|
1 /* |
|
2 * Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved. |
|
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
|
4 * |
|
5 * This code is free software; you can redistribute it and/or modify it |
|
6 * under the terms of the GNU General Public License version 2 only, as |
|
7 * published by the Free Software Foundation. |
|
8 * |
|
9 * This code is distributed in the hope that it will be useful, but WITHOUT |
|
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
12 * version 2 for more details (a copy is included in the LICENSE file that |
|
13 * accompanied this code). |
|
14 * |
|
15 * You should have received a copy of the GNU General Public License version |
|
16 * 2 along with this work; if not, write to the Free Software Foundation, |
|
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
18 * |
|
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, |
|
20 * CA 95054 USA or visit www.sun.com if you need additional information or |
|
21 * have any questions. |
|
22 * |
|
23 */ |
|
24 |
|
25 #include "incls/_precompiled.incl" |
|
26 #include "incls/_assembler_x86_32.cpp.incl" |
|
27 |
|
28 // Implementation of AddressLiteral |
|
29 |
|
30 AddressLiteral::AddressLiteral(address target, relocInfo::relocType rtype) { |
|
31 _is_lval = false; |
|
32 _target = target; |
|
33 switch (rtype) { |
|
34 case relocInfo::oop_type: |
|
35 // Oops are a special case. Normally they would be their own section |
|
36 // but in cases like icBuffer they are literals in the code stream that |
|
37 // we don't have a section for. We use none so that we get a literal address |
|
38 // which is always patchable. |
|
39 break; |
|
40 case relocInfo::external_word_type: |
|
41 _rspec = external_word_Relocation::spec(target); |
|
42 break; |
|
43 case relocInfo::internal_word_type: |
|
44 _rspec = internal_word_Relocation::spec(target); |
|
45 break; |
|
46 case relocInfo::opt_virtual_call_type: |
|
47 _rspec = opt_virtual_call_Relocation::spec(); |
|
48 break; |
|
49 case relocInfo::static_call_type: |
|
50 _rspec = static_call_Relocation::spec(); |
|
51 break; |
|
52 case relocInfo::runtime_call_type: |
|
53 _rspec = runtime_call_Relocation::spec(); |
|
54 break; |
|
55 case relocInfo::poll_type: |
|
56 case relocInfo::poll_return_type: |
|
57 _rspec = Relocation::spec_simple(rtype); |
|
58 break; |
|
59 case relocInfo::none: |
|
60 break; |
|
61 default: |
|
62 ShouldNotReachHere(); |
|
63 break; |
|
64 } |
|
65 } |
|
66 |
|
67 // Implementation of Address |
|
68 |
|
69 Address Address::make_array(ArrayAddress adr) { |
|
70 #ifdef _LP64 |
|
71 // Not implementable on 64bit machines |
|
72 // Should have been handled higher up the call chain. |
|
73 ShouldNotReachHere(); |
|
74 #else |
|
75 AddressLiteral base = adr.base(); |
|
76 Address index = adr.index(); |
|
77 assert(index._disp == 0, "must not have disp"); // maybe it can? |
|
78 Address array(index._base, index._index, index._scale, (intptr_t) base.target()); |
|
79 array._rspec = base._rspec; |
|
80 return array; |
|
81 #endif // _LP64 |
|
82 } |
|
83 |
|
84 #ifndef _LP64 |
|
85 |
|
86 // exceedingly dangerous constructor |
|
87 Address::Address(address loc, RelocationHolder spec) { |
|
88 _base = noreg; |
|
89 _index = noreg; |
|
90 _scale = no_scale; |
|
91 _disp = (intptr_t) loc; |
|
92 _rspec = spec; |
|
93 } |
|
94 #endif // _LP64 |
|
95 |
|
96 // Convert the raw encoding form into the form expected by the constructor for |
|
97 // Address. An index of 4 (rsp) corresponds to having no index, so convert |
|
98 // that to noreg for the Address constructor. |
|
99 Address Address::make_raw(int base, int index, int scale, int disp) { |
|
100 bool valid_index = index != rsp->encoding(); |
|
101 if (valid_index) { |
