1 /**************************************************************************
3 * Copyright (C) 1999-2005 Brian Paul All Rights Reserved.
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be included
13 * in all copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
16 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
19 * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
20 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
22 **************************************************************************/
24 #include "pipe/p_config.h"
26 #if defined(PIPE_ARCH_X86)
28 #include "pipe/p_compiler.h"
29 #include "pipe/p_debug.h"
30 #include "util/u_pointer.h"
32 #include "rtasm_execmem.h"
33 #include "rtasm_x86sse.h"
42 void x86_print_reg( struct x86_reg reg
)
44 if (reg
.mod
!= mod_REG
)
50 case reg_AX
: debug_printf( "EAX" ); break;
51 case reg_CX
: debug_printf( "ECX" ); break;
52 case reg_DX
: debug_printf( "EDX" ); break;
53 case reg_BX
: debug_printf( "EBX" ); break;
54 case reg_SP
: debug_printf( "ESP" ); break;
55 case reg_BP
: debug_printf( "EBP" ); break;
56 case reg_SI
: debug_printf( "ESI" ); break;
57 case reg_DI
: debug_printf( "EDI" ); break;
61 debug_printf( "MMX%u", reg
.idx
);
64 debug_printf( "XMM%u", reg
.idx
);
67 debug_printf( "fp%u", reg
.idx
);
71 if (reg
.mod
== mod_DISP8
||
72 reg
.mod
== mod_DISP32
)
73 debug_printf("+%d", reg
.disp
);
75 if (reg
.mod
!= mod_REG
)
81 #define DUMP_START() debug_printf( "\n" )
82 #define DUMP_END() debug_printf( "\n" )
85 const char *foo = __FUNCTION__; \
86 while (*foo && *foo != '_') \
90 debug_printf( "\n% 4x% 15s ", p->csr - p->store, foo ); \
93 #define DUMP_I( I ) do { \
95 debug_printf( "%u", I ); \
98 #define DUMP_R( R0 ) do { \
100 x86_print_reg( R0 ); \
103 #define DUMP_RR( R0, R1 ) do { \
105 x86_print_reg( R0 ); \
106 debug_printf( ", " ); \
107 x86_print_reg( R1 ); \
110 #define DUMP_RI( R0, I ) do { \
112 x86_print_reg( R0 ); \
113 debug_printf( ", %u", I ); \
116 #define DUMP_RRI( R0, R1, I ) do { \
118 x86_print_reg( R0 ); \
119 debug_printf( ", " ); \
120 x86_print_reg( R1 ); \
121 debug_printf( ", %u", I ); \
131 #define DUMP_RR( R0, R1 )
132 #define DUMP_RI( R0, I )
133 #define DUMP_RRI( R0, R1, I )
138 static void do_realloc( struct x86_function
*p
)
140 if (p
->store
== p
->error_overflow
) {
143 else if (p
->size
== 0) {
145 p
->store
= rtasm_exec_malloc(p
->size
);
149 uintptr_t used
= pointer_to_uintptr( p
->csr
) - pointer_to_uintptr( p
->store
);
150 unsigned char *tmp
= p
->store
;
152 p
->store
= rtasm_exec_malloc(p
->size
);
155 memcpy(p
->store
, tmp
, used
);
156 p
->csr
= p
->store
+ used
;
162 rtasm_exec_free(tmp
);
165 if (p
->store
== NULL
) {
166 p
->store
= p
->csr
= p
->error_overflow
;
167 p
->size
= sizeof(p
->error_overflow
);
171 /* Emit bytes to the instruction stream:
173 static unsigned char *reserve( struct x86_function
*p
, int bytes
)
175 if (p
->csr
+ bytes
- p
->store
> (int) p
->size
)
179 unsigned char *csr
= p
->csr
;
187 static void emit_1b( struct x86_function
*p
, char b0
)
189 char *csr
= (char *)reserve(p
, 1);
193 static void emit_1i( struct x86_function
*p
, int i0
)
195 int *icsr
= (int *)reserve(p
, sizeof(i0
));
199 static void emit_1ub( struct x86_function
*p
, unsigned char b0
)
201 unsigned char *csr
= reserve(p
, 1);
205 static void emit_2ub( struct x86_function
*p
, unsigned char b0
, unsigned char b1
)
207 unsigned char *csr
= reserve(p
, 2);
212 static void emit_3ub( struct x86_function
*p
, unsigned char b0
, unsigned char b1
, unsigned char b2
)
214 unsigned char *csr
= reserve(p
, 3);
221 /* Build a modRM byte + possible displacement. No treatment of SIB
222 * indexing. BZZT - no way to encode an absolute address.
