Adding two more people to the RELICENSING file.
[tinycc.git] / x86_64-gen.c
Commit [+]AuthorDateLineData
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001/*
2 * x86-64 code generator for TCC
3 *
4 * Copyright (c) 2008 Shinichiro Hamaji
5 *
6 * Based on i386-gen.c by Fabrice Bellard
7 *
8 * This library is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2 of the License, or (at your option) any later version.
12 *
13 * This library is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with this library; if not, write to the Free Software
20 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
21 */
22
88a3ccab grischka2009-12-20 01:53:49 +010023#ifdef TARGET_DEFS_ONLY
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090024
25/* number of available registers */
41b3c7a5 James Lyon2013-04-27 20:39:34 +010026#define NB_REGS 25
88a3ccab grischka2009-12-20 01:53:49 +010027#define NB_ASM_REGS 8
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090028
29/* a register can belong to several classes. The classes must be
30 sorted from more general to more precise (see gv2() code which does
31 assumptions on it). */
32#define RC_INT 0x0001 /* generic integer register */
33#define RC_FLOAT 0x0002 /* generic float register */
34#define RC_RAX 0x0004
35#define RC_RCX 0x0008
36#define RC_RDX 0x0010
cbce6d2b James Lyon2013-04-19 22:05:49 +010037#define RC_ST0 0x0080 /* only for long double */
0e5c0ee0 grischka2009-12-19 00:06:41 +010038#define RC_R8 0x0100
39#define RC_R9 0x0200
ae191c3a grischka2012-03-05 20:19:28 +010040#define RC_R10 0x0400
41#define RC_R11 0x0800
cbce6d2b
JL
James Lyon2013-04-19 22:05:49 +010042#define RC_XMM0 0x1000
43#define RC_XMM1 0x2000
44#define RC_XMM2 0x4000
45#define RC_XMM3 0x8000
46#define RC_XMM4 0x10000
47#define RC_XMM5 0x20000
1caee8ab
JL
James Lyon2013-04-25 22:30:53 +010048#define RC_XMM6 0x40000
49#define RC_XMM7 0x80000
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090050#define RC_IRET RC_RAX /* function return: integer register */
51#define RC_LRET RC_RDX /* function return: second integer register */
52#define RC_FRET RC_XMM0 /* function return: float register */
0e17671f James Lyon2013-04-19 15:33:16 +010053#define RC_QRET RC_XMM1 /* function return: second float register */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090054
55/* pretty names for the registers */
56enum {
57 TREG_RAX = 0,
58 TREG_RCX = 1,
59 TREG_RDX = 2,
41b3c7a5 James Lyon2013-04-27 20:39:34 +010060 TREG_RSP = 4,
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090061 TREG_RSI = 6,
62 TREG_RDI = 7,
946afd23 James Lyon2013-04-19 18:31:24 +010063
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090064 TREG_R8 = 8,
65 TREG_R9 = 9,
66 TREG_R10 = 10,
67 TREG_R11 = 11,
68
946afd23
JL
James Lyon2013-04-19 18:31:24 +010069 TREG_XMM0 = 16,
70 TREG_XMM1 = 17,
cbce6d2b
JL
James Lyon2013-04-19 22:05:49 +010071 TREG_XMM2 = 18,
72 TREG_XMM3 = 19,
73 TREG_XMM4 = 20,
74 TREG_XMM5 = 21,
75 TREG_XMM6 = 22,
76 TREG_XMM7 = 23,
946afd23 James Lyon2013-04-19 18:31:24 +010077
41b3c7a5 James Lyon2013-04-27 20:39:34 +010078 TREG_ST0 = 24,
946afd23
JL
James Lyon2013-04-19 18:31:24 +010079
80 TREG_MEM = 0x20,
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090081};
82
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +090083#define REX_BASE(reg) (((reg) >> 3) & 1)
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090084#define REG_VALUE(reg) ((reg) & 7)
85
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090086/* return registers for function */
87#define REG_IRET TREG_RAX /* single word int return register */
88#define REG_LRET TREG_RDX /* second word return register (for long long) */
89#define REG_FRET TREG_XMM0 /* float return register */
0e17671f James Lyon2013-04-19 15:33:16 +010090#define REG_QRET TREG_XMM1 /* second float return register */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +090091
92/* defined if function parameters must be evaluated in reverse order */
93#define INVERT_FUNC_PARAMS
94
95/* pointer size, in bytes */
96#define PTR_SIZE 8
97
98/* long double size and alignment, in bytes */
99#define LDOUBLE_SIZE 16
6ee366e7 James Lyon2013-04-26 16:42:12 +0100100#define LDOUBLE_ALIGN 16
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900101/* maximum alignment (for aligned attribute support) */
6ee366e7 James Lyon2013-04-26 16:42:12 +0100102#define MAX_ALIGN 16
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900103
104/******************************************************/
105/* ELF defines */
106
107#define EM_TCC_TARGET EM_X86_64
108
109/* relocation type for 32 bit data relocation */
5dadff3d
SH
Shinichiro Hamaji2009-07-19 06:42:23 +0900110#define R_DATA_32 R_X86_64_32
111#define R_DATA_PTR R_X86_64_64
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900112#define R_JMP_SLOT R_X86_64_JUMP_SLOT
113#define R_COPY R_X86_64_COPY
114
a913ee60
MM
Michael Matz2014-04-03 17:59:41 +0200115#define ELF_START_ADDR 0x400000
116#define ELF_PAGE_SIZE 0x200000
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900117
118/******************************************************/
88a3ccab grischka2009-12-20 01:53:49 +0100119#else /* ! TARGET_DEFS_ONLY */
120/******************************************************/
89ad24e7 gus knight2015-07-29 16:53:57 -0400121#include "tcc.h"
88a3ccab grischka2009-12-20 01:53:49 +0100122#include <assert.h>
123
946afd23 James Lyon2013-04-19 18:31:24 +0100124ST_DATA const int reg_classes[NB_REGS] = {
88a3ccab grischka2009-12-20 01:53:49 +0100125 /* eax */ RC_INT | RC_RAX,
126 /* ecx */ RC_INT | RC_RCX,
127 /* edx */ RC_INT | RC_RDX,
946afd23 James Lyon2013-04-19 18:31:24 +0100128 0,
41b3c7a5 James Lyon2013-04-27 20:39:34 +0100129 0,
0e5c0ee0 grischka2009-12-19 00:06:41 +0100130 0,
131 0,
132 0,
cbce6d2b
JL
James Lyon2013-04-19 22:05:49 +0100133 RC_R8,
134 RC_R9,
135 RC_R10,
136 RC_R11,
946afd23
JL
James Lyon2013-04-19 18:31:24 +0100137 0,
138 0,
139 0,
140 0,
141 /* xmm0 */ RC_FLOAT | RC_XMM0,
142 /* xmm1 */ RC_FLOAT | RC_XMM1,
cbce6d2b
JL
James Lyon2013-04-19 22:05:49 +0100143 /* xmm2 */ RC_FLOAT | RC_XMM2,
144 /* xmm3 */ RC_FLOAT | RC_XMM3,
145 /* xmm4 */ RC_FLOAT | RC_XMM4,
1caee8ab
JL
James Lyon2013-04-25 22:30:53 +0100146 /* xmm5 */ RC_FLOAT | RC_XMM5,
147 /* xmm6 an xmm7 are included so gv() can be used on them,
148 but they are not tagged with RC_FLOAT because they are
149 callee saved on Windows */
150 RC_XMM6,
41b3c7a5
JL
James Lyon2013-04-27 20:39:34 +0100151 RC_XMM7,
152 /* st0 */ RC_ST0
88a3ccab grischka2009-12-20 01:53:49 +0100153};
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900154
155static unsigned long func_sub_sp_offset;
156static int func_ret_sub;
157
158/* XXX: make it faster ? */
159void g(int c)
160{
161 int ind1;
162 ind1 = ind + 1;
163 if (ind1 > cur_text_section->data_allocated)
164 section_realloc(cur_text_section, ind1);
165 cur_text_section->data[ind] = c;
166 ind = ind1;
167}
168
169void o(unsigned int c)
170{
171 while (c) {
172 g(c);
173 c = c >> 8;
174 }
175}
176
1308e8eb grischka2009-12-19 22:08:37 +0100177void gen_le16(int v)
178{
179 g(v);
180 g(v >> 8);
181}
182
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900183void gen_le32(int c)
184{
185 g(c);
186 g(c >> 8);
187 g(c >> 16);
188 g(c >> 24);
189}
190
191void gen_le64(int64_t c)
192{
193 g(c);
194 g(c >> 8);
195 g(c >> 16);
196 g(c >> 24);
197 g(c >> 32);
198 g(c >> 40);
199 g(c >> 48);
200 g(c >> 56);
201}
202
070b86a8 grischka2009-12-20 02:19:51 +0100203void orex(int ll, int r, int r2, int b)
204{
205 if ((r & VT_VALMASK) >= VT_CONST)
206 r = 0;
207 if ((r2 & VT_VALMASK) >= VT_CONST)
208 r2 = 0;
209 if (ll || REX_BASE(r) || REX_BASE(r2))
210 o(0x40 | REX_BASE(r) | (REX_BASE(r2) << 2) | (ll << 3));
211 o(b);
212}
213
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900214/* output a symbol and patch all calls to it */
215void gsym_addr(int t, int a)
216{
217 int n, *ptr;
218 while (t) {
219 ptr = (int *)(cur_text_section->data + t);
220 n = *ptr; /* next value */
221 *ptr = a - t - 4;
222 t = n;
223 }
224}
225
226void gsym(int t)
227{
228 gsym_addr(t, ind);
229}
230
231/* psym is used to put an instruction with a data field which is a
232 reference to a symbol. It is in fact the same as oad ! */
233#define psym oad
234
235static int is64_type(int t)
236{
237 return ((t & VT_BTYPE) == VT_PTR ||
238 (t & VT_BTYPE) == VT_FUNC ||
239 (t & VT_BTYPE) == VT_LLONG);
240}
241
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900242/* instruction + 4 bytes data. Return the address of the data */
88a3ccab grischka2009-12-20 01:53:49 +0100243ST_FUNC int oad(int c, int s)
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900244{
245 int ind1;
246
247 o(c);
248 ind1 = ind + 4;
249 if (ind1 > cur_text_section->data_allocated)
250 section_realloc(cur_text_section, ind1);
251 *(int *)(cur_text_section->data + ind) = s;
252 s = ind;
253 ind = ind1;
254 return s;
255}
256
88a3ccab grischka2009-12-20 01:53:49 +0100257ST_FUNC void gen_addr32(int r, Sym *sym, int c)
1308e8eb grischka2009-12-19 22:08:37 +0100258{
259 if (r & VT_SYM)
260 greloc(cur_text_section, sym, ind, R_X86_64_32);
261 gen_le32(c);
262}
263
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900264/* output constant with relocation if 'r & VT_SYM' is true */
88a3ccab grischka2009-12-20 01:53:49 +0100265ST_FUNC void gen_addr64(int r, Sym *sym, int64_t c)
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900266{
267 if (r & VT_SYM)
268 greloc(cur_text_section, sym, ind, R_X86_64_64);
269 gen_le64(c);
270}
271
272/* output constant with relocation if 'r & VT_SYM' is true */
88a3ccab grischka2009-12-20 01:53:49 +0100273ST_FUNC void gen_addrpc32(int r, Sym *sym, int c)
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900274{
275 if (r & VT_SYM)
276 greloc(cur_text_section, sym, ind, R_X86_64_PC32);
277 gen_le32(c-4);
278}
279
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900280/* output got address with relocation */
281static void gen_gotpcrel(int r, Sym *sym, int c)
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900282{
45987579 grischka2009-07-18 22:05:49 +0200283#ifndef TCC_TARGET_PE
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900284 Section *sr;
285 ElfW(Rela) *rel;
286 greloc(cur_text_section, sym, ind, R_X86_64_GOTPCREL);
287 sr = cur_text_section->reloc;
288 rel = (ElfW(Rela) *)(sr->data + sr->data_offset - sizeof(ElfW(Rela)));
289 rel->r_addend = -4;
45987579 grischka2009-07-18 22:05:49 +0200290#else
6e0a658e grischka2014-05-08 17:32:29 +0200291 tcc_error("internal error: no GOT on PE: %s %x %x | %02x %02x %02x\n",
292 get_tok_str(sym->v, NULL), c, r,
45987579 grischka2009-07-18 22:05:49 +0200293 cur_text_section->data[ind-3],
294 cur_text_section->data[ind-2],
295 cur_text_section->data[ind-1]
296 );
297 greloc(cur_text_section, sym, ind, R_X86_64_PC32);
298#endif
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +0900299 gen_le32(0);
