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 "util/u_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);*/
375 if (p
->csr
- p
->store
<= -offset
) {
376 /* probably out of memory (using the error_overflow buffer) */
381 if (offset
<= 127 && offset
>= -128) {
382 emit_1ub(p
, 0x70 + cc
);
383 emit_1b(p
, (char) offset
);
386 offset
= label
- (x86_get_label(p
) + 6);
387 emit_2ub(p
, 0x0f, 0x80 + cc
);
392 /* Always use a 32bit offset for forward jumps:
394 int x86_jcc_forward( struct x86_function
*p
,
398 emit_2ub(p
, 0x0f, 0x80 + cc
);
400 return x86_get_label(p
);
403 int x86_jmp_forward( struct x86_function
*p
)
408 return x86_get_label(p
);
411 int x86_call_forward( struct x86_function
*p
)
417 return x86_get_label(p
);
420 /* Fixup offset from forward jump:
422 void x86_fixup_fwd_jump( struct x86_function
*p
,
425 *(int *)(p
->store
+ fixup
- 4) = x86_get_label(p
) - fixup
;
428 void x86_jmp( struct x86_function
*p
, int label
)
432 emit_1i(p
, label
- x86_get_label(p
) - 4);
435 void x86_call( struct x86_function
*p
, struct x86_reg reg
)
439 emit_modrm_noreg(p
, 2, reg
);
443 void x86_mov_reg_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
446 assert(dst
.file
== file_REG32
);
447 assert(dst
.mod
== mod_REG
);
448 emit_1ub(p
, 0xb8 + dst
.idx
);
453 * Immediate group 1 instructions.
456 x86_group1_imm( struct x86_function
*p
,
457 unsigned op
, struct x86_reg dst
, int imm
)
459 assert(dst
.file
== file_REG32
);
460 assert(dst
.mod
== mod_REG
);
461 if(-0x80 <= imm
&& imm
< 0x80) {
463 emit_modrm_noreg(p
, op
, dst
);
464 emit_1b(p
, (char)imm
);
468 emit_modrm_noreg(p
, op
, dst
);
473 void x86_add_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
476 x86_group1_imm(p
, 0, dst
, imm
);
479 void x86_or_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
482 x86_group1_imm(p
, 1, dst
, imm
);
485 void x86_and_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
488 x86_group1_imm(p
, 4, dst
, imm
);
491 void x86_sub_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
494 x86_group1_imm(p
, 5, dst
, imm
);
497 void x86_xor_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
500 x86_group1_imm(p
, 6, dst
, imm
);
503 void x86_cmp_imm( struct x86_function
*p
, struct x86_reg dst
, int imm
)
506 x86_group1_imm(p
, 7, dst
, imm
);
510 void x86_push( struct x86_function
*p
,
514 if (reg
.mod
== mod_REG
)
515 emit_1ub(p
, 0x50 + reg
.idx
);
519 emit_modrm_noreg(p
, 6, reg
);
523 p
->stack_offset
+= 4;
526 void x86_push_imm32( struct x86_function
*p
,
533 p
->stack_offset
+= 4;
537 void x86_pop( struct x86_function
*p
,
541 assert(reg
.mod
== mod_REG
);
542 emit_1ub(p
, 0x58 + reg
.idx
);
543 p
->stack_offset
-= 4;
546 void x86_inc( struct x86_function
*p
,
550 assert(reg
.mod
== mod_REG
);
551 emit_1ub(p
, 0x40 + reg
.idx
);
554 void x86_dec( struct x86_function
*p
,
558 assert(reg
.mod
== mod_REG
);
559 emit_1ub(p
, 0x48 + reg
.idx
);
562 void x86_ret( struct x86_function
*p
)
565 assert(p
->stack_offset
== 0);
569 void x86_retw( struct x86_function
*p
, unsigned short imm
)
572 emit_3ub(p
, 0xc2, imm
& 0xff, (imm
>> 8) & 0xff);
575 void x86_sahf( struct x86_function
*p
)
581 void x86_mov( struct x86_function
*p
,
586 emit_op_modrm( p
, 0x8b, 0x89, dst
, src
);
589 void x86_xor( struct x86_function
*p
,
594 emit_op_modrm( p
, 0x33, 0x31, dst
, src
);
597 void x86_cmp( struct x86_function
*p
,
602 emit_op_modrm( p
, 0x3b, 0x39, dst
, src
);
605 void x86_lea( struct x86_function
*p
,
611 emit_modrm( p
, dst
, src
);
614 void x86_test( struct x86_function
*p
,
620 emit_modrm( p
, dst
, src
);
623 void x86_add( struct x86_function
*p
,
628 emit_op_modrm(p
, 0x03, 0x01, dst
, src
);
631 /* Calculate EAX * src, results in EDX:EAX.
