb->VsrD(0), &env->fp_status)) { \
2685 vxsnan_flag = true; \
2687 t.VsrD(0) = xb->VsrD(0); \
2688 } else if (float64_is_zero(xa->VsrD(0)) && \
2689 float64_is_zero(xb->VsrD(0))) { \
2690 if (max) { \
2691 if (!float64_is_neg(xa->VsrD(0)) || \
2692 !float64_is_neg(xb->VsrD(0))) { \
2693 t.VsrD(0) = 0ULL; \
2694 } else { \
2695 t.VsrD(0) = 0x8000000000000000ULL; \
2697 } else { \
2698 if (float64_is_neg(xa->VsrD(0)) || \
2699 float64_is_neg(xb->VsrD(0))) { \
2700 t.VsrD(0) = 0x8000000000000000ULL; \
2701 } else { \
2702 t.VsrD(0) = 0ULL; \
2705 } else if ((max && \
2706 !float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status)) || \
2707 (!max && \
2708 float64_lt(xa->VsrD(0), xb->VsrD(0), &env->fp_status))) { \
2709 t.VsrD(0) = xa->VsrD(0); \
2710 } else { \
2711 t.VsrD(0) = xb->VsrD(0); \
2714 vex_flag = fpscr_ve & vxsnan_flag; \
2715 if (vxsnan_flag) { \
2716 float_invalid_op_vxsnan(env, GETPC()); \
2718 if (!vex_flag) { \
2719 *xt = t; \
2723 VSX_MAX_MINJ(xsmaxjdp, 1);
2724 VSX_MAX_MINJ(xsminjdp, 0);
2727 * VSX_CMP - VSX floating point compare
2728 * op - instruction mnemonic
2729 * nels - number of elements (1, 2 or 4)
2730 * tp - type (float32 or float64)
2731 * fld - vsr_t field (VsrD(*) or VsrW(*))
2732 * cmp - comparison operation
2733 * svxvc - set VXVC bit
2734 * exp - expected result of comparison
2736 #define VSX_CMP(op, nels, tp, fld, cmp, svxvc, exp) \
2737 uint32_t helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
2738 ppc_vsr_t *xa, ppc_vsr_t *xb) \
2740 ppc_vsr_t t = *xt; \
2741 uint32_t crf6 = 0; \
2742 int i; \
2743 int all_true = 1; \
2744 int all_false = 1; \
2746 for (i = 0; i < nels; i++) { \
2747 if (unlikely(tp##_is_any_nan(xa->fld) || \
2748 tp##_is_any_nan(xb->fld))) { \
2749 if (tp##_is_signaling_nan(xa->fld, &env->fp_status) || \
2750 tp##_is_signaling_nan(xb->fld, &env->fp_status)) { \
2751 float_invalid_op_vxsnan(env, GETPC()); \
2753 if (svxvc) { \
2754 float_invalid_op_vxvc(env, 0, GETPC()); \
2756 t.fld = 0; \
2757 all_true = 0; \
2758 } else { \
2759 if (tp##_##cmp(xb->fld, xa->fld, &env->fp_status) == exp) { \
2760 t.fld = -1; \
2761 all_false = 0; \
2762 } else { \
2763 t.fld = 0; \
2764 all_true = 0; \
2769 *xt = t; \
2770 crf6 = (all_true ? 0x8 : 0) | (all_false ? 0x2 : 0); \
2771 return crf6; \
2774 VSX_CMP(xvcmpeqdp, 2, float64, VsrD(i), eq, 0, 1)
2775 VSX_CMP(xvcmpgedp, 2, float64, VsrD(i), le, 1, 1)
2776 VSX_CMP(xvcmpgtdp, 2, float64, VsrD(i), lt, 1, 1)
2777 VSX_CMP(xvcmpnedp, 2, float64, VsrD(i), eq, 0, 0)
2778 VSX_CMP(xvcmpeqsp, 4, float32, VsrW(i), eq, 0, 1)
2779 VSX_CMP(xvcmpgesp, 4, float32, VsrW(i), le, 1, 1)
2780 VSX_CMP(xvcmpgtsp, 4, float32, VsrW(i), lt, 1, 1)
2781 VSX_CMP(xvcmpnesp, 4, float32, VsrW(i), eq, 0, 0)
2784 * VSX_CVT_FP_TO_FP - VSX floating point/floating point conversion
2785 * op - instruction mnemonic
2786 * nels - number of elements (1, 2 or 4)
2787 * stp - source type (float32 or float64)
2788 * ttp - target type (float32 or float64)
2789 * sfld - source vsr_t field
2790 * tfld - target vsr_t field (f32 or f64)
2791 * sfprf - set FPRF
2793 #define VSX_CVT_FP_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf) \
2794 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2796 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
2797 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
