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/>.
24 #if !defined(CONFIG_USER_ONLY)
25 #include "exec/softmmu_exec.h"
26 #endif /* !defined(CONFIG_USER_ONLY) */
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
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)
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
)
76 temp
.l
.lower
= cpu_ldq_data(env
, ptr
);
77 temp
.l
.upper
= cpu_lduw_data(env
, ptr
+ 8);
81 static inline void helper_fstt(CPUX86State
*env
, floatx80 f
, target_ulong ptr
)
86 cpu_stq_data(env
, ptr
, temp
.l
.lower
);
87 cpu_stw_data(env
, ptr
+ 8, temp
.l
.upper
);
92 static inline double floatx80_to_double(CPUX86State
*env
, floatx80 a
)
99 u
.f64
= floatx80_to_float64(a
, &env
->fp_status
);
103 static inline floatx80
double_to_floatx80(CPUX86State
*env
, double a
)
111 return float64_to_floatx80(u
.f64
, &env
->fp_status
);
114 static void fpu_set_exception(CPUX86State
*env
, int mask
)
117 if (env
->fpus
& (~env
->fpuc
& FPUC_EM
)) {
118 env
->fpus
|= FPUS_SE
| FPUS_B
;
122 static inline floatx80
helper_fdiv(CPUX86State
*env
, floatx80 a
, floatx80 b
)
124 if (floatx80_is_zero(b
)) {
125 fpu_set_exception(env
, FPUS_ZE
);
127 return floatx80_div(a
, b
, &env
->fp_status
);
130 static void fpu_raise_exception(CPUX86State
*env
)
132 if (env
->cr
[0] & CR0_NE_MASK
) {
133 raise_exception(env
, EXCP10_COPR
);
135 #if !defined(CONFIG_USER_ONLY)
142 void helper_flds_FT0(CPUX86State
*env
, uint32_t val
)
150 FT0
= float32_to_floatx80(u
.f
, &env
->fp_status
);
153 void helper_fldl_FT0(CPUX86State
*env
, uint64_t val
)
161 FT0
= float64_to_floatx80(u
.f
, &env
->fp_status
);
164 void helper_fildl_FT0(CPUX86State
*env
, int32_t val
)
166 FT0
= int32_to_floatx80(val
, &env
->fp_status
);
169 void helper_flds_ST0(CPUX86State
*env
, uint32_t val
)
177 new_fpstt
= (env
->fpstt
- 1) & 7;
179 env
->fpregs
[new_fpstt
].d
= float32_to_floatx80(u
.f
, &env
->fp_status
);
180 env
->fpstt
= new_fpstt
;
181 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
184 void helper_fldl_ST0(CPUX86State
*env
, uint64_t val
)
192 new_fpstt
= (env
->fpstt
- 1) & 7;
194 env
->fpregs
[new_fpstt
].d
= float64_to_floatx80(u
.f
, &env
->fp_status
);
195 env
->fpstt
= new_fpstt
;
196 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
199 void helper_fildl_ST0(CPUX86State
*env
, int32_t val
)
203 new_fpstt
= (env
->fpstt
- 1) & 7;
204 env
->fpregs
[new_fpstt
].d
= int32_to_floatx80(val
, &env
->fp_status
);
205 env
->fpstt
= new_fpstt
;
206 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
209 void helper_fildll_ST0(CPUX86State
*env
, int64_t val
)
213 new_fpstt
= (env
->fpstt
- 1) & 7;
214 env
->fpregs
[new_fpstt
].d
= int64_to_floatx80(val
, &env
->fp_status
);
215 env
->fpstt
= new_fpstt
;
216 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
219 uint32_t helper_fsts_ST0(CPUX86State
*env
)
226 u
.f
= floatx80_to_float32(ST0
, &env
->fp_status
);
230 uint64_t helper_fstl_ST0(CPUX86State
*env
)
237 u
.f