633 void x86_mul( struct x86_function
*p
,
638 emit_modrm_noreg(p
, 4, src
);
642 void x86_imul( struct x86_function
*p
,
647 emit_2ub(p
, X86_TWOB
, 0xAF);
648 emit_modrm(p
, dst
, src
);
652 void x86_sub( struct x86_function
*p
,
657 emit_op_modrm(p
, 0x2b, 0x29, dst
, src
);
660 void x86_or( struct x86_function
*p
,
665 emit_op_modrm( p
, 0x0b, 0x09, dst
, src
);
668 void x86_and( struct x86_function
*p
,
673 emit_op_modrm( p
, 0x23, 0x21, dst
, src
);
678 /***********************************************************************
682 void sse_prefetchnta( struct x86_function
*p
, struct x86_reg ptr
)
685 assert(ptr
.mod
!= mod_REG
);
686 emit_2ub(p
, 0x0f, 0x18);
687 emit_modrm_noreg(p
, 0, ptr
);
690 void sse_prefetch0( struct x86_function
*p
, struct x86_reg ptr
)
693 assert(ptr
.mod
!= mod_REG
);
694 emit_2ub(p
, 0x0f, 0x18);
695 emit_modrm_noreg(p
, 1, ptr
);
698 void sse_prefetch1( struct x86_function
*p
, struct x86_reg ptr
)
701 assert(ptr
.mod
!= mod_REG
);
702 emit_2ub(p
, 0x0f, 0x18);
703 emit_modrm_noreg(p
, 2, ptr
);
706 void sse_movntps( struct x86_function
*p
,
712 assert(dst
.mod
!= mod_REG
);
713 assert(src
.mod
== mod_REG
);
714 emit_2ub(p
, 0x0f, 0x2b);
715 emit_modrm(p
, src
, dst
);
721 void sse_movss( struct x86_function
*p
,
726 emit_2ub(p
, 0xF3, X86_TWOB
);
727 emit_op_modrm( p
, 0x10, 0x11, dst
, src
);
730 void sse_movaps( struct x86_function
*p
,
735 emit_1ub(p
, X86_TWOB
);
736 emit_op_modrm( p
, 0x28, 0x29, dst
, src
);
739 void sse_movups( struct x86_function
*p
,
744 emit_1ub(p
, X86_TWOB
);
745 emit_op_modrm( p
, 0x10, 0x11, dst
, src
);
748 void sse_movhps( struct x86_function
*p
,
753 assert(dst
.mod
!= mod_REG
|| src
.mod
!= mod_REG
);
754 emit_1ub(p
, X86_TWOB
);
755 emit_op_modrm( p
, 0x16, 0x17, dst
, src
); /* cf movlhps */
758 void sse_movlps( struct x86_function
*p
,
763 assert(dst
.mod
!= mod_REG
|| src
.mod
!= mod_REG
);
764 emit_1ub(p
, X86_TWOB
);
765 emit_op_modrm( p
, 0x12, 0x13, dst
, src
); /* cf movhlps */
768 void sse_maxps( struct x86_function
*p
,
773 emit_2ub(p
, X86_TWOB
, 0x5F);
774 emit_modrm( p
, dst
, src
);
777 void sse_maxss( struct x86_function
*p
,
782 emit_3ub(p
, 0xF3, X86_TWOB
, 0x5F);
783 emit_modrm( p
, dst
, src
);
786 void sse_divss( struct x86_function
*p
,
791 emit_3ub(p
, 0xF3, X86_TWOB
, 0x5E);
792 emit_modrm( p
, dst
, src
);
795 void sse_minps( struct x86_function
*p
,
800 emit_2ub(p
, X86_TWOB
, 0x5D);
801 emit_modrm( p
, dst
, src
);
804 void sse_subps( struct x86_function
*p
,
809 emit_2ub(p
, X86_TWOB
, 0x5C);