224 * This is the "/r" field in the x86 manuals...
226 static void emit_modrm( struct x86_function
*p
,
228 struct x86_reg regmem
)
230 unsigned char val
= 0;
232 assert(reg
.mod
== mod_REG
);
234 val
|= regmem
.mod
<< 6; /* mod field */
235 val
|= reg
.idx
<< 3; /* reg field */
236 val
|= regmem
.idx
; /* r/m field */
240 /* Oh-oh we've stumbled into the SIB thing.
242 if (regmem
.file
== file_REG32
&&
243 regmem
.idx
== reg_SP
&&
244 regmem
.mod
!= mod_REG
) {
245 emit_1ub(p
, 0x24); /* simplistic! */
248 switch (regmem
.mod
) {
253 emit_1b(p
, (char) regmem
.disp
);
256 emit_1i(p
, regmem
.disp
);
264 /* Emits the "/0".."/7" specialized versions of the modrm ("/r") bytes.
266 static void emit_modrm_noreg( struct x86_function
*p
,
268 struct x86_reg regmem
)
270 struct x86_reg dummy
= x86_make_reg(file_REG32
, op
);
271 emit_modrm(p
, dummy
, regmem
);
274 /* Many x86 instructions have two opcodes to cope with the situations
275 * where the destination is a register or memory reference
276 * respectively. This function selects the correct opcode based on
277 * the arguments presented.
279 static void emit_op_modrm( struct x86_function
*p
,
280 unsigned char op_dst_is_reg
,
281 unsigned char op_dst_is_mem
,
287 emit_1ub(p
, op_dst_is_reg
);
288 emit_modrm(p
, dst
, src
);
293 assert(src
.mod
== mod_REG
);
294 emit_1ub(p
, op_dst_is_mem
);
295 emit_modrm(p
, src
, dst
);
309 /* Create and manipulate registers and regmem values:
311 struct x86_reg
x86_make_reg( enum x86_reg_file file
,
312 enum x86_reg_name idx
)
324 struct x86_reg
x86_make_disp( struct x86_reg reg
,
327 assert(reg
.file
== file_REG32
);
329 if (reg
.mod
== mod_REG
)
334 if (reg
.disp
== 0 && reg
.idx
!= reg_BP
)
335 reg
.mod
= mod_INDIRECT
;
336 else if (reg
.disp
<= 127 && reg
.disp
>= -128)
339 reg
.mod
= mod_DISP32
;
344 struct x86_reg
x86_deref( struct x86_reg reg
)
346 return x86_make_disp(reg
, 0);
349 struct x86_reg
x86_get_base_reg( struct x86_reg reg
)
351 return x86_make_reg( reg
.file
, reg
.idx
);
354 int x86_get_label( struct x86_function
*p
)
356 return p
->csr
- p
->store
;
361 /***********************************************************************
366 void x86_jcc( struct x86_function
*p
,
370 int offset
= label
- (x86_get_label(p
) + 2);
374 assert(p
->csr
- p
->store
> -offset
);
377 if (offset
<= 127 && offset
>= -128) {
378 emit_1ub(p
, 0x70 + cc
);
379 emit_1b(p
, (char) offset
);
382 offset
= label
- (x86_get_label(p
) + 6);
383 emit_2ub(p
, 0x0f, 0x80 + cc
);
388 /* Always use a 32bit offset for forward jumps:
390 int x86_jcc_forward( struct x86_function
*p
,
394 emit_2ub(p
, 0x0f, 0x80 + cc
);
396 return x86_get_label(p
);
399 int x86_jmp_forward( struct x86_function
*p
)
404 return x86_get_label(p
);
407 int x86_call_forward( struct x86_function
*p
)
413 return x86_get_label(p
);
416 /* Fixup offset from forward jump:
418 void x86_fixup_fwd_jump( struct x86_function
*p
,
421 *(int *)(p
->store
+ fixup
- 4) = x86_get_label(p
) - fixup
;
424 void x86_jmp( struct x86_function
*p
, int label
)
428 emit_1i(p
, label
- x86_get_label(p
) - 4);
431 void x86_call( struct x86_function
*p
, struct x86_reg reg
)
435 emit_modrm_noreg(p
, 2, reg
);
439 void x86_mov_reg_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
442 assert(dst
.file
== file_REG32
);
443 assert(dst
.mod
== mod_REG
);
444 emit_1ub(p
, 0xb8 + dst
.idx
);
449 * Immediate group 1 instructions.