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900300 if (c) {
301 /* we use add c, %xxx for displacement */
070b86a8 grischka2009-12-20 02:19:51 +0100302 orex(1, r, 0, 0x81);
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900303 o(0xc0 + REG_VALUE(r));
304 gen_le32(c);
305 }
306}
307
308static void gen_modrm_impl(int op_reg, int r, Sym *sym, int c, int is_got)
309{
310 op_reg = REG_VALUE(op_reg) << 3;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900311 if ((r & VT_VALMASK) == VT_CONST) {
312 /* constant memory reference */
313 o(0x05 | op_reg);
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900314 if (is_got) {
315 gen_gotpcrel(r, sym, c);
316 } else {
317 gen_addrpc32(r, sym, c);
318 }
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900319 } else if ((r & VT_VALMASK) == VT_LOCAL) {
320 /* currently, we use only ebp as base */
321 if (c == (char)c) {
322 /* short reference */
323 o(0x45 | op_reg);
324 g(c);
325 } else {
326 oad(0x85 | op_reg, c);
327 }
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900328 } else if ((r & VT_VALMASK) >= TREG_MEM) {
329 if (c) {
330 g(0x80 | op_reg | REG_VALUE(r));
331 gen_le32(c);
332 } else {
333 g(0x00 | op_reg | REG_VALUE(r));
334 }
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900335 } else {
0e5c0ee0 grischka2009-12-19 00:06:41 +0100336 g(0x00 | op_reg | REG_VALUE(r));
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900337 }
338}
339
340/* generate a modrm reference. 'op_reg' contains the addtionnal 3
341 opcode bits */
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900342static void gen_modrm(int op_reg, int r, Sym *sym, int c)
343{
344 gen_modrm_impl(op_reg, r, sym, c, 0);
345}
346
347/* generate a modrm reference. 'op_reg' contains the addtionnal 3
348 opcode bits */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900349static void gen_modrm64(int opcode, int op_reg, int r, Sym *sym, int c)
350{
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +0900351 int is_got;
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +0900352 is_got = (op_reg & TREG_MEM) && !(sym->type.t & VT_STATIC);
b5486240 grischka2009-12-20 20:33:21 +0100353 orex(1, r, op_reg, opcode);
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +0900354 gen_modrm_impl(op_reg, r, sym, c, is_got);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900355}
356
357
358/* load 'r' from value 'sv' */
359void load(int r, SValue *sv)
360{
361 int v, t, ft, fc, fr;
362 SValue v1;
363
94bf4d2c grischka2009-12-19 22:08:52 +0100364#ifdef TCC_TARGET_PE
2341ee51 grischka2010-01-14 20:55:51 +0100365 SValue v2;
366 sv = pe_getimport(sv, &v2);
94bf4d2c grischka2009-12-19 22:08:52 +0100367#endif
368
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900369 fr = sv->r;
d0dae7f2 Thomas Preud'homme2014-02-07 22:31:44 +0800370 ft = sv->type.t & ~VT_DEFSIGN;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900371 fc = sv->c.ul;
372
45987579 grischka2009-07-18 22:05:49 +0200373#ifndef TCC_TARGET_PE
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900374 /* we use indirect access via got */
375 if ((fr & VT_VALMASK) == VT_CONST && (fr & VT_SYM) &&
376 (fr & VT_LVAL) && !(sv->sym->type.t & VT_STATIC)) {
377 /* use the result register as a temporal register */
378 int tr = r | TREG_MEM;
379 if (is_float(ft)) {
380 /* we cannot use float registers as a temporal register */
381 tr = get_reg(RC_INT) | TREG_MEM;
382 }
383 gen_modrm64(0x8b, tr, fr, sv->sym, 0);
384
385 /* load from the temporal register */
386 fr = tr | VT_LVAL;
387 }
45987579 grischka2009-07-18 22:05:49 +0200388#endif
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +0900389
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900390 v = fr & VT_VALMASK;
391 if (fr & VT_LVAL) {
070b86a8 grischka2009-12-20 02:19:51 +0100392 int b, ll;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900393 if (v == VT_LLOCAL) {
394 v1.type.t = VT_PTR;
395 v1.r = VT_LOCAL | VT_LVAL;
396 v1.c.ul = fc;
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900397 fr = r;
059aea5d Philip2015-04-23 18:08:28 +0000398 if (!(reg_classes[fr] & (RC_INT|RC_R11)))
ae191c3a grischka2012-03-05 20:19:28 +0100399 fr = get_reg(RC_INT);
400 load(fr, &v1);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900401 }
070b86a8 grischka2009-12-20 02:19:51 +0100402 ll = 0;
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900403 if ((ft & VT_BTYPE) == VT_FLOAT) {
946afd23
JL
James Lyon2013-04-19 18:31:24 +0100404 b = 0x6e0f66;
405 r = REG_VALUE(r); /* movd */
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900406 } else if ((ft & VT_BTYPE) == VT_DOUBLE) {
0e17671f James Lyon2013-04-19 15:33:16 +0100407 b = 0x7e0ff3; /* movq */
946afd23 James Lyon2013-04-19 18:31:24 +0100408 r = REG_VALUE(r);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900409 } else if ((ft & VT_BTYPE) == VT_LDOUBLE) {
070b86a8 grischka2009-12-20 02:19:51 +0100410 b = 0xdb, r = 5; /* fldt */
f6b50558 Thomas Preud'homme2013-06-14 16:18:16 +0200411 } else if ((ft & VT_TYPE) == VT_BYTE || (ft & VT_TYPE) == VT_BOOL) {
070b86a8 grischka2009-12-20 02:19:51 +0100412 b = 0xbe0f; /* movsbl */
413 } else if ((ft & VT_TYPE) == (VT_BYTE | VT_UNSIGNED)) {
414 b = 0xb60f; /* movzbl */
415 } else if ((ft & VT_TYPE) == VT_SHORT) {
416 b = 0xbf0f; /* movswl */
417 } else if ((ft & VT_TYPE) == (VT_SHORT | VT_UNSIGNED)) {
418 b = 0xb70f; /* movzwl */
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900419 } else {
0e17671f
JL
James Lyon2013-04-19 15:33:16 +0100420 assert(((ft & VT_BTYPE) == VT_INT) || ((ft & VT_BTYPE) == VT_LLONG)
421 || ((ft & VT_BTYPE) == VT_PTR) || ((ft & VT_BTYPE) == VT_ENUM)
422 || ((ft & VT_BTYPE) == VT_FUNC));
070b86a8 grischka2009-12-20 02:19:51 +0100423 ll = is64_type(ft);
424 b = 0x8b;
425 }
426 if (ll) {
427 gen_modrm64(b, r, fr, sv->sym, fc);
428 } else {
429 orex(ll, fr, r, b);
430 gen_modrm(r, fr, sv->sym, fc);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900431 }
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900432 } else {
433 if (v == VT_CONST) {
bb5e0df7 grischka2009-07-18 22:07:03 +0200434 if (fr & VT_SYM) {
435#ifdef TCC_TARGET_PE
070b86a8 grischka2009-12-20 02:19:51 +0100436 orex(1,0,r,0x8d);
bb5e0df7 grischka2009-07-18 22:07:03 +0200437 o(0x05 + REG_VALUE(r) * 8); /* lea xx(%rip), r */
438 gen_addrpc32(fr, sv->sym, fc);
439#else
440 if (sv->sym->type.t & VT_STATIC) {
070b86a8 grischka2009-12-20 02:19:51 +0100441 orex(1,0,r,0x8d);
bb5e0df7 grischka2009-07-18 22:07:03 +0200442 o(0x05 + REG_VALUE(r) * 8); /* lea xx(%rip), r */
443 gen_addrpc32(fr, sv->sym, fc);
444 } else {
070b86a8 grischka2009-12-20 02:19:51 +0100445 orex(1,0,r,0x8b);
bb5e0df7 grischka2009-07-18 22:07:03 +0200446 o(0x05 + REG_VALUE(r) * 8); /* mov xx(%rip), r */
1d0a5c25 Michael Matz2012-04-15 19:29:45 +0200447 gen_gotpcrel(r, sv->sym, fc);
bb5e0df7 grischka2009-07-18 22:07:03 +0200448 }
449#endif
450 } else if (is64_type(ft)) {
070b86a8 grischka2009-12-20 02:19:51 +0100451 orex(1,r,0, 0xb8 + REG_VALUE(r)); /* mov $xx, r */
bb5e0df7 grischka2009-07-18 22:07:03 +0200452 gen_le64(sv->c.ull);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900453 } else {
070b86a8 grischka2009-12-20 02:19:51 +0100454 orex(0,r,0, 0xb8 + REG_VALUE(r)); /* mov $xx, r */
bb5e0df7 grischka2009-07-18 22:07:03 +0200455 gen_le32(fc);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900456 }
457 } else if (v == VT_LOCAL) {
070b86a8 grischka2009-12-20 02:19:51 +0100458 orex(1,0,r,0x8d); /* lea xxx(%ebp), r */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900459 gen_modrm(r, VT_LOCAL, sv->sym, fc);
460 } else if (v == VT_CMP) {
070b86a8 grischka2009-12-20 02:19:51 +0100461 orex(0,r,0,0);
89ad24e7 gus knight2015-07-29 16:53:57 -0400462 if ((fc & ~0x100) != TOK_NE)
463 oad(0xb8 + REG_VALUE(r), 0); /* mov $0, r */
464 else
465 oad(0xb8 + REG_VALUE(r), 1); /* mov $1, r */
466 if (fc & 0x100)
467 {
468 /* This was a float compare. If the parity bit is
469 set the result was unordered, meaning false for everything
470 except TOK_NE, and true for TOK_NE. */
471 fc &= ~0x100;
472 o(0x037a + (REX_BASE(r) << 8));
473 }
070b86a8 grischka2009-12-20 02:19:51 +0100474 orex(0,r,0, 0x0f); /* setxx %br */
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900475 o(fc);
0e5c0ee0 grischka2009-12-19 00:06:41 +0100476 o(0xc0 + REG_VALUE(r));
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900477 } else if (v == VT_JMP || v == VT_JMPI) {
478 t = v & 1;
070b86a8 grischka2009-12-20 02:19:51 +0100479 orex(0,r,0,0);
0e5c0ee0 grischka2009-12-19 00:06:41 +0100480 oad(0xb8 + REG_VALUE(r), t); /* mov $1, r */
481 o(0x05eb + (REX_BASE(r) << 8)); /* jmp after */
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900482 gsym(fc);
070b86a8 grischka2009-12-20 02:19:51 +0100483 orex(0,r,0,0);
0e5c0ee0 grischka2009-12-19 00:06:41 +0100484 oad(0xb8 + REG_VALUE(r), t ^ 1); /* mov $0, r */
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900485 } else if (v != r) {
5c35ba66 James Lyon2013-04-24 02:19:15 +0100486 if ((r >= TREG_XMM0) && (r <= TREG_XMM7)) {
0e17671f
JL
James Lyon2013-04-19 15:33:16 +0100487 if (v == TREG_ST0) {
488 /* gen_cvt_ftof(VT_DOUBLE); */
489 o(0xf0245cdd); /* fstpl -0x10(%rsp) */
490 /* movsd -0x10(%rsp),%xmmN */
491 o(0x100ff2);
946afd23 James Lyon2013-04-19 18:31:24 +0100492 o(0x44 + REG_VALUE(r)*8); /* %xmmN */
0e17671f
JL
James Lyon2013-04-19 15:33:16 +0100493 o(0xf024);
494 } else {
5c35ba66 James Lyon2013-04-24 02:19:15 +0100495 assert((v >= TREG_XMM0) && (v <= TREG_XMM7));
0e17671f
JL
James Lyon2013-04-19 15:33:16 +0100496 if ((ft & VT_BTYPE) == VT_FLOAT) {
497 o(0x100ff3);
498 } else {
499 assert((ft & VT_BTYPE) == VT_DOUBLE);
500 o(0x100ff2);
501 }
946afd23 James Lyon2013-04-19 18:31:24 +0100502 o(0xc0 + REG_VALUE(v) + REG_VALUE(r)*8);
0e17671f James Lyon2013-04-19 15:33:16 +0100503 }
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900504 } else if (r == TREG_ST0) {
fdb3b10d Thomas Preud'homme2014-03-08 18:36:02 +0800505 assert((v >= TREG_XMM0) && (v <= TREG_XMM7));
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900506 /* gen_cvt_ftof(VT_LDOUBLE); */
946afd23 James Lyon2013-04-19 18:31:24 +0100507 /* movsd %xmmN,-0x10(%rsp) */
0e17671f James Lyon2013-04-19 15:33:16 +0100508 o(0x110ff2);
946afd23 James Lyon2013-04-19 18:31:24 +0100509 o(0x44 + REG_VALUE(r)*8); /* %xmmN */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900510 o(0xf024);
511 o(0xf02444dd); /* fldl -0x10(%rsp) */
512 } else {
070b86a8 grischka2009-12-20 02:19:51 +0100513 orex(1,r,v, 0x89);
0e5c0ee0 grischka2009-12-19 00:06:41 +0100514 o(0xc0 + REG_VALUE(r) + REG_VALUE(v) * 8); /* mov v, r */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900515 }
516 }
517 }
518}
519
520/* store register 'r' in lvalue 'v' */
521void store(int r, SValue *v)
522{
523 int fr, bt, ft, fc;
524 int op64 = 0;
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900525 /* store the REX prefix in this variable when PIC is enabled */