2798 ppc_vsr_t t = *xt; \
2799 int i; \
2801 for (i = 0; i < nels; i++) { \
2802 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2803 if (unlikely(stp##_is_signaling_nan(xb->sfld, \
2804 &env->fp_status))) { \
2805 float_invalid_op_vxsnan(env, GETPC()); \
2806 t.tfld = ttp##_snan_to_qnan(t.tfld); \
2808 if (sfprf) { \
2809 helper_compute_fprf_##ttp(env, t.tfld); \
2813 *xt = t; \
2814 do_float_check_status(env, GETPC()); \
2817 VSX_CVT_FP_TO_FP(xscvdpsp, 1, float64, float32, VsrD(0), VsrW(0), 1)
2818 VSX_CVT_FP_TO_FP(xscvspdp, 1, float32, float64, VsrW(0), VsrD(0), 1)
2819 VSX_CVT_FP_TO_FP(xvcvdpsp, 2, float64, float32, VsrD(i), VsrW(2 * i), 0)
2820 VSX_CVT_FP_TO_FP(xvcvspdp, 2, float32, float64, VsrW(2 * i), VsrD(i), 0)
2823 * VSX_CVT_FP_TO_FP_VECTOR - VSX floating point/floating point conversion
2824 * op - instruction mnemonic
2825 * nels - number of elements (1, 2 or 4)
2826 * stp - source type (float32 or float64)
2827 * ttp - target type (float32 or float64)
2828 * sfld - source vsr_t field
2829 * tfld - target vsr_t field (f32 or f64)
2830 * sfprf - set FPRF
2832 #define VSX_CVT_FP_TO_FP_VECTOR(op, nels, stp, ttp, sfld, tfld, sfprf) \
2833 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2835 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32]; \
2836 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32]; \
2837 ppc_vsr_t t = *xt; \
2838 int i; \
2840 for (i = 0; i < nels; i++) { \
2841 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
2842 if (unlikely(stp##_is_signaling_nan(xb->sfld, \
2843 &env->fp_status))) { \
2844 float_invalid_op_vxsnan(env, GETPC()); \
2845 t.tfld = ttp##_snan_to_qnan(t.tfld); \
2847 if (sfprf) { \
2848 helper_compute_fprf_##ttp(env, t.tfld); \
2852 *xt = t; \
2853 do_float_check_status(env, GETPC()); \
2856 VSX_CVT_FP_TO_FP_VECTOR(xscvdpqp, 1, float64, float128, VsrD(0), f128, 1)
2859 * VSX_CVT_FP_TO_FP_HP - VSX floating point/floating point conversion
2860 * involving one half precision value
2861 * op - instruction mnemonic
2862 * nels - number of elements (1, 2 or 4)
2863 * stp - source type
2864 * ttp - target type
2865 * sfld - source vsr_t field
2866 * tfld - target vsr_t field
2867 * sfprf - set FPRF
2869 #define VSX_CVT_FP_TO_FP_HP(op, nels, stp, ttp, sfld, tfld, sfprf) \
2870 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2872 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
2873 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
2874 ppc_vsr_t t = { }; \
2875 int i; \
2877 for (i = 0; i < nels; i++) { \
2878 t.tfld = stp##_to_##ttp(xb->sfld, 1, &env->fp_status); \
2879 if (unlikely(stp##_is_signaling_nan(xb->sfld, \
2880 &env->fp_status))) { \
2881 float_invalid_op_vxsnan(env, GETPC()); \
2882 t.tfld = ttp##_snan_to_qnan(t.tfld); \
2884 if (sfprf) { \
2885 helper_compute_fprf_##ttp(env, t.tfld); \
2889 *xt = t; \
2890 do_float_check_status(env, GETPC()); \
2893 VSX_CVT_FP_TO_FP_HP(xscvdphp, 1, float64, float16, VsrD(0), VsrH(3), 1)
2894 VSX_CVT_FP_TO_FP_HP(xscvhpdp, 1, float16, float64, VsrH(3), VsrD(0), 1)
2895 VSX_CVT_FP_TO_FP_HP(xvcvsphp, 4, float32, float16, VsrW(i), VsrH(2 * i + 1), 0)
2896 VSX_CVT_FP_TO_FP_HP(xvcvhpsp, 4, float16, float32, VsrH(2 * i + 1), VsrW(i), 0)
2899 * xscvqpdp isn't using VSX_CVT_FP_TO_FP() because xscvqpdpo will be
2900 * added to this later.