452 x86_group1_imm( struct x86_function
*p
,
453 unsigned op
, struct x86_reg dst
, int imm
)
455 assert(dst
.file
== file_REG32
);
456 assert(dst
.mod
== mod_REG
);
457 if(-0x80 <= imm
&& imm
< 0x80) {
459 emit_modrm_noreg(p
, op
, dst
);
460 emit_1b(p
, (char)imm
);
464 emit_modrm_noreg(p
, op
, dst
);
469 void x86_add_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
472 x86_group1_imm(p
, 0, dst
, imm
);
475 void x86_or_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
478 x86_group1_imm(p
, 1, dst
, imm
);
481 void x86_and_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
484 x86_group1_imm(p
, 4, dst
, imm
);
487 void x86_sub_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
490 x86_group1_imm(p
, 5, dst
, imm
);
493 void x86_xor_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
496 x86_group1_imm(p
, 6, dst
, imm
);
499 void x86_cmp_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
502 x86_group1_imm(p
, 7, dst
, imm
);
506 void x86_push( struct x86_function
*p
,
510 if (reg
.mod
== mod_REG
)
511 emit_1ub(p
, 0x50 + reg
.idx
);
515 emit_modrm_noreg(p
, 6, reg
);
519 p
->stack_offset
+= 4;
522 void x86_push_imm32( struct x86_function
*p
,
529 p
->stack_offset
+= 4;
533 void x86_pop( struct x86_function
*p
,
537 assert(reg
.mod
== mod_REG
);
538 emit_1ub(p
, 0x58 + reg
.idx
);
539 p
->stack_offset
-= 4;
542 void x86_inc( struct x86_function
*p
,
546 assert(reg
.mod
== mod_REG
);
547 emit_1ub(p
, 0x40 + reg
.idx
);
550 void x86_dec( struct x86_function
*p
,
554 assert(reg
.mod
== mod_REG
);
555 emit_1ub(p
, 0x48 + reg
.idx
);
558 void x86_ret( struct x86_function
*p
)
561 assert(p
->stack_offset
== 0);
565 void x86_retw( struct x86_function
*p
, unsigned short imm
)
568 emit_3ub(p
, 0xc2, imm
& 0xff, (imm
>> 8) & 0xff);
571 void x86_sahf( struct x86_function
*p
)
577 void x86_mov( struct x86_function
*p
,
582 emit_op_modrm( p
, 0x8b, 0x89, dst
, src
);
585 void x86_xor( struct x86_function
*p
,
590 emit_op_modrm( p
, 0x33, 0x31, dst
, src
);
593 void x86_cmp( struct x86_function
*p
,
598 emit_op_modrm( p
, 0x3b, 0x39, dst
, src
);
601 void x86_lea( struct x86_function
*p
,
607 emit_modrm( p
, dst
, src
);
610 void x86_test( struct x86_function
*p
,
616 emit_modrm( p
, dst
, src
);
619 void x86_add( struct x86_function
*p
,
624 emit_op_modrm(p
, 0x03, 0x01, dst
, src
);
627 /* Calculate EAX * src, results in EDX:EAX.
629 void x86_mul( struct x86_function
*p
,
634 emit_modrm_noreg(p
, 4, src
);
638 void x86_imul( struct x86_function
*p
,
643 emit_2ub(p
, X86_TWOB
, 0xAF);
644 emit_modrm(p
, dst
, src
);
648 void x86_sub( struct x86_function
*p
,
653 emit_op_modrm(p
, 0x2b, 0x29, dst
, src
);
656 void x86_or( struct x86_function
*p
,
661 emit_op_modrm( p
, 0x0b, 0x09, dst
, src
);
664 void x86_and( struct x86_function
*p
,
669 emit_op_modrm( p
, 0x23, 0x21, dst
, src
);
674 /***********************************************************************
679 void sse_movss( struct x86_function
*p
,
684 emit_2ub(p
, 0xF3, X86_TWOB
);
685 emit_op_modrm( p
, 0x10, 0x11, dst
, src
);
688 void sse_movaps( struct x86_function
*p
,
693 emit_1ub(p
, X86_TWOB
);
694 emit_op_modrm( p
, 0x28, 0x29, dst
, src
);
