Merge remote-tracking branch 'remotes/sstabellini/tags/xen-20160622-tag' into staging
[qemu/ar7.git] / target-i386 / fpu_helper.c
blob929489bfc81f2fd747a43f554c1cafbc21df51d6
1 /*
2 * x86 FPU, MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4/PNI helpers
4 * Copyright (c) 2003 Fabrice Bellard
6 * This library is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2 of the License, or (at your option) any later version.
11 * This library is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with this library; if not, see <http://www.gnu.org/licenses/>.
20 #include "qemu/osdep.h"
21 #include <math.h>
22 #include "cpu.h"
23 #include "exec/helper-proto.h"
24 #include "qemu/host-utils.h"
25 #include "exec/exec-all.h"
26 #include "exec/cpu_ldst.h"
28 #define FPU_RC_MASK 0xc00
29 #define FPU_RC_NEAR 0x000
30 #define FPU_RC_DOWN 0x400
31 #define FPU_RC_UP 0x800
32 #define FPU_RC_CHOP 0xc00
34 #define MAXTAN 9223372036854775808.0
36 /* the following deal with x86 long double-precision numbers */
37 #define MAXEXPD 0x7fff
38 #define EXPBIAS 16383
39 #define EXPD(fp) (fp.l.upper & 0x7fff)
40 #define SIGND(fp) ((fp.l.upper) & 0x8000)
41 #define MANTD(fp) (fp.l.lower)
42 #define BIASEXPONENT(fp) fp.l.upper = (fp.l.upper & ~(0x7fff)) | EXPBIAS
44 #define FPUS_IE (1 << 0)
45 #define FPUS_DE (1 << 1)
46 #define FPUS_ZE (1 << 2)
47 #define FPUS_OE (1 << 3)
48 #define FPUS_UE (1 << 4)
49 #define FPUS_PE (1 << 5)
50 #define FPUS_SF (1 << 6)
51 #define FPUS_SE (1 << 7)
52 #define FPUS_B (1 << 15)
54 #define FPUC_EM 0x3f
56 #define floatx80_lg2 make_floatx80(0x3ffd, 0x9a209a84fbcff799LL)
57 #define floatx80_l2e make_floatx80(0x3fff, 0xb8aa3b295c17f0bcLL)
58 #define floatx80_l2t make_floatx80(0x4000, 0xd49a784bcd1b8afeLL)
60 static inline void fpush(CPUX86State *env)
62 env->fpstt = (env->fpstt - 1) & 7;
63 env->fptags[env->fpstt] = 0; /* validate stack entry */
66 static inline void fpop(CPUX86State *env)
68 env->fptags[env->fpstt] = 1; /* invalidate stack entry */
69 env->fpstt = (env->fpstt + 1) & 7;
72 static inline floatx80 helper_fldt(CPUX86State *env, target_ulong ptr,
73 uintptr_t retaddr)
75 CPU_LDoubleU temp;
77 temp.l.lower = cpu_ldq_data_ra(env, ptr, retaddr);
78 temp.l.upper = cpu_lduw_data_ra(env, ptr + 8, retaddr);
79 return temp.d;
82 static inline void helper_fstt(CPUX86State *env, floatx80 f, target_ulong ptr,
83 uintptr_t retaddr)
85 CPU_LDoubleU temp;
87 temp.d = f;
88 cpu_stq_data_ra(env, ptr, temp.l.lower, retaddr);
89 cpu_stw_data_ra(env, ptr + 8, temp.l.upper, retaddr);
92 /* x87 FPU helpers */
94 static inline double floatx80_to_double(CPUX86State *env, floatx80 a)
96 union {
97 float64 f64;
98 double d;
99 } u;
101 u.f64 = floatx80_to_float64(a, &env->fp_status);
102 return u.d;
105 static inline floatx80 double_to_floatx80(CPUX86State *env, double a)
107 union {
108 float64 f64;
109 double d;
110 } u;
112 u.d = a;
113 return float64_to_floatx80(u.f64, &env->fp_status);
116 static void fpu_set_exception(CPUX86State *env, int mask)
118 env->fpus |= mask;
119 if (env->fpus & (~env->fpuc & FPUC_EM)) {
120 env->fpus |= FPUS_SE | FPUS_B;
124 static inline floatx80 helper_fdiv(CPUX86State *env, floatx80 a, floatx80 b)
126 if (floatx80_is_zero(b)) {
127 fpu_set_exception(env, FPUS_ZE);
129 return floatx80_div(a, b, &env->fp_status);
132 static void fpu_raise_exception(CPUX86State *env, uintptr_t retaddr)
134 if (env->cr[0] & CR0_NE_MASK) {
135 raise_exception_ra(env, EXCP10_COPR, retaddr);
137 #if !defined(CONFIG_USER_ONLY)
138 else {
139 cpu_set_ferr(env);
141 #endif
144 void helper_flds_FT0(CPUX86State *env, uint32_t val)
146 union {
147 float32 f;
148 uint32_t i;
149 } u;
151 u.i = val;
152 FT0 = float32_to_floatx80(u.f, &env->fp_status);
155 void helper_fldl_FT0(CPUX86State *env, uint64_t val)
157 union {
158 float64 f;
159 uint64_t i;
160 } u;
162 u.i = val;
163 FT0 = float64_to_floatx80(u.f, &env->fp_status);
166 void helper_fildl_FT0(CPUX86State *env, int32_t val)
168 FT0 = int32_to_floatx80(val, &env->fp_status);
171 void helper_flds_ST0(CPUX86State *env, uint32_t val)
173 int new_fpstt;
174 union {
175 float32 f;
176 uint32_t i;
177 } u;
179 new_fpstt = (env->fpstt - 1) & 7;
180 u.i = val;
181 env->fpregs[new_fpstt].d = float32_to_floatx80(u.f, &env->fp_status);
182 env->fpstt = new_fpstt;
183 env->fptags[new_fpstt] = 0; /* validate stack entry */
186 void helper_fldl_ST0(CPUX86State *env, uint64_t val)
188 int new_fpstt;
189 union {
190 float64 f;
191 uint64_t i;
192 } u;
194 new_fpstt = (env->fpstt - 1) & 7;
195 u.i = val;
196 env->fpregs[new_fpstt].d = float64_to_floatx80(u.f, &env->fp_status);
197 env->fpstt = new_fpstt;
198 env->fptags[new_fpstt] = 0; /* validate stack entry */
201 void helper_fildl_ST0(CPUX86State *env, int32_t val)
203 int new_fpstt;