2902 void helper_xscvqpdp(CPUPPCState *env, uint32_t opcode,
2903 ppc_vsr_t *xt, ppc_vsr_t *xb)
2905 ppc_vsr_t t = { };
2906 float_status tstat;
2908 tstat = env->fp_status;
2909 if (unlikely(Rc(opcode) != 0)) {
2910 tstat.float_rounding_mode = float_round_to_odd;
2913 t.VsrD(0) = float128_to_float64(xb->f128, &tstat);
2914 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
2915 if (unlikely(float128_is_signaling_nan(xb->f128, &tstat))) {
2916 float_invalid_op_vxsnan(env, GETPC());
2917 t.VsrD(0) = float64_snan_to_qnan(t.VsrD(0));
2919 helper_compute_fprf_float64(env, t.VsrD(0));
2921 *xt = t;
2922 do_float_check_status(env, GETPC());
2925 uint64_t helper_xscvdpspn(CPUPPCState *env, uint64_t xb)
2927 float_status tstat = env->fp_status;
2928 set_float_exception_flags(0, &tstat);
2930 return (uint64_t)float64_to_float32(xb, &tstat) << 32;
2933 uint64_t helper_xscvspdpn(CPUPPCState *env, uint64_t xb)
2935 float_status tstat = env->fp_status;
2936 set_float_exception_flags(0, &tstat);
2938 return float32_to_float64(xb >> 32, &tstat);
2942 * VSX_CVT_FP_TO_INT - VSX floating point to integer conversion
2943 * op - instruction mnemonic
2944 * nels - number of elements (1, 2 or 4)
2945 * stp - source type (float32 or float64)
2946 * ttp - target type (int32, uint32, int64 or uint64)
2947 * sfld - source vsr_t field
2948 * tfld - target vsr_t field
2949 * rnan - resulting NaN
2951 #define VSX_CVT_FP_TO_INT(op, nels, stp, ttp, sfld, tfld, rnan) \
2952 void helper_##op(CPUPPCState *env, uint32_t opcode) \
2954 int all_flags = env->fp_status.float_exception_flags, flags; \
2955 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
2956 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
2957 ppc_vsr_t t = *xt; \
2958 int i; \
2960 for (i = 0; i < nels; i++) { \
2961 env->fp_status.float_exception_flags = 0; \
2962 t.tfld = stp##_to_##ttp##_round_to_zero(xb->sfld, &env->fp_status); \
2963 flags = env->fp_status.float_exception_flags; \
2964 if (unlikely(flags & float_flag_invalid)) { \
2965 float_invalid_cvt(env, 0, GETPC(), stp##_classify(xb->sfld)); \
2966 t.tfld = rnan; \
2968 all_flags |= flags; \
2971 *xt = t; \
2972 env->fp_status.float_exception_flags = all_flags; \
2973 do_float_check_status(env, GETPC()); \
2976 VSX_CVT_FP_TO_INT(xscvdpsxds, 1, float64, int64, VsrD(0), VsrD(0), \
2977 0x8000000000000000ULL)
2978 VSX_CVT_FP_TO_INT(xscvdpsxws, 1, float64, int32, VsrD(0), VsrW(1), \
2979 0x80000000U)
2980 VSX_CVT_FP_TO_INT(xscvdpuxds, 1, float64, uint64, VsrD(0), VsrD(0), 0ULL)
2981 VSX_CVT_FP_TO_INT(xscvdpuxws, 1, float64, uint32, VsrD(0), VsrW(1), 0U)
2982 VSX_CVT_FP_TO_INT(xvcvdpsxds, 2, float64, int64, VsrD(i), VsrD(i), \
2983 0x8000000000000000ULL)
2984 VSX_CVT_FP_TO_INT(xvcvdpsxws, 2, float64, int32, VsrD(i), VsrW(2 * i), \
2985 0x80000000U)
2986 VSX_CVT_FP_TO_INT(xvcvdpuxds, 2, float64, uint64, VsrD(i), VsrD(i), 0ULL)
2987 VSX_CVT_FP_TO_INT(xvcvdpuxws, 2, float64, uint32, VsrD(i), VsrW(2 * i), 0U)
2988 VSX_CVT_FP_TO_INT(xvcvspsxds, 2, float32, int64, VsrW(2 * i), VsrD(i), \
2989 0x8000000000000000ULL)
2990 VSX_CVT_FP_TO_INT(xvcvspsxws, 4, float32, int32, VsrW(i), VsrW(i), 0x80000000U)