|
102 Address madr(as_Register(base), as_Register(index), (Address::ScaleFactor)scale, in_ByteSize(disp)); |
|
103 return madr; |
|
104 } else { |
|
105 Address madr(as_Register(base), noreg, Address::no_scale, in_ByteSize(disp)); |
|
106 return madr; |
|
107 } |
|
108 } |
|
109 |
|
110 // Implementation of Assembler |
|
111 |
|
112 int AbstractAssembler::code_fill_byte() { |
|
113 return (u_char)'\xF4'; // hlt |
|
114 } |
|
115 |
|
116 // make this go away someday |
|
117 void Assembler::emit_data(jint data, relocInfo::relocType rtype, int format) { |
|
118 if (rtype == relocInfo::none) |
|
119 emit_long(data); |
|
120 else emit_data(data, Relocation::spec_simple(rtype), format); |
|
121 } |
|
122 |
|
123 |
|
124 void Assembler::emit_data(jint data, RelocationHolder const& rspec, int format) { |
|
125 assert(imm32_operand == 0, "default format must be imm32 in this file"); |
|
126 assert(inst_mark() != NULL, "must be inside InstructionMark"); |
|
127 if (rspec.type() != relocInfo::none) { |
|
128 #ifdef ASSERT |
|
129 check_relocation(rspec, format); |
|
130 #endif |
|
131 // Do not use AbstractAssembler::relocate, which is not intended for |
|
132 // embedded words. Instead, relocate to the enclosing instruction. |
|
133 |
|
134 // hack. call32 is too wide for mask so use disp32 |
|
135 if (format == call32_operand) |
|
136 code_section()->relocate(inst_mark(), rspec, disp32_operand); |
|
137 else |
|
138 code_section()->relocate(inst_mark(), rspec, format); |
|
139 } |
|
140 emit_long(data); |
|
141 } |
|
142 |
|
143 |
|
144 void Assembler::emit_arith_b(int op1, int op2, Register dst, int imm8) { |
|
145 assert(dst->has_byte_register(), "must have byte register"); |
|
146 assert(isByte(op1) && isByte(op2), "wrong opcode"); |
|
147 assert(isByte(imm8), "not a byte"); |
|
148 assert((op1 & 0x01) == 0, "should be 8bit operation"); |
|
149 emit_byte(op1); |
|
150 emit_byte(op2 | dst->encoding()); |
|
151 emit_byte(imm8); |
|
152 } |
|
153 |
|
154 |
|
155 void Assembler::emit_arith(int op1, int op2, Register dst, int imm32) { |
|
156 assert(isByte(op1) && isByte(op2), "wrong opcode"); |
|
157 assert((op1 & 0x01) == 1, "should be 32bit operation"); |
|
158 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); |
|
159 if (is8bit(imm32)) { |
|
160 emit_byte(op1 | 0x02); // set sign bit |
|
161 emit_byte(op2 | dst->encoding()); |
|
162 emit_byte(imm32 & 0xFF); |
|
163 } else { |
|
164 emit_byte(op1); |
|
165 emit_byte(op2 | dst->encoding()); |
|
166 emit_long(imm32); |
|
167 } |
|
168 } |
|
169 |
|
170 // immediate-to-memory forms |
|
171 void Assembler::emit_arith_operand(int op1, Register rm, Address adr, int imm32) { |
|
172 assert((op1 & 0x01) == 1, "should be 32bit operation"); |
|
173 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); |
|
174 if (is8bit(imm32)) { |
|
175 emit_byte(op1 | 0x02); // set sign bit |
|
176 emit_operand(rm,adr); |
|
177 emit_byte(imm32 & 0xFF); |
|
178 } else { |
|
179 emit_byte(op1); |
|
180 emit_operand(rm,adr); |
|
181 emit_long(imm32); |
|
182 } |
|
183 } |
|
184 |
|
185 void Assembler::emit_arith(int op1, int op2, Register dst, jobject obj) { |
|
186 assert(isByte(op1) && isByte(op2), "wrong opcode"); |
|
187 assert((op1 & 0x01) == 1, "should be 32bit operation"); |
|
188 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); |
|
189 InstructionMark im(this); |
|
190 emit_byte(op1); |
|
191 emit_byte(op2 | dst->encoding()); |
|
192 emit_data((int)obj, relocInfo::oop_type, 0); |
|
193 } |
|
194 |
|
195 |
|
196 void Assembler::emit_arith(int op1, int op2, Register dst, Register src) { |
|
197 assert(isByte(op1) && isByte(op2), "wrong opcode"); |
|
198 emit_byte(op1); |
|
199 emit_byte(op2 | dst->encoding() << 3 | src->encoding()); |