= floatx80_to_float64(ST0
, &env
->fp_status
);
241 int32_t helper_fist_ST0(CPUX86State
*env
)
245 val
= floatx80_to_int32(ST0
, &env
->fp_status
);
246 if (val
!= (int16_t)val
) {
252 int32_t helper_fistl_ST0(CPUX86State
*env
)
256 val
= floatx80_to_int32(ST0
, &env
->fp_status
);
260 int64_t helper_fistll_ST0(CPUX86State
*env
)
264 val
= floatx80_to_int64(ST0
, &env
->fp_status
);
268 int32_t helper_fistt_ST0(CPUX86State
*env
)
272 val
= floatx80_to_int32_round_to_zero(ST0
, &env
->fp_status
);
273 if (val
!= (int16_t)val
) {
279 int32_t helper_fisttl_ST0(CPUX86State
*env
)
283 val
= floatx80_to_int32_round_to_zero(ST0
, &env
->fp_status
);
287 int64_t helper_fisttll_ST0(CPUX86State
*env
)
291 val
= floatx80_to_int64_round_to_zero(ST0
, &env
->fp_status
);
295 void helper_fldt_ST0(CPUX86State
*env
, target_ulong ptr
)
299 new_fpstt
= (env
->fpstt
- 1) & 7;
300 env
->fpregs
[new_fpstt
].d
= helper_fldt(env
, ptr
);
301 env
->fpstt
= new_fpstt
;
302 env
->fptags
[new_fpstt
] = 0; /* validate stack entry */
305 void helper_fstt_ST0(CPUX86State
*env
, target_ulong ptr
)
307 helper_fstt(env
, ST0
, ptr
);
310 void helper_fpush(CPUX86State
*env
)
315 void helper_fpop(CPUX86State
*env
)
320 void helper_fdecstp(CPUX86State
*env
)
322 env
->fpstt
= (env
->fpstt
- 1) & 7;
323 env
->fpus
&= ~0x4700;
326 void helper_fincstp(CPUX86State
*env
)
328 env
->fpstt
= (env
->fpstt
+ 1) & 7;
329 env
->fpus
&= ~0x4700;
334 void helper_ffree_STN(CPUX86State
*env
, int st_index
)
336 env
->fptags
[(env
->fpstt
+ st_index
) & 7] = 1;
339 void helper_fmov_ST0_FT0(CPUX86State
*env
)
344 void helper_fmov_FT0_STN(CPUX86State
*env
, int st_index
)
349 void helper_fmov_ST0_STN(CPUX86State
*env
, int st_index
)
354 void helper_fmov_STN_ST0(CPUX86State
*env
, int st_index
)
359 void helper_fxchg_ST0_STN(CPUX86State
*env
, int st_index
)
370 static const int fcom_ccval
[4] = {0x0100, 0x4000, 0x0000, 0x4500};
372 void helper_fcom_ST0_FT0(CPUX86State
*env
)
376 ret
= floatx80_compare(ST0
, FT0
, &env
->fp_status
);
377 env
->fpus
= (env
->fpus
& ~0x4500) | fcom_ccval
[ret
+ 1];
380 void helper_fucom_ST0_FT0(CPUX86State
*env
)
384 ret
= floatx80_compare_quiet(ST0
, FT0
, &env
->fp_status
);
385 env
->fpus
= (env
->fpus
& ~0x4500) | fcom_ccval
[ret
+ 1];
388 static const int fcomi_ccval
[4] = {CC_C
, CC_Z
, 0, CC_Z
| CC_P
| CC_C
};
390 void helper_fcomi_ST0_FT0(CPUX86State
*env
)
395 ret
= floatx80_compare(ST0
, FT0
, &env
->fp_status
);
396 eflags
= cpu_cc_compute_all(env
, CC_OP
);
397 eflags
= (eflags
& ~(CC_Z
| CC_P
| CC_C
)) | fcomi_ccval
[ret
+ 1];
401 void helper_fucomi_ST0_FT0(CPUX86State
*env
)
406 ret
= floatx80_compare_quiet(ST0
, FT0
, &env
->fp_status
);
407 eflags
= cpu_cc_compute_all(env
, CC_OP
);
408 eflags
= (eflags
& ~(CC_Z
| CC_P
| CC_C
)) | fcomi_ccval
[ret
+ 1];
412 void helper_fadd_ST0_FT0(CPUX86State
*env
)
414 ST0
= floatx80_add(ST0
, FT0
, &env
->fp_status
);
417 void helper_fmul_ST0_FT0(CPUX86State
*env
)
419 ST0
= floatx80_mul(ST0
