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"
23 #include "exec/helper-proto.h"
24 #include "qemu/host-utils.h"
25 #include "exec/cpu_ldst.h"
27 #define FPU_RC_MASK 0xc00
28 #define FPU_RC_NEAR 0x000
29 #define FPU_RC_DOWN 0x400
30 #define FPU_RC_UP 0x800
31 #define FPU_RC_CHOP 0xc00
33 #define MAXTAN 9223372036854775808.0
35 /* the following deal with x86 long double-precision numbers */
36 #define MAXEXPD 0x7fff
38 #define EXPD(fp) (fp.l.upper & 0x7fff)
39 #define SIGND(fp) ((fp.l.upper) & 0x8000)
40 #define MANTD(fp) (fp.l.lower)
41 #define BIASEXPONENT(fp) fp.l.upper = (fp.l.upper & ~(0x7fff)) | EXPBIAS
43 #define FPUS_IE (1 << 0)
44 #define FPUS_DE (1 << 1)
45 #define FPUS_ZE (1 << 2)
46 #define FPUS_OE (1 << 3)
47 #define FPUS_UE (1 << 4)
48 #define FPUS_PE (1 << 5)
49 #define FPUS_SF (1 << 6)
50 #define FPUS_SE (1 << 7)
51 #define FPUS_B (1 << 15)
55 #define floatx80_lg2 make_floatx80(0x3ffd, 0x9a209a84fbcff799LL)
56 #define floatx80_l2e make_floatx80(0x3fff, 0xb8aa3b295c17f0bcLL)
57 #define floatx80_l2t make_floatx80(0x4000, 0xd49a784bcd1b8afeLL)
59 static inline void fpush(CPUX86State
*env
)
61 env
->fpstt
= (env
->fpstt
- 1) & 7;
62 env
->fptags
[env
->fpstt
] = 0; /* validate stack entry */
65 static inline void fpop(CPUX86State
*env
)
67 env
->fptags
[env
->fpstt
] = 1; /* invalidate stack entry */
68 env
->fpstt
= (env
->fpstt
+ 1) & 7;
71 static inline floatx80
helper_fldt(CPUX86State
*env
, target_ulong ptr
,
76 temp
.l
.lower
= cpu_ldq_data_ra(env
, ptr
, retaddr
);
77 temp
.l
.upper
= cpu_lduw_data_ra(env
, ptr
+ 8, retaddr
);
81 static inline void helper_fstt(CPUX86State
*env
, floatx80 f
, target_ulong ptr
,
87 cpu_stq_data_ra(env
, ptr
, temp
.l
.lower
, retaddr
);
88 cpu_stw_data_ra(env
, ptr
+ 8, temp
.l
.upper
, retaddr
);
93 static inline double floatx80_to_double(CPUX86State
*env
, floatx80 a
)
100 u
.f64
= floatx80_to_float64(a
, &env
->fp_status
);
104 static inline floatx80
double_to_floatx80(CPUX86State
*env
, double a
)
112 return float64_to_floatx80(u
.f64
, &env
->fp_status
);
115 static void fpu_set_exception(CPUX86State
*env
, int mask
)
118 if (env
->fpus
& (~env
->fpuc
& FPUC_EM
)) {
119 env
->fpus
|= FPUS_SE
| FPUS_B
;
123 static inline floatx80
helper_fdiv(CPUX86State
*env
, floatx80 a
, floatx80 b
)
125 if (floatx80_is_zero(b
)) {
126 fpu_set_exception(env
, FPUS_ZE
);
128 return floatx80_div(a
, b
, &env
->fp_status
);
131 static void fpu_raise_exception(CPUX86State
*env
, uintptr_t retaddr
)
133 if (env
->cr
[0] & CR0_NE_MASK
) {
134 raise_exception_ra(env
, EXCP10_COPR
, retaddr
);
136 #if !defined(CONFIG_USER_ONLY)
143 void helper_flds_FT0(CPUX86State
*env
, uint32_t val
)
151 FT0
= float32_to_floatx80(u
.f
, &env
->fp_status
);
154 void helper_fldl_FT0(CPUX86State
*env
, uint64_t val
)
162 FT0
= float64_to_floatx80(u
.f
, &env
->fp_status
);
165 void helper_fildl_FT0(CPUX86State
*env
, int32_t val
)
167 FT0
= int32_to_floatx80(val
, &env
->fp_status
);
170 void helper_flds_ST0(CPUX86State
*env
, uint32_t val
)
178 new_fpstt
= (env
->fpstt
- 1) & 7;
180 env
->fpregs
[new_fpstt
].d
= float32_to_floatx80(u
.f
, &env
->fp_status
);
181 env
->fpstt
= new_fpstt
;
182 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
185 void helper_fldl_ST0(CPUX86State
*env
, uint64_t val
)
193 new_fpstt
= (env
->fpstt
- 1) & 7;
195 env
->fpregs
[new_fpstt
].d
= float64_to_floatx80(u
.f
, &env
->fp_status
);
196 env
->fpstt
= new_fpstt
;
197 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
200 void helper_fildl_ST0(CPUX86State
*env
, int32_t val
)
204 new_fpstt
= (env
->fpstt
- 1) & 7;
205 env
->fpregs
[new_fpstt
].d
= int32_to_floatx80(val
, &env
->fp_status
);
206 env
->fpstt
= new_fpstt
;
207 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
210 void helper_fildll_ST0(CPUX86State
*env
, int64_t val
)
214 new_fpstt
= (env
->fpstt
- 1) & 7;
215 env
->fpregs
[new_fpstt
].d