|
200 } |
|
201 |
|
202 |
|
203 void Assembler::emit_operand(Register reg, |
|
204 Register base, |
|
205 Register index, |
|
206 Address::ScaleFactor scale, |
|
207 int disp, |
|
208 RelocationHolder const& rspec) { |
|
209 |
|
210 relocInfo::relocType rtype = (relocInfo::relocType) rspec.type(); |
|
211 if (base->is_valid()) { |
|
212 if (index->is_valid()) { |
|
213 assert(scale != Address::no_scale, "inconsistent address"); |
|
214 // [base + index*scale + disp] |
|
215 if (disp == 0 && rtype == relocInfo::none && base != rbp) { |
|
216 // [base + index*scale] |
|
217 // [00 reg 100][ss index base] |
|
218 assert(index != rsp, "illegal addressing mode"); |
|
219 emit_byte(0x04 | reg->encoding() << 3); |
|
220 emit_byte(scale << 6 | index->encoding() << 3 | base->encoding()); |
|
221 } else if (is8bit(disp) && rtype == relocInfo::none) { |
|
222 // [base + index*scale + imm8] |
|
223 // [01 reg 100][ss index base] imm8 |
|
224 assert(index != rsp, "illegal addressing mode"); |
|
225 emit_byte(0x44 | reg->encoding() << 3); |
|
226 emit_byte(scale << 6 | index->encoding() << 3 | base->encoding()); |
|
227 emit_byte(disp & 0xFF); |
|
228 } else { |
|
229 // [base + index*scale + imm32] |
|
230 // [10 reg 100][ss index base] imm32 |
|
231 assert(index != rsp, "illegal addressing mode"); |
|
232 emit_byte(0x84 | reg->encoding() << 3); |
|
233 emit_byte(scale << 6 | index->encoding() << 3 | base->encoding()); |
|
234 emit_data(disp, rspec, disp32_operand); |
|
235 } |
|
236 } else if (base == rsp) { |
|
237 // [esp + disp] |
|
238 if (disp == 0 && rtype == relocInfo::none) { |
|
239 // [esp] |
|
240 // [00 reg 100][00 100 100] |
|
241 emit_byte(0x04 | reg->encoding() << 3); |
|
242 emit_byte(0x24); |
|
243 } else if (is8bit(disp) && rtype == relocInfo::none) { |
|
244 // [esp + imm8] |
|
245 // [01 reg 100][00 100 100] imm8 |
|
246 emit_byte(0x44 | reg->encoding() << 3); |
|
247 emit_byte(0x24); |
|
248 emit_byte(disp & 0xFF); |
|
249 } else { |
|
250 // [esp + imm32] |
|
251 // [10 reg 100][00 100 100] imm32 |
|
252 emit_byte(0x84 | reg->encoding() << 3); |
|
253 emit_byte(0x24); |
|
254 emit_data(disp, rspec, disp32_operand); |
|
255 } |
|
256 } else { |
|
257 // [base + disp] |
|
258 assert(base != rsp, "illegal addressing mode"); |
|
259 if (disp == 0 && rtype == relocInfo::none && base != rbp) { |
|
260 // [base] |
|
261 // [00 reg base] |
|
262 assert(base != rbp, "illegal addressing mode"); |
|
263 emit_byte(0x00 | reg->encoding() << 3 | base->encoding()); |
|
264 } else if (is8bit(disp) && rtype == relocInfo::none) { |
|
265 // [base + imm8] |
|
266 // [01 reg base] imm8 |
|
267 emit_byte(0x40 | reg->encoding() << 3 | base->encoding()); |
|
268 emit_byte(disp & 0xFF); |
|
269 } else { |
|
270 // [base + imm32] |
|
271 // [10 reg base] imm32 |
|
272 emit_byte(0x80 | reg->encoding() << 3 | base->encoding()); |
|
273 emit_data(disp, rspec, disp32_operand); |
|
274 } |
|
275 } |
|
276 } else { |
|
277 if (index->is_valid()) { |
|
278 assert(scale != Address::no_scale, "inconsistent address"); |
|
279 // [index*scale + disp] |
|
280 // [00 reg 100][ss index 101] imm32 |
|
281 assert(index != rsp, "illegal addressing mode"); |
|
282 emit_byte(0x04 | reg->encoding() << 3); |
|
283 emit_byte(scale << 6 | index->encoding() << 3 | 0x05); |
|
284 emit_data(disp, rspec, disp32_operand); |
|
285 } else { |
|
286 // [disp] |
|
287 // [00 reg 101] imm32 |
|
288 emit_byte(0x05 | reg->encoding() << 3); |
|
289 emit_data(disp, rspec, disp32_operand); |
|
290 } |
|
291 } |
|
292 } |
|
293 |
|
294 // Secret local extension to Assembler::WhichOperand: |
|
295 #define end_pc_operand (_WhichOperand_limit) |
|
296 |
|
297 address Assembler::locate_operand(address inst, WhichOperand which) { |