526 int pic = 0;
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900527
94bf4d2c grischka2009-12-19 22:08:52 +0100528#ifdef TCC_TARGET_PE
2341ee51 grischka2010-01-14 20:55:51 +0100529 SValue v2;
530 v = pe_getimport(v, &v2);
94bf4d2c grischka2009-12-19 22:08:52 +0100531#endif
532
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900533 ft = v->type.t;
534 fc = v->c.ul;
535 fr = v->r & VT_VALMASK;
536 bt = ft & VT_BTYPE;
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +0900537
45987579 grischka2009-07-18 22:05:49 +0200538#ifndef TCC_TARGET_PE
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900539 /* we need to access the variable via got */
540 if (fr == VT_CONST && (v->r & VT_SYM)) {
541 /* mov xx(%rip), %r11 */
542 o(0x1d8b4c);
543 gen_gotpcrel(TREG_R11, v->sym, v->c.ul);
544 pic = is64_type(bt) ? 0x49 : 0x41;
545 }
45987579 grischka2009-07-18 22:05:49 +0200546#endif
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +0900547
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900548 /* XXX: incorrect if float reg to reg */
549 if (bt == VT_FLOAT) {
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900550 o(0x66);
551 o(pic);
552 o(0x7e0f); /* movd */
946afd23 James Lyon2013-04-19 18:31:24 +0100553 r = REG_VALUE(r);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900554 } else if (bt == VT_DOUBLE) {
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900555 o(0x66);
556 o(pic);
557 o(0xd60f); /* movq */
946afd23 James Lyon2013-04-19 18:31:24 +0100558 r = REG_VALUE(r);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900559 } else if (bt == VT_LDOUBLE) {
560 o(0xc0d9); /* fld %st(0) */
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +0900561 o(pic);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900562 o(0xdb); /* fstpt */
563 r = 7;
564 } else {
565 if (bt == VT_SHORT)
566 o(0x66);
830b7533 Shinichiro Hamaji2009-04-13 00:27:04 +0900567 o(pic);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900568 if (bt == VT_BYTE || bt == VT_BOOL)
070b86a8 grischka2009-12-20 02:19:51 +0100569 orex(0, 0, r, 0x88);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900570 else if (is64_type(bt))
571 op64 = 0x89;
572 else
070b86a8 grischka2009-12-20 02:19:51 +0100573 orex(0, 0, r, 0x89);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +0900574 }
830b7533
SH
Shinichiro Hamaji2009-04-13 00:27:04 +0900575 if (pic) {
576 /* xxx r, (%r11) where xxx is mov, movq, fld, or etc */
577 if (op64)
578 o(op64);
579 o(3 + (r << 3));
580 } else if (op64) {
0e5c0ee0 grischka2009-12-19 00:06:41 +0100581 if (fr == VT_CONST || fr == VT_LOCAL || (v->r & VT_LVAL)) {
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900582 gen_modrm64(op64, r, v->r, v->sym, fc);
583 } else if (fr != r) {
584 /* XXX: don't we really come here? */
585 abort();
586 o(0xc0 + fr + r * 8); /* mov r, fr */
587 }
588 } else {
0e5c0ee0 grischka2009-12-19 00:06:41 +0100589 if (fr == VT_CONST || fr == VT_LOCAL || (v->r & VT_LVAL)) {
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900590 gen_modrm(r, v->r, v->sym, fc);
591 } else if (fr != r) {
592 /* XXX: don't we really come here? */
593 abort();
594 o(0xc0 + fr + r * 8); /* mov r, fr */
595 }
596 }
597}
598
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900599/* 'is_jmp' is '1' if it is a jump */
600static void gcall_or_jmp(int is_jmp)
601{
602 int r;
aacf65bb Philip2015-04-23 17:30:16 +0000603 if ((vtop->r & (VT_VALMASK | VT_LVAL)) == VT_CONST &&
89ad24e7 gus knight2015-07-29 16:53:57 -0400604 ((vtop->r & VT_SYM) || (vtop->c.ll-4) == (int)(vtop->c.ll-4))) {
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900605 /* constant case */
606 if (vtop->r & VT_SYM) {
607 /* relocation case */
6e0a658e grischka2014-05-08 17:32:29 +0200608#ifdef TCC_TARGET_PE
609 greloc(cur_text_section, vtop->sym, ind + 1, R_X86_64_PC32);
610#else
611 greloc(cur_text_section, vtop->sym, ind + 1, R_X86_64_PLT32);
612#endif
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +0900613 } else {
614 /* put an empty PC32 relocation */
615 put_elf_reloc(symtab_section, cur_text_section,
616 ind + 1, R_X86_64_PC32, 0);
617 }
618 oad(0xe8 + is_jmp, vtop->c.ul - 4); /* call/jmp im */
619 } else {
620 /* otherwise, indirect call */
621 r = TREG_R11;
622 load(r, vtop);
623 o(0x41); /* REX */
624 o(0xff); /* call/jmp *r */
625 o(0xd0 + REG_VALUE(r) + (is_jmp << 4));
626 }
627}
628
559675b9 seyko2015-04-10 15:17:22 +0300629#if defined(CONFIG_TCC_BCHECK)
630#ifndef TCC_TARGET_PE
631static addr_t func_bound_offset;
632static unsigned long func_bound_ind;
633#endif
634
635static void gen_static_call(int v)
636{
637 Sym *sym = external_global_sym(v, &func_old_type, 0);
638 oad(0xe8, -4);
639 greloc(cur_text_section, sym, ind-4, R_X86_64_PC32);
640}
641
642/* generate a bounded pointer addition */
643ST_FUNC void gen_bounded_ptr_add(void)
644{
645 /* save all temporary registers */
646 save_regs(0);
647
648 /* prepare fast x86_64 function call */
649 gv(RC_RAX);
650 o(0xc68948); // mov %rax,%rsi ## second arg in %rsi, this must be size
651 vtop--;
652
653 gv(RC_RAX);
654 o(0xc78948); // mov %rax,%rdi ## first arg in %rdi, this must be ptr
655 vtop--;
656
657 /* do a fast function call */
658 gen_static_call(TOK___bound_ptr_add);
659
660 /* returned pointer is in rax */
661 vtop++;
662 vtop->r = TREG_RAX | VT_BOUNDED;
663
664
665 /* relocation offset of the bounding function call point */
89ad24e7 gus knight2015-07-29 16:53:57 -0400666 vtop->c.ull = (cur_text_section->reloc->data_offset - sizeof(ElfW(Rela)));
559675b9 seyko2015-04-10 15:17:22 +0300667}
668
669/* patch pointer addition in vtop so that pointer dereferencing is
670 also tested */
671ST_FUNC void gen_bounded_ptr_deref(void)
672{
673 addr_t func;
674 int size, align;
675 ElfW(Rela) *rel;
676 Sym *sym;
677
678 size = 0;
679 /* XXX: put that code in generic part of tcc */
680 if (!is_float(vtop->type.t)) {
681 if (vtop->r & VT_LVAL_BYTE)
682 size = 1;
683 else if (vtop->r & VT_LVAL_SHORT)
684 size = 2;
685 }
686 if (!size)
687 size = type_size(&vtop->type, &align);
688 switch(size) {
689 case 1: func = TOK___bound_ptr_indir1; break;
690 case 2: func = TOK___bound_ptr_indir2; break;
691 case 4: func = TOK___bound_ptr_indir4; break;
692 case 8: func = TOK___bound_ptr_indir8; break;
693 case 12: func = TOK___bound_ptr_indir12; break;
694 case 16: func = TOK___bound_ptr_indir16; break;
695 default:
696 tcc_error("unhandled size when dereferencing bounded pointer");
697 func = 0;
698 break;
699 }
700
701 sym = external_global_sym(func, &func_old_type, 0);
702 if (!sym->c)
703 put_extern_sym(sym, NULL, 0, 0);
704
705 /* patch relocation */
706 /* XXX: find a better solution ? */
707
708 rel = (ElfW(Rela) *)(cur_text_section->reloc->data + vtop->c.ull);
709 rel->r_info = ELF64_R_INFO(sym->c, ELF64_R_TYPE(rel->r_info));
710}
711#endif
712
45987579 grischka2009-07-18 22:05:49 +0200713#ifdef TCC_TARGET_PE
50b040ef grischka2009-12-19 22:40:28 +0100714
45987579 grischka2009-07-18 22:05:49 +0200715#define REGN 4
5c35ba66 James Lyon2013-04-24 02:19:15 +0100716static const uint8_t arg_regs[REGN] = {
45987579 grischka2009-07-18 22:05:49 +0200717 TREG_RCX, TREG_RDX, TREG_R8, TREG_R9
718};
50b040ef grischka2009-12-19 22:40:28 +0100719
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100720/* Prepare arguments in R10 and R11 rather than RCX and RDX
721 because gv() will not ever use these */
722static int arg_prepare_reg(int idx) {
723 if (idx == 0 || idx == 1)
724 /* idx=0: r10, idx=1: r11 */
725 return idx + 10;
726 else
727 return arg_regs[idx];
728}
729
50b040ef grischka2009-12-19 22:40:28 +0100730static int func_scratch;
731
732/* Generate function call. The function address is pushed first, then
733 all the parameters in call order. This functions pops all the
734 parameters and the function address. */
735
736void gen_offs_sp(int b, int r, int d)
737{
070b86a8 grischka2009-12-20 02:19:51 +0100738 orex(1,0,r & 0x100 ? 0 : r, b);
50b040ef grischka2009-12-19 22:40:28 +0100739 if (d == (char)d) {
070b86a8 grischka2009-12-20 02:19:51 +0100740 o(0x2444 | (REG_VALUE(r) << 3));
50b040ef grischka2009-12-19 22:40:28 +0100741 g(d);
742 } else {
070b86a8 grischka2009-12-20 02:19:51 +0100743 o(0x2484 | (REG_VALUE(r) << 3));
50b040ef grischka2009-12-19 22:40:28 +0100744 gen_le32(d);
745 }
746}
747
dcec8673
TP
Thomas Preud'homme2013-11-22 09:27:15 +0800748/* Return the number of registers needed to return the struct, or 0 if
749 returning via struct pointer. */
ef3d38c5 gus knight2015-07-29 16:57:41 -0400750ST_FUNC int gfunc_sret(CType *vt, int variadic, CType *ret, int *ret_align, int *regsize)
be1b6ba7 grischka2013-04-30 00:33:34 +0200751{
5c35ba66 James Lyon2013-04-24 02:19:15 +0100752 int size, align;
50899e30 Michael Matz2015-03-09 00:19:59 +0100753 *regsize = 8;
be1b6ba7 grischka2013-04-30 00:33:34 +0200754 *ret_align = 1; // Never have to re-align return values for x86-64
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100755 size = type_size(vt, &align);
756 ret->ref = NULL;
757 if (size > 8) {
ef3d38c5 gus knight2015-07-29 16:57:41 -0400758 return 0;
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100759 } else if (size > 4) {
760 ret->t = VT_LLONG;
ef3d38c5 gus knight2015-07-29 16:57:41 -0400761 return 1;
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100762 } else if (size > 2) {
763 ret->t = VT_INT;
ef3d38c5 gus knight2015-07-29 16:57:41 -0400764 return 1;
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100765 } else if (size > 1) {
766 ret->t = VT_SHORT;
ef3d38c5 gus knight2015-07-29 16:57:41 -0400767 return 1;
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100768 } else {
769 ret->t = VT_BYTE;
ef3d38c5 gus knight2015-07-29 16:57:41 -0400770 return 1;
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100771 }
772}
773
0f5942c6
TP
Thomas Preud'homme2013-09-24 15:36:04 +0200774static int is_sse_float(int t) {
775 int bt;
776 bt = t & VT_BTYPE;
777 return bt == VT_DOUBLE || bt == VT_FLOAT;
778}
779
5c35ba66 James Lyon2013-04-24 02:19:15 +0100780int gfunc_arg_size(CType *type) {
be1b6ba7 grischka2013-04-30 00:33:34 +0200781 int align;
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100782 if (type->t & (VT_ARRAY|VT_BITFIELD))
783 return 8;
5c35ba66 James Lyon2013-04-24 02:19:15 +0100784 return type_size(type, &align);
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +0100785}
786
50b040ef grischka2009-12-19 22:40:28 +0100787void gfunc_call(int nb_args)
788{
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100789 int size, r, args_size, i, d, bt, struct_size;
790 int arg;
50b040ef grischka2009-12-19 22:40:28 +0100791
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100792 args_size = (nb_args < REGN ? REGN : nb_args) * PTR_SIZE;
793 arg = nb_args;
50b040ef grischka2009-12-19 22:40:28 +0100794
795 /* for struct arguments, we need to call memcpy and the function
796 call breaks register passing arguments we are preparing.