810 emit_modrm( p
, dst
, src
);
813 void sse_mulps( struct x86_function
*p
,
818 emit_2ub(p
, X86_TWOB
, 0x59);
819 emit_modrm( p
, dst
, src
);
822 void sse_mulss( struct x86_function
*p
,
827 emit_3ub(p
, 0xF3, X86_TWOB
, 0x59);
828 emit_modrm( p
, dst
, src
);
831 void sse_addps( struct x86_function
*p
,
836 emit_2ub(p
, X86_TWOB
, 0x58);
837 emit_modrm( p
, dst
, src
);
840 void sse_addss( struct x86_function
*p
,
845 emit_3ub(p
, 0xF3, X86_TWOB
, 0x58);
846 emit_modrm( p
, dst
, src
);
849 void sse_andnps( struct x86_function
*p
,
854 emit_2ub(p
, X86_TWOB
, 0x55);
855 emit_modrm( p
, dst
, src
);
858 void sse_andps( struct x86_function
*p
,
863 emit_2ub(p
, X86_TWOB
, 0x54);
864 emit_modrm( p
, dst
, src
);
867 void sse_rsqrtps( struct x86_function
*p
,
872 emit_2ub(p
, X86_TWOB
, 0x52);
873 emit_modrm( p
, dst
, src
);
876 void sse_rsqrtss( struct x86_function
*p
,
881 emit_3ub(p
, 0xF3, X86_TWOB
, 0x52);
882 emit_modrm( p
, dst
, src
);
886 void sse_movhlps( struct x86_function
*p
,
891 assert(dst
.mod
== mod_REG
&& src
.mod
== mod_REG
);
892 emit_2ub(p
, X86_TWOB
, 0x12);
893 emit_modrm( p
, dst
, src
);
896 void sse_movlhps( struct x86_function
*p
,
901 assert(dst
.mod
== mod_REG
&& src
.mod
== mod_REG
);
902 emit_2ub(p
, X86_TWOB
, 0x16);
903 emit_modrm( p
, dst
, src
);
906 void sse_orps( struct x86_function
*p
,
911 emit_2ub(p
, X86_TWOB
, 0x56);
912 emit_modrm( p
, dst
, src
);
915 void sse_xorps( struct x86_function
*p
,
920 emit_2ub(p
, X86_TWOB
, 0x57);
921 emit_modrm( p
, dst
, src
);
924 void sse_cvtps2pi( struct x86_function
*p
,
929 assert(dst
.file
== file_MMX
&&
930 (src
.file
== file_XMM
|| src
.mod
!= mod_REG
));
934 emit_2ub(p
, X86_TWOB
, 0x2d);
935 emit_modrm( p
, dst
, src
);
938 void sse2_cvtdq2ps( struct x86_function
*p
,
943 emit_2ub(p
, X86_TWOB
, 0x5b);
944 emit_modrm( p
, dst
, src
);
948 /* Shufps can also be used to implement a reduced swizzle when dest ==
951 void sse_shufps( struct x86_function
*p
,
956 DUMP_RRI( dst
, src
, shuf
);
957 emit_2ub(p
, X86_TWOB
, 0xC6);
958 emit_modrm(p
, dst
, src
);
962 void sse_unpckhps( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
965 emit_2ub( p
, X86_TWOB
, 0x15 );
966 emit_modrm( p
, dst
, src
);
969 void sse_unpcklps( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
972 emit_2ub( p
, X86_TWOB
, 0x14 );
973 emit_modrm( p
, dst
, src
);
976 void sse_cmpps( struct x86_function
*p
,
981 DUMP_RRI( dst
, src
, cc
);
982 emit_2ub(p
, X86_TWOB
, 0xC2);
983 emit_modrm(p
, dst
, src
);
987 void sse_pmovmskb( struct x86_function