= int64_to_floatx80(val
, &env
->fp_status
);
216 env
->fpstt
= new_fpstt
;
217 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
220 uint32_t helper_fsts_ST0(CPUX86State
*env
)
227 u
.f
= floatx80_to_float32(ST0
, &env
->fp_status
);
231 uint64_t helper_fstl_ST0(CPUX86State
*env
)
238 u
.f
= floatx80_to_float64(ST0
, &env
->fp_status
);
242 int32_t helper_fist_ST0(CPUX86State
*env
)
246 val
= floatx80_to_int32(ST0
, &env
->fp_status
);
247 if (val
!= (int16_t)val
) {
253 int32_t helper_fistl_ST0(CPUX86State
*env
)
256 signed char old_exp_flags
;
258 old_exp_flags
= get_float_exception_flags(&env
->fp_status
);
259 set_float_exception_flags(0, &env
->fp_status
);
261 val
= floatx80_to_int32(ST0
, &env
->fp_status
);
262 if (get_float_exception_flags(&env
->fp_status
) & float_flag_invalid
) {
265 set_float_exception_flags(get_float_exception_flags(&env
->fp_status
)
266 | old_exp_flags
, &env
->fp_status
);
270 int64_t helper_fistll_ST0(CPUX86State
*env
)
273 signed char old_exp_flags
;
275 old_exp_flags
= get_float_exception_flags(&env
->fp_status
);
276 set_float_exception_flags(0, &env
->fp_status
);
278 val
= floatx80_to_int64(ST0
, &env
->fp_status
);
279 if (get_float_exception_flags(&env
->fp_status
) & float_flag_invalid
) {
280 val
= 0x8000000000000000ULL
;
282 set_float_exception_flags(get_float_exception_flags(&env
->fp_status
)
283 | old_exp_flags
, &env
->fp_status
);
287 int32_t helper_fistt_ST0(CPUX86State
*env
)
291 val
= floatx80_to_int32_round_to_zero(ST0
, &env
->fp_status
);
292 if (val
!= (int16_t)val
) {
298 int32_t helper_fisttl_ST0(CPUX86State
*env
)
302 val
= floatx80_to_int32_round_to_zero(ST0
, &env
->fp_status
);
306 int64_t helper_fisttll_ST0(CPUX86State
*env
)
310 val
= floatx80_to_int64_round_to_zero(ST0
, &env
->fp_status
);
314 void helper_fldt_ST0(CPUX86State
*env
, target_ulong ptr
)
318 new_fpstt
= (env
->fpstt
- 1) & 7;
319 env
->fpregs
[new_fpstt
].d
= helper_fldt(env
, ptr
, GETPC());
320 env
->fpstt
= new_fpstt
;
321 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
324 void helper_fstt_ST0(CPUX86State
*env
, target_ulong ptr
)
326 helper_fstt(env
, ST0
, ptr
, GETPC());
329 void helper_fpush(CPUX86State
*env
)
334 void helper_fpop(CPUX86State
*env
)
339 void helper_fdecstp(CPUX86State
*env
)
341 env
->fpstt
= (env
->fpstt
- 1) & 7;
342 env
->fpus
&= ~0x4700;
345 void helper_fincstp(CPUX86State
*env
)
347 env
->fpstt
= (env
->fpstt
+ 1) & 7;
348 env
->fpus
&= ~0x4700;
353 void helper_ffree_STN(CPUX86State
*env
, int st_index
)
355 env
->fptags
[(env
->fpstt
+ st_index
) & 7] = 1;
358 void helper_fmov_ST0_FT0(CPUX86State
*env
)
363 void helper_fmov_FT0_STN(CPUX86State
*env
, int st_index
)
368 void helper_fmov_ST0_STN(CPUX86State
*env
, int st_index
)
373 void helper_fmov_STN_ST0(CPUX86State
*env
, int st_index
)
378 void helper_fxchg_ST0_STN(CPUX86State
*env
, int st_index
)
389 static const int fcom_ccval
[4] = {0x0100, 0x4000, 0x0000, 0x4500};
391 void helper_fcom_ST0_FT0(CPUX86State
*env
)
395 ret
= floatx80_compare(ST0
, FT0
, &env
->fp_status
);
396 env
->fpus
= (env
->fpus
& ~0x4500) | fcom_ccval
[ret
+ 1];
399 void helper_fucom_ST0_FT0(CPUX86State
*env
)
403 ret
= floatx80_compare_quiet(ST0
, FT0
, &env
->fp_status
);
404 env
->fpus
= (env
->fpus
& ~0x4500) | fcom_ccval
[ret
+ 1];
407 static const int fcomi_ccval
[4] = {CC_C
, CC_Z
, 0, CC_Z
| CC_P
| CC_C
};
409 void helper_fcomi_ST0_FT0(CPUX86State
*env
)
414 ret
= floatx80_compare(ST0
, FT0
, &env
->fp_status
);
415 eflags
= cpu_cc_compute_all(env
, CC_OP
);
416 eflags
= (eflags
& ~(CC_Z
| CC_P
| CC_C
)) | fcomi_ccval
[ret
+ 1];
420 void helper_fucomi_ST0_FT0(CPUX86State
*env
)
425 ret
= floatx80_compare_quiet(ST0
, FT0
, &env
->fp_status
);
426 eflags
= cpu_cc_compute_all(env
, CC_OP
);
427 eflags
= (eflags
& ~(CC_Z
| CC_P
| CC_C
)) | fcomi_ccval
[ret
+ 1];
431 void helper_fadd_ST0_FT0(CPUX86State
*env
)
433 ST0
= floatx80_add(ST0