205 new_fpstt = (env->fpstt - 1) & 7;
206 env->fpregs[new_fpstt].d = int32_to_floatx80(val, &env->fp_status);
207 env->fpstt = new_fpstt;
208 env->fptags[new_fpstt] = 0; /* validate stack entry */
211 void helper_fildll_ST0(CPUX86State *env, int64_t val)
213 int new_fpstt;
215 new_fpstt = (env->fpstt - 1) & 7;
216 env->fpregs[new_fpstt].d = int64_to_floatx80(val, &env->fp_status);
217 env->fpstt = new_fpstt;
218 env->fptags[new_fpstt] = 0; /* validate stack entry */
221 uint32_t helper_fsts_ST0(CPUX86State *env)
223 union {
224 float32 f;
225 uint32_t i;
226 } u;
228 u.f = floatx80_to_float32(ST0, &env->fp_status);
229 return u.i;
232 uint64_t helper_fstl_ST0(CPUX86State *env)
234 union {
235 float64 f;
236 uint64_t i;
237 } u;
239 u.f = floatx80_to_float64(ST0, &env->fp_status);
240 return u.i;
243 int32_t helper_fist_ST0(CPUX86State *env)
245 int32_t val;
247 val = floatx80_to_int32(ST0, &env->fp_status);
248 if (val != (int16_t)val) {
249 val = -32768;
251 return val;
254 int32_t helper_fistl_ST0(CPUX86State *env)
256 int32_t val;
257 signed char old_exp_flags;
259 old_exp_flags = get_float_exception_flags(&env->fp_status);
260 set_float_exception_flags(0, &env->fp_status);
262 val = floatx80_to_int32(ST0, &env->fp_status);
263 if (get_float_exception_flags(&env->fp_status) & float_flag_invalid) {
264 val = 0x80000000;
266 set_float_exception_flags(get_float_exception_flags(&env->fp_status)
267 | old_exp_flags, &env->fp_status);
268 return val;
271 int64_t helper_fistll_ST0(CPUX86State *env)
273 int64_t val;
274 signed char old_exp_flags;
276 old_exp_flags = get_float_exception_flags(&env->fp_status);
277 set_float_exception_flags(0, &env->fp_status);
279 val = floatx80_to_int64(ST0, &env->fp_status);
280 if (get_float_exception_flags(&env->fp_status) & float_flag_invalid) {
281 val = 0x8000000000000000ULL;
283 set_float_exception_flags(get_float_exception_flags(&env->fp_status)
284 | old_exp_flags, &env->fp_status);
285 return val;
288 int32_t helper_fistt_ST0(CPUX86State *env)
290 int32_t val;
292 val = floatx80_to_int32_round_to_zero(ST0, &env->fp_status);
293 if (val != (int16_t)val) {
294 val = -32768;
296 return val;
299 int32_t helper_fisttl_ST0(CPUX86State *env)
301 return floatx80_to_int32_round_to_zero(ST0, &env->fp_status);
304 int64_t helper_fisttll_ST0(CPUX86State *env)
306 return floatx80_to_int64_round_to_zero(ST0, &env->fp_status);
309 void helper_fldt_ST0(CPUX86State *env, target_ulong ptr)
311 int new_fpstt;
313 new_fpstt = (env->fpstt - 1) & 7;
314 env->fpregs[new_fpstt].d = helper_fldt(env, ptr, GETPC());
315 env->fpstt = new_fpstt;
316 env->fptags[new_fpstt] = 0; /* validate stack entry */
319 void helper_fstt_ST0(CPUX86State *env, target_ulong ptr)
321 helper_fstt(env, ST0, ptr, GETPC());
324 void helper_fpush(CPUX86State *env)
326 fpush(env);
329 void helper_fpop(CPUX86State *env)
331 fpop(env);
334 void helper_fdecstp(CPUX86State *env)
336 env->fpstt = (env->fpstt - 1) & 7;
337 env->fpus &= ~0x4700;
340 void helper_fincstp(CPUX86State *env)
342 env->fpstt = (env->fpstt + 1) & 7;
343 env->fpus &= ~0x4700;
346 /* FPU move */
348 void helper_ffree_STN(CPUX86State *env, int st_index)
350 env->fptags[(env->fpstt + st_index) & 7] = 1;
353 void helper_fmov_ST0_FT0(CPUX86State *env)
355 ST0 = FT0;
358 void helper_fmov_FT0_STN(CPUX86State *env, int st_index)
360 FT0 = ST(st_index);
363 void helper_fmov_ST0_STN(CPUX86State *env, int st_index)
365 ST0 = ST(st_index);
368 void helper_fmov_STN_ST0(CPUX86State *env, int st_index)
370 ST(st_index) = ST0;
373 void helper_fxchg_ST0_STN(CPUX86State *env, int st_index)
375 floatx80 tmp;
377 tmp = ST(st_index);
378 ST(st_index) = ST0;
379 ST0 = tmp;
382 /* FPU operations */
384 static const int fcom_ccval[4] = {0x0100, 0x4000, 0x0000, 0x4500};
386 void helper_fcom_ST0_FT0(CPUX86State *env)
388 int ret;
390 ret = floatx80_compare(ST0, FT0, &env->fp_status);
391 env->fpus = (env->fpus & ~0x4500) | fcom_ccval[ret + 1];
394 void helper_fucom_ST0_FT0(CPUX86State *env)
396 int ret;
398 ret = floatx80_compare_quiet(ST0, FT0, &env->fp_status);
399 env->fpus = (env->fpus & ~0x4500) | fcom_ccval[ret + 1];
402 static const int fcomi_ccval[4] = {CC_C, CC_Z, 0, CC_Z | CC_P | CC_C};
404 void helper_fcomi_ST0_FT0(CPUX86State *env)
406 int eflags;
407 int ret;
409 ret = floatx80_compare(ST0, FT0, &env->fp_status);
410 eflags = cpu_cc_compute_all(env, CC_OP);
411 eflags = (eflags & ~(CC_Z | CC_P | CC_C)) | fcomi_ccval[ret + 1];
412 CC_SRC = eflags;
415 void helper_fucomi_ST0_FT0(CPUX86State *env)
417 int eflags;
418 int ret;
420 ret = floatx80_compare_quiet(ST0, FT0, &env->fp_status);
421 eflags = cpu_cc_compute_all(env, CC_OP);
422 eflags = (eflags & ~(CC_Z | CC_P | CC_C)) | fcomi_ccval[ret + 1];
423 CC_SRC = eflags;
426 void helper_fadd_ST0_FT0(CPUX86State *env)