|
298 // Decode the given instruction, and return the address of |
|
299 // an embedded 32-bit operand word. |
|
300 |
|
301 // If "which" is disp32_operand, selects the displacement portion |
|
302 // of an effective address specifier. |
|
303 // If "which" is imm32_operand, selects the trailing immediate constant. |
|
304 // If "which" is call32_operand, selects the displacement of a call or jump. |
|
305 // Caller is responsible for ensuring that there is such an operand, |
|
306 // and that it is 32 bits wide. |
|
307 |
|
308 // If "which" is end_pc_operand, find the end of the instruction. |
|
309 |
|
310 address ip = inst; |
|
311 |
|
312 debug_only(bool has_imm32 = false); |
|
313 int tail_size = 0; // other random bytes (#32, #16, etc.) at end of insn |
|
314 |
|
315 again_after_prefix: |
|
316 switch (0xFF & *ip++) { |
|
317 |
|
318 // These convenience macros generate groups of "case" labels for the switch. |
|
319 #define REP4(x) (x)+0: case (x)+1: case (x)+2: case (x)+3 |
|
320 #define REP8(x) (x)+0: case (x)+1: case (x)+2: case (x)+3: \ |
|
321 case (x)+4: case (x)+5: case (x)+6: case (x)+7 |
|
322 #define REP16(x) REP8((x)+0): \ |
|
323 case REP8((x)+8) |
|
324 |
|
325 case CS_segment: |
|
326 case SS_segment: |
|
327 case DS_segment: |
|
328 case ES_segment: |
|
329 case FS_segment: |
|
330 case GS_segment: |
|
331 assert(ip == inst+1, "only one prefix allowed"); |
|
332 goto again_after_prefix; |
|
333 |
|
334 case 0xFF: // pushl a; decl a; incl a; call a; jmp a |
|
335 case 0x88: // movb a, r |
|
336 case 0x89: // movl a, r |
|
337 case 0x8A: // movb r, a |
|
338 case 0x8B: // movl r, a |
|
339 case 0x8F: // popl a |
|
340 break; |
|
341 |
|
342 case 0x68: // pushl #32(oop?) |
|
343 if (which == end_pc_operand) return ip + 4; |
|
344 assert(which == imm32_operand, "pushl has no disp32"); |
|
345 return ip; // not produced by emit_operand |
|
346 |
|
347 case 0x66: // movw ... (size prefix) |
|
348 switch (0xFF & *ip++) { |
|
349 case 0x8B: // movw r, a |
|
350 case 0x89: // movw a, r |
|
351 break; |
|
352 case 0xC7: // movw a, #16 |
|
353 tail_size = 2; // the imm16 |
|
354 break; |
|
355 case 0x0F: // several SSE/SSE2 variants |
|
356 ip--; // reparse the 0x0F |
|
357 goto again_after_prefix; |
|
358 default: |
|
359 ShouldNotReachHere(); |
|
360 } |
|
361 break; |
|
362 |
|
363 case REP8(0xB8): // movl r, #32(oop?) |
|
364 if (which == end_pc_operand) return ip + 4; |
|
365 assert(which == imm32_operand || which == disp32_operand, ""); |
|
366 return ip; |
|
367 |
|
368 case 0x69: // imul r, a, #32 |
|
369 case 0xC7: // movl a, #32(oop?) |
|
370 tail_size = 4; |
|
371 debug_only(has_imm32 = true); // has both kinds of operands! |
|
372 break; |
|
373 |
|
374 case 0x0F: // movx..., etc. |
|
375 switch (0xFF & *ip++) { |
|
376 case 0x12: // movlps |
|
377 case 0x28: // movaps |
|
378 case 0x2E: // ucomiss |
|
379 case 0x2F: // comiss |
|
380 case 0x54: // andps |
|
381 case 0x55: // andnps |
|
382 case 0x56: // orps |
|
383 case 0x57: // xorps |
|
384 case 0x6E: // movd |
|
385 case 0x7E: // movd |
|
386 case 0xAE: // ldmxcsr a |
|
387 // amd side says it these have both operands but that doesn't |
|
388 // appear to be true. |
|
389 // debug_only(has_imm32 = true); // has both kinds of operands! |
|
390 break; |
|
391 |
|
392 case 0xAD: // shrd r, a, %cl |
|
393 case 0xAF: // imul r, a |
|
394 case 0xBE: // movsxb r, a |
|
395 case 0xBF: // movsxw r, a |
|
396 case 0xB6: // movzxb r, a |
|
397 case 0xB7: // movzxw r, a |
|
398 case REP16(0x40): // cmovl cc, r, a |
|
399 case 0xB0: // cmpxchgb |
|
400 case 0xB1: // cmpxchg |
|
401 case 0xC1: // xaddl |
|
402 case 0xC7: // cmpxchg8 |
|
403 case REP16(0x90): // setcc a |
|
404 // fall out of the switch to decode the address |