, FT0
, &env
->fp_status
);
436 void helper_fmul_ST0_FT0(CPUX86State
*env
)
438 ST0
= floatx80_mul(ST0
, FT0
, &env
->fp_status
);
441 void helper_fsub_ST0_FT0(CPUX86State
*env
)
443 ST0
= floatx80_sub(ST0
, FT0
, &env
->fp_status
);
446 void helper_fsubr_ST0_FT0(CPUX86State
*env
)
448 ST0
= floatx80_sub(FT0
, ST0
, &env
->fp_status
);
451 void helper_fdiv_ST0_FT0(CPUX86State
*env
)
453 ST0
= helper_fdiv(env
, ST0
, FT0
);
456 void helper_fdivr_ST0_FT0(CPUX86State
*env
)
458 ST0
= helper_fdiv(env
, FT0
, ST0
);
461 /* fp operations between STN and ST0 */
463 void helper_fadd_STN_ST0(CPUX86State
*env
, int st_index
)
465 ST(st_index
) = floatx80_add(ST(st_index
), ST0
, &env
->fp_status
);
468 void helper_fmul_STN_ST0(CPUX86State
*env
, int st_index
)
470 ST(st_index
) = floatx80_mul(ST(st_index
), ST0
, &env
->fp_status
);
473 void helper_fsub_STN_ST0(CPUX86State
*env
, int st_index
)
475 ST(st_index
) = floatx80_sub(ST(st_index
), ST0
, &env
->fp_status
);
478 void helper_fsubr_STN_ST0(CPUX86State
*env
, int st_index
)
480 ST(st_index
) = floatx80_sub(ST0
, ST(st_index
), &env
->fp_status
);
483 void helper_fdiv_STN_ST0(CPUX86State
*env
, int st_index
)
488 *p
= helper_fdiv(env
, *p
, ST0
);
491 void helper_fdivr_STN_ST0(CPUX86State
*env
, int st_index
)
496 *p
= helper_fdiv(env
, ST0
, *p
);
499 /* misc FPU operations */
500 void helper_fchs_ST0(CPUX86State
*env
)
502 ST0
= floatx80_chs(ST0
);
505 void helper_fabs_ST0(CPUX86State
*env
)
507 ST0
= floatx80_abs(ST0
);
510 void helper_fld1_ST0(CPUX86State
*env
)
515 void helper_fldl2t_ST0(CPUX86State
*env
)
520 void helper_fldl2e_ST0(CPUX86State
*env
)
525 void helper_fldpi_ST0(CPUX86State
*env
)
530 void helper_fldlg2_ST0(CPUX86State
*env
)
535 void helper_fldln2_ST0(CPUX86State
*env
)
540 void helper_fldz_ST0(CPUX86State
*env
)
545 void helper_fldz_FT0(CPUX86State
*env
)
550 uint32_t helper_fnstsw(CPUX86State
*env
)
552 return (env
->fpus
& ~0x3800) | (env
->fpstt
& 0x7) << 11;
555 uint32_t helper_fnstcw(CPUX86State
*env
)
560 void update_fp_status(CPUX86State
*env
)
564 /* set rounding mode */
565 switch (env
->fpuc
& FPU_RC_MASK
) {
568 rnd_type
= float_round_nearest_even
;
571 rnd_type
= float_round_down
;
574 rnd_type
= float_round_up
;
577 rnd_type
= float_round_to_zero
;
580 set_float_rounding_mode(rnd_type
, &env
->fp_status
);
581 switch ((env
->fpuc
>> 8) & 3) {
593 set_floatx80_rounding_precision(rnd_type
, &env
->fp_status
);
596 void helper_fldcw(CPUX86State
*env
, uint32_t val
)
598 cpu_set_fpuc(env
, val
);
601 void helper_fclex(CPUX86State
*env
)
606 void helper_fwait(CPUX86State
*env
)
608 if (env
->fpus
& FPUS_SE
) {
609 fpu_raise_exception(env
, GETPC());
613 void helper_fninit(CPUX86State
*env
)
617 cpu_set_fpuc(env
, 0x37f);
630 void helper_fbld_ST0(CPUX86State
*env
, target_ulong ptr
)
638 for (i
= 8; i
>= 0; i
--) {
639 v
= cpu_ldub_data_ra(env
, ptr
+ i
, GETPC());
640 val
= (val
* 100) + ((v
>> 4) * 10) + (v
& 0xf);
642 tmp
= int64_to_floatx80(val
, &env
->fp_status
);
643 if (cpu_ldub_data_ra(env
, ptr
+ 9, GETPC()) & 0x80) {
644 tmp
= floatx80_chs(tmp
);
650 void helper_fbst_ST0(CPUX86State
*env
, target_ulong ptr
)
653 target_ulong mem_ref
, mem_end
;
656 val
= floatx80_to_int64(ST0
, &env
->fp_status
);
658 mem_end
= mem_ref
+ 9;
660 cpu_stb_data_ra(env
, mem_end
, 0x80, GETPC());
663 cpu_stb_data_ra(env
, mem_end
, 0x00, GETPC());
665 while (mem_ref
< mem_end
) {
671 v
= ((v
/ 10) << 4) | (v
% 10);
672 cpu_stb_data_ra(env
, mem_ref
++, v
, GETPC());
674 while (mem_ref
< mem_end
) {
675 cpu_stb_data_ra(env
, mem_ref
++, 0, GETPC());
679 void helper_f2xm1(CPUX86State
*env
)
681 double val
= floatx80_to_double(env
, ST0
);
683 val
= pow(2.0, val
) - 1.0;
684 ST0
= double_to_floatx80(env
, val
);
687 void helper_fyl2x(CPUX86State
*env
)
689 double fptemp
= floatx80_to_double(env
, ST0
);
692 fptemp
= log(fptemp
) / log(2.0); /* log2(ST) */