428 ST0 = floatx80_add(ST0, FT0, &env->fp_status);
431 void helper_fmul_ST0_FT0(CPUX86State *env)
433 ST0 = floatx80_mul(ST0, FT0, &env->fp_status);
436 void helper_fsub_ST0_FT0(CPUX86State *env)
438 ST0 = floatx80_sub(ST0, FT0, &env->fp_status);
441 void helper_fsubr_ST0_FT0(CPUX86State *env)
443 ST0 = floatx80_sub(FT0, ST0, &env->fp_status);
446 void helper_fdiv_ST0_FT0(CPUX86State *env)
448 ST0 = helper_fdiv(env, ST0, FT0);
451 void helper_fdivr_ST0_FT0(CPUX86State *env)
453 ST0 = helper_fdiv(env, FT0, ST0);
456 /* fp operations between STN and ST0 */
458 void helper_fadd_STN_ST0(CPUX86State *env, int st_index)
460 ST(st_index) = floatx80_add(ST(st_index), ST0, &env->fp_status);
463 void helper_fmul_STN_ST0(CPUX86State *env, int st_index)
465 ST(st_index) = floatx80_mul(ST(st_index), ST0, &env->fp_status);
468 void helper_fsub_STN_ST0(CPUX86State *env, int st_index)
470 ST(st_index) = floatx80_sub(ST(st_index), ST0, &env->fp_status);
473 void helper_fsubr_STN_ST0(CPUX86State *env, int st_index)
475 ST(st_index) = floatx80_sub(ST0, ST(st_index), &env->fp_status);
478 void helper_fdiv_STN_ST0(CPUX86State *env, int st_index)
480 floatx80 *p;
482 p = &ST(st_index);
483 *p = helper_fdiv(env, *p, ST0);
486 void helper_fdivr_STN_ST0(CPUX86State *env, int st_index)
488 floatx80 *p;
490 p = &ST(st_index);
491 *p = helper_fdiv(env, ST0, *p);
494 /* misc FPU operations */
495 void helper_fchs_ST0(CPUX86State *env)
497 ST0 = floatx80_chs(ST0);
500 void helper_fabs_ST0(CPUX86State *env)
502 ST0 = floatx80_abs(ST0);
505 void helper_fld1_ST0(CPUX86State *env)
507 ST0 = floatx80_one;
510 void helper_fldl2t_ST0(CPUX86State *env)
512 ST0 = floatx80_l2t;
515 void helper_fldl2e_ST0(CPUX86State *env)
517 ST0 = floatx80_l2e;
520 void helper_fldpi_ST0(CPUX86State *env)
522 ST0 = floatx80_pi;
525 void helper_fldlg2_ST0(CPUX86State *env)
527 ST0 = floatx80_lg2;
530 void helper_fldln2_ST0(CPUX86State *env)
532 ST0 = floatx80_ln2;
535 void helper_fldz_ST0(CPUX86State *env)
537 ST0 = floatx80_zero;
540 void helper_fldz_FT0(CPUX86State *env)
542 FT0 = floatx80_zero;
545 uint32_t helper_fnstsw(CPUX86State *env)
547 return (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
550 uint32_t helper_fnstcw(CPUX86State *env)
552 return env->fpuc;
555 void update_fp_status(CPUX86State *env)
557 int rnd_type;
559 /* set rounding mode */
560 switch (env->fpuc & FPU_RC_MASK) {
561 default:
562 case FPU_RC_NEAR:
563 rnd_type = float_round_nearest_even;
564 break;
565 case FPU_RC_DOWN:
566 rnd_type = float_round_down;
567 break;
568 case FPU_RC_UP:
569 rnd_type = float_round_up;
570 break;
571 case FPU_RC_CHOP:
572 rnd_type = float_round_to_zero;
573 break;
575 set_float_rounding_mode(rnd_type, &env->fp_status);
576 switch ((env->fpuc >> 8) & 3) {
577 case 0:
578 rnd_type = 32;
579 break;
580 case 2:
581 rnd_type = 64;
582 break;
583 case 3:
584 default:
585 rnd_type = 80;
586 break;
588 set_floatx80_rounding_precision(rnd_type, &env->fp_status);
591 void helper_fldcw(CPUX86State *env, uint32_t val)
593 cpu_set_fpuc(env, val);
596 void helper_fclex(CPUX86State *env)
598 env->fpus &= 0x7f00;
601 void helper_fwait(CPUX86State *env)
603 if (env->fpus & FPUS_SE) {
604 fpu_raise_exception(env, GETPC());
608 void helper_fninit(CPUX86State *env)
610 env->fpus = 0;
611 env->fpstt = 0;
612 cpu_set_fpuc(env, 0x37f);
613 env->fptags[0] = 1;
614 env->fptags[1] = 1;
615 env->fptags[2] = 1;
616 env->fptags[3] = 1;
617 env->fptags[4] = 1;
618 env->fptags[5] = 1;
619 env->fptags[6] = 1;
620 env->fptags[7] = 1;
623 /* BCD ops */
625 void helper_fbld_ST0(CPUX86State *env, target_ulong ptr)
627 floatx80 tmp;
628 uint64_t val;
629 unsigned int v;
630 int i;
632 val = 0;
633 for (i = 8; i >= 0; i--) {
634 v = cpu_ldub_data_ra(env, ptr + i, GETPC());
635 val = (val * 100) + ((v >> 4) * 10) + (v & 0xf);
637 tmp = int64_to_floatx80(val, &env->fp_status);
638 if (cpu_ldub_data_ra(env, ptr + 9, GETPC()) & 0x80) {
639 tmp = floatx80_chs(tmp);
641 fpush(env);
642 ST0 = tmp;
645 void helper_fbst_ST0(CPUX86State *env, target_ulong ptr)
647 int v;
648 target_ulong mem_ref, mem_end;
649 int64_t val;
651 val = floatx80_to_int64(ST0, &env->fp_status);
652 mem_ref = ptr;
653 mem_end = mem_ref + 9;
654 if (val < 0) {
655 cpu_stb_data_ra(env, mem_end, 0x80, GETPC());
656 val = -val;
657 } else {
658 cpu_stb_data_ra(env, mem_end, 0x00, GETPC());
660 while (mem_ref < mem_end) {
661 if (val == 0) {
662 break;
664 v = val % 100;
665 val = val / 100;
666 v = ((v / 10) << 4) | (v % 10);
667 cpu_stb_data_ra(env, mem_ref++, v, GETPC());
669 while (mem_ref < mem_end) {
670 cpu_stb_data_ra(env, mem_ref++, 0, GETPC());
674 void helper_f2xm1(CPUX86State *env)
676 double val = floatx80_to_double(env, ST0);
678 val = pow(2.0, val) - 1.0;
679 ST0 = double_to_floatx80(env, val);