*p
,
992 emit_3ub(p
, 0x66, X86_TWOB
, 0xD7);
993 emit_modrm(p
, dst
, src
);
996 void sse_movmskps( struct x86_function
*p
,
1000 DUMP_RR( dst
, src
);
1001 emit_2ub(p
, X86_TWOB
, 0x50);
1002 emit_modrm(p
, dst
, src
);
1005 /***********************************************************************
1010 * Perform a reduced swizzle:
1012 void sse2_pshufd( struct x86_function
*p
,
1017 DUMP_RRI( dst
, src
, shuf
);
1018 emit_3ub(p
, 0x66, X86_TWOB
, 0x70);
1019 emit_modrm(p
, dst
, src
);
1023 void sse2_cvttps2dq( struct x86_function
*p
,
1025 struct x86_reg src
)
1027 DUMP_RR( dst
, src
);
1028 emit_3ub( p
, 0xF3, X86_TWOB
, 0x5B );
1029 emit_modrm( p
, dst
, src
);
1032 void sse2_cvtps2dq( struct x86_function
*p
,
1034 struct x86_reg src
)
1036 DUMP_RR( dst
, src
);
1037 emit_3ub(p
, 0x66, X86_TWOB
, 0x5B);
1038 emit_modrm( p
, dst
, src
);
1041 void sse2_packssdw( struct x86_function
*p
,
1043 struct x86_reg src
)
1045 DUMP_RR( dst
, src
);
1046 emit_3ub(p
, 0x66, X86_TWOB
, 0x6B);
1047 emit_modrm( p
, dst
, src
);
1050 void sse2_packsswb( struct x86_function
*p
,
1052 struct x86_reg src
)
1054 DUMP_RR( dst
, src
);
1055 emit_3ub(p
, 0x66, X86_TWOB
, 0x63);
1056 emit_modrm( p
, dst
, src
);
1059 void sse2_packuswb( struct x86_function
*p
,
1061 struct x86_reg src
)
1063 DUMP_RR( dst
, src
);
1064 emit_3ub(p
, 0x66, X86_TWOB
, 0x67);
1065 emit_modrm( p
, dst
, src
);
1068 void sse2_punpcklbw( struct x86_function
*p
,
1070 struct x86_reg src
)
1072 DUMP_RR( dst
, src
);
1073 emit_3ub(p
, 0x66, X86_TWOB
, 0x60);
1074 emit_modrm( p
, dst
, src
);
1078 void sse2_rcpps( struct x86_function
*p
,
1080 struct x86_reg src
)
1082 DUMP_RR( dst
, src
);
1083 emit_2ub(p
, X86_TWOB
, 0x53);
1084 emit_modrm( p
, dst
, src
);
1087 void sse2_rcpss( struct x86_function
*p
,
1089 struct x86_reg src
)
1091 DUMP_RR( dst
, src
);
1092 emit_3ub(p
, 0xF3, X86_TWOB
, 0x53);
1093 emit_modrm( p
, dst
, src
);
1096 void sse2_movd( struct x86_function
*p
,
1098 struct x86_reg src
)
1100 DUMP_RR( dst
, src
);
1101 emit_2ub(p
, 0x66, X86_TWOB
);
1102 emit_op_modrm( p
, 0x6e, 0x7e, dst
, src
);
1108 /***********************************************************************
1111 static void note_x87_pop( struct x86_function
*p
)
1114 assert(p
->x87_stack
>= 0);
1117 static void note_x87_push( struct x86_function
*p
)
1120 assert(p
->x87_stack
<= 7);
1123 void x87_assert_stack_empty( struct x86_function
*p
)
1125 assert (p
->x87_stack
== 0);
1129 void x87_fist( struct x86_function
*p
, struct x86_reg dst
)