697 void sse_movups( struct x86_function
*p
,
702 emit_1ub(p
, X86_TWOB
);
703 emit_op_modrm( p
, 0x10, 0x11, dst
, src
);
706 void sse_movhps( struct x86_function
*p
,
711 assert(dst
.mod
!= mod_REG
|| src
.mod
!= mod_REG
);
712 emit_1ub(p
, X86_TWOB
);
713 emit_op_modrm( p
, 0x16, 0x17, dst
, src
); /* cf movlhps */
716 void sse_movlps( struct x86_function
*p
,
721 assert(dst
.mod
!= mod_REG
|| src
.mod
!= mod_REG
);
722 emit_1ub(p
, X86_TWOB
);
723 emit_op_modrm( p
, 0x12, 0x13, dst
, src
); /* cf movhlps */
726 void sse_maxps( struct x86_function
*p
,
731 emit_2ub(p
, X86_TWOB
, 0x5F);
732 emit_modrm( p
, dst
, src
);
735 void sse_maxss( struct x86_function
*p
,
740 emit_3ub(p
, 0xF3, X86_TWOB
, 0x5F);
741 emit_modrm( p
, dst
, src
);
744 void sse_divss( struct x86_function
*p
,
749 emit_3ub(p
, 0xF3, X86_TWOB
, 0x5E);
750 emit_modrm( p
, dst
, src
);
753 void sse_minps( struct x86_function
*p
,
758 emit_2ub(p
, X86_TWOB
, 0x5D);
759 emit_modrm( p
, dst
, src
);
762 void sse_subps( struct x86_function
*p
,
767 emit_2ub(p
, X86_TWOB
, 0x5C);
768 emit_modrm( p
, dst
, src
);
771 void sse_mulps( struct x86_function
*p
,
776 emit_2ub(p
, X86_TWOB
, 0x59);
777 emit_modrm( p
, dst
, src
);
780 void sse_mulss( struct x86_function
*p
,
785 emit_3ub(p
, 0xF3, X86_TWOB
, 0x59);
786 emit_modrm( p
, dst
, src
);
789 void sse_addps( struct x86_function
*p
,
794 emit_2ub(p
, X86_TWOB
, 0x58);
795 emit_modrm( p
, dst
, src
);
798 void sse_addss( struct x86_function
*p
,
803 emit_3ub(p
, 0xF3, X86_TWOB
, 0x58);
804 emit_modrm( p
, dst
, src
);
807 void sse_andnps( struct x86_function
*p
,
812 emit_2ub(p
, X86_TWOB
, 0x55);
813 emit_modrm( p
, dst
, src
);
816 void sse_andps( struct x86_function
*p
,
821 emit_2ub(p
, X86_TWOB
, 0x54);
822 emit_modrm( p
, dst
, src
);
825 void sse_rsqrtps( struct x86_function
*p
,
830 emit_2ub(p
, X86_TWOB
, 0x52);
831 emit_modrm( p
, dst
, src
);
834 void sse_rsqrtss( struct x86_function
*p
,
839 emit_3ub(p
, 0xF3, X86_TWOB
, 0x52);
840 emit_modrm( p
, dst
, src
);
844 void sse_movhlps( struct x86_function
*p
,
849 assert(dst
.mod
== mod_REG
&& src
.mod
== mod_REG
);
850 emit_2ub(p
, X86_TWOB
, 0x12);
851 emit_modrm( p
, dst
, src
);
854 void sse_movlhps( struct x86_function
*p
,
859 assert(dst
.mod
== mod_REG
&& src
.mod
== mod_REG
);
860 emit_2ub(p
, X86_TWOB
, 0x16);
861 emit_modrm( p
, dst
, src
);
864 void sse_orps( struct x86_function
*p
,
869 emit_2ub(p
, X86_TWOB
, 0x56);
870 emit_modrm( p
, dst
, src
);
873 void sse_xorps( struct x86_function
*p
,
878 emit_2ub(p
, X86_TWOB
, 0x57);
879 emit_modrm( p
, dst
, src
);
882 void sse_cvtps2pi( struct x86_function
*p
,
887 assert(dst
.file
== file_MMX
&&
888 (src
.file
== file_XMM
|| src
.mod
!= mod_REG
));
892 emit_2ub(p
, X86_TWOB
, 0x2d);
893 emit_modrm( p
, dst
, src
);
896 void sse2_cvtdq2ps( struct x86_function
*p
,
901 emit_2ub(p
, X86_TWOB
, 0x5b);
902 emit_modrm( p
, dst
, src
);
906 /* Shufps can also be used to implement a reduced swizzle when dest ==
909 void sse_shufps( struct x86_function
*p
,
914 DUMP_RRI( dst
, src
, shuf
);
915 emit_2ub(p
, X86_TWOB
, 0xC6);
916 emit_modrm(p
, dst
, src
);
920 void sse_unpckhps( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