682 void helper_fyl2x(CPUX86State *env)
684 double fptemp = floatx80_to_double(env, ST0);
686 if (fptemp > 0.0) {
687 fptemp = log(fptemp) / log(2.0); /* log2(ST) */
688 fptemp *= floatx80_to_double(env, ST1);
689 ST1 = double_to_floatx80(env, fptemp);
690 fpop(env);
691 } else {
692 env->fpus &= ~0x4700;
693 env->fpus |= 0x400;
697 void helper_fptan(CPUX86State *env)
699 double fptemp = floatx80_to_double(env, ST0);
701 if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) {
702 env->fpus |= 0x400;
703 } else {
704 fptemp = tan(fptemp);
705 ST0 = double_to_floatx80(env, fptemp);
706 fpush(env);
707 ST0 = floatx80_one;
708 env->fpus &= ~0x400; /* C2 <-- 0 */
709 /* the above code is for |arg| < 2**52 only */
713 void helper_fpatan(CPUX86State *env)
715 double fptemp, fpsrcop;
717 fpsrcop = floatx80_to_double(env, ST1);
718 fptemp = floatx80_to_double(env, ST0);
719 ST1 = double_to_floatx80(env, atan2(fpsrcop, fptemp));
720 fpop(env);
723 void helper_fxtract(CPUX86State *env)
725 CPU_LDoubleU temp;
727 temp.d = ST0;
729 if (floatx80_is_zero(ST0)) {
730 /* Easy way to generate -inf and raising division by 0 exception */
731 ST0 = floatx80_div(floatx80_chs(floatx80_one), floatx80_zero,
732 &env->fp_status);
733 fpush(env);
734 ST0 = temp.d;
735 } else {
736 int expdif;
738 expdif = EXPD(temp) - EXPBIAS;
739 /* DP exponent bias */
740 ST0 = int32_to_floatx80(expdif, &env->fp_status);
741 fpush(env);
742 BIASEXPONENT(temp);
743 ST0 = temp.d;
747 void helper_fprem1(CPUX86State *env)
749 double st0, st1, dblq, fpsrcop, fptemp;
750 CPU_LDoubleU fpsrcop1, fptemp1;
751 int expdif;
752 signed long long int q;
754 st0 = floatx80_to_double(env, ST0);
755 st1 = floatx80_to_double(env, ST1);
757 if (isinf(st0) || isnan(st0) || isnan(st1) || (st1 == 0.0)) {
758 ST0 = double_to_floatx80(env, 0.0 / 0.0); /* NaN */
759 env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
760 return;
763 fpsrcop = st0;
764 fptemp = st1;
765 fpsrcop1.d = ST0;
766 fptemp1.d = ST1;
767 expdif = EXPD(fpsrcop1) - EXPD(fptemp1);
769 if (expdif < 0) {
770 /* optimisation? taken from the AMD docs */
771 env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
772 /* ST0 is unchanged */
773 return;
776 if (expdif < 53) {
777 dblq = fpsrcop / fptemp;
778 /* round dblq towards nearest integer */
779 dblq = rint(dblq);
780 st0 = fpsrcop - fptemp * dblq;
782 /* convert dblq to q by truncating towards zero */
783 if (dblq < 0.0) {
784 q = (signed long long int)(-dblq);
785 } else {
786 q = (signed long long int)dblq;
789 env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
790 /* (C0,C3,C1) <-- (q2,q1,q0) */
791 env->fpus |= (q & 0x4) << (8 - 2); /* (C0) <-- q2 */
792 env->fpus |= (q & 0x2) << (14 - 1); /* (C3) <-- q1 */
793 env->fpus |= (q & 0x1) << (9 - 0); /* (C1) <-- q0 */
794 } else {
795 env->fpus |= 0x400; /* C2 <-- 1 */
796 fptemp = pow(2.0, expdif - 50);
797 fpsrcop = (st0 / st1) / fptemp;
798 /* fpsrcop = integer obtained by chopping */
799 fpsrcop = (fpsrcop < 0.0) ?
800 -(floor(fabs(fpsrcop))) : floor(fpsrcop);
801 st0 -= (st1 * fpsrcop * fptemp);
803 ST0 = double_to_floatx80(env, st0);
806 void helper_fprem(CPUX86State *env)
808 double st0, st1, dblq, fpsrcop, fptemp;
809 CPU_LDoubleU fpsrcop1, fptemp1;
810 int expdif;
811 signed long long int q;
813 st0 = floatx80_to_double(env, ST0);
814 st1 = floatx80_to_double(env, ST1);
816 if (isinf(st0) || isnan(st0) || isnan(st1) || (st1 == 0.0)) {
817 ST0 = double_to_floatx80(env, 0.0 / 0.0); /* NaN */
818 env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
819 return;
822 fpsrcop = st0;
823 fptemp = st1;
824 fpsrcop1.d = ST0;
825 fptemp1.d = ST1;
826 expdif = EXPD(fpsrcop1) - EXPD(fptemp1);
828 if (expdif < 0) {
829 /* optimisation? taken from the AMD docs */
830 env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
831 /* ST0 is unchanged */
832 return;
835 if (expdif < 53) {
836 dblq = fpsrcop / fptemp; /* ST0 / ST1 */
837 /* round dblq towards zero */
838 dblq = (dblq < 0.0) ? ceil(dblq) : floor(dblq);
839 st0 = fpsrcop - fptemp * dblq; /* fpsrcop is ST0 */
841 /* convert dblq to q by truncating towards zero */
842 if (dblq < 0.0) {
843 q = (signed long long int)(-dblq);
844 } else {
845 q = (signed long long int)dblq;
848 env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
849 /* (C0,C3,C1) <-- (q2,q1,q0) */
850 env->fpus |= (q & 0x4) << (8 - 2); /* (C0) <-- q2 */
851 env->fpus |= (q & 0x2) << (14 - 1); /* (C3) <-- q1 */
852 env->fpus |= (q & 0x1) << (9 - 0); /* (C1) <-- q0 */
853 } else {
854 int N = 32 + (expdif % 32); /* as per AMD docs */
856 env->fpus |= 0x400; /* C2 <-- 1 */
857 fptemp = pow(2.0, (double)(expdif - N));
858 fpsrcop = (st0 / st1) / fptemp;
859 /* fpsrcop = integer obtained by chopping */
860 fpsrcop = (fpsrcop < 0.0) ?