|
405 break; |
|
406 case 0xAC: // shrd r, a, #8 |
|
407 tail_size = 1; // the imm8 |
|
408 break; |
|
409 case REP16(0x80): // jcc rdisp32 |
|
410 if (which == end_pc_operand) return ip + 4; |
|
411 assert(which == call32_operand, "jcc has no disp32 or imm32"); |
|
412 return ip; |
|
413 default: |
|
414 ShouldNotReachHere(); |
|
415 } |
|
416 break; |
|
417 |
|
418 case 0x81: // addl a, #32; addl r, #32 |
|
419 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl |
|
420 // in the case of cmpl, the imm32 might be an oop |
|
421 tail_size = 4; |
|
422 debug_only(has_imm32 = true); // has both kinds of operands! |
|
423 break; |
|
424 |
|
425 case 0x85: // test r/m, r |
|
426 break; |
|
427 |
|
428 case 0x83: // addl a, #8; addl r, #8 |
|
429 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl |
|
430 tail_size = 1; |
|
431 break; |
|
432 |
|
433 case 0x9B: |
|
434 switch (0xFF & *ip++) { |
|
435 case 0xD9: // fnstcw a |
|
436 break; |
|
437 default: |
|
438 ShouldNotReachHere(); |
|
439 } |
|
440 break; |
|
441 |
|
442 case REP4(0x00): // addb a, r; addl a, r; addb r, a; addl r, a |
|
443 case REP4(0x10): // adc... |
|
444 case REP4(0x20): // and... |
|
445 case REP4(0x30): // xor... |
|
446 case REP4(0x08): // or... |
|
447 case REP4(0x18): // sbb... |
|
448 case REP4(0x28): // sub... |
|
449 case REP4(0x38): // cmp... |
|
450 case 0xF7: // mull a |
|
451 case 0x8D: // leal r, a |
|
452 case 0x87: // xchg r, a |
|
453 break; |
|
454 |
|
455 case 0xC1: // sal a, #8; sar a, #8; shl a, #8; shr a, #8 |
|
456 case 0xC6: // movb a, #8 |
|
457 case 0x80: // cmpb a, #8 |
|
458 case 0x6B: // imul r, a, #8 |
|
459 tail_size = 1; // the imm8 |
|
460 break; |
|
461 |
|
462 case 0xE8: // call rdisp32 |
|
463 case 0xE9: // jmp rdisp32 |
|
464 if (which == end_pc_operand) return ip + 4; |
|
465 assert(which == call32_operand, "call has no disp32 or imm32"); |
|
466 return ip; |
|
467 |
|
468 case 0xD1: // sal a, 1; sar a, 1; shl a, 1; shr a, 1 |
|
469 case 0xD3: // sal a, %cl; sar a, %cl; shl a, %cl; shr a, %cl |
|
470 case 0xD9: // fld_s a; fst_s a; fstp_s a; fldcw a |
|
471 case 0xDD: // fld_d a; fst_d a; fstp_d a |
|
472 case 0xDB: // fild_s a; fistp_s a; fld_x a; fstp_x a |
|
473 case 0xDF: // fild_d a; fistp_d a |
|
474 case 0xD8: // fadd_s a; fsubr_s a; fmul_s a; fdivr_s a; fcomp_s a |
|
475 case 0xDC: // fadd_d a; fsubr_d a; fmul_d a; fdivr_d a; fcomp_d a |
|
476 case 0xDE: // faddp_d a; fsubrp_d a; fmulp_d a; fdivrp_d a; fcompp_d a |
|
477 break; |
|
478 |
|
479 case 0xF3: // For SSE |
|
480 case 0xF2: // For SSE2 |
|
481 ip++; ip++; |
|
482 break; |
|
483 |
|
484 default: |
|
485 ShouldNotReachHere(); |
|
486 |
|
487 #undef REP8 |
|
488 #undef REP16 |
|
489 } |
|
490 |
|
491 assert(which != call32_operand, "instruction is not a call, jmp, or jcc"); |
|
492 assert(which != imm32_operand || has_imm32, "instruction has no imm32 field"); |
|
493 |
|
494 // parse the output of emit_operand |
|
495 int op2 = 0xFF & *ip++; |
|
496 int base = op2 & 0x07; |
|
497 int op3 = -1; |
|
498 const int b100 = 4; |
|
499 const int b101 = 5; |
|
500 if (base == b100 && (op2 >> 6) != 3) { |
|
501 op3 = 0xFF & *ip++; |
|
502 base = op3 & 0x07; // refetch the base |
|
503 } |
|
504 // now ip points at the disp (if any) |
|
505 |
|
506 switch (op2 >> 6) { |
|
507 case 0: |
|
508 // [00 reg 100][ss index base] |
|
509 // [00 reg 100][00 100 rsp] |
|
510 // [00 reg base] |
|
511 // [00 reg 100][ss index 101][disp32] |
|
512 // [00 reg 101] [disp32] |
|
513 |
|
514 if (base == b101) { |
|
515 if (which == disp32_operand) |
|
516 return ip; // caller wants the disp32 |
|
517 ip += 4; // skip the disp32 |
|
518 } |
|
519 break; |
|
520 |
|
521 case 1: |
|
522 // [01 reg 100][ss index base][disp8] |