, FT0
, &env
->fp_status
);
422 void helper_fsub_ST0_FT0(CPUX86State
*env
)
424 ST0
= floatx80_sub(ST0
, FT0
, &env
->fp_status
);
427 void helper_fsubr_ST0_FT0(CPUX86State
*env
)
429 ST0
= floatx80_sub(FT0
, ST0
, &env
->fp_status
);
432 void helper_fdiv_ST0_FT0(CPUX86State
*env
)
434 ST0
= helper_fdiv(env
, ST0
, FT0
);
437 void helper_fdivr_ST0_FT0(CPUX86State
*env
)
439 ST0
= helper_fdiv(env
, FT0
, ST0
);
442 /* fp operations between STN and ST0 */
444 void helper_fadd_STN_ST0(CPUX86State
*env
, int st_index
)
446 ST(st_index
) = floatx80_add(ST(st_index
), ST0
, &env
->fp_status
);
449 void helper_fmul_STN_ST0(CPUX86State
*env
, int st_index
)
451 ST(st_index
) = floatx80_mul(ST(st_index
), ST0
, &env
->fp_status
);
454 void helper_fsub_STN_ST0(CPUX86State
*env
, int st_index
)
456 ST(st_index
) = floatx80_sub(ST(st_index
), ST0
, &env
->fp_status
);
459 void helper_fsubr_STN_ST0(CPUX86State
*env
, int st_index
)
461 ST(st_index
) = floatx80_sub(ST0
, ST(st_index
), &env
->fp_status
);
464 void helper_fdiv_STN_ST0(CPUX86State
*env
, int st_index
)
469 *p
= helper_fdiv(env
, *p
, ST0
);
472 void helper_fdivr_STN_ST0(CPUX86State
*env
, int st_index
)
477 *p
= helper_fdiv(env
, ST0
, *p
);
480 /* misc FPU operations */
481 void helper_fchs_ST0(CPUX86State
*env
)
483 ST0
= floatx80_chs(ST0
);
486 void helper_fabs_ST0(CPUX86State
*env
)
488 ST0
= floatx80_abs(ST0
);
491 void helper_fld1_ST0(CPUX86State
*env
)
496 void helper_fldl2t_ST0(CPUX86State
*env
)
501 void helper_fldl2e_ST0(CPUX86State
*env
)
506 void helper_fldpi_ST0(CPUX86State
*env
)
511 void helper_fldlg2_ST0(CPUX86State
*env
)
516 void helper_fldln2_ST0(CPUX86State
*env
)
521 void helper_fldz_ST0(CPUX86State
*env
)
526 void helper_fldz_FT0(CPUX86State
*env
)
531 uint32_t helper_fnstsw(CPUX86State
*env
)
533 return (env
->fpus
& ~0x3800) | (env
->fpstt
& 0x7) << 11;
536 uint32_t helper_fnstcw(CPUX86State
*env
)
541 static void update_fp_status(CPUX86State
*env
)
545 /* set rounding mode */
546 switch (env
->fpuc
& FPU_RC_MASK
) {
549 rnd_type
= float_round_nearest_even
;
552 rnd_type
= float_round_down
;
555 rnd_type
= float_round_up
;
558 rnd_type
= float_round_to_zero
;
561 set_float_rounding_mode(rnd_type
, &env
->fp_status
);
562 switch ((env
->fpuc
>> 8) & 3) {
574 set_floatx80_rounding_precision(rnd_type
, &env
->fp_status
);
577 void helper_fldcw(CPUX86State
*env
, uint32_t val
)
580 update_fp_status(env
);
583 void helper_fclex(CPUX86State
*env
)
588 void helper_fwait(CPUX86State
*env
)
590 if (env
->fpus
& FPUS_SE
) {
591 fpu_raise_exception(env
);
595 void helper_fninit(CPUX86State
*env
)
612 void helper_fbld_ST0(CPUX86State
*env
, target_ulong ptr
)
620 for (i
= 8; i
>= 0; i
--) {
621 v
= cpu_ldub_data(env
, ptr
+ i
);
622 val
= (val
* 100) + ((v
>> 4) * 10) + (v
& 0xf);
624 tmp
= int64_to_floatx80(val
, &env
->fp_status
);
625 if (cpu_ldub_data(env
, ptr
+ 9) & 0x80) {
632 void helper_fbst_ST0(CPUX86State
*env
, target_ulong ptr
)
635 target_ulong mem_ref