861 -(floor(fabs(fpsrcop))) : floor(fpsrcop);
862 st0 -= (st1 * fpsrcop * fptemp);
864 ST0 = double_to_floatx80(env, st0);
867 void helper_fyl2xp1(CPUX86State *env)
869 double fptemp = floatx80_to_double(env, ST0);
871 if ((fptemp + 1.0) > 0.0) {
872 fptemp = log(fptemp + 1.0) / log(2.0); /* log2(ST + 1.0) */
873 fptemp *= floatx80_to_double(env, ST1);
874 ST1 = double_to_floatx80(env, fptemp);
875 fpop(env);
876 } else {
877 env->fpus &= ~0x4700;
878 env->fpus |= 0x400;
882 void helper_fsqrt(CPUX86State *env)
884 if (floatx80_is_neg(ST0)) {
885 env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
886 env->fpus |= 0x400;
888 ST0 = floatx80_sqrt(ST0, &env->fp_status);
891 void helper_fsincos(CPUX86State *env)
893 double fptemp = floatx80_to_double(env, ST0);
895 if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) {
896 env->fpus |= 0x400;
897 } else {
898 ST0 = double_to_floatx80(env, sin(fptemp));
899 fpush(env);
900 ST0 = double_to_floatx80(env, cos(fptemp));
901 env->fpus &= ~0x400; /* C2 <-- 0 */
902 /* the above code is for |arg| < 2**63 only */
906 void helper_frndint(CPUX86State *env)
908 ST0 = floatx80_round_to_int(ST0, &env->fp_status);
911 void helper_fscale(CPUX86State *env)
913 if (floatx80_is_any_nan(ST1)) {
914 ST0 = ST1;
915 } else {
916 int n = floatx80_to_int32_round_to_zero(ST1, &env->fp_status);
917 ST0 = floatx80_scalbn(ST0, n, &env->fp_status);
921 void helper_fsin(CPUX86State *env)
923 double fptemp = floatx80_to_double(env, ST0);
925 if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) {
926 env->fpus |= 0x400;
927 } else {
928 ST0 = double_to_floatx80(env, sin(fptemp));
929 env->fpus &= ~0x400; /* C2 <-- 0 */
930 /* the above code is for |arg| < 2**53 only */
934 void helper_fcos(CPUX86State *env)
936 double fptemp = floatx80_to_double(env, ST0);
938 if ((fptemp > MAXTAN) || (fptemp < -MAXTAN)) {
939 env->fpus |= 0x400;
940 } else {
941 ST0 = double_to_floatx80(env, cos(fptemp));
942 env->fpus &= ~0x400; /* C2 <-- 0 */
943 /* the above code is for |arg| < 2**63 only */
947 void helper_fxam_ST0(CPUX86State *env)
949 CPU_LDoubleU temp;
950 int expdif;
952 temp.d = ST0;
954 env->fpus &= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
955 if (SIGND(temp)) {
956 env->fpus |= 0x200; /* C1 <-- 1 */
959 /* XXX: test fptags too */
960 expdif = EXPD(temp);
961 if (expdif == MAXEXPD) {
962 if (MANTD(temp) == 0x8000000000000000ULL) {
963 env->fpus |= 0x500; /* Infinity */
964 } else {
965 env->fpus |= 0x100; /* NaN */
967 } else if (expdif == 0) {
968 if (MANTD(temp) == 0) {
969 env->fpus |= 0x4000; /* Zero */
970 } else {
971 env->fpus |= 0x4400; /* Denormal */
973 } else {
974 env->fpus |= 0x400;
978 static void do_fstenv(CPUX86State *env, target_ulong ptr, int data32,
979 uintptr_t retaddr)
981 int fpus, fptag, exp, i;
982 uint64_t mant;
983 CPU_LDoubleU tmp;
985 fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
986 fptag = 0;
987 for (i = 7; i >= 0; i--) {
988 fptag <<= 2;
989 if (env->fptags[i]) {
990 fptag |= 3;
991 } else {
992 tmp.d = env->fpregs[i].d;
993 exp = EXPD(tmp);
994 mant = MANTD(tmp);
995 if (exp == 0 && mant == 0) {
996 /* zero */
997 fptag |= 1;
998 } else if (exp == 0 || exp == MAXEXPD
999 || (mant & (1LL << 63)) == 0) {
1000 /* NaNs, infinity, denormal */
1001 fptag |= 2;
1005 if (data32) {
1006 /* 32 bit */
1007 cpu_stl_data_ra(env, ptr, env->fpuc, retaddr);
1008 cpu_stl_data_ra(env, ptr + 4, fpus, retaddr);
1009 cpu_stl_data_ra(env, ptr + 8, fptag, retaddr);
1010 cpu_stl_data_ra(env, ptr + 12, 0, retaddr); /* fpip */
1011 cpu_stl_data_ra(env, ptr + 16, 0, retaddr); /* fpcs */
1012 cpu_stl_data_ra(env, ptr + 20, 0, retaddr); /* fpoo */
1013 cpu_stl_data_ra(env, ptr + 24, 0, retaddr); /* fpos */
1014 } else {
1015 /* 16 bit */
1016 cpu_stw_data_ra(env, ptr, env->fpuc, retaddr);
1017 cpu_stw_data_ra(env, ptr + 2, fpus, retaddr);
1018 cpu_stw_data_ra(env, ptr + 4, fptag, retaddr);
1019 cpu_stw_data_ra(env, ptr + 6, 0, retaddr);
1020 cpu_stw_data_ra(env, ptr + 8, 0, retaddr);
1021 cpu_stw_data_ra(env, ptr + 10, 0, retaddr);
1022 cpu_stw_data_ra(env, ptr + 12, 0, retaddr);
1026 void helper_fstenv(CPUX86State *env, target_ulong ptr, int data32)
1028 do_fstenv(env, ptr, data32, GETPC());
1031 static void do_fldenv(CPUX86State *env, target_ulong ptr, int data32,
1032 uintptr_t retaddr)
1034 int i, fpus, fptag;
1036 if (data32) {
1037 cpu_set_fpuc(env, cpu_lduw_data_ra(env, ptr, retaddr));
1038 fpus = cpu_lduw_data_ra(env, ptr + 4, retaddr);
1039 fptag = cpu_lduw_data_ra(env, ptr + 8, retaddr);
1040 } else {
1041 cpu_set_fpuc(env, cpu_lduw_data_ra(env, ptr, retaddr));
1042 fpus = cpu_lduw_data_ra(env, ptr + 2, retaddr);
1043 fptag = cpu_lduw_data_ra(env, ptr + 4, retaddr);
1045 env->fpstt = (fpus >> 11) & 7;
1046 env->fpus = fpus & ~0x3800;
1047 for (i = 0; i < 8; i++) {
1048 env->fptags[i] = ((fptag & 3) == 3);
1049 fptag >>= 2;
1053 void helper_fldenv(CPUX86State *env, target_ulong ptr, int data32)
1055 do_fldenv(env, ptr, data32, GETPC());
1058 void helper_fsave(CPUX86State *env, target_ulong ptr, int data32)
1060 floatx80 tmp;
1061 int i;
1063 do_fstenv(env, ptr, data32, GETPC());
1065 ptr += (14 << data32);
1066 for (i = 0; i < 8; i++) {
1067 tmp = ST(i);
1068 helper_fstt(env, tmp, ptr, GETPC());
1069 ptr += 10;
1072 /* fninit */
1073 env->fpus = 0;
1074 env->fpstt = 0;
1075 cpu_set_fpuc(env, 0x37f);
1076 env->fptags[0] = 1;
1077 env->fptags[1] = 1;
1078 env->fptags[2] = 1;
1079 env->fptags[3] = 1;
1080 env->fptags[4] = 1;
1081 env->fptags[5] = 1;
1082 env->fptags[6] = 1;
1083 env->fptags[7] = 1;
1086 void helper_frstor(CPUX86State *env, target_ulong ptr, int data32)
1088 floatx80 tmp;
1089 int i;
1091 do_fldenv(env, ptr, data32, GETPC());
1092 ptr += (14 << data32);
1094 for (i = 0; i < 8; i++) {
1095 tmp = helper_fldt(env, ptr, GETPC());
1096 ST(i) = tmp;
1097 ptr += 10;
1101 #if defined(CONFIG_USER_ONLY)
1102 void cpu_x86_fsave(CPUX86State *env, target_ulong ptr, int data32)
1104 helper_fsave(env, ptr, data32);
1107 void cpu_x86_frstor(CPUX86State *env, target_ulong ptr, int data32)
1109 helper_frstor(env, ptr, data32);
1111 #endif
1113 static void do_xsave_fpu(CPUX86State *env, target_ulong ptr, uintptr_t ra)
1115 int fpus, fptag, i;
1116 target_ulong addr;
1118 fpus = (env->fpus & ~0x3800) | (env->fpstt & 0x7) << 11;
1119 fptag = 0;
1120 for (i = 0; i < 8; i++) {
1121 fptag |= (env->fptags[i] << i);
1123 cpu_stw_data_ra(env, ptr, env->fpuc, ra);
1124 cpu_stw_data_ra(env, ptr + 2, fpus, ra);
1125 cpu_stw_data_ra(env, ptr + 4, fptag ^ 0xff, ra);
1127 /* In 32-bit mode this is eip, sel, dp, sel.