797 So, we process arguments which will be passed by stack first. */
8d107d9f grischka2010-12-17 13:15:01 +0100798 struct_size = args_size;
50b040ef grischka2009-12-19 22:40:28 +0100799 for(i = 0; i < nb_args; i++) {
be1b6ba7 grischka2013-04-30 00:33:34 +0200800 SValue *sv;
89ad24e7 gus knight2015-07-29 16:53:57 -0400801
be1b6ba7 grischka2013-04-30 00:33:34 +0200802 --arg;
803 sv = &vtop[-i];
50b040ef grischka2009-12-19 22:40:28 +0100804 bt = (sv->type.t & VT_BTYPE);
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100805 size = gfunc_arg_size(&sv->type);
806
807 if (size <= 8)
808 continue; /* arguments smaller than 8 bytes passed in registers or on stack */
809
50b040ef grischka2009-12-19 22:40:28 +0100810 if (bt == VT_STRUCT) {
50b040ef grischka2009-12-19 22:40:28 +0100811 /* align to stack align size */
9d347f87 Shinichiro Hamaji2010-08-27 02:49:09 +0900812 size = (size + 15) & ~15;
50b040ef grischka2009-12-19 22:40:28 +0100813 /* generate structure store */
814 r = get_reg(RC_INT);
8d107d9f grischka2010-12-17 13:15:01 +0100815 gen_offs_sp(0x8d, r, struct_size);
816 struct_size += size;
50b040ef grischka2009-12-19 22:40:28 +0100817
818 /* generate memcpy call */
819 vset(&sv->type, r | VT_LVAL, 0);
4a01eb09 grischka2009-12-19 22:41:26 +0100820 vpushv(sv);
50b040ef grischka2009-12-19 22:40:28 +0100821 vstore();
822 --vtop;
50b040ef grischka2009-12-19 22:40:28 +0100823 } else if (bt == VT_LDOUBLE) {
50b040ef grischka2009-12-19 22:40:28 +0100824 gv(RC_ST0);
8d107d9f grischka2010-12-17 13:15:01 +0100825 gen_offs_sp(0xdb, 0x107, struct_size);
826 struct_size += 16;
50b040ef grischka2009-12-19 22:40:28 +0100827 }
828 }
829
8d107d9f grischka2010-12-17 13:15:01 +0100830 if (func_scratch < struct_size)
831 func_scratch = struct_size;
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100832
833 arg = nb_args;
8d107d9f grischka2010-12-17 13:15:01 +0100834 struct_size = args_size;
835
50b040ef grischka2009-12-19 22:40:28 +0100836 for(i = 0; i < nb_args; i++) {
5c35ba66 James Lyon2013-04-24 02:19:15 +0100837 --arg;
50b040ef grischka2009-12-19 22:40:28 +0100838 bt = (vtop->type.t & VT_BTYPE);
8d107d9f grischka2010-12-17 13:15:01 +0100839
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100840 size = gfunc_arg_size(&vtop->type);
841 if (size > 8) {
8d107d9f grischka2010-12-17 13:15:01 +0100842 /* align to stack align size */
843 size = (size + 15) & ~15;
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100844 if (arg >= REGN) {
845 d = get_reg(RC_INT);
8d107d9f grischka2010-12-17 13:15:01 +0100846 gen_offs_sp(0x8d, d, struct_size);
5c35ba66 James Lyon2013-04-24 02:19:15 +0100847 gen_offs_sp(0x89, d, arg*8);
8d107d9f grischka2010-12-17 13:15:01 +0100848 } else {
5c35ba66 James Lyon2013-04-24 02:19:15 +0100849 d = arg_prepare_reg(arg);
8d107d9f grischka2010-12-17 13:15:01 +0100850 gen_offs_sp(0x8d, d, struct_size);
851 }
852 struct_size += size;
50b040ef grischka2009-12-19 22:40:28 +0100853 } else {
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100854 if (is_sse_float(vtop->type.t)) {
855 gv(RC_XMM0); /* only use one float register */
856 if (arg >= REGN) {
857 /* movq %xmm0, j*8(%rsp) */
858 gen_offs_sp(0xd60f66, 0x100, arg*8);
859 } else {
860 /* movaps %xmm0, %xmmN */
861 o(0x280f);
862 o(0xc0 + (arg << 3));
863 d = arg_prepare_reg(arg);
864 /* mov %xmm0, %rxx */
865 o(0x66);
866 orex(1,d,0, 0x7e0f);
867 o(0xc0 + REG_VALUE(d));
868 }
50b040ef grischka2009-12-19 22:40:28 +0100869 } else {
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100870 if (bt == VT_STRUCT) {
871 vtop->type.ref = NULL;
872 vtop->type.t = size > 4 ? VT_LLONG : size > 2 ? VT_INT
873 : size > 1 ? VT_SHORT : VT_BYTE;
874 }
89ad24e7 gus knight2015-07-29 16:53:57 -0400875
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100876 r = gv(RC_INT);
877 if (arg >= REGN) {
878 gen_offs_sp(0x89, r, arg*8);
0e5c0ee0 grischka2009-12-19 00:06:41 +0100879 } else {
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100880 d = arg_prepare_reg(arg);
881 orex(1,d,r,0x89); /* mov */
882 o(0xc0 + REG_VALUE(r) * 8 + REG_VALUE(d));
50b040ef grischka2009-12-19 22:40:28 +0100883 }
884 }
885 }
886 vtop--;
887 }
0e5c0ee0 grischka2009-12-19 00:06:41 +0100888 save_regs(0);
89ad24e7 gus knight2015-07-29 16:53:57 -0400889
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100890 /* Copy R10 and R11 into RCX and RDX, respectively */
891 if (nb_args > 0) {
892 o(0xd1894c); /* mov %r10, %rcx */
893 if (nb_args > 1) {
894 o(0xda894c); /* mov %r11, %rdx */
895 }
896 }
89ad24e7 gus knight2015-07-29 16:53:57 -0400897
50b040ef grischka2009-12-19 22:40:28 +0100898 gcall_or_jmp(0);
899 vtop--;
900}
901
902
903#define FUNC_PROLOG_SIZE 11
904
905/* generate function prolog of type 't' */
906void gfunc_prolog(CType *func_type)
907{
5c35ba66 James Lyon2013-04-24 02:19:15 +0100908 int addr, reg_param_index, bt, size;
50b040ef grischka2009-12-19 22:40:28 +0100909 Sym *sym;
910 CType *type;
911
912 func_ret_sub = 0;
913 func_scratch = 0;
914 loc = 0;
915
916 addr = PTR_SIZE * 2;
917 ind += FUNC_PROLOG_SIZE;
918 func_sub_sp_offset = ind;
919 reg_param_index = 0;
920
921 sym = func_type->ref;
922
923 /* if the function returns a structure, then add an
924 implicit pointer parameter */
925 func_vt = sym->type;
8efaa711 Thomas Preud'homme2014-01-06 22:27:39 +0800926 func_var = (sym->c == FUNC_ELLIPSIS);
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100927 size = gfunc_arg_size(&func_vt);
928 if (size > 8) {
50b040ef grischka2009-12-19 22:40:28 +0100929 gen_modrm64(0x89, arg_regs[reg_param_index], VT_LOCAL, NULL, addr);
5c35ba66 James Lyon2013-04-24 02:19:15 +0100930 func_vc = addr;
50b040ef grischka2009-12-19 22:40:28 +0100931 reg_param_index++;
5c35ba66 James Lyon2013-04-24 02:19:15 +0100932 addr += 8;
50b040ef grischka2009-12-19 22:40:28 +0100933 }
934
935 /* define parameters */
936 while ((sym = sym->next) != NULL) {
937 type = &sym->type;
938 bt = type->t & VT_BTYPE;
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100939 size = gfunc_arg_size(type);
940 if (size > 8) {
941 if (reg_param_index < REGN) {
942 gen_modrm64(0x89, arg_regs[reg_param_index], VT_LOCAL, NULL, addr);
943 }
8d107d9f grischka2010-12-17 13:15:01 +0100944 sym_push(sym->v & ~SYM_FIELD, type, VT_LOCAL | VT_LVAL | VT_REF, addr);
945 } else {
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100946 if (reg_param_index < REGN) {
947 /* save arguments passed by register */
948 if ((bt == VT_FLOAT) || (bt == VT_DOUBLE)) {
949 o(0xd60f66); /* movq */
950 gen_modrm(reg_param_index, VT_LOCAL, NULL, addr);
951 } else {
952 gen_modrm64(0x89, arg_regs[reg_param_index], VT_LOCAL, NULL, addr);
953 }
954 }
8d107d9f grischka2010-12-17 13:15:01 +0100955 sym_push(sym->v & ~SYM_FIELD, type, VT_LOCAL | VT_LVAL, addr);
956 }
5c35ba66 James Lyon2013-04-24 02:19:15 +0100957 addr += 8;
50b040ef grischka2009-12-19 22:40:28 +0100958 reg_param_index++;
50b040ef grischka2009-12-19 22:40:28 +0100959 }
960
961 while (reg_param_index < REGN) {
5c35ba66 James Lyon2013-04-24 02:19:15 +0100962 if (func_type->ref->c == FUNC_ELLIPSIS) {
50b040ef grischka2009-12-19 22:40:28 +0100963 gen_modrm64(0x89, arg_regs[reg_param_index], VT_LOCAL, NULL, addr);
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +0100964 addr += 8;
965 }
50b040ef grischka2009-12-19 22:40:28 +0100966 reg_param_index++;
50b040ef grischka2009-12-19 22:40:28 +0100967 }
50b040ef grischka2009-12-19 22:40:28 +0100968}
969
970/* generate function epilog */
971void gfunc_epilog(void)
972{
973 int v, saved_ind;
974
975 o(0xc9); /* leave */
976 if (func_ret_sub == 0) {
977 o(0xc3); /* ret */
978 } else {
979 o(0xc2); /* ret n */
980 g(func_ret_sub);
981 g(func_ret_sub >> 8);
982 }
983
984 saved_ind = ind;
985 ind = func_sub_sp_offset - FUNC_PROLOG_SIZE;
986 /* align local size to word & save local variables */
987 v = (func_scratch + -loc + 15) & -16;
988
50b040ef grischka2009-12-19 22:40:28 +0100989 if (v >= 4096) {
990 Sym *sym = external_global_sym(TOK___chkstk, &func_old_type, 0);
991 oad(0xb8, v); /* mov stacksize, %eax */
992 oad(0xe8, -4); /* call __chkstk, (does the stackframe too) */
993 greloc(cur_text_section, sym, ind-4, R_X86_64_PC32);
994 o(0x90); /* fill for FUNC_PROLOG_SIZE = 11 bytes */
995 } else {
996 o(0xe5894855); /* push %rbp, mov %rsp, %rbp */
997 o(0xec8148); /* sub rsp, stacksize */
998 gen_le32(v);
999 }
df4c0892 grischka2011-07-14 19:09:49 +02001000
1001 cur_text_section->data_offset = saved_ind;
1002 pe_add_unwind_data(ind, saved_ind, v);
1003 ind = cur_text_section->data_offset;
50b040ef grischka2009-12-19 22:40:28 +01001004}
1005
45987579 grischka2009-07-18 22:05:49 +02001006#else
50b040ef grischka2009-12-19 22:40:28 +01001007
1008static void gadd_sp(int val)
1009{
1010 if (val == (char)val) {
1011 o(0xc48348);
1012 g(val);
1013 } else {
1014 oad(0xc48148, val); /* add $xxx, %rsp */
1015 }
1016}
1017
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001018typedef enum X86_64_Mode {
1019 x86_64_mode_none,
1020 x86_64_mode_memory,
1021 x86_64_mode_integer,
1022 x86_64_mode_sse,
1023 x86_64_mode_x87
1024} X86_64_Mode;
1025
3fe2a95d grischka2014-01-07 14:57:07 +01001026static X86_64_Mode classify_x86_64_merge(X86_64_Mode a, X86_64_Mode b)
1027{
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001028 if (a == b)
1029 return a;
1030 else if (a == x86_64_mode_none)
1031 return b;
1032 else if (b == x86_64_mode_none)
1033 return a;
1034 else if ((a == x86_64_mode_memory) || (b == x86_64_mode_memory))
1035 return x86_64_mode_memory;
1036 else if ((a == x86_64_mode_integer) || (b == x86_64_mode_integer))
1037 return x86_64_mode_integer;
1038 else if ((a == x86_64_mode_x87) || (b == x86_64_mode_x87))
1039 return x86_64_mode_memory;
1040 else
1041 return x86_64_mode_sse;
1042}
1043
ef3d38c5 gus knight2015-07-29 16:57:41 -04001044static X86_64_Mode classify_x86_64_inner(CType *ty)
3fe2a95d grischka2014-01-07 14:57:07 +01001045{
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001046 X86_64_Mode mode;
1047 Sym *f;
89ad24e7 gus knight2015-07-29 16:53:57 -04001048
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001049 switch (ty->t & VT_BTYPE) {
1050 case VT_VOID: return x86_64_mode_none;
89ad24e7 gus knight2015-07-29 16:53:57 -04001051
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001052 case VT_INT:
1053 case VT_BYTE:
1054 case VT_SHORT:
1055 case VT_LLONG:
1056 case VT_BOOL:
1057 case VT_PTR:
0e17671f James Lyon2013-04-19 15:33:16 +01001058 case VT_FUNC:
55ea6d3f James Lyon2013-04-19 00:40:48 +01001059 case VT_ENUM: return x86_64_mode_integer;
89ad24e7 gus knight2015-07-29 16:53:57 -04001060
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001061 case VT_FLOAT:
1062 case VT_DOUBLE: return x86_64_mode_sse;
89ad24e7 gus knight2015-07-29 16:53:57 -04001063
55ea6d3f James Lyon2013-04-19 00:40:48 +01001064 case VT_LDOUBLE: return x86_64_mode_x87;
89ad24e7 gus knight2015-07-29 16:53:57 -04001065
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001066 case VT_STRUCT:
1067 f = ty->ref;
1068
55ea6d3f James Lyon2013-04-19 00:40:48 +01001069 mode = x86_64_mode_none;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001070 for (f = f->next; f; f = f->next)
1071 mode = classify_x86_64_merge(mode, classify_x86_64_inner(&f->type));
89ad24e7 gus knight2015-07-29 16:53:57 -04001072
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001073 return mode;
1074 }
89ad24e7 gus knight2015-07-29 16:53:57 -04001075
0e17671f James Lyon2013-04-19 15:33:16 +01001076 assert(0);
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001077}
1078
ef3d38c5 gus knight2015-07-29 16:57:41 -04001079static X86_64_Mode classify_x86_64_arg(CType *ty, CType *ret, int *psize, int *palign, int *reg_count)
3fe2a95d grischka2014-01-07 14:57:07 +01001080{
55ea6d3f James Lyon2013-04-19 00:40:48 +01001081 X86_64_Mode mode;
0f5942c6 Thomas Preud'homme2013-09-24 15:36:04 +02001082 int size, align, ret_t = 0;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001083
b961ba53 James Lyon2013-04-19 11:08:12 +01001084 if (ty->t & (VT_BITFIELD|VT_ARRAY)) {
55ea6d3f James Lyon2013-04-19 00:40:48 +01001085 *psize = 8;
59b8007f Thomas Preud'homme2014-01-03 08:41:34 +08001086 *palign = 8;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001087 *reg_count = 1;
b961ba53
JL
James Lyon2013-04-19 11:08:12 +01001088 ret_t = ty->t;
1089 mode = x86_64_mode_integer;
1090 } else {
1091 size = type_size(ty, &align);
1092 *psize = (size + 7) & ~7;
6ee366e7 James Lyon2013-04-26 16:42:12 +01001093 *palign = (align + 7) & ~7;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001094
b961ba53
JL
James Lyon2013-04-19 11:08:12 +01001095 if (size > 16) {
1096 mode = x86_64_mode_memory;
1097 } else {
ef3d38c5 gus knight2015-07-29 16:57:41 -04001098 mode = classify_x86_64_inner(ty);