1133 emit_modrm_noreg(p
, 2, dst
);
1136 void x87_fistp( struct x86_function
*p
, struct x86_reg dst
)
1140 emit_modrm_noreg(p
, 3, dst
);
1144 void x87_fild( struct x86_function
*p
, struct x86_reg arg
)
1148 emit_modrm_noreg(p
, 0, arg
);
1152 void x87_fldz( struct x86_function
*p
)
1155 emit_2ub(p
, 0xd9, 0xee);
1160 void x87_fldcw( struct x86_function
*p
, struct x86_reg arg
)
1163 assert(arg
.file
== file_REG32
);
1164 assert(arg
.mod
!= mod_REG
);
1166 emit_modrm_noreg(p
, 5, arg
);
1169 void x87_fld1( struct x86_function
*p
)
1172 emit_2ub(p
, 0xd9, 0xe8);
1176 void x87_fldl2e( struct x86_function
*p
)
1179 emit_2ub(p
, 0xd9, 0xea);
1183 void x87_fldln2( struct x86_function
*p
)
1186 emit_2ub(p
, 0xd9, 0xed);
1190 void x87_fwait( struct x86_function
*p
)
1196 void x87_fnclex( struct x86_function
*p
)
1199 emit_2ub(p
, 0xdb, 0xe2);
1202 void x87_fclex( struct x86_function
*p
)
1208 void x87_fcmovb( struct x86_function
*p
, struct x86_reg arg
)
1211 assert(arg
.file
== file_x87
);
1212 emit_2ub(p
, 0xda, 0xc0+arg
.idx
);
1215 void x87_fcmove( struct x86_function
*p
, struct x86_reg arg
)
1218 assert(arg
.file
== file_x87
);
1219 emit_2ub(p
, 0xda, 0xc8+arg
.idx
);
1222 void x87_fcmovbe( struct x86_function
*p
, struct x86_reg arg
)
1225 assert(arg
.file
== file_x87
);
1226 emit_2ub(p
, 0xda, 0xd0+arg
.idx
);
1229 void x87_fcmovnb( struct x86_function
*p
, struct x86_reg arg
)
1232 assert(arg
.file
== file_x87
);
1233 emit_2ub(p
, 0xdb, 0xc0+arg
.idx
);
1236 void x87_fcmovne( struct x86_function
*p
, struct x86_reg arg
)
1239 assert(arg
.file
== file_x87
);
1240 emit_2ub(p
, 0xdb, 0xc8+arg
.idx
);
1243 void x87_fcmovnbe( struct x86_function
*p
, struct x86_reg arg
)
1246 assert(arg
.file
== file_x87
);
1247 emit_2ub(p
, 0xdb, 0xd0+arg
.idx
);
1252 static void x87_arith_op( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg arg
,
1253 unsigned char dst0ub0
,
1254 unsigned char dst0ub1
,
1255 unsigned char arg0ub0
,
1256 unsigned char arg0ub1
,
1257 unsigned char argmem_noreg
)
1259 assert(dst
.file
== file_x87
);
1261 if (arg
.file
== file_x87
) {
1263 emit_2ub(p
, dst0ub0
, dst0ub1
+arg
.idx
);
1264 else if (arg
.idx
== 0)
1265 emit_2ub(p
, arg0ub0
, arg0ub1
+arg
.idx
);
1269 else if (dst
.idx
== 0) {
1270 assert(arg
.file
== file_REG32
);
1272 emit_modrm_noreg(p
, argmem_noreg
, arg
);
1278 void x87_fmul( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1280 DUMP_RR( dst
, src
);
1281 x87_arith_op(p
, dst
, src
,
1287 void x87_fsub( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1289 DUMP_RR( dst