923 emit_2ub( p
, X86_TWOB
, 0x15 );
924 emit_modrm( p
, dst
, src
);
927 void sse_unpcklps( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
930 emit_2ub( p
, X86_TWOB
, 0x14 );
931 emit_modrm( p
, dst
, src
);
934 void sse_cmpps( struct x86_function
*p
,
939 DUMP_RRI( dst
, src
, cc
);
940 emit_2ub(p
, X86_TWOB
, 0xC2);
941 emit_modrm(p
, dst
, src
);
945 void sse_pmovmskb( struct x86_function
*p
,
950 emit_3ub(p
, 0x66, X86_TWOB
, 0xD7);
951 emit_modrm(p
, dst
, src
);
954 /***********************************************************************
959 * Perform a reduced swizzle:
961 void sse2_pshufd( struct x86_function
*p
,
966 DUMP_RRI( dst
, src
, shuf
);
967 emit_3ub(p
, 0x66, X86_TWOB
, 0x70);
968 emit_modrm(p
, dst
, src
);
972 void sse2_cvttps2dq( struct x86_function
*p
,
977 emit_3ub( p
, 0xF3, X86_TWOB
, 0x5B );
978 emit_modrm( p
, dst
, src
);
981 void sse2_cvtps2dq( struct x86_function
*p
,
986 emit_3ub(p
, 0x66, X86_TWOB
, 0x5B);
987 emit_modrm( p
, dst
, src
);
990 void sse2_packssdw( struct x86_function
*p
,
995 emit_3ub(p
, 0x66, X86_TWOB
, 0x6B);
996 emit_modrm( p
, dst
, src
);
999 void sse2_packsswb( struct x86_function
*p
,
1001 struct x86_reg src
)
1003 DUMP_RR( dst
, src
);
1004 emit_3ub(p
, 0x66, X86_TWOB
, 0x63);
1005 emit_modrm( p
, dst
, src
);
1008 void sse2_packuswb( struct x86_function
*p
,
1010 struct x86_reg src
)
1012 DUMP_RR( dst
, src
);
1013 emit_3ub(p
, 0x66, X86_TWOB
, 0x67);
1014 emit_modrm( p
, dst
, src
);
1017 void sse2_punpcklbw( struct x86_function
*p
,
1019 struct x86_reg src
)
1021 DUMP_RR( dst
, src
);
1022 emit_3ub(p
, 0x66, X86_TWOB
, 0x60);
1023 emit_modrm( p
, dst
, src
);
1027 void sse2_rcpps( struct x86_function
*p
,
1029 struct x86_reg src
)
1031 DUMP_RR( dst
, src
);
1032 emit_2ub(p
, X86_TWOB
, 0x53);
1033 emit_modrm( p
, dst
, src
);
1036 void sse2_rcpss( struct x86_function
*p
,
1038 struct x86_reg src
)
1040 DUMP_RR( dst
, src
);
1041 emit_3ub(p
, 0xF3, X86_TWOB
, 0x53);
1042 emit_modrm( p
, dst
, src
);
1045 void sse2_movd( struct x86_function
*p
,
1047 struct x86_reg src
)
1049 DUMP_RR( dst
, src
);
1050 emit_2ub(p
, 0x66, X86_TWOB
);
1051 emit_op_modrm( p
, 0x6e, 0x7e, dst
, src
);
1057 /***********************************************************************
1060 static void note_x87_pop( struct x86_function
*p
)
1063 assert(p
->x87_stack
>= 0);
1066 static void note_x87_push( struct x86_function
*p
)
1069 assert(p
->x87_stack
<= 7);
1072 void x87_assert_stack_empty( struct x86_function
*p
)
1074 assert (p
->x87_stack
== 0);
1078 void x87_fist( struct x86_function
*p
, struct x86_reg dst
)
1082 emit_modrm_noreg(p
, 2, dst
);
1085 void x87_fistp( struct x86_function
*p
, struct x86_reg dst
)
1089 emit_modrm_noreg(p
, 3, dst
);
1093 void x87_fild( struct x86_function
*p
, struct x86_reg arg
)
1097 emit_modrm_noreg(p
, 0, arg
);
1101 void x87_fldz( struct x86_function
*p
)
1104 emit_2ub(p
, 0xd9, 0xee);
1109 void x87_fldcw( struct x86_function
*p
, struct x86_reg arg
)
1112 assert(arg
.file
== file_REG32
);
1113 assert(arg
.mod
!= mod_REG
);
1115 emit_modrm_noreg(p
, 5, arg
);
1118 void x87_fld1( struct x86_function
*p
)
1121 emit_2ub(p
, 0xd9, 0xe8);