, mem_end
;
638 val
= floatx80_to_int64(ST0
, &env
->fp_status
);
640 mem_end
= mem_ref
+ 9;
642 cpu_stb_data(env
, mem_end
, 0x80);
645 cpu_stb_data(env
, mem_end
, 0x00);
647 while (mem_ref
< mem_end
) {
653 v
= ((v
/ 10) << 4) | (v
% 10);
654 cpu_stb_data(env
, mem_ref
++, v
);
656 while (mem_ref
< mem_end
) {
657 cpu_stb_data(env
, mem_ref
++, 0);
661 void helper_f2xm1(CPUX86State
*env
)
663 double val
= floatx80_to_double(env
, ST0
);
665 val
= pow(2.0, val
) - 1.0;
666 ST0
= double_to_floatx80(env
, val
);
669 void helper_fyl2x(CPUX86State
*env
)
671 double fptemp
= floatx80_to_double(env
, ST0
);
674 fptemp
= log(fptemp
) / log(2.0); /* log2(ST) */
675 fptemp
*= floatx80_to_double(env
, ST1
);
676 ST1
= double_to_floatx80(env
, fptemp
);
679 env
->fpus
&= ~0x4700;
684 void helper_fptan(CPUX86State
*env
)
686 double fptemp
= floatx80_to_double(env
, ST0
);
688 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
691 fptemp
= tan(fptemp
);
692 ST0
= double_to_floatx80(env
, fptemp
);
695 env
->fpus
&= ~0x400; /* C2 <-- 0 */
696 /* the above code is for |arg| < 2**52 only */
700 void helper_fpatan(CPUX86State
*env
)
702 double fptemp
, fpsrcop
;
704 fpsrcop
= floatx80_to_double(env
, ST1
);
705 fptemp
= floatx80_to_double(env
, ST0
);
706 ST1
= double_to_floatx80(env
, atan2(fpsrcop
, fptemp
));
710 void helper_fxtract(CPUX86State
*env
)
716 if (floatx80_is_zero(ST0
)) {
717 /* Easy way to generate -inf and raising division by 0 exception */
718 ST0
= floatx80_div(floatx80_chs(floatx80_one
), floatx80_zero
,
725 expdif
= EXPD(temp
) - EXPBIAS
;
726 /* DP exponent bias */
727 ST0
= int32_to_floatx80(expdif
, &env
->fp_status
);
734 void helper_fprem1(CPUX86State
*env
)
736 double st0
, st1
, dblq
, fpsrcop
, fptemp
;
737 CPU_LDoubleU fpsrcop1
, fptemp1
;
739 signed long long int q
;
741 st0
= floatx80_to_double(env
, ST0
);
742 st1
= floatx80_to_double(env
, ST1
);
744 if (isinf(st0
) || isnan(st0
) || isnan(st1
) || (st1
== 0.0)) {
745 ST0
= double_to_floatx80(env
, 0.0 / 0.0); /* NaN */
746 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
754 expdif
= EXPD(fpsrcop1
) - EXPD(fptemp1
);
757 /* optimisation? taken from the AMD docs */
758 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
759 /* ST0 is unchanged */
764 dblq
= fpsrcop
/ fptemp
;
765 /* round dblq towards nearest integer */
767 st0
= fpsrcop
- fptemp
* dblq
;
769 /* convert dblq to q by truncating towards zero */
771 q
= (signed long long int)(-dblq
);
773 q
= (signed long long int)dblq
;
776 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
777 /* (C0,C3,C1) <-- (q2,q1,q0) */
778 env
->fpus
|= (q
& 0x4) << (8 - 2); /* (C0) <-- q2 */
779 env
->fpus
|= (q
& 0x2) << (14 - 1); /* (C3) <-- q1 */
780 env
->fpus
|= (q
& 0x1) << (9 - 0); /* (C1) <-- q0 */
782 env
->fpus
|= 0x400; /* C2 <-- 1 */
783 fptemp
= pow(2.0, expdif
- 50);
784 fpsrcop
= (st0
/ st1
) / fptemp
;
785 /* fpsrcop = integer obtained by chopping */
786 fpsrcop
= (fpsrcop
< 0.0) ?