, src
);
1290 x87_arith_op(p
, dst
, src
,
1296 void x87_fsubr( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1298 DUMP_RR( dst
, src
);
1299 x87_arith_op(p
, dst
, src
,
1305 void x87_fadd( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1307 DUMP_RR( dst
, src
);
1308 x87_arith_op(p
, dst
, src
,
1314 void x87_fdiv( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1316 DUMP_RR( dst
, src
);
1317 x87_arith_op(p
, dst
, src
,
1323 void x87_fdivr( struct x86_function
*p
, struct x86_reg dst
, struct x86_reg src
)
1325 DUMP_RR( dst
, src
);
1326 x87_arith_op(p
, dst
, src
,
1332 void x87_fmulp( struct x86_function
*p
, struct x86_reg dst
)
1335 assert(dst
.file
== file_x87
);
1336 assert(dst
.idx
>= 1);
1337 emit_2ub(p
, 0xde, 0xc8+dst
.idx
);
1341 void x87_fsubp( struct x86_function
*p
, struct x86_reg dst
)
1344 assert(dst
.file
== file_x87
);
1345 assert(dst
.idx
>= 1);
1346 emit_2ub(p
, 0xde, 0xe8+dst
.idx
);
1350 void x87_fsubrp( struct x86_function
*p
, struct x86_reg dst
)
1353 assert(dst
.file
== file_x87
);
1354 assert(dst
.idx
>= 1);
1355 emit_2ub(p
, 0xde, 0xe0+dst
.idx
);
1359 void x87_faddp( struct x86_function
*p
, struct x86_reg dst
)
1362 assert(dst
.file
== file_x87
);
1363 assert(dst
.idx
>= 1);
1364 emit_2ub(p
, 0xde, 0xc0+dst
.idx
);
1368 void x87_fdivp( struct x86_function
*p
, struct x86_reg dst
)
1371 assert(dst
.file
== file_x87
);
1372 assert(dst
.idx
>= 1);
1373 emit_2ub(p
, 0xde, 0xf8+dst
.idx
);
1377 void x87_fdivrp( struct x86_function
*p
, struct x86_reg dst
)
1380 assert(dst
.file
== file_x87
);
1381 assert(dst
.idx
>= 1);
1382 emit_2ub(p
, 0xde, 0xf0+dst
.idx
);
1386 void x87_ftst( struct x86_function
*p
)
1389 emit_2ub(p
, 0xd9, 0xe4);
1392 void x87_fucom( struct x86_function
*p
, struct x86_reg arg
)
1395 assert(arg
.file
== file_x87
);
1396 emit_2ub(p
, 0xdd, 0xe0+arg
.idx
);
1399 void x87_fucomp( struct x86_function
*p
, struct x86_reg arg
)
1402 assert(arg
.file
== file_x87
);
1403 emit_2ub(p
, 0xdd, 0xe8+arg
.idx
);
1407 void x87_fucompp( struct x86_function
*p
)
1410 emit_2ub(p
, 0xda, 0xe9);
1411 note_x87_pop(p
); /* pop twice */
1412 note_x87_pop(p
); /* pop twice */
1415 void x87_fxch( struct x86_function
*p
, struct x86_reg arg
)
1418 assert(arg
.file
== file_x87
);
1419 emit_2ub(p
, 0xd9, 0xc8+arg
.idx
);
1422 void x87_fabs( struct x86_function
*p
)
1425 emit_2ub(p
, 0xd9, 0xe1);
1428 void x87_fchs( struct x86_function
*p
)
1431 emit_2ub(p
, 0xd9, 0xe0);
1434 void x87_fcos( struct x86_function
*p
)
1437 emit_2ub(p