1099 switch (mode) {
1100 case x86_64_mode_integer:
1101 if (size > 8) {
1102 *reg_count = 2;
1103 ret_t = VT_QLONG;
b961ba53 James Lyon2013-04-19 11:08:12 +01001104 } else {
ef3d38c5 gus knight2015-07-29 16:57:41 -04001105 *reg_count = 1;
1106 ret_t = (size > 4) ? VT_LLONG : VT_INT;
b961ba53 James Lyon2013-04-19 11:08:12 +01001107 }
ef3d38c5 gus knight2015-07-29 16:57:41 -04001108 break;
1109
1110 case x86_64_mode_x87:
1111 *reg_count = 1;
1112 ret_t = VT_LDOUBLE;
1113 break;
b961ba53 James Lyon2013-04-19 11:08:12 +01001114
ef3d38c5 gus knight2015-07-29 16:57:41 -04001115 case x86_64_mode_sse:
1116 if (size > 8) {
1117 *reg_count = 2;
1118 ret_t = VT_QFLOAT;
4e04f67c seyko2015-05-14 07:32:24 +03001119 } else {
ef3d38c5 gus knight2015-07-29 16:57:41 -04001120 *reg_count = 1;
1121 ret_t = (size > 4) ? VT_DOUBLE : VT_FLOAT;
b961ba53 James Lyon2013-04-19 11:08:12 +01001122 }
ef3d38c5 gus knight2015-07-29 16:57:41 -04001123 break;
1124 default: break; /* nothing to be done for x86_64_mode_memory and x86_64_mode_none*/
b961ba53
JL
James Lyon2013-04-19 11:08:12 +01001125 }
1126 }
55ea6d3f James Lyon2013-04-19 00:40:48 +01001127 }
ef3d38c5 gus knight2015-07-29 16:57:41 -04001128
b961ba53
JL
James Lyon2013-04-19 11:08:12 +01001129 if (ret) {
1130 ret->ref = NULL;
1131 ret->t = ret_t;
55ea6d3f James Lyon2013-04-19 00:40:48 +01001132 }
ef3d38c5 gus knight2015-07-29 16:57:41 -04001133
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001134 return mode;
1135}
1136
3fe2a95d grischka2014-01-07 14:57:07 +01001137ST_FUNC int classify_x86_64_va_arg(CType *ty)
1138{
b961ba53
JL
James Lyon2013-04-19 11:08:12 +01001139 /* This definition must be synced with stdarg.h */
1140 enum __va_arg_type {
1141 __va_gen_reg, __va_float_reg, __va_stack
1142 };
ef3d38c5 gus knight2015-07-29 16:57:41 -04001143 int size, align, reg_count;
1144 X86_64_Mode mode = classify_x86_64_arg(ty, NULL, &size, &align, &reg_count);
55ea6d3f James Lyon2013-04-19 00:40:48 +01001145 switch (mode) {
b961ba53
JL
James Lyon2013-04-19 11:08:12 +01001146 default: return __va_stack;
1147 case x86_64_mode_integer: return __va_gen_reg;
1148 case x86_64_mode_sse: return __va_float_reg;
55ea6d3f James Lyon2013-04-19 00:40:48 +01001149 }
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001150}
1151
ef3d38c5 gus knight2015-07-29 16:57:41 -04001152/* Return the number of registers needed to return the struct, or 0 if
1153 returning via struct pointer. */
1154ST_FUNC int gfunc_sret(CType *vt, int variadic, CType *ret, int *ret_align, int *regsize)
4e04f67c seyko2015-05-14 07:32:24 +03001155{
ef3d38c5 gus knight2015-07-29 16:57:41 -04001156 int size, align, reg_count;
55ea6d3f James Lyon2013-04-19 00:40:48 +01001157 *ret_align = 1; // Never have to re-align return values for x86-64
50899e30 Michael Matz2015-03-09 00:19:59 +01001158 *regsize = 8;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001159 return (classify_x86_64_arg(vt, ret, &size, &align, &reg_count) != x86_64_mode_memory);
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001160}
1161
45987579 grischka2009-07-18 22:05:49 +02001162#define REGN 6
c998985c grischka2009-07-18 22:07:42 +02001163static const uint8_t arg_regs[REGN] = {
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001164 TREG_RDI, TREG_RSI, TREG_RDX, TREG_RCX, TREG_R8, TREG_R9
1165};
45987579 grischka2009-07-18 22:05:49 +02001166
0e17671f
JL
James Lyon2013-04-19 15:33:16 +01001167static int arg_prepare_reg(int idx) {
1168 if (idx == 2 || idx == 3)
1169 /* idx=2: r10, idx=3: r11 */
1170 return idx + 8;
1171 else
1172 return arg_regs[idx];
1173}
1174
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001175/* Generate function call. The function address is pushed first, then
1176 all the parameters in call order. This functions pops all the
1177 parameters and the function address. */
1178void gfunc_call(int nb_args)
1179{
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001180 X86_64_Mode mode;
1181 CType type;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001182 int size, align, r, args_size, stack_adjust, run_start, run_end, i, reg_count;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001183 int nb_reg_args = 0;
1184 int nb_sse_args = 0;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001185 int sse_reg, gen_reg;
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001186
6ee366e7 James Lyon2013-04-26 16:42:12 +01001187 /* calculate the number of integer/float register arguments */
ef3d38c5 gus knight2015-07-29 16:57:41 -04001188 for(i = 0; i < nb_args; i++) {
1189 mode = classify_x86_64_arg(&vtop[-i].type, NULL, &size, &align, &reg_count);
1190 if (mode == x86_64_mode_sse)
1191 nb_sse_args += reg_count;
1192 else if (mode == x86_64_mode_integer)
1193 nb_reg_args += reg_count;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001194 }
1195
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001196 /* arguments are collected in runs. Each run is a collection of 8-byte aligned arguments
1197 and ended by a 16-byte aligned argument. This is because, from the point of view of
1198 the callee, argument alignment is computed from the bottom up. */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001199 /* for struct arguments, we need to call memcpy and the function
1200 call breaks register passing arguments we are preparing.
1201 So, we process arguments which will be passed by stack first. */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001202 gen_reg = nb_reg_args;
1203 sse_reg = nb_sse_args;
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001204 run_start = 0;
1205 args_size = 0;
1206 while (run_start != nb_args) {
1207 int run_gen_reg = gen_reg, run_sse_reg = sse_reg;
89ad24e7 gus knight2015-07-29 16:53:57 -04001208
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001209 run_end = nb_args;
1210 stack_adjust = 0;
1211 for(i = run_start; (i < nb_args) && (run_end == nb_args); i++) {
ef3d38c5 gus knight2015-07-29 16:57:41 -04001212 mode = classify_x86_64_arg(&vtop[-i].type, NULL, &size, &align, &reg_count);
1213 switch (mode) {
1214 case x86_64_mode_memory:
1215 case x86_64_mode_x87:
1216 stack_arg:
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001217 if (align == 16)
1218 run_end = i;
cbce6d2b James Lyon2013-04-19 22:05:49 +01001219 else
6ee366e7 James Lyon2013-04-26 16:42:12 +01001220 stack_adjust += size;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001221 break;
1222
1223 case x86_64_mode_sse:
1224 sse_reg -= reg_count;
1225 if (sse_reg + reg_count > 8) goto stack_arg;
1226 break;
1227
1228 case x86_64_mode_integer:
1229 gen_reg -= reg_count;
1230 if (gen_reg + reg_count > REGN) goto stack_arg;
1231 break;
1232 default: break; /* nothing to be done for x86_64_mode_none */
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001233 }
1234 }
89ad24e7 gus knight2015-07-29 16:53:57 -04001235
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001236 gen_reg = run_gen_reg;
1237 sse_reg = run_sse_reg;
89ad24e7 gus knight2015-07-29 16:53:57 -04001238
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001239 /* adjust stack to align SSE boundary */
1240 if (stack_adjust &= 15) {
1241 /* fetch cpu flag before the following sub will change the value */
1242 if (vtop >= vstack && (vtop->r & VT_VALMASK) == VT_CMP)
1243 gv(RC_INT);
1244
1245 stack_adjust = 16 - stack_adjust;
1246 o(0x48);
1247 oad(0xec81, stack_adjust); /* sub $xxx, %rsp */
1248 args_size += stack_adjust;
1249 }
89ad24e7 gus knight2015-07-29 16:53:57 -04001250
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001251 for(i = run_start; i < run_end;) {
1252 /* Swap argument to top, it will possibly be changed here,
1253 and might use more temps. At the end of the loop we keep
1254 in on the stack and swap it back to its original position
1255 if it is a register. */
ef3d38c5 gus knight2015-07-29 16:57:41 -04001256 SValue tmp = vtop[0];
1257 int arg_stored = 1;
1258
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001259 vtop[0] = vtop[-i];
1260 vtop[-i] = tmp;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001261 mode = classify_x86_64_arg(&vtop->type, NULL, &size, &align, &reg_count);
89ad24e7 gus knight2015-07-29 16:53:57 -04001262
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001263 switch (vtop->type.t & VT_BTYPE) {
1264 case VT_STRUCT:
ef3d38c5 gus knight2015-07-29 16:57:41 -04001265 if (mode == x86_64_mode_sse) {
1266 if (sse_reg > 8)
1267 sse_reg -= reg_count;
1268 else
1269 arg_stored = 0;
1270 } else if (mode == x86_64_mode_integer) {
1271 if (gen_reg > REGN)
1272 gen_reg -= reg_count;
1273 else
1274 arg_stored = 0;
1275 }
1276
1277 if (arg_stored) {
1278 /* allocate the necessary size on stack */
1279 o(0x48);
1280 oad(0xec81, size); /* sub $xxx, %rsp */
1281 /* generate structure store */
1282 r = get_reg(RC_INT);
1283 orex(1, r, 0, 0x89); /* mov %rsp, r */
1284 o(0xe0 + REG_VALUE(r));
1285 vset(&vtop->type, r | VT_LVAL, 0);
1286 vswap();
1287 vstore();
1288 args_size += size;
1289 }
6ee366e7 James Lyon2013-04-26 16:42:12 +01001290 break;
89ad24e7 gus knight2015-07-29 16:53:57 -04001291
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001292 case VT_LDOUBLE:
1293 assert(0);
1294 break;
89ad24e7 gus knight2015-07-29 16:53:57 -04001295
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001296 case VT_FLOAT:
1297 case VT_DOUBLE:
ef3d38c5 gus knight2015-07-29 16:57:41 -04001298 assert(mode == x86_64_mode_sse);
1299 if (sse_reg > 8) {
1300 --sse_reg;
1301 r = gv(RC_FLOAT);
1302 o(0x50); /* push $rax */
1303 /* movq %xmmN, (%rsp) */
1304 o(0xd60f66);
1305 o(0x04 + REG_VALUE(r)*8);
1306 o(0x24);
1307 args_size += size;
1308 } else {
1309 arg_stored = 0;
1310 }
6ee366e7 James Lyon2013-04-26 16:42:12 +01001311 break;
89ad24e7 gus knight2015-07-29 16:53:57 -04001312
6ee366e7 James Lyon2013-04-26 16:42:12 +01001313 default:
ef3d38c5 gus knight2015-07-29 16:57:41 -04001314 assert(mode == x86_64_mode_integer);
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001315 /* simple type */
1316 /* XXX: implicit cast ? */
ef3d38c5 gus knight2015-07-29 16:57:41 -04001317 if (gen_reg > REGN) {
1318 --gen_reg;
1319 r = gv(RC_INT);
1320 orex(0,r,0,0x50 + REG_VALUE(r)); /* push r */
1321 args_size += size;
1322 } else {
1323 arg_stored = 0;
1324 }
6ee366e7 James Lyon2013-04-26 16:42:12 +01001325 break;
cbce6d2b James Lyon2013-04-19 22:05:49 +01001326 }
ef3d38c5 gus knight2015-07-29 16:57:41 -04001327
1328 /* And swap the argument back to it's original position. */
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001329 tmp = vtop[0];
1330 vtop[0] = vtop[-i];
1331 vtop[-i] = tmp;
1332
ef3d38c5 gus knight2015-07-29 16:57:41 -04001333 if (arg_stored) {
1334 vrotb(i+1);
1335 assert((vtop->type.t == tmp.type.t) && (vtop->r == tmp.r));
1336 vpop();
1337 --nb_args;
1338 --run_end;
1339 } else {
1340 ++i;
1341 }
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001342 }
1343
1344 /* handle 16 byte aligned arguments at end of run */
1345 run_start = i = run_end;
1346 while (i < nb_args) {
1347 /* Rotate argument to top since it will always be popped */
ef3d38c5 gus knight2015-07-29 16:57:41 -04001348 mode = classify_x86_64_arg(&vtop[-i].type, NULL, &size, &align, &reg_count);
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001349 if (align != 16)
1350 break;
1351
1352 vrotb(i+1);
89ad24e7 gus knight2015-07-29 16:53:57 -04001353
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001354 if ((vtop->type.t & VT_BTYPE) == VT_LDOUBLE) {
1355 gv(RC_ST0);
1356 oad(0xec8148, size); /* sub $xxx, %rsp */
1357 o(0x7cdb); /* fstpt 0(%rsp) */
1358 g(0x24);
1359 g(0x00);
1360 args_size += size;
1361 } else {
1362 assert(mode == x86_64_mode_memory);
1363
cbce6d2b
JL
James Lyon2013-04-19 22:05:49 +01001364 /* allocate the necessary size on stack */
1365 o(0x48);
1366 oad(0xec81, size); /* sub $xxx, %rsp */
1367 /* generate structure store */
1368 r = get_reg(RC_INT);
1369 orex(1, r, 0, 0x89); /* mov %rsp, r */
1370 o(0xe0 + REG_VALUE(r));
1371 vset(&vtop->type, r | VT_LVAL, 0);
1372 vswap();
1373 vstore();
1374 args_size += size;
1375 }
89ad24e7 gus knight2015-07-29 16:53:57 -04001376
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001377 vpop();
1378 --nb_args;
cbce6d2b James Lyon2013-04-19 22:05:49 +01001379 }
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001380 }
ef3d38c5 gus knight2015-07-29 16:57:41 -04001381
86ac6b9b Michael Matz2012-04-15 05:12:43 +02001382 /* XXX This should be superfluous. */
b5486240 grischka2009-12-20 20:33:21 +01001383 save_regs(0); /* save used temporary registers */
070b86a8 grischka2009-12-20 02:19:51 +01001384
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001385 /* then, we prepare register passing arguments.