1125 void x87_fldl2e( struct x86_function
*p
)
1128 emit_2ub(p
, 0xd9, 0xea);
1132 void x87_fldln2( struct x86_function
*p
)
1135 emit_2ub(p
, 0xd9, 0xed);
1139 void x87_fwait( struct x86_function
*p
)
1145 void x87_fnclex( struct x86_function
*p
)
1148 emit_2ub(p
, 0xdb, 0xe2);
1151 void x87_fclex( struct x86_function
*p
)
1157 void x87_fcmovb( struct x86_function
*p
, struct x86_reg arg
)
1160 assert(arg
.file
== file_x87
);
1161 emit_2ub(p
, 0xda, 0xc0+arg
.idx
);
1164 void x87_fcmove( struct x86_function
*p
, struct x86_reg arg
)
1167 assert(arg
.file
== file_x87
);
1168 emit_2ub(p
, 0xda, 0xc8+arg
.idx
);
1171 void x87_fcmovbe( struct x86_function
*p
, struct x86_reg arg
)
1174 assert(arg
.file
== file_x87
);
1175 emit_2ub(p
, 0xda, 0xd0+arg
.idx
);
1178 void x87_fcmovnb( struct x86_function
*p
, struct x86_reg arg
)
1181 assert(arg
.file
== file_x87
);
1182 emit_2ub(p
, 0xdb, 0xc0+arg
.idx
);
1185 void x87_fcmovne( struct x86_function
*p
, struct x86_reg arg
)
1188 assert(arg
.file
== file_x87
);
1189 emit_2ub(p
, 0xdb, 0xc8+arg
.idx
);
1192 void x87_fcmovnbe( struct x86_function
*p
, struct x86_reg arg
)
1195 assert(arg
.file
== file_x87
);
1196 emit_2ub(p
, 0xdb, 0xd0+arg
.idx
);
1201 static void x87_arith_op( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg arg
,
1202 unsigned char dst0ub0
,
1203 unsigned char dst0ub1
,
1204 unsigned char arg0ub0
,
1205 unsigned char arg0ub1
,
1206 unsigned char argmem_noreg
)
1208 assert(dst
.file
== file_x87
);
1210 if (arg
.file
== file_x87
) {
1212 emit_2ub(p
, dst0ub0
, dst0ub1
+arg
.idx
);
1213 else if (arg
.idx
== 0)
1214 emit_2ub(p
, arg0ub0
, arg0ub1
+arg
.idx
);
1218 else if (dst
.idx
== 0) {
1219 assert(arg
.file
== file_REG32
);
1221 emit_modrm_noreg(p
, argmem_noreg
, arg
);
1227 void x87_fmul( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1229 DUMP_RR( dst
, src
);
1230 x87_arith_op(p
, dst
, src
,
1236 void x87_fsub( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1238 DUMP_RR( dst
, src
);
1239 x87_arith_op(p
, dst
, src
,
1245 void x87_fsubr( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1247 DUMP_RR( dst
, src
);
1248 x87_arith_op(p
, dst
, src
,
1254 void x87_fadd( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1256 DUMP_RR( dst
, src
);
1257 x87_arith_op(p
, dst
, src
,
1263 void x87_fdiv( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1265 DUMP_RR( dst
, src
);
1266 x87_arith_op(p
, dst
, src
,
1272 void x87_fdivr( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1274 DUMP_RR( dst
, src
);
1275 x87_arith_op(p
, dst
, src
,
1281 void x87_fmulp( struct x86_function
*p
, struct x86_reg dst
)
1284 assert(dst
.file
== file_x87
);
1285 assert(dst
.idx
>= 1);
1286 emit_2ub(p
, 0xde, 0xc8+dst
.idx
);
1290 void x87_fsubp( struct x86_function
*p
, struct x86_reg dst
)
1293 assert(dst
.file
== file_x87
);
1294 assert(dst
.idx
>= 1);
1295 emit_2ub(p
, 0xde, 0xe8+dst
.idx
);
1299 void x87_fsubrp( struct x86_function
*p
, struct x86_reg dst
)
1302 assert(dst
.file
== file_x87
);
1303 assert(dst
.idx
>= 1);
1304 emit_2ub(p
, 0xde, 0xe0+dst
.idx
);
1308 void x87_faddp( struct x86_function
*p
, struct x86_reg dst
)
1311 assert(dst
.file
== file_x87