693 fptemp
*= floatx80_to_double(env
, ST1
);
694 ST1
= double_to_floatx80(env
, fptemp
);
697 env
->fpus
&= ~0x4700;
702 void helper_fptan(CPUX86State
*env
)
704 double fptemp
= floatx80_to_double(env
, ST0
);
706 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
709 fptemp
= tan(fptemp
);
710 ST0
= double_to_floatx80(env
, fptemp
);
713 env
->fpus
&= ~0x400; /* C2 <-- 0 */
714 /* the above code is for |arg| < 2**52 only */
718 void helper_fpatan(CPUX86State
*env
)
720 double fptemp
, fpsrcop
;
722 fpsrcop
= floatx80_to_double(env
, ST1
);
723 fptemp
= floatx80_to_double(env
, ST0
);
724 ST1
= double_to_floatx80(env
, atan2(fpsrcop
, fptemp
));
728 void helper_fxtract(CPUX86State
*env
)
734 if (floatx80_is_zero(ST0
)) {
735 /* Easy way to generate -inf and raising division by 0 exception */
736 ST0
= floatx80_div(floatx80_chs(floatx80_one
), floatx80_zero
,
743 expdif
= EXPD(temp
) - EXPBIAS
;
744 /* DP exponent bias */
745 ST0
= int32_to_floatx80(expdif
, &env
->fp_status
);
752 void helper_fprem1(CPUX86State
*env
)
754 double st0
, st1
, dblq
, fpsrcop
, fptemp
;
755 CPU_LDoubleU fpsrcop1
, fptemp1
;
757 signed long long int q
;
759 st0
= floatx80_to_double(env
, ST0
);
760 st1
= floatx80_to_double(env
, ST1
);
762 if (isinf(st0
) || isnan(st0
) || isnan(st1
) || (st1
== 0.0)) {
763 ST0
= double_to_floatx80(env
, 0.0 / 0.0); /* NaN */
764 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
772 expdif
= EXPD(fpsrcop1
) - EXPD(fptemp1
);
775 /* optimisation? taken from the AMD docs */
776 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
777 /* ST0 is unchanged */
782 dblq
= fpsrcop
/ fptemp
;
783 /* round dblq towards nearest integer */
785 st0
= fpsrcop
- fptemp
* dblq
;
787 /* convert dblq to q by truncating towards zero */
789 q
= (signed long long int)(-dblq
);
791 q
= (signed long long int)dblq
;
794 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
795 /* (C0,C3,C1) <-- (q2,q1,q0) */
796 env
->fpus
|= (q
& 0x4) << (8 - 2); /* (C0) <-- q2 */
797 env
->fpus
|= (q
& 0x2) << (14 - 1); /* (C3) <-- q1 */
798 env
->fpus
|= (q
& 0x1) << (9 - 0); /* (C1) <-- q0 */
800 env
->fpus
|= 0x400; /* C2 <-- 1 */
801 fptemp
= pow(2.0, expdif
- 50);
802 fpsrcop
= (st0
/ st1
) / fptemp
;
803 /* fpsrcop = integer obtained by chopping */
804 fpsrcop
= (fpsrcop
< 0.0) ?
805 -(floor(fabs(fpsrcop
))) : floor(fpsrcop
);
806 st0
-= (st1
* fpsrcop
* fptemp
);
808 ST0
= double_to_floatx80(env
, st0
);
811 void helper_fprem(CPUX86State
*env
)
813 double st0
, st1
, dblq
, fpsrcop
, fptemp
;
814 CPU_LDoubleU fpsrcop1
, fptemp1
;
816 signed long long int q
;
818 st0
= floatx80_to_double(env
, ST0
);
819 st1
= floatx80_to_double(env
, ST1
);
821 if (isinf(st0
) || isnan(st0
) || isnan(st1
) || (st1
== 0.0)) {
822 ST0
= double_to_floatx80(env
, 0.0 / 0.0); /* NaN */
823 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
831 expdif
= EXPD(fpsrcop1
) - EXPD(fptemp1
);
834 /* optimisation? taken from the AMD docs */
835 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
836 /* ST0 is unchanged */
841 dblq
= fpsrcop
/ fptemp
; /* ST0 / ST1 */
842 /* round dblq towards zero */
843 dblq
= (dblq
< 0.0) ? ceil(dblq
) : floor(dblq
);
844 st0
= fpsrcop
- fptemp
* dblq
; /* fpsrcop is ST0 */
846 /* convert dblq to q by truncating towards zero */
848 q
= (signed long long int)(-dblq
);
850 q
= (signed long long int)dblq
;
853 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
854 /* (C0,C3,C1) <-- (q2,q1,q0) */
855 env
->fpus
|= (q
& 0x4) << (8 - 2); /* (C0) <-- q2 */
856 env
->fpus
|= (q
& 0x2) << (14 - 1); /* (C3) <-- q1 */
857 env
->fpus
|= (q
& 0x1) << (9 - 0); /* (C1) <-- q0 */
859 int N
= 32 + (expdif
% 32); /* as per AMD docs */
861 env
->fpus
|= 0x400; /* C2 <-- 1 */
862 fptemp
= pow(2.0, (double)(expdif
- N
));
863 fpsrcop
= (st0
/ st1
) / fptemp
;
864 /* fpsrcop = integer obtained by chopping */
865 fpsrcop
= (fpsrcop
< 0.0) ?