787 -(floor(fabs(fpsrcop
))) : floor(fpsrcop
);
788 st0
-= (st1
* fpsrcop
* fptemp
);
790 ST0
= double_to_floatx80(env
, st0
);
793 void helper_fprem(CPUX86State
*env
)
795 double st0
, st1
, dblq
, fpsrcop
, fptemp
;
796 CPU_LDoubleU fpsrcop1
, fptemp1
;
798 signed long long int q
;
800 st0
= floatx80_to_double(env
, ST0
);
801 st1
= floatx80_to_double(env
, ST1
);
803 if (isinf(st0
) || isnan(st0
) || isnan(st1
) || (st1
== 0.0)) {
804 ST0
= double_to_floatx80(env
, 0.0 / 0.0); /* NaN */
805 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
813 expdif
= EXPD(fpsrcop1
) - EXPD(fptemp1
);
816 /* optimisation? taken from the AMD docs */
817 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
818 /* ST0 is unchanged */
823 dblq
= fpsrcop
/ fptemp
; /* ST0 / ST1 */
824 /* round dblq towards zero */
825 dblq
= (dblq
< 0.0) ? ceil(dblq
) : floor(dblq
);
826 st0
= fpsrcop
- fptemp
* dblq
; /* fpsrcop is ST0 */
828 /* convert dblq to q by truncating towards zero */
830 q
= (signed long long int)(-dblq
);
832 q
= (signed long long int)dblq
;
835 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
836 /* (C0,C3,C1) <-- (q2,q1,q0) */
837 env
->fpus
|= (q
& 0x4) << (8 - 2); /* (C0) <-- q2 */
838 env
->fpus
|= (q
& 0x2) << (14 - 1); /* (C3) <-- q1 */
839 env
->fpus
|= (q
& 0x1) << (9 - 0); /* (C1) <-- q0 */
841 int N
= 32 + (expdif
% 32); /* as per AMD docs */
843 env
->fpus
|= 0x400; /* C2 <-- 1 */
844 fptemp
= pow(2.0, (double)(expdif
- N
));
845 fpsrcop
= (st0
/ st1
) / fptemp
;
846 /* fpsrcop = integer obtained by chopping */
847 fpsrcop
= (fpsrcop
< 0.0) ?
848 -(floor(fabs(fpsrcop
))) : floor(fpsrcop
);
849 st0
-= (st1
* fpsrcop
* fptemp
);
851 ST0
= double_to_floatx80(env
, st0
);
854 void helper_fyl2xp1(CPUX86State
*env
)
856 double fptemp
= floatx80_to_double(env
, ST0
);
858 if ((fptemp
+ 1.0) > 0.0) {
859 fptemp
= log(fptemp
+ 1.0) / log(2.0); /* log2(ST + 1.0) */
860 fptemp
*= floatx80_to_double(env
, ST1
);
861 ST1
= double_to_floatx80(env
, fptemp
);
864 env
->fpus
&= ~0x4700;
869 void helper_fsqrt(CPUX86State
*env
)
871 if (floatx80_is_neg(ST0
)) {
872 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
875 ST0
= floatx80_sqrt(ST0
, &env
->fp_status
);
878 void helper_fsincos(CPUX86State
*env
)
880 double fptemp
= floatx80_to_double(env
, ST0
);
882 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
885 ST0
= double_to_floatx80(env
, sin(fptemp
));
887 ST0
= double_to_floatx80(env
, cos(fptemp
));
888 env
->fpus
&= ~0x400; /* C2 <-- 0 */
889 /* the above code is for |arg| < 2**63 only */
893 void helper_frndint(CPUX86State