1128 In 64-bit mode this is rip, rdp.
1129 But in either case we don't write actual data, just zeros. */
1130 cpu_stq_data_ra(env, ptr + 0x08, 0, ra); /* eip+sel; rip */
1131 cpu_stq_data_ra(env, ptr + 0x10, 0, ra); /* edp+sel; rdp */
1133 addr = ptr + 0x20;
1134 for (i = 0; i < 8; i++) {
1135 floatx80 tmp = ST(i);
1136 helper_fstt(env, tmp, addr, ra);
1137 addr += 16;
1141 static void do_xsave_mxcsr(CPUX86State *env, target_ulong ptr, uintptr_t ra)
1143 cpu_stl_data_ra(env, ptr + 0x18, env->mxcsr, ra); /* mxcsr */
1144 cpu_stl_data_ra(env, ptr + 0x1c, 0x0000ffff, ra); /* mxcsr_mask */
1147 static void do_xsave_sse(CPUX86State *env, target_ulong ptr, uintptr_t ra)
1149 int i, nb_xmm_regs;
1150 target_ulong addr;
1152 if (env->hflags & HF_CS64_MASK) {
1153 nb_xmm_regs = 16;
1154 } else {
1155 nb_xmm_regs = 8;
1158 addr = ptr + 0xa0;
1159 for (i = 0; i < nb_xmm_regs; i++) {
1160 cpu_stq_data_ra(env, addr, env->xmm_regs[i].ZMM_Q(0), ra);
1161 cpu_stq_data_ra(env, addr + 8, env->xmm_regs[i].ZMM_Q(1), ra);
1162 addr += 16;
1166 static void do_xsave_bndregs(CPUX86State *env, target_ulong addr, uintptr_t ra)
1168 int i;
1170 for (i = 0; i < 4; i++, addr += 16) {
1171 cpu_stq_data_ra(env, addr, env->bnd_regs[i].lb, ra);
1172 cpu_stq_data_ra(env, addr + 8, env->bnd_regs[i].ub, ra);
1176 static void do_xsave_bndcsr(CPUX86State *env, target_ulong addr, uintptr_t ra)
1178 cpu_stq_data_ra(env, addr, env->bndcs_regs.cfgu, ra);
1179 cpu_stq_data_ra(env, addr + 8, env->bndcs_regs.sts, ra);
1182 static void do_xsave_pkru(CPUX86State *env, target_ulong addr, uintptr_t ra)
1184 cpu_stq_data_ra(env, addr, env->pkru, ra);
1187 void helper_fxsave(CPUX86State *env, target_ulong ptr)
1189 uintptr_t ra = GETPC();
1191 /* The operand must be 16 byte aligned */
1192 if (ptr & 0xf) {
1193 raise_exception_ra(env, EXCP0D_GPF, ra);
1196 do_xsave_fpu(env, ptr, ra);
1198 if (env->cr[4] & CR4_OSFXSR_MASK) {
1199 do_xsave_mxcsr(env, ptr, ra);
1200 /* Fast FXSAVE leaves out the XMM registers */
1201 if (!(env->efer & MSR_EFER_FFXSR)
1202 || (env->hflags & HF_CPL_MASK)
1203 || !(env->hflags & HF_LMA_MASK)) {
1204 do_xsave_sse(env, ptr, ra);
1209 static uint64_t get_xinuse(CPUX86State *env)
1211 uint64_t inuse = -1;
1213 /* For the most part, we don't track XINUSE. We could calculate it
1214 here for all components, but it's probably less work to simply
1215 indicate in use. That said, the state of BNDREGS is important
1216 enough to track in HFLAGS, so we might as well use that here. */
1217 if ((env->hflags & HF_MPX_IU_MASK) == 0) {
1218 inuse &= ~XSTATE_BNDREGS_MASK;
1220 return inuse;
1223 static void do_xsave(CPUX86State *env, target_ulong ptr, uint64_t rfbm,
1224 uint64_t inuse, uint64_t opt, uintptr_t ra)
1226 uint64_t old_bv, new_bv;
1228 /* The OS must have enabled XSAVE. */
1229 if (!(env->cr[4] & CR4_OSXSAVE_MASK)) {
1230 raise_exception_ra(env, EXCP06_ILLOP, ra);
1233 /* The operand must be 64 byte aligned. */
1234 if (ptr & 63) {
1235 raise_exception_ra(env, EXCP0D_GPF, ra);
1238 /* Never save anything not enabled by XCR0. */
1239 rfbm &= env->xcr0;
1240 opt &= rfbm;
1242 if (opt & XSTATE_FP_MASK) {
1243 do_xsave_fpu(env, ptr, ra);
1245 if (rfbm & XSTATE_SSE_MASK) {
1246 /* Note that saving MXCSR is not suppressed by XSAVEOPT. */
1247 do_xsave_mxcsr(env, ptr, ra);
1249 if (opt & XSTATE_SSE_MASK) {
1250 do_xsave_sse(env, ptr, ra);
1252 if (opt & XSTATE_BNDREGS_MASK) {
1253 target_ulong off = x86_ext_save_areas[XSTATE_BNDREGS_BIT].offset;
1254 do_xsave_bndregs(env, ptr + off, ra);
1256 if (opt & XSTATE_BNDCSR_MASK) {
1257 target_ulong off = x86_ext_save_areas[XSTATE_BNDCSR_BIT].offset;
1258 do_xsave_bndcsr(env, ptr + off, ra);
1260 if (opt & XSTATE_PKRU_MASK) {
1261 target_ulong off = x86_ext_save_areas[XSTATE_PKRU_BIT].offset;
1262 do_xsave_pkru(env, ptr + off, ra);
1265 /* Update the XSTATE_BV field. */
1266 old_bv = cpu_ldq_data_ra(env, ptr + 512, ra);
1267 new_bv = (old_bv & ~rfbm) | (inuse & rfbm);
1268 cpu_stq_data_ra(env, ptr + 512, new_bv, ra);
1271 void helper_xsave(CPUX86State *env, target_ulong ptr, uint64_t rfbm)
1273 do_xsave(env, ptr, rfbm, get_xinuse(env), -1, GETPC());
1276 void helper_xsaveopt(CPUX86State *env, target_ulong ptr, uint64_t rfbm)
1278 uint64_t inuse = get_xinuse(env);