866 -(floor(fabs(fpsrcop
))) : floor(fpsrcop
);
867 st0
-= (st1
* fpsrcop
* fptemp
);
869 ST0
= double_to_floatx80(env
, st0
);
872 void helper_fyl2xp1(CPUX86State
*env
)
874 double fptemp
= floatx80_to_double(env
, ST0
);
876 if ((fptemp
+ 1.0) > 0.0) {
877 fptemp
= log(fptemp
+ 1.0) / log(2.0); /* log2(ST + 1.0) */
878 fptemp
*= floatx80_to_double(env
, ST1
);
879 ST1
= double_to_floatx80(env
, fptemp
);
882 env
->fpus
&= ~0x4700;
887 void helper_fsqrt(CPUX86State
*env
)
889 if (floatx80_is_neg(ST0
)) {
890 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
893 ST0
= floatx80_sqrt(ST0
, &env
->fp_status
);
896 void helper_fsincos(CPUX86State
*env
)
898 double fptemp
= floatx80_to_double(env
, ST0
);
900 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
903 ST0
= double_to_floatx80(env
, sin(fptemp
));
905 ST0
= double_to_floatx80(env
, cos(fptemp
));
906 env
->fpus
&= ~0x400; /* C2 <-- 0 */
907 /* the above code is for |arg| < 2**63 only */
911 void helper_frndint(CPUX86State
*env
)
913 ST0
= floatx80_round_to_int(ST0
, &env
->fp_status
);
916 void helper_fscale(CPUX86State
*env
)
918 if (floatx80_is_any_nan(ST1
)) {
921 int n
= floatx80_to_int32_round_to_zero(ST1
, &env
->fp_status
);
922 ST0
= floatx80_scalbn(ST0
, n
, &env
->fp_status
);
926 void helper_fsin(CPUX86State
*env
)
928 double fptemp
= floatx80_to_double(env
, ST0
);
930 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
933 ST0
= double_to_floatx80(env
, sin(fptemp
));
934 env
->fpus
&= ~0x400; /* C2 <-- 0 */
935 /* the above code is for |arg| < 2**53 only */
939 void helper_fcos(CPUX86State
*env
)
941 double fptemp
= floatx80_to_double(env
, ST0
);
943 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
946 ST0
= double_to_floatx80(env
, cos(fptemp
));
947 env
->fpus
&= ~0x400; /* C2 <-- 0 */
948 /* the above code is for |arg| < 2**63 only */
952 void helper_fxam_ST0(CPUX86State
*env
)
959 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
961 env
->fpus
|= 0x200; /* C1 <-- 1 */
964 /* XXX: test fptags too */
966 if (expdif
== MAXEXPD
) {
967 if (MANTD(temp
) == 0x8000000000000000ULL
) {
968 env
->fpus
|= 0x500; /* Infinity */
970 env
->fpus
|= 0x100; /* NaN */
972 } else if (expdif
== 0) {
973 if (MANTD(temp
) == 0) {
974 env
->fpus
|= 0x4000; /* Zero */
976 env
->fpus
|= 0x4400; /* Denormal */
983 static void do_fstenv(CPUX86State
*env
, target_ulong ptr
, int data32
,
986 int fpus
, fptag
, exp
, i
;
990 fpus
= (env
->fpus
& ~0x3800) | (env
->fpstt
& 0x7) << 11;
992 for (i
= 7; i
>= 0; i
--) {
994 if (env
->fptags
[i
]) {
997 tmp
.d
= env
->fpregs
[i
].d
;
1000 if (exp
== 0 && mant
== 0) {
1003 } else if (exp
== 0 || exp
== MAXEXPD
1004 || (mant
& (1LL << 63)) == 0) {
1005 /* NaNs, infinity, denormal */
1012 cpu_stl_data_ra(env
, ptr
, env
->fpuc
, retaddr
);
1013 cpu_stl_data_ra(env
, ptr
+ 4, fpus
, retaddr
);
1014 cpu_stl_data_ra(env
, ptr
+ 8, fptag
, retaddr
);
1015 cpu_stl_data_ra(env
, ptr
+ 12, 0, retaddr
); /* fpip */
1016 cpu_stl_data_ra(env
, ptr
+ 16, 0, retaddr
); /* fpcs */
1017 cpu_stl_data_ra(env
, ptr
+ 20, 0, retaddr
); /* fpoo */
1018 cpu_stl_data_ra(env
, ptr
+ 24, 0, retaddr
); /* fpos */
1021 cpu_stw_data_ra(env
, ptr
, env
->fpuc
, retaddr
);
1022 cpu_stw_data_ra(env
, ptr
+ 2, fpus
, retaddr
);
1023 cpu_stw_data_ra(env
, ptr
+ 4, fptag
, retaddr
);
1024 cpu_stw_data_ra(env
, ptr
+ 6, 0, retaddr
);
1025 cpu_stw_data_ra(env
, ptr
+ 8, 0, retaddr
);
1026 cpu_stw_data_ra(env
, ptr
+ 10, 0, retaddr
);
1027 cpu_stw_data_ra(env
, ptr
+ 12, 0, retaddr
);
1031 void helper_fstenv(CPUX86State
*env
, target_ulong ptr
, int data32
)
1033 do_fstenv(env
, ptr
, data32
, GETPC());
1036 static void do_fldenv(CPUX86State
*env
, target_ulong ptr
, int data32
,