, 0xd9, 0xff);
1441 void x87_fprndint( struct x86_function
*p
)
1444 emit_2ub(p
, 0xd9, 0xfc);
1447 void x87_fscale( struct x86_function
*p
)
1450 emit_2ub(p
, 0xd9, 0xfd);
1453 void x87_fsin( struct x86_function
*p
)
1456 emit_2ub(p
, 0xd9, 0xfe);
1459 void x87_fsincos( struct x86_function
*p
)
1462 emit_2ub(p
, 0xd9, 0xfb);
1465 void x87_fsqrt( struct x86_function
*p
)
1468 emit_2ub(p
, 0xd9, 0xfa);
1471 void x87_fxtract( struct x86_function
*p
)
1474 emit_2ub(p
, 0xd9, 0xf4);
1479 * Restrictions: -1.0 <= st0 <= 1.0
1481 void x87_f2xm1( struct x86_function
*p
)
1484 emit_2ub(p
, 0xd9, 0xf0);
1487 /* st1 = st1 * log2(st0);
1490 void x87_fyl2x( struct x86_function
*p
)
1493 emit_2ub(p
, 0xd9, 0xf1);
1497 /* st1 = st1 * log2(st0 + 1.0);
1500 * A fast operation, with restrictions: -.29 < st0 < .29
1502 void x87_fyl2xp1( struct x86_function
*p
)
1505 emit_2ub(p
, 0xd9, 0xf9);
1510 void x87_fld( struct x86_function
*p
, struct x86_reg arg
)
1513 if (arg
.file
== file_x87
)
1514 emit_2ub(p
, 0xd9, 0xc0 + arg
.idx
);
1517 emit_modrm_noreg(p
, 0, arg
);
1522 void x87_fst( struct x86_function
*p
, struct x86_reg dst
)
1525 if (dst
.file
== file_x87
)
1526 emit_2ub(p
, 0xdd, 0xd0 + dst
.idx
);
1529 emit_modrm_noreg(p
, 2, dst
);
1533 void x87_fstp( struct x86_function
*p
, struct x86_reg dst
)
1536 if (dst
.file
== file_x87
)
1537 emit_2ub(p
, 0xdd, 0xd8 + dst
.idx
);
1540 emit_modrm_noreg(p
, 3, dst
);
1545 void x87_fpop( struct x86_function
*p
)
1547 x87_fstp( p
, x86_make_reg( file_x87
, 0 ));
1551 void x87_fcom( struct x86_function
*p
, struct x86_reg dst
)
1554 if (dst
.file
== file_x87
)
1555 emit_2ub(p
, 0xd8, 0xd0 + dst
.idx
);
1558 emit_modrm_noreg(p
, 2, dst
);
1563 void x87_fcomp( struct x86_function
*p
, struct x86_reg dst
)
1566 if (dst
.file
== file_x87
)
1567 emit_2ub(p
, 0xd8, 0xd8 + dst
.idx
);
1570 emit_modrm_noreg(p
, 3, dst
);
1575 void x87_fcomi( struct x86_function
*p
, struct x86_reg arg
)
1578 emit_2ub(p
, 0xdb, 0xf0+arg
.idx
);
1581 void x87_fcomip( struct x86_function
*p
, struct x86_reg arg
)
1584 emit_2ub(p
, 0xdb, 0xf0+arg
.idx
);
1589 void x87_fnstsw( struct x86_function
*p
, struct x86_reg dst
)
1592 assert(dst
.file
== file_REG32
);
1594 if (dst
.idx
== reg_AX
&&
1596 emit_2ub(p
, 0xdf, 0xe0);
1599 emit_modrm_noreg(p
, 7, dst
);
1604 void x87_fnstcw( struct x86_function
*p
, struct x86_reg dst
)
1607 assert(dst
.file
== file_REG32
);
1609 emit_1ub(p
, 0x9b); /* WAIT -- needed? */
1611 emit_modrm_noreg(p
, 7, dst
);
1617 /***********************************************************************