1386 Note that we cannot set RDX and RCX in this loop because gv()
1387 may break these temporary registers. Let's use R10 and R11
1388 instead of them */
cbce6d2b
JL
James Lyon2013-04-19 22:05:49 +01001389 assert(gen_reg <= REGN);
1390 assert(sse_reg <= 8);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001391 for(i = 0; i < nb_args; i++) {
ef3d38c5 gus knight2015-07-29 16:57:41 -04001392 mode = classify_x86_64_arg(&vtop->type, &type, &size, &align, &reg_count);
55ea6d3f James Lyon2013-04-19 00:40:48 +01001393 /* Alter stack entry type so that gv() knows how to treat it */
ef3d38c5 gus knight2015-07-29 16:57:41 -04001394 vtop->type = type;
1395 if (mode == x86_64_mode_sse) {
1396 if (reg_count == 2) {
1397 sse_reg -= 2;
1398 gv(RC_FRET); /* Use pair load into xmm0 & xmm1 */
1399 if (sse_reg) { /* avoid redundant movaps %xmm0, %xmm0 */
1400 /* movaps %xmm0, %xmmN */
1401 o(0x280f);
1402 o(0xc0 + (sse_reg << 3));
1403 /* movaps %xmm1, %xmmN */
1404 o(0x280f);
1405 o(0xc1 + ((sse_reg+1) << 3));
1406 }
1407 } else {
1408 assert(reg_count == 1);
5c35ba66
JL
James Lyon2013-04-24 02:19:15 +01001409 --sse_reg;
1410 /* Load directly to register */
1411 gv(RC_XMM0 << sse_reg);
ef3d38c5 gus knight2015-07-29 16:57:41 -04001412 }
1413 } else if (mode == x86_64_mode_integer) {
1414 /* simple type */
1415 /* XXX: implicit cast ? */
1416 int d;
1417 gen_reg -= reg_count;
1418 r = gv(RC_INT);
1419 d = arg_prepare_reg(gen_reg);
1420 orex(1,d,r,0x89); /* mov */
1421 o(0xc0 + REG_VALUE(r) * 8 + REG_VALUE(d));
1422 if (reg_count == 2) {
1423 d = arg_prepare_reg(gen_reg+1);
1424 orex(1,d,vtop->r2,0x89); /* mov */
1425 o(0xc0 + REG_VALUE(vtop->r2) * 8 + REG_VALUE(d));
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001426 }
1427 }
1428 vtop--;
1429 }
cbce6d2b
JL
James Lyon2013-04-19 22:05:49 +01001430 assert(gen_reg == 0);
1431 assert(sse_reg == 0);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001432
86ac6b9b
MM
Michael Matz2012-04-15 05:12:43 +02001433 /* We shouldn't have many operands on the stack anymore, but the
1434 call address itself is still there, and it might be in %eax
1435 (or edx/ecx) currently, which the below writes would clobber.
1436 So evict all remaining operands here. */
1437 save_regs(0);
1438
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001439 /* Copy R10 and R11 into RDX and RCX, respectively */
1440 if (nb_reg_args > 2) {
1441 o(0xd2894c); /* mov %r10, %rdx */
1442 if (nb_reg_args > 3) {
1443 o(0xd9894c); /* mov %r11, %rcx */
1444 }
1445 }
1446
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001447 oad(0xb8, nb_sse_args < 8 ? nb_sse_args : 8); /* mov nb_sse_args, %eax */
1448 gcall_or_jmp(0);
1449 if (args_size)
1450 gadd_sp(args_size);
1451 vtop--;
1452}
1453
50b040ef grischka2009-12-19 22:40:28 +01001454
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001455#define FUNC_PROLOG_SIZE 11
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001456
1457static void push_arg_reg(int i) {
1458 loc -= 8;
1459 gen_modrm64(0x89, arg_regs[i], VT_LOCAL, NULL, loc);
1460}
1461
1462/* generate function prolog of type 't' */
1463void gfunc_prolog(CType *func_type)
1464{
55ea6d3f James Lyon2013-04-19 00:40:48 +01001465 X86_64_Mode mode;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001466 int i, addr, align, size, reg_count;
0f5942c6 Thomas Preud'homme2013-09-24 15:36:04 +02001467 int param_addr = 0, reg_param_index, sse_param_index;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001468 Sym *sym;
1469 CType *type;
1470
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001471 sym = func_type->ref;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001472 addr = PTR_SIZE * 2;
1473 loc = 0;
1474 ind += FUNC_PROLOG_SIZE;
1475 func_sub_sp_offset = ind;
50b040ef grischka2009-12-19 22:40:28 +01001476 func_ret_sub = 0;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001477
1478 if (func_type->ref->c == FUNC_ELLIPSIS) {
1479 int seen_reg_num, seen_sse_num, seen_stack_size;
1480 seen_reg_num = seen_sse_num = 0;
1481 /* frame pointer and return address */
1482 seen_stack_size = PTR_SIZE * 2;
1483 /* count the number of seen parameters */
1484 sym = func_type->ref;
1485 while ((sym = sym->next) != NULL) {
1486 type = &sym->type;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001487 mode = classify_x86_64_arg(type, NULL, &size, &align, &reg_count);
b961ba53
JL
James Lyon2013-04-19 11:08:12 +01001488 switch (mode) {
1489 default:
6ee366e7
JL
James Lyon2013-04-26 16:42:12 +01001490 stack_arg:
1491 seen_stack_size = ((seen_stack_size + align - 1) & -align) + size;
b961ba53 James Lyon2013-04-19 11:08:12 +01001492 break;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001493
b961ba53 James Lyon2013-04-19 11:08:12 +01001494 case x86_64_mode_integer:
ef3d38c5 gus knight2015-07-29 16:57:41 -04001495 if (seen_reg_num + reg_count <= 8) {
1496 seen_reg_num += reg_count;
1497 } else {
b961ba53 James Lyon2013-04-19 11:08:12 +01001498 seen_reg_num = 8;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001499 goto stack_arg;
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001500 }
ef3d38c5 gus knight2015-07-29 16:57:41 -04001501 break;
1502
1503 case x86_64_mode_sse:
1504 if (seen_sse_num + reg_count <= 8) {
1505 seen_sse_num += reg_count;
1506 } else {
b961ba53 James Lyon2013-04-19 11:08:12 +01001507 seen_sse_num = 8;
4e04f67c seyko2015-05-14 07:32:24 +03001508 goto stack_arg;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001509 }
b961ba53 James Lyon2013-04-19 11:08:12 +01001510 break;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001511 }
1512 }
1513
1514 loc -= 16;
1515 /* movl $0x????????, -0x10(%rbp) */
1516 o(0xf045c7);
1517 gen_le32(seen_reg_num * 8);
1518 /* movl $0x????????, -0xc(%rbp) */
1519 o(0xf445c7);
1520 gen_le32(seen_sse_num * 16 + 48);
1521 /* movl $0x????????, -0x8(%rbp) */
1522 o(0xf845c7);
1523 gen_le32(seen_stack_size);
1524
1525 /* save all register passing arguments */
1526 for (i = 0; i < 8; i++) {
1527 loc -= 16;
1528 o(0xd60f66); /* movq */
1529 gen_modrm(7 - i, VT_LOCAL, NULL, loc);
1530 /* movq $0, loc+8(%rbp) */
1531 o(0x85c748);
1532 gen_le32(loc + 8);
1533 gen_le32(0);
1534 }
45987579 grischka2009-07-18 22:05:49 +02001535 for (i = 0; i < REGN; i++) {
1536 push_arg_reg(REGN-1-i);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001537 }
1538 }
1539
1540 sym = func_type->ref;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001541 reg_param_index = 0;
1542 sse_param_index = 0;
1543
1544 /* if the function returns a structure, then add an
1545 implicit pointer parameter */
1546 func_vt = sym->type;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001547 mode = classify_x86_64_arg(&func_vt, NULL, &size, &align, &reg_count);
55ea6d3f James Lyon2013-04-19 00:40:48 +01001548 if (mode == x86_64_mode_memory) {
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001549 push_arg_reg(reg_param_index);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001550 func_vc = loc;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001551 reg_param_index++;
1552 }
1553 /* define parameters */
1554 while ((sym = sym->next) != NULL) {
1555 type = &sym->type;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001556 mode = classify_x86_64_arg(type, NULL, &size, &align, &reg_count);
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001557 switch (mode) {
1558 case x86_64_mode_sse:
ef3d38c5 gus knight2015-07-29 16:57:41 -04001559 if (sse_param_index + reg_count <= 8) {
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001560 /* save arguments passed by register */
ef3d38c5 gus knight2015-07-29 16:57:41 -04001561 loc -= reg_count * 8;
0e17671f James Lyon2013-04-19 15:33:16 +01001562 param_addr = loc;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001563 for (i = 0; i < reg_count; ++i) {
55ea6d3f James Lyon2013-04-19 00:40:48 +01001564 o(0xd60f66); /* movq */
ef3d38c5 gus knight2015-07-29 16:57:41 -04001565 gen_modrm(sse_param_index, VT_LOCAL, NULL, param_addr + i*8);
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001566 ++sse_param_index;
1567 }
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001568 } else {
6ee366e7 James Lyon2013-04-26 16:42:12 +01001569 addr = (addr + align - 1) & -align;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001570 param_addr = addr;
1571 addr += size;
1572 }
55ea6d3f James Lyon2013-04-19 00:40:48 +01001573 break;
89ad24e7 gus knight2015-07-29 16:53:57 -04001574
55ea6d3f
JL
James Lyon2013-04-19 00:40:48 +01001575 case x86_64_mode_memory:
1576 case x86_64_mode_x87:
6ee366e7 James Lyon2013-04-26 16:42:12 +01001577 addr = (addr + align - 1) & -align;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001578 param_addr = addr;
1579 addr += size;
55ea6d3f James Lyon2013-04-19 00:40:48 +01001580 break;
ef3d38c5 gus knight2015-07-29 16:57:41 -04001581
1582 case x86_64_mode_integer: {
1583 if (reg_param_index + reg_count <= REGN) {
1584 /* save arguments passed by register */
1585 loc -= reg_count * 8;
1586 param_addr = loc;
1587 for (i = 0; i < reg_count; ++i) {
1588 gen_modrm64(0x89, arg_regs[reg_param_index], VT_LOCAL, NULL, param_addr + i*8);
1589 ++reg_param_index;
1590 }
1591 } else {
1592 addr = (addr + align - 1) & -align;
1593 param_addr = addr;
1594 addr += size;
1595 }
1596 break;
1597 }
89ad24e7 gus knight2015-07-29 16:53:57 -04001598 default: break; /* nothing to be done for x86_64_mode_none */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001599 }
1600 sym_push(sym->v & ~SYM_FIELD, type,
1601 VT_LOCAL | VT_LVAL, param_addr);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001602 }
559675b9 seyko2015-04-10 15:17:22 +03001603
1604#ifdef CONFIG_TCC_BCHECK
1605 /* leave some room for bound checking code */
1606 if (tcc_state->do_bounds_check) {
1607 func_bound_offset = lbounds_section->data_offset;
1608 func_bound_ind = ind;
1609 oad(0xb8, 0); /* lbound section pointer */
89ad24e7 gus knight2015-07-29 16:53:57 -04001610 o(0xc78948); /* mov %rax,%rdi ## first arg in %rdi, this must be ptr */
1611 oad(0xb8, 0); /* call to function */
559675b9 seyko2015-04-10 15:17:22 +03001612 }
1613#endif
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001614}
1615
1616/* generate function epilog */
1617void gfunc_epilog(void)
1618{
1619 int v, saved_ind;
1620
559675b9 seyko2015-04-10 15:17:22 +03001621#ifdef CONFIG_TCC_BCHECK
1622 if (tcc_state->do_bounds_check
89ad24e7 gus knight2015-07-29 16:53:57 -04001623 && func_bound_offset != lbounds_section->data_offset)
559675b9 seyko2015-04-10 15:17:22 +03001624 {
1625 addr_t saved_ind;
1626 addr_t *bounds_ptr;
1627 Sym *sym_data;
1628
1629 /* add end of table info */
1630 bounds_ptr = section_ptr_add(lbounds_section, sizeof(addr_t));
1631 *bounds_ptr = 0;
1632
1633 /* generate bound local allocation */
ef3d38c5 gus knight2015-07-29 16:57:41 -04001634 sym_data = get_sym_ref(&char_pointer_type, lbounds_section,
559675b9 seyko2015-04-10 15:17:22 +03001635 func_bound_offset, lbounds_section->data_offset);
1636 saved_ind = ind;
1637 ind = func_bound_ind;
1638 greloc(cur_text_section, sym_data, ind + 1, R_386_32);
1639 ind = ind + 5 + 3;
1640 gen_static_call(TOK___bound_local_new);
1641 ind = saved_ind;
1642
1643 /* generate bound check local freeing */
1644 o(0x5250); /* save returned value, if any */
1645 greloc(cur_text_section, sym_data, ind + 1, R_386_32);
1646 oad(0xb8, 0); /* mov xxx, %rax */
89ad24e7 gus knight2015-07-29 16:53:57 -04001647 o(0xc78948); /* mov %rax,%rdi ## first arg in %rdi, this must be ptr */
559675b9 seyko2015-04-10 15:17:22 +03001648 gen_static_call(TOK___bound_local_delete);
1649 o(0x585a); /* restore returned value, if any */
1650 }
1651#endif
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001652 o(0xc9); /* leave */
1653 if (func_ret_sub == 0) {
1654 o(0xc3); /* ret */
1655 } else {
1656 o(0xc2); /* ret n */
1657 g(func_ret_sub);
1658 g(func_ret_sub >> 8);
1659 }
1660 /* align local size to word & save local variables */
1661 v = (-loc + 15) & -16;
1662 saved_ind = ind;
1663 ind = func_sub_sp_offset - FUNC_PROLOG_SIZE;
50b040ef grischka2009-12-19 22:40:28 +01001664 o(0xe5894855); /* push %rbp, mov %rsp, %rbp */
1665 o(0xec8148); /* sub rsp, stacksize */
1666 gen_le32(v);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001667 ind = saved_ind;
1668}
1669
50b040ef grischka2009-12-19 22:40:28 +01001670#endif /* not PE */
1671
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001672/* generate a jump to a label */
1673int gjmp(int t)
1674{
1675 return psym(0xe9, t);
1676}
1677
1678/* generate a jump to a fixed address */
1679void gjmp_addr(int a)
1680{
1681 int r;
1682 r = a - ind - 2;
1683 if (r == (char)r) {
1684 g(0xeb);
1685 g(r);
1686 } else {
1687 oad(0xe9, a - ind - 5);
1688 }
1689}
1690
1691/* generate a test. set 'inv' to invert test. Stack entry is popped */
1692int gtst(int inv, int t)
1693{
1694 int v, *p;
1695
1696 v = vtop->r & VT_VALMASK;
1697 if (v == VT_CMP) {
1698 /* fast case : can jump directly since flags are set */
89ad24e7 gus knight2015-07-29 16:53:57 -04001699 if (vtop->c.i & 0x100)
1700 {
1701 /* This was a float compare. If the parity flag is set
1702 the result was unordered. For anything except != this
1703 means false and we don't jump (anding both conditions).