1042 cpu_set_fpuc(env
, cpu_lduw_data_ra(env
, ptr
, retaddr
));
1043 fpus
= cpu_lduw_data_ra(env
, ptr
+ 4, retaddr
);
1044 fptag
= cpu_lduw_data_ra(env
, ptr
+ 8, retaddr
);
1046 cpu_set_fpuc(env
, cpu_lduw_data_ra(env
, ptr
, retaddr
));
1047 fpus
= cpu_lduw_data_ra(env
, ptr
+ 2, retaddr
);
1048 fptag
= cpu_lduw_data_ra(env
, ptr
+ 4, retaddr
);
1050 env
->fpstt
= (fpus
>> 11) & 7;
1051 env
->fpus
= fpus
& ~0x3800;
1052 for (i
= 0; i
< 8; i
++) {
1053 env
->fptags
[i
] = ((fptag
& 3) == 3);
1058 void helper_fldenv(CPUX86State
*env
, target_ulong ptr
, int data32
)
1060 do_fldenv(env
, ptr
, data32
, GETPC());
1063 void helper_fsave(CPUX86State
*env
, target_ulong ptr
, int data32
)
1068 do_fstenv(env
, ptr
, data32
, GETPC());
1070 ptr
+= (14 << data32
);
1071 for (i
= 0; i
< 8; i
++) {
1073 helper_fstt(env
, tmp
, ptr
, GETPC());
1080 cpu_set_fpuc(env
, 0x37f);
1091 void helper_frstor(CPUX86State
*env
, target_ulong ptr
, int data32
)
1096 do_fldenv(env
, ptr
, data32
, GETPC());
1097 ptr
+= (14 << data32
);
1099 for (i
= 0; i
< 8; i
++) {
1100 tmp
= helper_fldt(env
, ptr
, GETPC());
1106 #if defined(CONFIG_USER_ONLY)
1107 void cpu_x86_fsave(CPUX86State
*env
, target_ulong ptr
, int data32
)
1109 helper_fsave(env
, ptr
, data32
);
1112 void cpu_x86_frstor(CPUX86State
*env
, target_ulong ptr
, int data32
)
1114 helper_frstor(env
, ptr
, data32
);
1118 static void do_xsave_fpu(CPUX86State
*env
, target_ulong ptr
, uintptr_t ra
)
1123 fpus
= (env
->fpus
& ~0x3800) | (env
->fpstt
& 0x7) << 11;
1125 for (i
= 0; i
< 8; i
++) {
1126 fptag
|= (env
->fptags
[i
] << i
);
1128 cpu_stw_data_ra(env
, ptr
, env
->fpuc
, ra
);
1129 cpu_stw_data_ra(env
, ptr
+ 2, fpus
, ra
);
1130 cpu_stw_data_ra(env
, ptr
+ 4, fptag
^ 0xff, ra
);
1132 /* In 32-bit mode this is eip, sel, dp, sel.
1133 In 64-bit mode this is rip, rdp.
1134 But in either case we don't write actual data, just zeros. */
1135 cpu_stq_data_ra(env
, ptr
+ 0x08, 0, ra
); /* eip+sel; rip */
1136 cpu_stq_data_ra(env
, ptr
+ 0x10, 0, ra
); /* edp+sel; rdp */
1139 for (i
= 0; i
< 8; i
++) {
1140 floatx80 tmp
= ST(i
);
1141 helper_fstt(env
, tmp
, addr
, ra
);
1146 static void do_xsave_mxcsr(CPUX86State
*env
, target_ulong ptr
, uintptr_t ra
)
1148 cpu_stl_data_ra(env
, ptr
+ 0x18, env
->mxcsr
, ra
); /* mxcsr */
1149 cpu_stl_data_ra(env
, ptr
+ 0x1c, 0x0000ffff, ra
); /* mxcsr_mask */
1152 static void do_xsave_sse(CPUX86State
*env
, target_ulong ptr
, uintptr_t ra
)
1157 if (env
->hflags
& HF_CS64_MASK
) {
1164 for (i
= 0; i
< nb_xmm_regs
; i
++) {
1165 cpu_stq_data_ra(env
, addr
, env
->xmm_regs
[i
].ZMM_Q(0), ra
);
1166 cpu_stq_data_ra(env
, addr
+ 8, env
->xmm_regs
[i
].ZMM_Q(1), ra
);
1171 static void do_xsave_bndregs(CPUX86State
*env
, target_ulong addr
, uintptr_t ra
)
1175 for (i
= 0; i
< 4; i
++, addr
+= 16) {
1176 cpu_stq_data_ra(env
, addr
, env
->bnd_regs
[i
].lb
, ra
);
1177 cpu_stq_data_ra(env
, addr
+ 8, env
->bnd_regs
[i
].ub
, ra
);
1181 static void do_xsave_bndcsr(CPUX86State
*env
, target_ulong addr
, uintptr_t ra
)
1183 cpu_stq_data_ra(env
, addr
, env
->bndcs_regs
.cfgu
, ra
);
1184 cpu_stq_data_ra(env
, addr
+ 8, env
->bndcs_regs
.sts
, ra
);
1187 void helper_fxsave(CPUX86State
*env
, target_ulong ptr
)
1189 uintptr_t ra
= GETPC();
1191 /* The operand must be 16 byte aligned */
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
;
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. */
1235 raise_exception_ra(env
, EXCP0D_GPF
, ra
);
1238 /* Never save anything not enabled by XCR0. */
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
);
1261 /* Update the XSTATE_BV field. */
1262 old_bv
= cpu_ldq_data_ra(env
, ptr
+ 512, ra
);
1263 new_bv
= (old_bv
& ~rfbm
) | (inuse
& rfbm
);
1264 cpu_stq_data_ra(env