2991 VSX_CVT_FP_TO_INT(xvcvspuxds, 2, float32, uint64, VsrW(2 * i), VsrD(i), 0ULL)
2992 VSX_CVT_FP_TO_INT(xvcvspuxws, 4, float32, uint32, VsrW(i), VsrW(i), 0U)
2995 * VSX_CVT_FP_TO_INT_VECTOR - VSX floating point to integer conversion
2996 * op - instruction mnemonic
2997 * stp - source type (float32 or float64)
2998 * ttp - target type (int32, uint32, int64 or uint64)
2999 * sfld - source vsr_t field
3000 * tfld - target vsr_t field
3001 * rnan - resulting NaN
3003 #define VSX_CVT_FP_TO_INT_VECTOR(op, stp, ttp, sfld, tfld, rnan) \
3004 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3006 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32]; \
3007 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32]; \
3008 ppc_vsr_t t = { }; \
3010 t.tfld = stp##_to_##ttp##_round_to_zero(xb->sfld, &env->fp_status); \
3011 if (env->fp_status.float_exception_flags & float_flag_invalid) { \
3012 float_invalid_cvt(env, 0, GETPC(), stp##_classify(xb->sfld)); \
3013 t.tfld = rnan; \
3016 *xt = t; \
3017 do_float_check_status(env, GETPC()); \
3020 VSX_CVT_FP_TO_INT_VECTOR(xscvqpsdz, float128, int64, f128, VsrD(0), \
3021 0x8000000000000000ULL)
3023 VSX_CVT_FP_TO_INT_VECTOR(xscvqpswz, float128, int32, f128, VsrD(0), \
3024 0xffffffff80000000ULL)
3025 VSX_CVT_FP_TO_INT_VECTOR(xscvqpudz, float128, uint64, f128, VsrD(0), 0x0ULL)
3026 VSX_CVT_FP_TO_INT_VECTOR(xscvqpuwz, float128, uint32, f128, VsrD(0), 0x0ULL)
3029 * VSX_CVT_INT_TO_FP - VSX integer to floating point conversion
3030 * op - instruction mnemonic
3031 * nels - number of elements (1, 2 or 4)
3032 * stp - source type (int32, uint32, int64 or uint64)
3033 * ttp - target type (float32 or float64)
3034 * sfld - source vsr_t field
3035 * tfld - target vsr_t field
3036 * jdef - definition of the j index (i or 2*i)
3037 * sfprf - set FPRF
3039 #define VSX_CVT_INT_TO_FP(op, nels, stp, ttp, sfld, tfld, sfprf, r2sp) \
3040 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3042 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
3043 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
3044 ppc_vsr_t t = *xt; \
3045 int i; \
3047 for (i = 0; i < nels; i++) { \
3048 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
3049 if (r2sp) { \
3050 t.tfld = helper_frsp(env, t.tfld); \
3052 if (sfprf) { \
3053 helper_compute_fprf_float64(env, t.tfld); \
3057 *xt = t; \
3058 do_float_check_status(env, GETPC()); \
3061 VSX_CVT_INT_TO_FP(xscvsxddp, 1, int64, float64, VsrD(0), VsrD(0), 1, 0)
3062 VSX_CVT_INT_TO_FP(xscvuxddp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 0)
3063 VSX_CVT_INT_TO_FP(xscvsxdsp, 1, int64, float64, VsrD(0), VsrD(0), 1, 1)
3064 VSX_CVT_INT_TO_FP(xscvuxdsp, 1, uint64, float64, VsrD(0), VsrD(0), 1, 1)
3065 VSX_CVT_INT_TO_FP(xvcvsxddp, 2, int64, float64, VsrD(i), VsrD(i), 0, 0)
3066 VSX_CVT_INT_TO_FP(xvcvuxddp, 2, uint64, float64, VsrD(i), VsrD(i), 0, 0)
3067 VSX_CVT_INT_TO_FP(xvcvsxwdp, 2, int32, float64, VsrW(2 * i), VsrD(i), 0, 0)
3068 VSX_CVT_INT_TO_FP(xvcvuxwdp, 2, uint64, float64, VsrW(2 * i), VsrD(i), 0, 0)
3069 VSX_CVT_INT_TO_FP(xvcvsxdsp, 2, int64, float32, VsrD(i), VsrW(2 * i), 0, 0)