|
523 // [01 reg 100][00 100 rsp][disp8] |
|
524 // [01 reg base] [disp8] |
|
525 ip += 1; // skip the disp8 |
|
526 break; |
|
527 |
|
528 case 2: |
|
529 // [10 reg 100][ss index base][disp32] |
|
530 // [10 reg 100][00 100 rsp][disp32] |
|
531 // [10 reg base] [disp32] |
|
532 if (which == disp32_operand) |
|
533 return ip; // caller wants the disp32 |
|
534 ip += 4; // skip the disp32 |
|
535 break; |
|
536 |
|
537 case 3: |
|
538 // [11 reg base] (not a memory addressing mode) |
|
539 break; |
|
540 } |
|
541 |
|
542 if (which == end_pc_operand) { |
|
543 return ip + tail_size; |
|
544 } |
|
545 |
|
546 assert(which == imm32_operand, "instruction has only an imm32 field"); |
|
547 return ip; |
|
548 } |
|
549 |
|
550 address Assembler::locate_next_instruction(address inst) { |
|
551 // Secretly share code with locate_operand: |
|
552 return locate_operand(inst, end_pc_operand); |
|
553 } |
|
554 |
|
555 |
|
556 #ifdef ASSERT |
|
557 void Assembler::check_relocation(RelocationHolder const& rspec, int format) { |
|
558 address inst = inst_mark(); |
|
559 assert(inst != NULL && inst < pc(), "must point to beginning of instruction"); |
|
560 address opnd; |
|
561 |
|
562 Relocation* r = rspec.reloc(); |
|
563 if (r->type() == relocInfo::none) { |
|
564 return; |
|
565 } else if (r->is_call() || format == call32_operand) { |
|
566 // assert(format == imm32_operand, "cannot specify a nonzero format"); |
|
567 opnd = locate_operand(inst, call32_operand); |
|
568 } else if (r->is_data()) { |
|
569 assert(format == imm32_operand || format == disp32_operand, "format ok"); |
|
570 opnd = locate_operand(inst, (WhichOperand)format); |
|
571 } else { |
|
572 assert(format == imm32_operand, "cannot specify a format"); |
|
573 return; |
|
574 } |
|
575 assert(opnd == pc(), "must put operand where relocs can find it"); |
|
576 } |
|
577 #endif |
|
578 |
|
579 |
|
580 |
|
581 void Assembler::emit_operand(Register reg, Address adr) { |
|
582 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec); |
|
583 } |
|
584 |
|
585 |
|
586 void Assembler::emit_farith(int b1, int b2, int i) { |
|
587 assert(isByte(b1) && isByte(b2), "wrong opcode"); |
|
588 assert(0 <= i && i < 8, "illegal stack offset"); |
|
589 emit_byte(b1); |
|
590 emit_byte(b2 + i); |
|
591 } |
|
592 |
|
593 |
|
594 void Assembler::pushad() { |
|
595 emit_byte(0x60); |
|
596 } |
|
597 |
|
598 void Assembler::popad() { |
|
599 emit_byte(0x61); |
|
600 } |
|
601 |
|
602 void Assembler::pushfd() { |
|
603 emit_byte(0x9C); |
|
604 } |
|
605 |
|
606 void Assembler::popfd() { |
|
607 emit_byte(0x9D); |
|
608 } |
|
609 |
|
610 void Assembler::pushl(int imm32) { |
|
611 emit_byte(0x68); |
|
612 emit_long(imm32); |
|
613 } |
|
614 |
|
615 #ifndef _LP64 |
|
616 void Assembler::push_literal32(int32_t imm32, RelocationHolder const& rspec) { |
|
617 InstructionMark im(this); |
|
618 emit_byte(0x68); |
|
619 emit_data(imm32, rspec, 0); |
|
620 } |
|
621 #endif // _LP64 |
|
622 |
|
623 void Assembler::pushl(Register src) { |
|
624 emit_byte(0x50 | src->encoding()); |
|
625 } |
|
626 |
|
627 |
|
628 void Assembler::pushl(Address src) { |
|
629 InstructionMark im(this); |
|
630 emit_byte(0xFF); |
|
631 emit_operand(rsi, src); |
|
632 } |
|
633 |
|
634 void Assembler::popl(Register dst) { |
|
635 emit_byte(0x58 | dst->encoding()); |
|
636 } |
|
637 |
|
638 |
|
639 void Assembler::popl(Address dst) { |
|
640 InstructionMark im(this); |
|
641 emit_byte(0x8F); |
|
642 emit_operand(rax, dst); |
|
643 } |
|
644 |
|
645 |
|
646 void Assembler::prefix(Prefix p) { |
|
647 a_byte(p); |
|
648 } |
|
649 |
|
650 |
|
651 void Assembler::movb(Register dst, Address src) { |
|
652 assert(dst->has_byte_register(), "must have byte register"); |