1704 For != this means true (oring both).
1705 Take care about inverting the test. We need to jump
1706 to our target if the result was unordered and test wasn't NE,
1707 otherwise if unordered we don't want to jump. */
1708 vtop->c.i &= ~0x100;
1709 if (!inv == (vtop->c.i != TOK_NE))
1710 o(0x067a); /* jp +6 */
1711 else
1712 {
1713 g(0x0f);
1714 t = psym(0x8a, t); /* jp t */
1715 }
1716 }
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001717 g(0x0f);
1718 t = psym((vtop->c.i - 16) ^ inv, t);
2437ccdc seyko2015-03-03 15:51:09 +03001719 } else if (v == VT_JMP || v == VT_JMPI) {
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001720 /* && or || optimization */
1721 if ((v & 1) == inv) {
1722 /* insert vtop->c jump list in t */
1723 p = &vtop->c.i;
1724 while (*p != 0)
1725 p = (int *)(cur_text_section->data + *p);
1726 *p = t;
1727 t = vtop->c.i;
1728 } else {
1729 t = gjmp(t);
1730 gsym(vtop->c.i);
1731 }
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001732 }
1733 vtop--;
1734 return t;
1735}
1736
1737/* generate an integer binary operation */
1738void gen_opi(int op)
1739{
1740 int r, fr, opc, c;
070b86a8 grischka2009-12-20 02:19:51 +01001741 int ll, uu, cc;
1742
1743 ll = is64_type(vtop[-1].type.t);
1744 uu = (vtop[-1].type.t & VT_UNSIGNED) != 0;
1745 cc = (vtop->r & (VT_VALMASK | VT_LVAL | VT_SYM)) == VT_CONST;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001746
1747 switch(op) {
1748 case '+':
1749 case TOK_ADDC1: /* add with carry generation */
1750 opc = 0;
1751 gen_op8:
070b86a8 grischka2009-12-20 02:19:51 +01001752 if (cc && (!ll || (int)vtop->c.ll == vtop->c.ll)) {
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001753 /* constant case */
1754 vswap();
1755 r = gv(RC_INT);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001756 vswap();
1757 c = vtop->c.i;
1758 if (c == (char)c) {
1759 /* XXX: generate inc and dec for smaller code ? */
070b86a8 grischka2009-12-20 02:19:51 +01001760 orex(ll, r, 0, 0x83);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001761 o(0xc0 | (opc << 3) | REG_VALUE(r));
1762 g(c);
1763 } else {
070b86a8 grischka2009-12-20 02:19:51 +01001764 orex(ll, r, 0, 0x81);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001765 oad(0xc0 | (opc << 3) | REG_VALUE(r), c);
1766 }
1767 } else {
1768 gv2(RC_INT, RC_INT);
1769 r = vtop[-1].r;
1770 fr = vtop[0].r;
070b86a8 grischka2009-12-20 02:19:51 +01001771 orex(ll, r, fr, (opc << 3) | 0x01);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001772 o(0xc0 + REG_VALUE(r) + REG_VALUE(fr) * 8);
1773 }
1774 vtop--;
1775 if (op >= TOK_ULT && op <= TOK_GT) {
1776 vtop->r = VT_CMP;
1777 vtop->c.i = op;
1778 }
1779 break;
1780 case '-':
1781 case TOK_SUBC1: /* sub with carry generation */
1782 opc = 5;
1783 goto gen_op8;
1784 case TOK_ADDC2: /* add with carry use */
1785 opc = 2;
1786 goto gen_op8;
1787 case TOK_SUBC2: /* sub with carry use */
1788 opc = 3;
1789 goto gen_op8;
1790 case '&':
1791 opc = 4;
1792 goto gen_op8;
1793 case '^':
1794 opc = 6;
1795 goto gen_op8;
1796 case '|':
1797 opc = 1;
1798 goto gen_op8;
1799 case '*':
1800 gv2(RC_INT, RC_INT);
1801 r = vtop[-1].r;
1802 fr = vtop[0].r;
070b86a8 grischka2009-12-20 02:19:51 +01001803 orex(ll, fr, r, 0xaf0f); /* imul fr, r */
1804 o(0xc0 + REG_VALUE(fr) + REG_VALUE(r) * 8);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001805 vtop--;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001806 break;
1807 case TOK_SHL:
1808 opc = 4;
1809 goto gen_shift;
1810 case TOK_SHR:
1811 opc = 5;
1812 goto gen_shift;
1813 case TOK_SAR:
1814 opc = 7;
1815 gen_shift:
1816 opc = 0xc0 | (opc << 3);
070b86a8 grischka2009-12-20 02:19:51 +01001817 if (cc) {
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001818 /* constant case */
1819 vswap();
1820 r = gv(RC_INT);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001821 vswap();
070b86a8 grischka2009-12-20 02:19:51 +01001822 orex(ll, r, 0, 0xc1); /* shl/shr/sar $xxx, r */
1823 o(opc | REG_VALUE(r));
1824 g(vtop->c.i & (ll ? 63 : 31));
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001825 } else {
1826 /* we generate the shift in ecx */
1827 gv2(RC_INT, RC_RCX);
1828 r = vtop[-1].r;
070b86a8 grischka2009-12-20 02:19:51 +01001829 orex(ll, r, 0, 0xd3); /* shl/shr/sar %cl, r */
1830 o(opc | REG_VALUE(r));
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001831 }
1832 vtop--;
1833 break;
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001834 case TOK_UDIV:
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001835 case TOK_UMOD:
070b86a8 grischka2009-12-20 02:19:51 +01001836 uu = 1;
1837 goto divmod;
1838 case '/':
1839 case '%':
1840 case TOK_PDIV:
1841 uu = 0;
1842 divmod:
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001843 /* first operand must be in eax */
1844 /* XXX: need better constraint for second operand */
1845 gv2(RC_RAX, RC_RCX);
1846 r = vtop[-1].r;
1847 fr = vtop[0].r;
1848 vtop--;
1849 save_reg(TREG_RDX);
070b86a8 grischka2009-12-20 02:19:51 +01001850 orex(ll, 0, 0, uu ? 0xd231 : 0x99); /* xor %edx,%edx : cqto */
1851 orex(ll, fr, 0, 0xf7); /* div fr, %eax */
1852 o((uu ? 0xf0 : 0xf8) + REG_VALUE(fr));
1853 if (op == '%' || op == TOK_UMOD)
1854 r = TREG_RDX;
1855 else
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09001856 r = TREG_RAX;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001857 vtop->r = r;
1858 break;
1859 default:
1860 opc = 7;
1861 goto gen_op8;
1862 }
1863}
1864
1865void gen_opl(int op)
1866{
1867 gen_opi(op);
1868}
1869
1870/* generate a floating point operation 'v = t1 op t2' instruction. The
1871 two operands are guaranted to have the same floating point type */
946afd23 James Lyon2013-04-19 18:31:24 +01001872/* XXX: need to use ST1 too */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001873void gen_opf(int op)
1874{
1875 int a, ft, fc, swapped, r;
1876 int float_type =
946afd23 James Lyon2013-04-19 18:31:24 +01001877 (vtop->type.t & VT_BTYPE) == VT_LDOUBLE ? RC_ST0 : RC_FLOAT;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001878
1879 /* convert constants to memory references */
1880 if ((vtop[-1].r & (VT_VALMASK | VT_LVAL)) == VT_CONST) {
1881 vswap();
1882 gv(float_type);
1883 vswap();
1884 }
1885 if ((vtop[0].r & (VT_VALMASK | VT_LVAL)) == VT_CONST)
1886 gv(float_type);
1887
1888 /* must put at least one value in the floating point register */
1889 if ((vtop[-1].r & VT_LVAL) &&
1890 (vtop[0].r & VT_LVAL)) {
1891 vswap();
1892 gv(float_type);
1893 vswap();
1894 }
1895 swapped = 0;
1896 /* swap the stack if needed so that t1 is the register and t2 is
1897 the memory reference */
1898 if (vtop[-1].r & VT_LVAL) {
1899 vswap();
1900 swapped = 1;
1901 }
1902 if ((vtop->type.t & VT_BTYPE) == VT_LDOUBLE) {
1903 if (op >= TOK_ULT && op <= TOK_GT) {
1904 /* load on stack second operand */
1905 load(TREG_ST0, vtop);
1906 save_reg(TREG_RAX); /* eax is used by FP comparison code */
1907 if (op == TOK_GE || op == TOK_GT)
1908 swapped = !swapped;
1909 else if (op == TOK_EQ || op == TOK_NE)
1910 swapped = 0;
1911 if (swapped)
1912 o(0xc9d9); /* fxch %st(1) */
e0e9a2a2
TP
Thomas Preud'homme2013-12-31 23:40:21 +08001913 if (op == TOK_EQ || op == TOK_NE)
1914 o(0xe9da); /* fucompp */
1915 else
1916 o(0xd9de); /* fcompp */
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001917 o(0xe0df); /* fnstsw %ax */
1918 if (op == TOK_EQ) {
1919 o(0x45e480); /* and $0x45, %ah */
1920 o(0x40fC80); /* cmp $0x40, %ah */
1921 } else if (op == TOK_NE) {
1922 o(0x45e480); /* and $0x45, %ah */
1923 o(0x40f480); /* xor $0x40, %ah */
1924 op = TOK_NE;
1925 } else if (op == TOK_GE || op == TOK_LE) {
1926 o(0x05c4f6); /* test $0x05, %ah */
1927 op = TOK_EQ;
1928 } else {
1929 o(0x45c4f6); /* test $0x45, %ah */
1930 op = TOK_EQ;
1931 }
1932 vtop--;
1933 vtop->r = VT_CMP;
1934 vtop->c.i = op;
1935 } else {
1936 /* no memory reference possible for long double operations */
1937 load(TREG_ST0, vtop);
1938 swapped = !swapped;
1939
1940 switch(op) {
1941 default:
1942 case '+':
1943 a = 0;
1944 break;
1945 case '-':
1946 a = 4;
1947 if (swapped)
1948 a++;
1949 break;
1950 case '*':
1951 a = 1;
1952 break;
1953 case '/':
1954 a = 6;
1955 if (swapped)
1956 a++;
1957 break;
1958 }
1959 ft = vtop->type.t;
1960 fc = vtop->c.ul;
1961 o(0xde); /* fxxxp %st, %st(1) */
1962 o(0xc1 + (a << 3));
1963 vtop--;
1964 }
1965 } else {
1966 if (op >= TOK_ULT && op <= TOK_GT) {
1967 /* if saved lvalue, then we must reload it */
1968 r = vtop->r;
1969 fc = vtop->c.ul;
1970 if ((r & VT_VALMASK) == VT_LLOCAL) {
1971 SValue v1;
1972 r = get_reg(RC_INT);
a42b0291 Michael Matz2012-06-10 09:01:13 +02001973 v1.type.t = VT_PTR;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09001974 v1.r = VT_LOCAL | VT_LVAL;
1975 v1.c.ul = fc;
1976 load(r, &v1);
1977 fc = 0;
1978 }
1979
1980 if (op == TOK_EQ || op == TOK_NE) {
1981 swapped = 0;
1982 } else {
1983 if (op == TOK_LE || op == TOK_LT)
1984 swapped = !swapped;
1985 if (op == TOK_LE || op == TOK_GE) {
1986 op = 0x93; /* setae */
1987 } else {
1988 op = 0x97; /* seta */
1989 }
1990 }
1991
1992 if (swapped) {
946afd23
JL
James Lyon2013-04-19 18:31:24 +01001993 gv(RC_FLOAT);
1994 vswap();
1995 }
1996 assert(!(vtop[-1].r & VT_LVAL));
89ad24e7 gus knight2015-07-29 16:53:57 -04001997
946afd23
JL
James Lyon2013-04-19 18:31:24 +01001998 if ((vtop->type.t & VT_BTYPE) == VT_DOUBLE)
1999 o(0x66);
e0e9a2a2
TP
Thomas Preud'homme2013-12-31 23:40:21 +08002000 if (op == TOK_EQ || op == TOK_NE)
2001 o(0x2e0f); /* ucomisd */
2002 else
2003 o(0x2f0f); /* comisd */
2004
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002005 if (vtop->r & VT_LVAL) {
2006 gen_modrm(vtop[-1].r, r, vtop->sym, fc);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002007 } else {