);
1312 assert(dst
.idx
>= 1);
1313 emit_2ub(p
, 0xde, 0xc0+dst
.idx
);
1317 void x87_fdivp( struct x86_function
*p
, struct x86_reg dst
)
1320 assert(dst
.file
== file_x87
);
1321 assert(dst
.idx
>= 1);
1322 emit_2ub(p
, 0xde, 0xf8+dst
.idx
);
1326 void x87_fdivrp( struct x86_function
*p
, struct x86_reg dst
)
1329 assert(dst
.file
== file_x87
);
1330 assert(dst
.idx
>= 1);
1331 emit_2ub(p
, 0xde, 0xf0+dst
.idx
);
1335 void x87_ftst( struct x86_function
*p
)
1338 emit_2ub(p
, 0xd9, 0xe4);
1341 void x87_fucom( struct x86_function
*p
, struct x86_reg arg
)
1344 assert(arg
.file
== file_x87
);
1345 emit_2ub(p
, 0xdd, 0xe0+arg
.idx
);
1348 void x87_fucomp( struct x86_function
*p
, struct x86_reg arg
)
1351 assert(arg
.file
== file_x87
);
1352 emit_2ub(p
, 0xdd, 0xe8+arg
.idx
);
1356 void x87_fucompp( struct x86_function
*p
)
1359 emit_2ub(p
, 0xda, 0xe9);
1360 note_x87_pop(p
); /* pop twice */
1361 note_x87_pop(p
); /* pop twice */
1364 void x87_fxch( struct x86_function
*p
, struct x86_reg arg
)
1367 assert(arg
.file
== file_x87
);
1368 emit_2ub(p
, 0xd9, 0xc8+arg
.idx
);
1371 void x87_fabs( struct x86_function
*p
)
1374 emit_2ub(p
, 0xd9, 0xe1);
1377 void x87_fchs( struct x86_function
*p
)
1380 emit_2ub(p
, 0xd9, 0xe0);
1383 void x87_fcos( struct x86_function
*p
)
1386 emit_2ub(p
, 0xd9, 0xff);
1390 void x87_fprndint( struct x86_function
*p
)
1393 emit_2ub(p
, 0xd9, 0xfc);
1396 void x87_fscale( struct x86_function
*p
)
1399 emit_2ub(p
, 0xd9, 0xfd);
1402 void x87_fsin( struct x86_function
*p
)
1405 emit_2ub(p
, 0xd9, 0xfe);
1408 void x87_fsincos( struct x86_function
*p
)
1411 emit_2ub(p
, 0xd9, 0xfb);
1414 void x87_fsqrt( struct x86_function
*p
)
1417 emit_2ub(p
, 0xd9, 0xfa);
1420 void x87_fxtract( struct x86_function
*p
)
1423 emit_2ub(p
, 0xd9, 0xf4);
1428 * Restrictions: -1.0 <= st0 <= 1.0
1430 void x87_f2xm1( struct x86_function
*p
)
1433 emit_2ub(p
, 0xd9, 0xf0);
1436 /* st1 = st1 * log2(st0);
1439 void x87_fyl2x( struct x86_function
*p
)
1442 emit_2ub(p
, 0xd9, 0xf1);
1446 /* st1 = st1 * log2(st0 + 1.0);
1449 * A fast operation, with restrictions: -.29 < st0 < .29
1451 void x87_fyl2xp1( struct x86_function
*p
)
1454 emit_2ub(p
, 0xd9, 0xf9);
1459 void x87_fld( struct x86_function
*p
, struct x86_reg arg
)
1462 if (arg
.file
== file_x87
)
1463 emit_2ub(p
, 0xd9, 0xc0 + arg
.idx
);
1466 emit_modrm_noreg(p
, 0, arg
);
1471 void x87_fst( struct x86_function
*p
, struct x86_reg dst
)
1474 if (dst
.file
== file_x87
)
1475 emit_2ub(p
, 0xdd, 0xd0 + dst
.idx
);
1478 emit_modrm_noreg(p
, 2, dst
);
1482 void x87_fstp( struct x86_function
*p
, struct x86_reg dst
)
1485 if (dst
.file
== file_x87
)
1486 emit_2ub(p
, 0xdd, 0xd8 + dst
.idx
);
1489 emit_modrm_noreg(p
, 3, dst
);
1494 void x87_fpop( struct x86_function
*p
)
1496 x87_fstp( p
, x86_make_reg( file_x87
, 0 ));
1500 void x87_fcom( struct x86_function
*p
, struct x86_reg dst
)
1503 if (dst
.file
== file_x87
)
1504 emit_2ub(p
, 0xd8, 0xd0 + dst
.idx
);
1507 emit_modrm_noreg(p
, 2, dst
);
1512 void x87_fcomp( struct x86_function
*p
, struct x86_reg dst
)
1515 if (dst
.file
== file_x87
)
1516 emit_2ub(p
, 0xd8, 0xd8 + dst
.idx
);
1519 emit_modrm_noreg(p
, 3, dst
);