1621 void mmx_emms( struct x86_function
*p
)
1624 assert(p
->need_emms
);
1625 emit_2ub(p
, 0x0f, 0x77);
1629 void mmx_packssdw( struct x86_function
*p
,
1631 struct x86_reg src
)
1633 DUMP_RR( dst
, src
);
1634 assert(dst
.file
== file_MMX
&&
1635 (src
.file
== file_MMX
|| src
.mod
!= mod_REG
));
1639 emit_2ub(p
, X86_TWOB
, 0x6b);
1640 emit_modrm( p
, dst
, src
);
1643 void mmx_packuswb( struct x86_function
*p
,
1645 struct x86_reg src
)
1647 DUMP_RR( dst
, src
);
1648 assert(dst
.file
== file_MMX
&&
1649 (src
.file
== file_MMX
|| src
.mod
!= mod_REG
));
1653 emit_2ub(p
, X86_TWOB
, 0x67);
1654 emit_modrm( p
, dst
, src
);
1657 void mmx_movd( struct x86_function
*p
,
1659 struct x86_reg src
)
1661 DUMP_RR( dst
, src
);
1663 emit_1ub(p
, X86_TWOB
);
1664 emit_op_modrm( p
, 0x6e, 0x7e, dst
, src
);
1667 void mmx_movq( struct x86_function
*p
,
1669 struct x86_reg src
)
1671 DUMP_RR( dst
, src
);
1673 emit_1ub(p
, X86_TWOB
);
1674 emit_op_modrm( p
, 0x6f, 0x7f, dst
, src
);
1678 /***********************************************************************
1683 void x86_cdecl_caller_push_regs( struct x86_function
*p
)
1685 x86_push(p
, x86_make_reg(file_REG32
, reg_AX
));
1686 x86_push(p
, x86_make_reg(file_REG32
, reg_CX
));
1687 x86_push(p
, x86_make_reg(file_REG32
, reg_DX
));
1690 void x86_cdecl_caller_pop_regs( struct x86_function
*p
)
1692 x86_pop(p
, x86_make_reg(file_REG32
, reg_DX
));
1693 x86_pop(p
, x86_make_reg(file_REG32
, reg_CX
));
1694 x86_pop(p
, x86_make_reg(file_REG32
, reg_AX
));
1698 /* Retreive a reference to one of the function arguments, taking into
1699 * account any push/pop activity:
1701 struct x86_reg
x86_fn_arg( struct x86_function
*p
,
1704 return x86_make_disp(x86_make_reg(file_REG32
, reg_SP
),
1705 p
->stack_offset
+ arg
* 4); /* ??? */
1709 void x86_init_func( struct x86_function
*p
)
1717 void x86_init_func_size( struct x86_function
*p
, unsigned code_size
)
1719 p
->size
= code_size
;
1720 p
->store
= rtasm_exec_malloc(code_size
);
1721 if (p
->store
== NULL
) {
1722 p
->store
= p
->error_overflow
;
1728 void x86_release_func( struct x86_function
*p
)
1730 if (p
->store
&& p
->store
!= p
->error_overflow
)
1731 rtasm_exec_free(p
->store
);
1739 void (*x86_get_func( struct x86_function
*p
))(void)
1742 if (DISASSEM
&& p
->store
)
1743 debug_printf("disassemble %p %p\n", p
->store
, p
->csr
);
1745 if (p
->store
== p
->error_overflow
)
1746 return (void (*)(void)) NULL
;
1748 return (void (*)(void)) p
->store
;
1753 void x86sse_dummy( void )