|
653 InstructionMark im(this); |
|
654 emit_byte(0x8A); |
|
655 emit_operand(dst, src); |
|
656 } |
|
657 |
|
658 |
|
659 void Assembler::movb(Address dst, int imm8) { |
|
660 InstructionMark im(this); |
|
661 emit_byte(0xC6); |
|
662 emit_operand(rax, dst); |
|
663 emit_byte(imm8); |
|
664 } |
|
665 |
|
666 |
|
667 void Assembler::movb(Address dst, Register src) { |
|
668 assert(src->has_byte_register(), "must have byte register"); |
|
669 InstructionMark im(this); |
|
670 emit_byte(0x88); |
|
671 emit_operand(src, dst); |
|
672 } |
|
673 |
|
674 |
|
675 void Assembler::movw(Address dst, int imm16) { |
|
676 InstructionMark im(this); |
|
677 |
|
678 emit_byte(0x66); // switch to 16-bit mode |
|
679 emit_byte(0xC7); |
|
680 emit_operand(rax, dst); |
|
681 emit_word(imm16); |
|
682 } |
|
683 |
|
684 |
|
685 void Assembler::movw(Register dst, Address src) { |
|
686 InstructionMark im(this); |
|
687 emit_byte(0x66); |
|
688 emit_byte(0x8B); |
|
689 emit_operand(dst, src); |
|
690 } |
|
691 |
|
692 |
|
693 void Assembler::movw(Address dst, Register src) { |
|
694 InstructionMark im(this); |
|
695 emit_byte(0x66); |
|
696 emit_byte(0x89); |
|
697 emit_operand(src, dst); |
|
698 } |
|
699 |
|
700 |
|
701 void Assembler::movl(Register dst, int imm32) { |
|
702 emit_byte(0xB8 | dst->encoding()); |
|
703 emit_long(imm32); |
|
704 } |
|
705 |
|
706 #ifndef _LP64 |
|
707 void Assembler::mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec) { |
|
708 |
|
709 InstructionMark im(this); |
|
710 emit_byte(0xB8 | dst->encoding()); |
|
711 emit_data((int)imm32, rspec, 0); |
|
712 } |
|
713 #endif // _LP64 |
|
714 |
|
715 void Assembler::movl(Register dst, Register src) { |
|
716 emit_byte(0x8B); |
|
717 emit_byte(0xC0 | (dst->encoding() << 3) | src->encoding()); |
|
718 } |
|
719 |
|
720 |
|
721 void Assembler::movl(Register dst, Address src) { |
|
722 InstructionMark im(this); |
|
723 emit_byte(0x8B); |
|
724 emit_operand(dst, src); |
|
725 } |
|
726 |
|
727 |
|
728 void Assembler::movl(Address dst, int imm32) { |
|
729 InstructionMark im(this); |
|
730 emit_byte(0xC7); |
|
731 emit_operand(rax, dst); |
|
732 emit_long(imm32); |
|
733 } |
|
734 |
|
735 #ifndef _LP64 |
|
736 void Assembler::mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec) { |
|
737 InstructionMark im(this); |
|
738 emit_byte(0xC7); |
|
739 emit_operand(rax, dst); |
|
740 emit_data((int)imm32, rspec, 0); |
|
741 } |
|
742 #endif // _LP64 |
|
743 |
|
744 void Assembler::movl(Address dst, Register src) { |
|
745 InstructionMark im(this); |
|
746 emit_byte(0x89); |
|
747 emit_operand(src, dst); |
|
748 } |
|
749 |
|
750 void Assembler::movsxb(Register dst, Address src) { |
|
751 InstructionMark im(this); |
|
752 emit_byte(0x0F); |
|
753 emit_byte(0xBE); |
|
754 emit_operand(dst, src); |
|
755 } |
|
756 |
|
757 void Assembler::movsxb(Register dst, Register src) { |
|
758 assert(src->has_byte_register(), "must have byte register"); |
|
759 emit_byte(0x0F); |
|
760 emit_byte(0xBE); |
|
761 emit_byte(0xC0 | (dst->encoding() << 3) | src->encoding()); |
|
762 } |
|
763 |
|
764 |
|
765 void Assembler::movsxw(Register dst, Address src) { |
|
766 InstructionMark im(this); |
|
767 emit_byte(0x0F); |
|
768 emit_byte(0xBF); |
|
769 emit_operand(dst, src); |
|
770 } |
|
771 |
|
772 |
|
773 void Assembler::movsxw(Register dst, Register src) { |
|
774 emit_byte(0x0F); |
|
775 emit_byte(0xBF); |
|
776 emit_byte(0xC0 | (dst->encoding() << 3) | src->encoding()); |
|
777 } |
|
778 |
|
779 |
|
780 void Assembler::movzxb(Register dst, Address src) { |
|
781 InstructionMark im(this); |
|
782 emit_byte(0x0F); |
|
783 emit_byte(0xB6); |
|
784 emit_operand(dst, src); |
|
785 } |
|
786 |
|
787 |
|
788 void Assembler::movzxb(Register dst, Register src) { |