1524 void x87_fcomi( struct x86_function
*p
, struct x86_reg arg
)
1527 emit_2ub(p
, 0xdb, 0xf0+arg
.idx
);
1530 void x87_fcomip( struct x86_function
*p
, struct x86_reg arg
)
1533 emit_2ub(p
, 0xdb, 0xf0+arg
.idx
);
1538 void x87_fnstsw( struct x86_function
*p
, struct x86_reg dst
)
1541 assert(dst
.file
== file_REG32
);
1543 if (dst
.idx
== reg_AX
&&
1545 emit_2ub(p
, 0xdf, 0xe0);
1548 emit_modrm_noreg(p
, 7, dst
);
1553 void x87_fnstcw( struct x86_function
*p
, struct x86_reg dst
)
1556 assert(dst
.file
== file_REG32
);
1558 emit_1ub(p
, 0x9b); /* WAIT -- needed? */
1560 emit_modrm_noreg(p
, 7, dst
);
1566 /***********************************************************************
1570 void mmx_emms( struct x86_function
*p
)
1573 assert(p
->need_emms
);
1574 emit_2ub(p
, 0x0f, 0x77);
1578 void mmx_packssdw( struct x86_function
*p
,
1580 struct x86_reg src
)
1582 DUMP_RR( dst
, src
);
1583 assert(dst
.file
== file_MMX
&&
1584 (src
.file
== file_MMX
|| src
.mod
!= mod_REG
));
1588 emit_2ub(p
, X86_TWOB
, 0x6b);
1589 emit_modrm( p
, dst
, src
);
1592 void mmx_packuswb( struct x86_function
*p
,
1594 struct x86_reg src
)
1596 DUMP_RR( dst
, src
);
1597 assert(dst
.file
== file_MMX
&&
1598 (src
.file
== file_MMX
|| src
.mod
!= mod_REG
));
1602 emit_2ub(p
, X86_TWOB
, 0x67);
1603 emit_modrm( p
, dst
, src
);
1606 void mmx_movd( struct x86_function
*p
,
1608 struct x86_reg src
)
1610 DUMP_RR( dst
, src
);
1612 emit_1ub(p
, X86_TWOB
);
1613 emit_op_modrm( p
, 0x6e, 0x7e, dst
, src
);
1616 void mmx_movq( struct x86_function
*p
,
1618 struct x86_reg src
)
1620 DUMP_RR( dst
, src
);
1622 emit_1ub(p
, X86_TWOB
);
1623 emit_op_modrm( p
, 0x6f, 0x7f, dst
, src
);
1627 /***********************************************************************
1632 void x86_cdecl_caller_push_regs( struct x86_function
*p
)
1634 x86_push(p
, x86_make_reg(file_REG32
, reg_AX
));
1635 x86_push(p
, x86_make_reg(file_REG32
, reg_CX
));
1636 x86_push(p
, x86_make_reg(file_REG32
, reg_DX
));
1639 void x86_cdecl_caller_pop_regs( struct x86_function
*p
)
1641 x86_pop(p
, x86_make_reg(file_REG32
, reg_DX
));
1642 x86_pop(p
, x86_make_reg(file_REG32
, reg_CX
));
1643 x86_pop(p
, x86_make_reg(file_REG32
, reg_AX
));
1647 /* Retreive a reference to one of the function arguments, taking into
1648 * account any push/pop activity:
1650 struct x86_reg
x86_fn_arg( struct x86_function
*p
,
1653 return x86_make_disp(x86_make_reg(file_REG32
, reg_SP
),
1654 p
->stack_offset
+ arg
* 4); /* ??? */
1658 void x86_init_func( struct x86_function
*p
)
1666 void x86_init_func_size( struct x86_function
*p
, unsigned code_size
)
1668 p
->size
= code_size
;
1669 p
->store
= rtasm_exec_malloc(code_size
);
1670 if (p
->store
== NULL
) {
1671 p
->store
= p
->error_overflow
;
1677 void x86_release_func( struct x86_function
*p
)
1679 if (p
->store
&& p
->store
!= p
->error_overflow
)
1680 rtasm_exec_free(p
->store
);
1688 void (*x86_get_func( struct x86_function
*p
))(void)
1691 if (DISASSEM
&& p
->store
)
1692 debug_printf("disassemble %p %p\n", p
->store
, p
->csr
);
1694 if (p
->store
== p
->error_overflow
)
1695 return (void (*)(void)) NULL
;
1697 return (void (*)(void)) p
->store
;
1702 void x86sse_dummy( void )