3070 VSX_CVT_INT_TO_FP(xvcvuxdsp, 2, uint64, float32, VsrD(i), VsrW(2 * i), 0, 0)
3071 VSX_CVT_INT_TO_FP(xvcvsxwsp, 4, int32, float32, VsrW(i), VsrW(i), 0, 0)
3072 VSX_CVT_INT_TO_FP(xvcvuxwsp, 4, uint32, float32, VsrW(i), VsrW(i), 0, 0)
3075 * VSX_CVT_INT_TO_FP_VECTOR - VSX integer to floating point conversion
3076 * op - instruction mnemonic
3077 * stp - source type (int32, uint32, int64 or uint64)
3078 * ttp - target type (float32 or float64)
3079 * sfld - source vsr_t field
3080 * tfld - target vsr_t field
3082 #define VSX_CVT_INT_TO_FP_VECTOR(op, stp, ttp, sfld, tfld) \
3083 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3085 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32]; \
3086 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32]; \
3087 ppc_vsr_t t = *xt; \
3089 t.tfld = stp##_to_##ttp(xb->sfld, &env->fp_status); \
3090 helper_compute_fprf_##ttp(env, t.tfld); \
3092 *xt = t; \
3093 do_float_check_status(env, GETPC()); \
3096 VSX_CVT_INT_TO_FP_VECTOR(xscvsdqp, int64, float128, VsrD(0), f128)
3097 VSX_CVT_INT_TO_FP_VECTOR(xscvudqp, uint64, float128, VsrD(0), f128)
3100 * For "use current rounding mode", define a value that will not be
3101 * one of the existing rounding model enums.
3103 #define FLOAT_ROUND_CURRENT (float_round_nearest_even + float_round_down + \
3104 float_round_up + float_round_to_zero)
3107 * VSX_ROUND - VSX floating point round
3108 * op - instruction mnemonic
3109 * nels - number of elements (1, 2 or 4)
3110 * tp - type (float32 or float64)
3111 * fld - vsr_t field (VsrD(*) or VsrW(*))
3112 * rmode - rounding mode
3113 * sfprf - set FPRF
3115 #define VSX_ROUND(op, nels, tp, fld, rmode, sfprf) \
3116 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3118 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
3119 ppc_vsr_t *xb = &env->vsr[xB(opcode)]; \
3120 ppc_vsr_t t = *xt; \
3121 int i; \
3123 if (rmode != FLOAT_ROUND_CURRENT) { \
3124 set_float_rounding_mode(rmode, &env->fp_status); \
3127 for (i = 0; i < nels; i++) { \
3128 if (unlikely(tp##_is_signaling_nan(xb->fld, \
3129 &env->fp_status))) { \
3130 float_invalid_op_vxsnan(env, GETPC()); \
3131 t.fld = tp##_snan_to_qnan(xb->fld); \
3132 } else { \
3133 t.fld = tp##_round_to_int(xb->fld, &env->fp_status); \
3135 if (sfprf) { \
3136 helper_compute_fprf_float64(env, t.fld); \
3140 /* \
3141 * If this is not a "use current rounding mode" instruction, \
3142 * then inhibit setting of the XX bit and restore rounding \
3143 * mode from FPSCR \
3144 */ \
3145 if (rmode != FLOAT_ROUND_CURRENT) { \
3146 fpscr_set_rounding_mode(env); \
3147 env->fp_status.float_exception_flags &= ~float_flag_inexact; \
3150 *xt = t; \
3151 do_float_check_status(env, GETPC()); \
3154 VSX_ROUND(xsrdpi, 1, float64, VsrD(0), float_round_ties_away, 1)
3155 VSX_ROUND(xsrdpic, 1, float64, VsrD(0), FLOAT_ROUND_CURRENT, 1)
3156 VSX_ROUND(xsrdpim, 1, float64, VsrD(0), float_round_down, 1)
3157 VSX_ROUND(xsrdpip, 1, float64, VsrD(0), float_round_up, 1)
3158 VSX_ROUND(xsrdpiz, 1, float64, VsrD(0), float_round_to_zero, 1)
3160 VSX_ROUND(xvrdpi, 2, float64, VsrD(i), float_round_ties_away, 0)
3161 VSX_ROUND(xvrdpic, 2, float64, VsrD(i), FLOAT_ROUND_CURRENT, 0)