946afd23 James Lyon2013-04-19 18:31:24 +01002008 o(0xc0 + REG_VALUE(vtop[0].r) + REG_VALUE(vtop[-1].r)*8);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002009 }
2010
2011 vtop--;
2012 vtop->r = VT_CMP;
2daae0dc Michael Matz2012-05-13 02:21:51 +02002013 vtop->c.i = op | 0x100;
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002014 } else {
946afd23 James Lyon2013-04-19 18:31:24 +01002015 assert((vtop->type.t & VT_BTYPE) != VT_LDOUBLE);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002016 switch(op) {
2017 default:
2018 case '+':
2019 a = 0;
2020 break;
2021 case '-':
2022 a = 4;
2023 break;
2024 case '*':
2025 a = 1;
2026 break;
2027 case '/':
2028 a = 6;
2029 break;
2030 }
2031 ft = vtop->type.t;
2032 fc = vtop->c.ul;
946afd23 James Lyon2013-04-19 18:31:24 +01002033 assert((ft & VT_BTYPE) != VT_LDOUBLE);
89ad24e7 gus knight2015-07-29 16:53:57 -04002034
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002035 r = vtop->r;
2036 /* if saved lvalue, then we must reload it */
2037 if ((vtop->r & VT_VALMASK) == VT_LLOCAL) {
2038 SValue v1;
2039 r = get_reg(RC_INT);
2040 v1.type.t = VT_PTR;
2041 v1.r = VT_LOCAL | VT_LVAL;
2042 v1.c.ul = fc;
2043 load(r, &v1);
2044 fc = 0;
2045 }
89ad24e7 gus knight2015-07-29 16:53:57 -04002046
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002047 assert(!(vtop[-1].r & VT_LVAL));
2048 if (swapped) {
2049 assert(vtop->r & VT_LVAL);
2050 gv(RC_FLOAT);
2051 vswap();
2052 }
89ad24e7 gus knight2015-07-29 16:53:57 -04002053
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002054 if ((ft & VT_BTYPE) == VT_DOUBLE) {
2055 o(0xf2);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002056 } else {
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002057 o(0xf3);
2058 }
2059 o(0x0f);
2060 o(0x58 + a);
89ad24e7 gus knight2015-07-29 16:53:57 -04002061
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002062 if (vtop->r & VT_LVAL) {
2063 gen_modrm(vtop[-1].r, r, vtop->sym, fc);
2064 } else {
2065 o(0xc0 + REG_VALUE(vtop[0].r) + REG_VALUE(vtop[-1].r)*8);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002066 }
946afd23 James Lyon2013-04-19 18:31:24 +01002067
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002068 vtop--;
2069 }
2070 }
2071}
2072
2073/* convert integers to fp 't' type. Must handle 'int', 'unsigned int'
2074 and 'long long' cases. */
2075void gen_cvt_itof(int t)
2076{
2077 if ((t & VT_BTYPE) == VT_LDOUBLE) {
2078 save_reg(TREG_ST0);
2079 gv(RC_INT);
2080 if ((vtop->type.t & VT_BTYPE) == VT_LLONG) {
2081 /* signed long long to float/double/long double (unsigned case
2082 is handled generically) */
2083 o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
2084 o(0x242cdf); /* fildll (%rsp) */
2085 o(0x08c48348); /* add $8, %rsp */
2086 } else if ((vtop->type.t & (VT_BTYPE | VT_UNSIGNED)) ==
2087 (VT_INT | VT_UNSIGNED)) {
2088 /* unsigned int to float/double/long double */
2089 o(0x6a); /* push $0 */
2090 g(0x00);
2091 o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
2092 o(0x242cdf); /* fildll (%rsp) */
2093 o(0x10c48348); /* add $16, %rsp */
2094 } else {
2095 /* int to float/double/long double */
2096 o(0x50 + (vtop->r & VT_VALMASK)); /* push r */
2097 o(0x2404db); /* fildl (%rsp) */
2098 o(0x08c48348); /* add $8, %rsp */
2099 }
2100 vtop->r = TREG_ST0;
2101 } else {
946afd23 James Lyon2013-04-19 18:31:24 +01002102 int r = get_reg(RC_FLOAT);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002103 gv(RC_INT);
946afd23 James Lyon2013-04-19 18:31:24 +01002104 o(0xf2 + ((t & VT_BTYPE) == VT_FLOAT?1:0));
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002105 if ((vtop->type.t & (VT_BTYPE | VT_UNSIGNED)) ==
2106 (VT_INT | VT_UNSIGNED) ||
2107 (vtop->type.t & VT_BTYPE) == VT_LLONG) {
2108 o(0x48); /* REX */
2109 }
2110 o(0x2a0f);
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002111 o(0xc0 + (vtop->r & VT_VALMASK) + REG_VALUE(r)*8); /* cvtsi2sd */
2112 vtop->r = r;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002113 }
2114}
2115
2116/* convert from one floating point type to another */
2117void gen_cvt_ftof(int t)
2118{
2119 int ft, bt, tbt;
2120
2121 ft = vtop->type.t;
2122 bt = ft & VT_BTYPE;
2123 tbt = t & VT_BTYPE;
89ad24e7 gus knight2015-07-29 16:53:57 -04002124
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002125 if (bt == VT_FLOAT) {
946afd23 James Lyon2013-04-19 18:31:24 +01002126 gv(RC_FLOAT);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002127 if (tbt == VT_DOUBLE) {
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002128 o(0x140f); /* unpcklps */
2129 o(0xc0 + REG_VALUE(vtop->r)*9);
2130 o(0x5a0f); /* cvtps2pd */
2131 o(0xc0 + REG_VALUE(vtop->r)*9);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002132 } else if (tbt == VT_LDOUBLE) {
946afd23 James Lyon2013-04-19 18:31:24 +01002133 save_reg(RC_ST0);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002134 /* movss %xmm0,-0x10(%rsp) */
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002135 o(0x110ff3);
2136 o(0x44 + REG_VALUE(vtop->r)*8);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002137 o(0xf024);
2138 o(0xf02444d9); /* flds -0x10(%rsp) */
2139 vtop->r = TREG_ST0;
2140 }
2141 } else if (bt == VT_DOUBLE) {
946afd23 James Lyon2013-04-19 18:31:24 +01002142 gv(RC_FLOAT);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002143 if (tbt == VT_FLOAT) {
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002144 o(0x140f66); /* unpcklpd */
2145 o(0xc0 + REG_VALUE(vtop->r)*9);
2146 o(0x5a0f66); /* cvtpd2ps */
2147 o(0xc0 + REG_VALUE(vtop->r)*9);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002148 } else if (tbt == VT_LDOUBLE) {
946afd23 James Lyon2013-04-19 18:31:24 +01002149 save_reg(RC_ST0);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002150 /* movsd %xmm0,-0x10(%rsp) */
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002151 o(0x110ff2);
2152 o(0x44 + REG_VALUE(vtop->r)*8);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002153 o(0xf024);
2154 o(0xf02444dd); /* fldl -0x10(%rsp) */
2155 vtop->r = TREG_ST0;
2156 }
2157 } else {
be1b6ba7 grischka2013-04-30 00:33:34 +02002158 int r;
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002159 gv(RC_ST0);
be1b6ba7 grischka2013-04-30 00:33:34 +02002160 r = get_reg(RC_FLOAT);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002161 if (tbt == VT_DOUBLE) {
2162 o(0xf0245cdd); /* fstpl -0x10(%rsp) */
2163 /* movsd -0x10(%rsp),%xmm0 */
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002164 o(0x100ff2);
2165 o(0x44 + REG_VALUE(r)*8);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002166 o(0xf024);
946afd23 James Lyon2013-04-19 18:31:24 +01002167 vtop->r = r;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002168 } else if (tbt == VT_FLOAT) {
2169 o(0xf0245cd9); /* fstps -0x10(%rsp) */
2170 /* movss -0x10(%rsp),%xmm0 */
946afd23
JL
James Lyon2013-04-19 18:31:24 +01002171 o(0x100ff3);
2172 o(0x44 + REG_VALUE(r)*8);
0a9873aa Shinichiro Hamaji2008-12-02 03:19:25 +09002173 o(0xf024);
946afd23 James Lyon2013-04-19 18:31:24 +01002174 vtop->r = r;
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002175 }
2176 }
2177}
2178
2179/* convert fp to int 't' type */
2180void gen_cvt_ftoi(int t)
2181{
2182 int ft, bt, size, r;
2183 ft = vtop->type.t;
2184 bt = ft & VT_BTYPE;
2185 if (bt == VT_LDOUBLE) {
2186 gen_cvt_ftof(VT_DOUBLE);
2187 bt = VT_DOUBLE;
2188 }
2189
946afd23 James Lyon2013-04-19 18:31:24 +01002190 gv(RC_FLOAT);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002191 if (t != VT_INT)
2192 size = 8;
2193 else
2194 size = 4;
2195
2196 r = get_reg(RC_INT);
2197 if (bt == VT_FLOAT) {
2198 o(0xf3);
2199 } else if (bt == VT_DOUBLE) {
2200 o(0xf2);
2201 } else {
2202 assert(0);
2203 }
070b86a8 grischka2009-12-20 02:19:51 +01002204 orex(size == 8, r, 0, 0x2c0f); /* cvttss2si or cvttsd2si */
946afd23 James Lyon2013-04-19 18:31:24 +01002205 o(0xc0 + REG_VALUE(vtop->r) + REG_VALUE(r)*8);
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002206 vtop->r = r;
2207}
2208
2209/* computed goto support */
2210void ggoto(void)
2211{
2212 gcall_or_jmp(1);
2213 vtop--;
2214}
2215
41b3c7a5
JL
James Lyon2013-04-27 20:39:34 +01002216/* Save the stack pointer onto the stack and return the location of its address */
2217ST_FUNC void gen_vla_sp_save(int addr) {
2218 /* mov %rsp,addr(%rbp)*/
2219 gen_modrm64(0x89, TREG_RSP, VT_LOCAL, NULL, addr);
2220}
2221
2222/* Restore the SP from a location on the stack */
2223ST_FUNC void gen_vla_sp_restore(int addr) {
2224 gen_modrm64(0x8b, TREG_RSP, VT_LOCAL, NULL, addr);
2225}
2226
2227/* Subtract from the stack pointer, and push the resulting value onto the stack */
2228ST_FUNC void gen_vla_alloc(CType *type, int align) {
2229#ifdef TCC_TARGET_PE
2230 /* alloca does more than just adjust %rsp on Windows */
2231 vpush_global_sym(&func_old_type, TOK_alloca);
2232 vswap(); /* Move alloca ref past allocation size */
2233 gfunc_call(1);
2234 vset(type, REG_IRET, 0);
2235#else
ef3d38c5 gus knight2015-07-29 16:57:41 -04002236 int r;
2237 r = gv(RC_INT); /* allocation size */
41b3c7a5
JL
James Lyon2013-04-27 20:39:34 +01002238 /* sub r,%rsp */
2239 o(0x2b48);
2240 o(0xe0 | REG_VALUE(r));
2241 /* We align to 16 bytes rather than align */
2242 /* and ~15, %rsp */
2243 o(0xf0e48348);
41b3c7a5 James Lyon2013-04-27 20:39:34 +01002244 vpop();
41b3c7a5
JL
James Lyon2013-04-27 20:39:34 +01002245#endif
2246}
2247
2248
0a9873aa
SH
Shinichiro Hamaji2008-12-02 03:19:25 +09002249/* end of x86-64 code generator */
2250/*************************************************************/
88a3ccab grischka2009-12-20 01:53:49 +01002251#endif /* ! TARGET_DEFS_ONLY */
2252/******************************************************/