*env
)
895 ST0
= floatx80_round_to_int(ST0
, &env
->fp_status
);
898 void helper_fscale(CPUX86State
*env
)
900 if (floatx80_is_any_nan(ST1
)) {
903 int n
= floatx80_to_int32_round_to_zero(ST1
, &env
->fp_status
);
904 ST0
= floatx80_scalbn(ST0
, n
, &env
->fp_status
);
908 void helper_fsin(CPUX86State
*env
)
910 double fptemp
= floatx80_to_double(env
, ST0
);
912 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
915 ST0
= double_to_floatx80(env
, sin(fptemp
));
916 env
->fpus
&= ~0x400; /* C2 <-- 0 */
917 /* the above code is for |arg| < 2**53 only */
921 void helper_fcos(CPUX86State
*env
)
923 double fptemp
= floatx80_to_double(env
, ST0
);
925 if ((fptemp
> MAXTAN
) || (fptemp
< -MAXTAN
)) {
928 ST0
= double_to_floatx80(env
, cos(fptemp
));
929 env
->fpus
&= ~0x400; /* C2 <-- 0 */
930 /* the above code is for |arg| < 2**63 only */
934 void helper_fxam_ST0(CPUX86State
*env
)
941 env
->fpus
&= ~0x4700; /* (C3,C2,C1,C0) <-- 0000 */
943 env
->fpus
|= 0x200; /* C1 <-- 1 */
946 /* XXX: test fptags too */
948 if (expdif
== MAXEXPD
) {
949 if (MANTD(temp
) == 0x8000000000000000ULL
) {
950 env
->fpus
|= 0x500; /* Infinity */
952 env
->fpus
|= 0x100; /* NaN */
954 } else if (expdif
== 0) {
955 if (MANTD(temp
) == 0) {
956 env
->fpus
|= 0x4000; /* Zero */
958 env
->fpus
|= 0x4400; /* Denormal */
965 void helper_fstenv(CPUX86State
*env
, target_ulong ptr
, int data32
)
967 int fpus
, fptag
, exp
, i
;
971 fpus
= (env
->fpus
& ~0x3800) | (env
->fpstt
& 0x7) << 11;
973 for (i
= 7; i
>= 0; i
--) {
975 if (env
->fptags
[i
]) {
978 tmp
.d
= env
->fpregs
[i
].d
;
981 if (exp
== 0 && mant
== 0) {
984 } else if (exp
== 0 || exp
== MAXEXPD
985 || (mant
& (1LL << 63)) == 0) {
986 /* NaNs, infinity, denormal */
993 cpu_stl_data(env
, ptr
, env
->fpuc
);
994 cpu_stl_data(env
, ptr
+ 4, fpus
);
995 cpu_stl_data(env
, ptr
+ 8, fptag
);
996 cpu_stl_data(env
, ptr
+ 12, 0); /* fpip */
997 cpu_stl_data(env
, ptr
+ 16, 0); /* fpcs */
998 cpu_stl_data(env
, ptr
+ 20, 0); /* fpoo */
999 cpu_stl_data(env
, ptr
+ 24, 0); /* fpos */
1002 cpu_stw_data(env
, ptr
, env
->fpuc
);
1003 cpu_stw_data(env
, ptr
+ 2, fpus
);
1004 cpu_stw_data(env
, ptr
+ 4, fptag
);
1005 cpu_stw_data(env
, ptr
+ 6, 0);
1006 cpu_stw_data(env
, ptr
+ 8, 0);
1007 cpu_stw_data(env
, ptr
+ 10, 0);
1008 cpu_stw_data(env
, ptr
+ 12, 0);
1012 void helper_fldenv(CPUX86State
*env
, target_ulong ptr
, int data32
)
1017 env
->fpuc
= cpu_lduw_data(env
, ptr
);
1018 fpus
= cpu_lduw_data(env
, ptr
+ 4);
1019 fptag
= cpu_lduw_data(env
, ptr
+ 8);
1021 env
->fpuc
= cpu_lduw_data(env