, ptr
+ 512, new_bv
, ra
);
1267 void helper_xsave(CPUX86State
*env
, target_ulong ptr
, uint64_t rfbm
)
1269 do_xsave(env
, ptr
, rfbm
, get_xinuse(env
), -1, GETPC());
1272 void helper_xsaveopt(CPUX86State
*env
, target_ulong ptr
, uint64_t rfbm
)
1274 uint64_t inuse
= get_xinuse(env
);
1275 do_xsave(env
, ptr
, rfbm
, inuse
, inuse
, GETPC());
1278 static void do_xrstor_fpu(CPUX86State
*env
, target_ulong ptr
, uintptr_t ra
)
1283 cpu_set_fpuc(env
, cpu_lduw_data_ra(env
, ptr
, ra
));
1284 fpus
= cpu_lduw_data_ra(env
, ptr
+ 2, ra
);
1285 fptag
= cpu_lduw_data_ra(env
, ptr
+ 4, ra
);
1286 env
->fpstt
= (fpus
>> 11) & 7;
1287 env
->fpus
= fpus
& ~0x3800;
1289 for (i
= 0; i
< 8; i
++) {
1290 env
->fptags
[i
] = ((fptag
>> i
) & 1);
1294 for (i
= 0; i
< 8; i
++) {
1295 floatx80 tmp
= helper_fldt(env
, addr
, ra
);
1301 static void do_xrstor_mxcsr(CPUX86State
*env
, target_ulong ptr
, uintptr_t ra
)
1303 cpu_set_mxcsr(env
, cpu_ldl_data_ra(env
, ptr
+ 0x18, ra
));
1306 static void do_xrstor_sse(CPUX86State
*env
, target_ulong ptr
, uintptr_t ra
)
1311 if (env
->hflags
& HF_CS64_MASK
) {
1318 for (i
= 0; i
< nb_xmm_regs
; i
++) {
1319 env
->xmm_regs
[i
].ZMM_Q(0) = cpu_ldq_data_ra(env
, addr
, ra
);
1320 env
->xmm_regs
[i
].ZMM_Q(1) = cpu_ldq_data_ra(env
, addr
+ 8, ra
);
1325 static void do_xrstor_bndregs(CPUX86State
*env
, target_ulong addr
, uintptr_t ra
)
1329 for (i
= 0; i
< 4; i
++, addr
+= 16) {
1330 env
->bnd_regs
[i
].lb
= cpu_ldq_data_ra(env
, addr
, ra
);
1331 env
->bnd_regs
[i
].ub
= cpu_ldq_data_ra(env
, addr
+ 8, ra
);
1335 static void do_xrstor_bndcsr(CPUX86State
*env
, target_ulong addr
, uintptr_t ra
)
1337 /* FIXME: Extend highest implemented bit of linear address. */
1338 env
->bndcs_regs
.cfgu
= cpu_ldq_data_ra(env
, addr
, ra
);
1339 env
->bndcs_regs
.sts
= cpu_ldq_data_ra(env
, addr
+ 8, ra
);
1342 void helper_fxrstor(CPUX86State
*env
, target_ulong ptr
)
1344 uintptr_t ra
= GETPC();
1346 /* The operand must be 16 byte aligned */
1348 raise_exception_ra(env
, EXCP0D_GPF
, ra
);
1351 do_xrstor_fpu(env
, ptr
, ra
);
1353 if (env
->cr
[4] & CR4_OSFXSR_MASK
) {
1354 do_xrstor_mxcsr(env
, ptr
, ra
);
1355 /* Fast FXRSTOR leaves out the XMM registers */
1356 if (!(env
->efer
& MSR_EFER_FFXSR
)
1357 || (env
->hflags
& HF_CPL_MASK
)
1358 || !(env
->hflags
& HF_LMA_MASK
)) {
1359 do_xrstor_sse(env
, ptr
, ra
);
1364 void helper_xrstor(CPUX86State
*env
, target_ulong ptr
, uint64_t rfbm
)
1366 uintptr_t ra
= GETPC();
1367 uint64_t xstate_bv
, xcomp_bv0
, xcomp_bv1
;
1371 /* The OS must have enabled XSAVE. */
1372 if (!(env
->cr
[4] & CR4_OSXSAVE_MASK
)) {
1373 raise_exception_ra(env
, EXCP06_ILLOP
, ra
);
1376 /* The operand must be 64 byte aligned. */
1378 raise_exception_ra(env
, EXCP0D_GPF
, ra
);
1381 xstate_bv
= cpu_ldq_data_ra(env
, ptr
+ 512, ra
);
1383 if ((int64_t)xstate_bv
< 0) {
1384 /* FIXME: Compact form. */
1385 raise_exception_ra(env
, EXCP0D_GPF
, ra
);
1388 /* Standard form. */
1390 /* The XSTATE field must not set bits not present in XCR0. */
1391 if (xstate_bv
& ~env
->xcr0
) {
1392 raise_exception_ra(env
, EXCP0D_GPF
, ra
);
1395 /* The XCOMP field must be zero. */
1396 xcomp_bv0
= cpu_ldq_data_ra(env
, ptr
+ 520, ra
);
1397 xcomp_bv1
= cpu_ldq_data_ra(env
, ptr
+ 528, ra
);
1398 if (xcomp_bv0
|| xcomp_bv1
) {
1399 raise_exception_ra(env
, EXCP0D_GPF
, ra
);
1402 if (rfbm
& XSTATE_FP_MASK
) {
1403 if (xstate_bv
& XSTATE_FP_MASK
) {
1404 do_xrstor_fpu(env
, ptr
, ra
);
1407 memset(env
->fpregs
, 0, sizeof(env
->fpregs
));
1410 if (rfbm
& XSTATE_SSE_MASK
) {
1411 /* Note that the standard form of XRSTOR loads MXCSR from memory
1412 whether or not the XSTATE_BV bit is set. */
1413 do_xrstor_mxcsr(env
, ptr
, ra
);
1414 if (xstate_bv