3162 VSX_ROUND(xvrdpim, 2, float64, VsrD(i), float_round_down, 0)
3163 VSX_ROUND(xvrdpip, 2, float64, VsrD(i), float_round_up, 0)
3164 VSX_ROUND(xvrdpiz, 2, float64, VsrD(i), float_round_to_zero, 0)
3166 VSX_ROUND(xvrspi, 4, float32, VsrW(i), float_round_ties_away, 0)
3167 VSX_ROUND(xvrspic, 4, float32, VsrW(i), FLOAT_ROUND_CURRENT, 0)
3168 VSX_ROUND(xvrspim, 4, float32, VsrW(i), float_round_down, 0)
3169 VSX_ROUND(xvrspip, 4, float32, VsrW(i), float_round_up, 0)
3170 VSX_ROUND(xvrspiz, 4, float32, VsrW(i), float_round_to_zero, 0)
3172 uint64_t helper_xsrsp(CPUPPCState *env, uint64_t xb)
3174 helper_reset_fpstatus(env);
3176 uint64_t xt = helper_frsp(env, xb);
3178 helper_compute_fprf_float64(env, xt);
3179 do_float_check_status(env, GETPC());
3180 return xt;
3183 #define VSX_XXPERM(op, indexed) \
3184 void helper_##op(CPUPPCState *env, ppc_vsr_t *xt, \
3185 ppc_vsr_t *xa, ppc_vsr_t *pcv) \
3187 ppc_vsr_t t = *xt; \
3188 int i, idx; \
3190 for (i = 0; i < 16; i++) { \
3191 idx = pcv->VsrB(i) & 0x1F; \
3192 if (indexed) { \
3193 idx = 31 - idx; \
3195 t.VsrB(i) = (idx <= 15) ? xa->VsrB(idx) \
3196 : xt->VsrB(idx - 16); \
3198 *xt = t; \
3201 VSX_XXPERM(xxperm, 0)
3202 VSX_XXPERM(xxpermr, 1)
3204 void helper_xvxsigsp(CPUPPCState *env, uint32_t opcode)
3206 ppc_vsr_t *xt = &env->vsr[xT(opcode)];
3207 ppc_vsr_t *xb = &env->vsr[xB(opcode)];
3208 ppc_vsr_t t = { };
3209 uint32_t exp, i, fraction;
3211 for (i = 0; i < 4; i++) {
3212 exp = (xb->VsrW(i) >> 23) & 0xFF;
3213 fraction = xb->VsrW(i) & 0x7FFFFF;
3214 if (exp != 0 && exp != 255) {
3215 t.VsrW(i) = fraction | 0x00800000;
3216 } else {
3217 t.VsrW(i) = fraction;
3220 *xt = t;
3224 * VSX_TEST_DC - VSX floating point test data class
3225 * op - instruction mnemonic
3226 * nels - number of elements (1, 2 or 4)
3227 * xbn - VSR register number
3228 * tp - type (float32 or float64)
3229 * fld - vsr_t field (VsrD(*) or VsrW(*))
3230 * tfld - target vsr_t field (VsrD(*) or VsrW(*))
3231 * fld_max - target field max
3232 * scrf - set result in CR and FPCC
3234 #define VSX_TEST_DC(op, nels, xbn, tp, fld, tfld, fld_max, scrf) \
3235 void helper_##op(CPUPPCState *env, uint32_t opcode) \
3237 ppc_vsr_t *xt = &env->vsr[xT(opcode)]; \
3238 ppc_vsr_t *xb = &env->vsr[xbn]; \
3239 ppc_vsr_t t = { }; \
3240 uint32_t i, sign, dcmx; \
3241 uint32_t cc, match = 0; \
3243 if (!scrf) { \
3244 dcmx = DCMX_XV(opcode); \
3245 } else { \
3246 t = *xt; \
3247 dcmx = DCMX(opcode); \
3250 for (i = 0; i < nels; i++) { \
3251 sign = tp##_is_neg(xb->fld); \
3252 if (tp##_is_any_nan(xb->fld)) { \
3253 match = extract32(dcmx, 6, 1); \
3254 } else if (tp##_is_infinity(xb->fld)) { \
3255 match = extract32(dcmx, 4 + !sign, 1); \
3256 } else if (tp##_is_zero(xb->fld)) { \
3257 match = extract32(dcmx, 2 + !sign, 1); \
3258 } else if (tp##_is_zero_or_denormal(xb->fld)) { \
3259 match = extract32(dcmx, 0 + !sign, 1); \
3262 if (scrf) { \
3263 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT; \
3264 env->fpscr &= ~(0x0F << FPSCR_FPRF); \
3265 env->fpscr |= cc << FPSCR_FPRF; \
3266 env->crf[BF(opcode)] = cc; \
3267 } else { \
3268 t.tfld = match ? fld_max : 0; \
3270 match = 0; \