, ptr
);
1022 fpus
= cpu_lduw_data(env
, ptr
+ 2);
1023 fptag
= cpu_lduw_data(env
, ptr
+ 4);
1025 env
->fpstt
= (fpus
>> 11) & 7;
1026 env
->fpus
= fpus
& ~0x3800;
1027 for (i
= 0; i
< 8; i
++) {
1028 env
->fptags
[i
] = ((fptag
& 3) == 3);
1033 void helper_fsave(CPUX86State
*env
, target_ulong ptr
, int data32
)
1038 helper_fstenv(env
, ptr
, data32
);
1040 ptr
+= (14 << data32
);
1041 for (i
= 0; i
< 8; i
++) {
1043 helper_fstt(env
, tmp
, ptr
);
1061 void helper_frstor(CPUX86State
*env
, target_ulong ptr
, int data32
)
1066 helper_fldenv(env
, ptr
, data32
);
1067 ptr
+= (14 << data32
);
1069 for (i
= 0; i
< 8; i
++) {
1070 tmp
= helper_fldt(env
, ptr
);
1076 #if defined(CONFIG_USER_ONLY)
1077 void cpu_x86_fsave(CPUX86State
*env
, target_ulong ptr
, int data32
)
1079 helper_fsave(env
, ptr
, data32
);
1082 void cpu_x86_frstor(CPUX86State
*env
, target_ulong ptr
, int data32
)
1084 helper_frstor(env
, ptr
, data32
);
1088 void helper_fxsave(CPUX86State
*env
, target_ulong ptr
, int data64
)
1090 int fpus
, fptag
, i
, nb_xmm_regs
;
1094 /* The operand must be 16 byte aligned */
1096 raise_exception(env
, EXCP0D_GPF
);
1099 fpus
= (env
->fpus
& ~0x3800) | (env
->fpstt
& 0x7) << 11;
1101 for (i
= 0; i
< 8; i
++) {
1102 fptag
|= (env
->fptags
[i
] << i
);
1104 cpu_stw_data(env
, ptr
, env
->fpuc
);
1105 cpu_stw_data(env
, ptr
+ 2, fpus
);
1106 cpu_stw_data(env
, ptr
+ 4, fptag
^ 0xff);
1107 #ifdef TARGET_X86_64
1109 cpu_stq_data(env
, ptr
+ 0x08, 0); /* rip */
1110 cpu_stq_data(env
, ptr
+ 0x10, 0); /* rdp */
1114 cpu_stl_data(env
, ptr
+ 0x08, 0); /* eip */
1115 cpu_stl_data(env
, ptr
+ 0x0c, 0); /* sel */
1116 cpu_stl_data(env
, ptr
+ 0x10, 0); /* dp */
1117 cpu_stl_data(env
, ptr
+ 0x14, 0); /* sel */
1121 for (i
= 0; i
< 8; i
++) {
1123 helper_fstt(env
, tmp
, addr
);
1127 if (env
->cr
[4] & CR4_OSFXSR_MASK
) {
1128 /* XXX: finish it */
1129 cpu_stl_data(env
, ptr
+ 0x18, env
->mxcsr
); /* mxcsr */
1130 cpu_stl_data(env
, ptr
+ 0x1c, 0x0000ffff); /* mxcsr_mask */
1131 if (env
->hflags
& HF_CS64_MASK
) {
1137 /* Fast FXSAVE leaves out the XMM registers */
1138 if (!(env
->efer
& MSR_EFER_FFXSR
)
1139 || (env
->hflags
& HF_CPL_MASK
)
1140 || !(env
->hflags
& HF_LMA_MASK
)) {
1141 for (i
= 0; i
< nb_xmm_regs
; i
++) {
1142 cpu_stq_data(env
, addr
, env
->xmm_regs
[i
].XMM_Q(0));
1143 cpu_stq_data(env
, addr
+ 8, env
->xmm_regs
[i
].XMM_Q(1));
1150 void helper_fxrstor(CPUX86State
*env
, target_ulong ptr
, int data64
)
1152 int i
, fpus
, fptag
, nb_xmm_regs
;
1156 /* The operand must be 16 byte aligned */