1279 do_xsave(env, ptr, rfbm, inuse, inuse, GETPC());
1282 static void do_xrstor_fpu(CPUX86State *env, target_ulong ptr, uintptr_t ra)
1284 int i, fpus, fptag;
1285 target_ulong addr;
1287 cpu_set_fpuc(env, cpu_lduw_data_ra(env, ptr, ra));
1288 fpus = cpu_lduw_data_ra(env, ptr + 2, ra);
1289 fptag = cpu_lduw_data_ra(env, ptr + 4, ra);
1290 env->fpstt = (fpus >> 11) & 7;
1291 env->fpus = fpus & ~0x3800;
1292 fptag ^= 0xff;
1293 for (i = 0; i < 8; i++) {
1294 env->fptags[i] = ((fptag >> i) & 1);
1297 addr = ptr + 0x20;
1298 for (i = 0; i < 8; i++) {
1299 floatx80 tmp = helper_fldt(env, addr, ra);
1300 ST(i) = tmp;
1301 addr += 16;
1305 static void do_xrstor_mxcsr(CPUX86State *env, target_ulong ptr, uintptr_t ra)
1307 cpu_set_mxcsr(env, cpu_ldl_data_ra(env, ptr + 0x18, ra));
1310 static void do_xrstor_sse(CPUX86State *env, target_ulong ptr, uintptr_t ra)
1312 int i, nb_xmm_regs;
1313 target_ulong addr;
1315 if (env->hflags & HF_CS64_MASK) {
1316 nb_xmm_regs = 16;
1317 } else {
1318 nb_xmm_regs = 8;
1321 addr = ptr + 0xa0;
1322 for (i = 0; i < nb_xmm_regs; i++) {
1323 env->xmm_regs[i].ZMM_Q(0) = cpu_ldq_data_ra(env, addr, ra);
1324 env->xmm_regs[i].ZMM_Q(1) = cpu_ldq_data_ra(env, addr + 8, ra);
1325 addr += 16;
1329 static void do_xrstor_bndregs(CPUX86State *env, target_ulong addr, uintptr_t ra)
1331 int i;
1333 for (i = 0; i < 4; i++, addr += 16) {
1334 env->bnd_regs[i].lb = cpu_ldq_data_ra(env, addr, ra);
1335 env->bnd_regs[i].ub = cpu_ldq_data_ra(env, addr + 8, ra);
1339 static void do_xrstor_bndcsr(CPUX86State *env, target_ulong addr, uintptr_t ra)
1341 /* FIXME: Extend highest implemented bit of linear address. */
1342 env->bndcs_regs.cfgu = cpu_ldq_data_ra(env, addr, ra);
1343 env->bndcs_regs.sts = cpu_ldq_data_ra(env, addr + 8, ra);
1346 static void do_xrstor_pkru(CPUX86State *env, target_ulong addr, uintptr_t ra)
1348 env->pkru = cpu_ldq_data_ra(env, addr, ra);
1351 void helper_fxrstor(CPUX86State *env, target_ulong ptr)
1353 uintptr_t ra = GETPC();
1355 /* The operand must be 16 byte aligned */
1356 if (ptr & 0xf) {
1357 raise_exception_ra(env, EXCP0D_GPF, ra);
1360 do_xrstor_fpu(env, ptr, ra);
1362 if (env->cr[4] & CR4_OSFXSR_MASK) {
1363 do_xrstor_mxcsr(env, ptr, ra);
1364 /* Fast FXRSTOR leaves out the XMM registers */
1365 if (!(env->efer & MSR_EFER_FFXSR)
1366 || (env->hflags & HF_CPL_MASK)
1367 || !(env->hflags & HF_LMA_MASK)) {
1368 do_xrstor_sse(env, ptr, ra);
1373 void helper_xrstor(CPUX86State *env, target_ulong ptr, uint64_t rfbm)
1375 uintptr_t ra = GETPC();
1376 uint64_t xstate_bv, xcomp_bv0, xcomp_bv1;
1378 rfbm &= env->xcr0;
1380 /* The OS must have enabled XSAVE. */
1381 if (!(env->cr[4] & CR4_OSXSAVE_MASK)) {
1382 raise_exception_ra(env, EXCP06_ILLOP, ra);
1385 /* The operand must be 64 byte aligned. */
1386 if (ptr & 63) {
1387 raise_exception_ra(env, EXCP0D_GPF, ra);
1390 xstate_bv = cpu_ldq_data_ra(env, ptr + 512, ra);
1392 if ((int64_t)xstate_bv < 0) {
1393 /* FIXME: Compact form. */
1394 raise_exception_ra(env, EXCP0D_GPF, ra);
1397 /* Standard form. */
1399 /* The XSTATE field must not set bits not present in XCR0. */
1400 if (xstate_bv & ~env->xcr0) {
1401 raise_exception_ra(env, EXCP0D_GPF, ra);
1404 /* The XCOMP field must be zero. */
1405 xcomp_bv0 = cpu_ldq_data_ra(env, ptr + 520, ra);
1406 xcomp_bv1 = cpu_ldq_data_ra(env, ptr + 528, ra);
1407 if (xcomp_bv0 || xcomp_bv1) {
1408 raise_exception_ra(env, EXCP0D_GPF, ra);
1411 if (rfbm & XSTATE_FP_MASK) {
1412 if (xstate_bv & XSTATE_FP_MASK) {
1413 do_xrstor_fpu(env, ptr, ra);
1414 } else {
1415 helper_fninit(env);
1416 memset(env->fpregs, 0, sizeof(env->fpregs));
1419 if (rfbm & XSTATE_SSE_MASK) {
1420 /* Note that the standard form of XRSTOR loads MXCSR from memory
1421 whether or not the XSTATE_BV bit is set. */
1422 do_xrstor_mxcsr(env, ptr, ra);
1423 if (xstate_bv & XSTATE_SSE_MASK) {
1424 do_xrstor_sse(env, ptr, ra);
1425 } else {
1426 /* ??? When AVX is implemented, we may have to be more
1427 selective in the clearing. */
1428 memset(env->xmm_regs, 0, sizeof(env->xmm_regs));
1431 if (rfbm & XSTATE_BNDREGS_MASK) {
1432 if (xstate_bv & XSTATE_BNDREGS_MASK) {
1433 target_ulong off = x86_ext_save_areas[XSTATE_BNDREGS_BIT].offset;
1434 do_xrstor_bndregs(env, ptr + off, ra);