3272 if (!scrf) { \
3273 *xt = t; \
3277 VSX_TEST_DC(xvtstdcdp, 2, xB(opcode), float64, VsrD(i), VsrD(i), UINT64_MAX, 0)
3278 VSX_TEST_DC(xvtstdcsp, 4, xB(opcode), float32, VsrW(i), VsrW(i), UINT32_MAX, 0)
3279 VSX_TEST_DC(xststdcdp, 1, xB(opcode), float64, VsrD(0), VsrD(0), 0, 1)
3280 VSX_TEST_DC(xststdcqp, 1, (rB(opcode) + 32), float128, f128, VsrD(0), 0, 1)
3282 void helper_xststdcsp(CPUPPCState *env, uint32_t opcode)
3284 ppc_vsr_t *xb = &env->vsr[xB(opcode)];
3285 uint32_t dcmx, sign, exp;
3286 uint32_t cc, match = 0, not_sp = 0;
3288 dcmx = DCMX(opcode);
3289 exp = (xb->VsrD(0) >> 52) & 0x7FF;
3291 sign = float64_is_neg(xb->VsrD(0));
3292 if (float64_is_any_nan(xb->VsrD(0))) {
3293 match = extract32(dcmx, 6, 1);
3294 } else if (float64_is_infinity(xb->VsrD(0))) {
3295 match = extract32(dcmx, 4 + !sign, 1);
3296 } else if (float64_is_zero(xb->VsrD(0))) {
3297 match = extract32(dcmx, 2 + !sign, 1);
3298 } else if (float64_is_zero_or_denormal(xb->VsrD(0)) ||
3299 (exp > 0 && exp < 0x381)) {
3300 match = extract32(dcmx, 0 + !sign, 1);
3303 not_sp = !float64_eq(xb->VsrD(0),
3304 float32_to_float64(
3305 float64_to_float32(xb->VsrD(0), &env->fp_status),
3306 &env->fp_status), &env->fp_status);
3308 cc = sign << CRF_LT_BIT | match << CRF_EQ_BIT | not_sp << CRF_SO_BIT;
3309 env->fpscr &= ~(0x0F << FPSCR_FPRF);
3310 env->fpscr |= cc << FPSCR_FPRF;
3311 env->crf[BF(opcode)] = cc;
3314 void helper_xsrqpi(CPUPPCState *env, uint32_t opcode)
3316 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32];
3317 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32];
3318 ppc_vsr_t t = { };
3319 uint8_t r = Rrm(opcode);
3320 uint8_t ex = Rc(opcode);
3321 uint8_t rmc = RMC(opcode);
3322 uint8_t rmode = 0;
3323 float_status tstat;
3325 helper_reset_fpstatus(env);
3327 if (r == 0 && rmc == 0) {
3328 rmode = float_round_ties_away;
3329 } else if (r == 0 && rmc == 0x3) {
3330 rmode = fpscr_rn;
3331 } else if (r == 1) {
3332 switch (rmc) {
3333 case 0:
3334 rmode = float_round_nearest_even;
3335 break;
3336 case 1:
3337 rmode = float_round_to_zero;
3338 break;
3339 case 2:
3340 rmode = float_round_up;
3341 break;
3342 case 3:
3343 rmode = float_round_down;
3344 break;
3345 default:
3346 abort();
3350 tstat = env->fp_status;
3351 set_float_exception_flags(0, &tstat);
3352 set_float_rounding_mode(rmode, &tstat);
3353 t.f128 = float128_round_to_int(xb->f128, &tstat);
3354 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3356 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3357 if (float128_is_signaling_nan(xb->f128, &tstat)) {
3358 float_invalid_op_vxsnan(env, GETPC());
3359 t.f128 = float128_snan_to_qnan(t.f128);
3363 if (ex == 0 && (tstat.float_exception_flags & float_flag_inexact)) {
3364 env->fp_status.float_exception_flags &= ~float_flag_inexact;
3367 helper_compute_fprf_float128(env, t.f128);
3368 do_float_check_status(env, GETPC());
3369 *xt = t;
3372 void helper_xsrqpxp(CPUPPCState *env, uint32_t opcode)
3374 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32];
3375 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32];
3376 ppc_vsr_t t = { };
3377 uint8_t r = Rrm(opcode);