1158 raise_exception(env
, EXCP0D_GPF
);
1161 env
->fpuc
= cpu_lduw_data(env
, ptr
);
1162 fpus
= cpu_lduw_data(env
, ptr
+ 2);
1163 fptag
= cpu_lduw_data(env
, ptr
+ 4);
1164 env
->fpstt
= (fpus
>> 11) & 7;
1165 env
->fpus
= fpus
& ~0x3800;
1167 for (i
= 0; i
< 8; i
++) {
1168 env
->fptags
[i
] = ((fptag
>> i
) & 1);
1172 for (i
= 0; i
< 8; i
++) {
1173 tmp
= helper_fldt(env
, addr
);
1178 if (env
->cr
[4] & CR4_OSFXSR_MASK
) {
1179 /* XXX: finish it */
1180 env
->mxcsr
= cpu_ldl_data(env
, ptr
+ 0x18);
1181 /* cpu_ldl_data(env, ptr + 0x1c); */
1182 if (env
->hflags
& HF_CS64_MASK
) {
1188 /* Fast FXRESTORE leaves out the XMM registers */
1189 if (!(env
->efer
& MSR_EFER_FFXSR
)
1190 || (env
->hflags
& HF_CPL_MASK
)
1191 || !(env
->hflags
& HF_LMA_MASK
)) {
1192 for (i
= 0; i
< nb_xmm_regs
; i
++) {
1193 env
->xmm_regs
[i
].XMM_Q(0) = cpu_ldq_data(env
, addr
);
1194 env
->xmm_regs
[i
].XMM_Q(1) = cpu_ldq_data(env
, addr
+ 8);
1201 void cpu_get_fp80(uint64_t *pmant
, uint16_t *pexp
, floatx80 f
)
1206 *pmant
= temp
.l
.lower
;
1207 *pexp
= temp
.l
.upper
;
1210 floatx80
cpu_set_fp80(uint64_t mant
, uint16_t upper
)
1214 temp
.l
.upper
= upper
;
1215 temp
.l
.lower
= mant
;
1220 /* XXX: optimize by storing fptt and fptags in the static cpu state */
1222 #define SSE_DAZ 0x0040
1223 #define SSE_RC_MASK 0x6000
1224 #define SSE_RC_NEAR 0x0000
1225 #define SSE_RC_DOWN 0x2000
1226 #define SSE_RC_UP 0x4000
1227 #define SSE_RC_CHOP 0x6000
1228 #define SSE_FZ 0x8000
1230 static void update_sse_status(CPUX86State
*env
)
1234 /* set rounding mode */
1235 switch (env
->mxcsr
& SSE_RC_MASK
) {
1238 rnd_type
= float_round_nearest_even
;
1241 rnd_type
= float_round_down
;
1244 rnd_type
= float_round_up
;
1247 rnd_type
= float_round_to_zero
;
1250 set_float_rounding_mode(rnd_type
, &env
->sse_status
);
1252 /* set denormals are zero */
1253 set_flush_inputs_to_zero((env
->mxcsr
& SSE_DAZ
) ? 1 : 0, &env
->sse_status
);
1255 /* set flush to zero */
1256 set_flush_to_zero((env
->mxcsr
& SSE_FZ
) ? 1 : 0, &env
->fp_status
);
1259 void helper_ldmxcsr(CPUX86State
*env
, uint32_t val
)
1262 update_sse_status(env
);
1265 void helper_enter_mmx(CPUX86State
*env
)
1268 *(uint32_t *)(env
->fptags
) = 0;
1269 *(uint32_t *)(env
->fptags
+ 4) = 0;
1272 void helper_emms(CPUX86State
*env
)
1274 /* set to empty state */
1275 *(uint32_t *)(env
->fptags
) = 0x01010101;
1276 *(uint32_t *)(env
->fptags
+ 4) = 0x01010101;
1280 void helper_movq(CPUX86State
*env
, void *d
, void *s
)
1282 *(uint64_t *)d
= *(uint64_t *)s
;
1286 #include "ops_sse.h"
1289 #include "ops_sse.h"