1435 env->hflags |= HF_MPX_IU_MASK;
1436 } else {
1437 memset(env->bnd_regs, 0, sizeof(env->bnd_regs));
1438 env->hflags &= ~HF_MPX_IU_MASK;
1441 if (rfbm & XSTATE_BNDCSR_MASK) {
1442 if (xstate_bv & XSTATE_BNDCSR_MASK) {
1443 target_ulong off = x86_ext_save_areas[XSTATE_BNDCSR_BIT].offset;
1444 do_xrstor_bndcsr(env, ptr + off, ra);
1445 } else {
1446 memset(&env->bndcs_regs, 0, sizeof(env->bndcs_regs));
1448 cpu_sync_bndcs_hflags(env);
1450 if (rfbm & XSTATE_PKRU_MASK) {
1451 uint64_t old_pkru = env->pkru;
1452 if (xstate_bv & XSTATE_PKRU_MASK) {
1453 target_ulong off = x86_ext_save_areas[XSTATE_PKRU_BIT].offset;
1454 do_xrstor_pkru(env, ptr + off, ra);
1455 } else {
1456 env->pkru = 0;
1458 if (env->pkru != old_pkru) {
1459 CPUState *cs = CPU(x86_env_get_cpu(env));
1460 tlb_flush(cs, 1);
1465 uint64_t helper_xgetbv(CPUX86State *env, uint32_t ecx)
1467 /* The OS must have enabled XSAVE. */
1468 if (!(env->cr[4] & CR4_OSXSAVE_MASK)) {
1469 raise_exception_ra(env, EXCP06_ILLOP, GETPC());
1472 switch (ecx) {
1473 case 0:
1474 return env->xcr0;
1475 case 1:
1476 if (env->features[FEAT_XSAVE] & CPUID_XSAVE_XGETBV1) {
1477 return env->xcr0 & get_xinuse(env);
1479 break;
1481 raise_exception_ra(env, EXCP0D_GPF, GETPC());
1484 void helper_xsetbv(CPUX86State *env, uint32_t ecx, uint64_t mask)
1486 uint32_t dummy, ena_lo, ena_hi;
1487 uint64_t ena;
1489 /* The OS must have enabled XSAVE. */
1490 if (!(env->cr[4] & CR4_OSXSAVE_MASK)) {
1491 raise_exception_ra(env, EXCP06_ILLOP, GETPC());
1494 /* Only XCR0 is defined at present; the FPU may not be disabled. */
1495 if (ecx != 0 || (mask & XSTATE_FP_MASK) == 0) {
1496 goto do_gpf;
1499 /* Disallow enabling unimplemented features. */
1500 cpu_x86_cpuid(env, 0x0d, 0, &ena_lo, &dummy, &dummy, &ena_hi);
1501 ena = ((uint64_t)ena_hi << 32) | ena_lo;
1502 if (mask & ~ena) {
1503 goto do_gpf;
1506 /* Disallow enabling only half of MPX. */
1507 if ((mask ^ (mask * (XSTATE_BNDCSR_MASK / XSTATE_BNDREGS_MASK)))
1508 & XSTATE_BNDCSR_MASK) {
1509 goto do_gpf;
1512 env->xcr0 = mask;
1513 cpu_sync_bndcs_hflags(env);
1514 return;
1516 do_gpf:
1517 raise_exception_ra(env, EXCP0D_GPF, GETPC());
1520 void cpu_get_fp80(uint64_t *pmant, uint16_t *pexp, floatx80 f)
1522 CPU_LDoubleU temp;
1524 temp.d = f;
1525 *pmant = temp.l.lower;
1526 *pexp = temp.l.upper;
1529 floatx80 cpu_set_fp80(uint64_t mant, uint16_t upper)
1531 CPU_LDoubleU temp;
1533 temp.l.upper = upper;
1534 temp.l.lower = mant;
1535 return temp.d;
1538 /* MMX/SSE */
1539 /* XXX: optimize by storing fptt and fptags in the static cpu state */
1541 #define SSE_DAZ 0x0040
1542 #define SSE_RC_MASK 0x6000
1543 #define SSE_RC_NEAR 0x0000
1544 #define SSE_RC_DOWN 0x2000
1545 #define SSE_RC_UP 0x4000
1546 #define SSE_RC_CHOP 0x6000
1547 #define SSE_FZ 0x8000
1549 void cpu_set_mxcsr(CPUX86State *env, uint32_t mxcsr)
1551 int rnd_type;
1553 env->mxcsr = mxcsr;
1555 /* set rounding mode */
1556 switch (mxcsr & SSE_RC_MASK) {
1557 default:
1558 case SSE_RC_NEAR:
1559 rnd_type = float_round_nearest_even;
1560 break;
1561 case SSE_RC_DOWN:
1562 rnd_type = float_round_down;
1563 break;
1564 case SSE_RC_UP:
1565 rnd_type = float_round_up;
1566 break;
1567 case SSE_RC_CHOP:
1568 rnd_type = float_round_to_zero;
1569 break;
1571 set_float_rounding_mode(rnd_type, &env->sse_status);
1573 /* set denormals are zero */
1574 set_flush_inputs_to_zero((mxcsr & SSE_DAZ) ? 1 : 0, &env->sse_status);
1576 /* set flush to zero */
1577 set_flush_to_zero((mxcsr & SSE_FZ) ? 1 : 0, &env->fp_status);
1580 void cpu_set_fpuc(CPUX86State *env, uint16_t val)
1582 env->fpuc = val;
1583 update_fp_status(env);
1586 void helper_ldmxcsr(CPUX86State *env, uint32_t val)
1588 cpu_set_mxcsr(env, val);
1591 void helper_enter_mmx(CPUX86State *env)
1593 env->fpstt = 0;
1594 *(uint32_t *)(env->fptags) = 0;
1595 *(uint32_t *)(env->fptags + 4) = 0;
1598 void helper_emms(CPUX86State *env)
1600 /* set to empty state */
1601 *(uint32_t *)(env->fptags) = 0x01010101;
1602 *(uint32_t *)(env->fptags + 4) = 0x01010101;
1605 /* XXX: suppress */
1606 void helper_movq(CPUX86State *env, void *d, void *s)
1608 *(uint64_t *)d = *(uint64_t *)s;
1611 #define SHIFT 0
1612 #include "ops_sse.h"
1614 #define SHIFT 1
1615 #include "ops_sse.h"