3378 uint8_t rmc = RMC(opcode);
3379 uint8_t rmode = 0;
3380 floatx80 round_res;
3381 float_status tstat;
3383 helper_reset_fpstatus(env);
3385 if (r == 0 && rmc == 0) {
3386 rmode = float_round_ties_away;
3387 } else if (r == 0 && rmc == 0x3) {
3388 rmode = fpscr_rn;
3389 } else if (r == 1) {
3390 switch (rmc) {
3391 case 0:
3392 rmode = float_round_nearest_even;
3393 break;
3394 case 1:
3395 rmode = float_round_to_zero;
3396 break;
3397 case 2:
3398 rmode = float_round_up;
3399 break;
3400 case 3:
3401 rmode = float_round_down;
3402 break;
3403 default:
3404 abort();
3408 tstat = env->fp_status;
3409 set_float_exception_flags(0, &tstat);
3410 set_float_rounding_mode(rmode, &tstat);
3411 round_res = float128_to_floatx80(xb->f128, &tstat);
3412 t.f128 = floatx80_to_float128(round_res, &tstat);
3413 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3415 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3416 if (float128_is_signaling_nan(xb->f128, &tstat)) {
3417 float_invalid_op_vxsnan(env, GETPC());
3418 t.f128 = float128_snan_to_qnan(t.f128);
3422 helper_compute_fprf_float128(env, t.f128);
3423 *xt = t;
3424 do_float_check_status(env, GETPC());
3427 void helper_xssqrtqp(CPUPPCState *env, uint32_t opcode)
3429 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32];
3430 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32];
3431 ppc_vsr_t t = { };
3432 float_status tstat;
3434 helper_reset_fpstatus(env);
3436 tstat = env->fp_status;
3437 if (unlikely(Rc(opcode) != 0)) {
3438 tstat.float_rounding_mode = float_round_to_odd;
3441 set_float_exception_flags(0, &tstat);
3442 t.f128 = float128_sqrt(xb->f128, &tstat);
3443 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3445 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3446 if (float128_is_signaling_nan(xb->f128, &tstat)) {
3447 float_invalid_op_vxsnan(env, GETPC());
3448 t.f128 = float128_snan_to_qnan(xb->f128);
3449 } else if (float128_is_quiet_nan(xb->f128, &tstat)) {
3450 t.f128 = xb->f128;
3451 } else if (float128_is_neg(xb->f128) && !float128_is_zero(xb->f128)) {
3452 float_invalid_op_vxsqrt(env, 1, GETPC());
3453 t.f128 = float128_default_nan(&env->fp_status);
3457 helper_compute_fprf_float128(env, t.f128);
3458 *xt = t;
3459 do_float_check_status(env, GETPC());
3462 void helper_xssubqp(CPUPPCState *env, uint32_t opcode)
3464 ppc_vsr_t *xt = &env->vsr[rD(opcode) + 32];
3465 ppc_vsr_t *xa = &env->vsr[rA(opcode) + 32];
3466 ppc_vsr_t *xb = &env->vsr[rB(opcode) + 32];
3467 ppc_vsr_t t = *xt;
3468 float_status tstat;
3470 helper_reset_fpstatus(env);
3472 tstat = env->fp_status;
3473 if (unlikely(Rc(opcode) != 0)) {
3474 tstat.float_rounding_mode = float_round_to_odd;
3477 set_float_exception_flags(0, &tstat);
3478 t.f128 = float128_sub(xa->f128, xb->f128, &tstat);
3479 env->fp_status.float_exception_flags |= tstat.float_exception_flags;
3481 if (unlikely(tstat.float_exception_flags & float_flag_invalid)) {
3482 float_invalid_op_addsub(env, 1, GETPC(),
3483 float128_classify(xa->f128) |
3484 float128_classify(xb->f128));
3487 helper_compute_fprf_float128(env, t.f128);
3488 *xt = t;
3489 do_float_check_status(env, GETPC());