& XSTATE_SSE_MASK
) {
1415 do_xrstor_sse(env
, ptr
, ra
);
1417 /* ??? When AVX is implemented, we may have to be more
1418 selective in the clearing. */
1419 memset(env
->xmm_regs
, 0, sizeof(env
->xmm_regs
));
1422 if (rfbm
& XSTATE_BNDREGS_MASK
) {
1423 if (xstate_bv
& XSTATE_BNDREGS_MASK
) {
1424 target_ulong off
= x86_ext_save_areas
[XSTATE_BNDREGS_BIT
].offset
;
1425 do_xrstor_bndregs(env
, ptr
+ off
, ra
);
1426 env
->hflags
|= HF_MPX_IU_MASK
;
1428 memset(env
->bnd_regs
, 0, sizeof(env
->bnd_regs
));
1429 env
->hflags
&= ~HF_MPX_IU_MASK
;
1432 if (rfbm
& XSTATE_BNDCSR_MASK
) {
1433 if (xstate_bv
& XSTATE_BNDCSR_MASK
) {
1434 target_ulong off
= x86_ext_save_areas
[XSTATE_BNDCSR_BIT
].offset
;
1435 do_xrstor_bndcsr(env
, ptr
+ off
, ra
);
1437 memset(&env
->bndcs_regs
, 0, sizeof(env
->bndcs_regs
));
1439 cpu_sync_bndcs_hflags(env
);
1443 uint64_t helper_xgetbv(CPUX86State
*env
, uint32_t ecx
)
1445 /* The OS must have enabled XSAVE. */
1446 if (!(env
->cr
[4] & CR4_OSXSAVE_MASK
)) {
1447 raise_exception_ra(env
, EXCP06_ILLOP
, GETPC());
1454 if (env
->features
[FEAT_XSAVE
] & CPUID_XSAVE_XGETBV1
) {
1455 return env
->xcr0
& get_xinuse(env
);
1459 raise_exception_ra(env
, EXCP0D_GPF
, GETPC());
1462 void helper_xsetbv(CPUX86State
*env
, uint32_t ecx
, uint64_t mask
)
1464 uint32_t dummy
, ena_lo
, ena_hi
;
1467 /* The OS must have enabled XSAVE. */
1468 if (!(env
->cr
[4] & CR4_OSXSAVE_MASK
)) {
1469 raise_exception_ra(env
, EXCP06_ILLOP
, GETPC());
1472 /* Only XCR0 is defined at present; the FPU may not be disabled. */
1473 if (ecx
!= 0 || (mask
& XSTATE_FP_MASK
) == 0) {
1477 /* Disallow enabling unimplemented features. */
1478 cpu_x86_cpuid(env
, 0x0d, 0, &ena_lo
, &dummy
, &dummy
, &ena_hi
);
1479 ena
= ((uint64_t)ena_hi
<< 32) | ena_lo
;
1484 /* Disallow enabling only half of MPX. */
1485 if ((mask
^ (mask
* (XSTATE_BNDCSR_MASK
/ XSTATE_BNDREGS_MASK
)))
1486 & XSTATE_BNDCSR_MASK
) {
1491 cpu_sync_bndcs_hflags(env
);
1495 raise_exception_ra(env
, EXCP0D_GPF
, GETPC());
1498 void cpu_get_fp80(uint64_t *pmant
, uint16_t *pexp
, floatx80 f
)
1503 *pmant
= temp
.l
.lower
;
1504 *pexp
= temp
.l
.upper
;
1507 floatx80
cpu_set_fp80(uint64_t mant
, uint16_t upper
)
1511 temp
.l
.upper
= upper
;
1512 temp
.l
.lower
= mant
;
1517 /* XXX: optimize by storing fptt and fptags in the static cpu state */
1519 #define SSE_DAZ 0x0040
1520 #define SSE_RC_MASK 0x6000
1521 #define SSE_RC_NEAR 0x0000
1522 #define SSE_RC_DOWN 0x2000
1523 #define SSE_RC_UP 0x4000
1524 #define SSE_RC_CHOP 0x6000
1525 #define SSE_FZ 0x8000
1527 void cpu_set_mxcsr(CPUX86State
*env
, uint32_t mxcsr
)
1533 /* set rounding mode */
1534 switch (mxcsr
& SSE_RC_MASK
) {
1537 rnd_type
= float_round_nearest_even
;
1540 rnd_type
= float_round_down
;
1543 rnd_type
= float_round_up
;
1546 rnd_type
= float_round_to_zero
;
1549 set_float_rounding_mode(rnd_type
, &env
->sse_status
);
1551 /* set denormals are zero */
1552 set_flush_inputs_to_zero((mxcsr
& SSE_DAZ
) ? 1 : 0, &env
->sse_status
);
1554 /* set flush to zero */
1555 set_flush_to_zero((mxcsr
& SSE_FZ
) ? 1 : 0, &env
->fp_status
);
1558 void cpu_set_fpuc(CPUX86State
*env
, uint16_t val
)
1561 update_fp_status(env
);
1564 void helper_ldmxcsr(CPUX86State
*env
, uint32_t val
)
1566 cpu_set_mxcsr(env
, val
);
1569 void helper_enter_mmx(CPUX86State
*env
)
1572 *(uint32_t *)(env
->fptags
) = 0;
1573 *(uint32_t *)(env
->fptags
+ 4) = 0;
1576 void helper_emms(CPUX86State
*env
)
1578 /* set to empty state */
1579 *(uint32_t *)(env
->fptags
) = 0x01010101;
1580 *(uint32_t *)(env
->fptags
+ 4) = 0x01010101;
1584 void helper_movq(CPUX86State
*env
, void *d
, void *s
)
1586 *(uint64_t *)d
= *(uint64_t *)s
;
1590 #include "ops_sse.h"
1593 #include "ops_sse.h"