|
789 assert(src->has_byte_register(), "must have byte register"); |
|
790 emit_byte(0x0F); |
|
791 emit_byte(0xB6); |
|
792 emit_byte(0xC0 | (dst->encoding() << 3) | src->encoding()); |
|
793 } |
|
794 |
|
795 |
|
796 void Assembler::movzxw(Register dst, Address src) { |
|
797 InstructionMark im(this); |
|
798 emit_byte(0x0F); |
|
799 emit_byte(0xB7); |
|
800 emit_operand(dst, src); |
|
801 } |
|
802 |
|
803 |
|
804 void Assembler::movzxw(Register dst, Register src) { |
|
805 emit_byte(0x0F); |
|
806 emit_byte(0xB7); |
|
807 emit_byte(0xC0 | (dst->encoding() << 3) | src->encoding()); |
|
808 } |
|
809 |
|
810 |
|
811 void Assembler::cmovl(Condition cc, Register dst, Register src) { |
|
812 guarantee(VM_Version::supports_cmov(), "illegal instruction"); |
|
813 emit_byte(0x0F); |
|
814 emit_byte(0x40 | cc); |
|
815 emit_byte(0xC0 | (dst->encoding() << 3) | src->encoding()); |
|
816 } |
|
817 |
|
818 |
|
819 void Assembler::cmovl(Condition cc, Register dst, Address src) { |
|
820 guarantee(VM_Version::supports_cmov(), "illegal instruction"); |
|
821 // The code below seems to be wrong - however the manual is inconclusive |
|
822 // do not use for now (remember to enable all callers when fixing this) |
|
823 Unimplemented(); |
|
824 // wrong bytes? |
|
825 InstructionMark im(this); |
|
826 emit_byte(0x0F); |
|
827 emit_byte(0x40 | cc); |
|
828 emit_operand(dst, src); |
|
829 } |
|
830 |
|
831 |
|
832 void Assembler::prefetcht0(Address src) { |
|
833 assert(VM_Version::supports_sse(), "must support"); |
|
834 InstructionMark im(this); |
|
835 emit_byte(0x0F); |
|
836 emit_byte(0x18); |
|
837 emit_operand(rcx, src); // 1, src |
|
838 } |
|
839 |
|
840 |
|
841 void Assembler::prefetcht1(Address src) { |
|
842 assert(VM_Version::supports_sse(), "must support"); |
|
843 InstructionMark im(this); |
|
844 emit_byte(0x0F); |
|
845 emit_byte(0x18); |
|
846 emit_operand(rdx, src); // 2, src |
|
847 } |
|
848 |
|
849 |
|
850 void Assembler::prefetcht2(Address src) { |
|
851 assert(VM_Version::supports_sse(), "must support"); |
|
852 InstructionMark im(this); |
|
853 emit_byte(0x0F); |
|
854 emit_byte(0x18); |
|
855 emit_operand(rbx, src); // 3, src |
|
856 } |
|
857 |
|
858 |
|
859 void Assembler::prefetchnta(Address src) { |
|
860 assert(VM_Version::supports_sse2(), "must support"); |
|
861 InstructionMark im(this); |
|
862 emit_byte(0x0F); |
|
863 emit_byte(0x18); |
|
864 emit_operand(rax, src); // 0, src |
|
865 } |
|
866 |
|
867 |
|
868 void Assembler::prefetchw(Address src) { |
|
869 assert(VM_Version::supports_3dnow(), "must support"); |
|
870 InstructionMark im(this); |
|
871 emit_byte(0x0F); |
|
872 emit_byte(0x0D); |
|
873 emit_operand(rcx, src); // 1, src |
|
874 } |
|
875 |
|
876 |
|
877 void Assembler::prefetchr(Address src) { |
|
878 assert(VM_Version::supports_3dnow(), "must support"); |
|
879 InstructionMark im(this); |
|
880 emit_byte(0x0F); |
|
881 emit_byte(0x0D); |
|
882 emit_operand(rax, src); // 0, src |
|
883 } |
|
884 |
|
885 |
|
886 void Assembler::adcl(Register dst, int imm32) { |
|
887 emit_arith(0x81, 0xD0, dst, imm32); |
|
888 } |
|
889 |
|
890 |
|
891 void Assembler::adcl(Register dst, Address src) { |
|
892 InstructionMark im(this); |
|
893 emit_byte(0x13); |
|
894 emit_operand(dst, src); |
|
895 } |
|
896 |
|
897 |
|
898 void Assembler::adcl(Register dst, Register src) { |
|
899 emit_arith(0x13, 0xC0, dst, src); |
|
900 } |
|
901 |
|
902 |
|
903 void Assembler::addl(Address dst, int imm32) { |
|
904 InstructionMark im(this); |
|
905 emit_arith_operand(0x81,rax,dst,imm32); |
|
906 } |
|
907 |
|
908 |
|
909 void Assembler::addl(Address dst, Register src) { |
|
910 InstructionMark im(this); |
|
911 emit_byte(0x01); |
|
912 emit_operand(src, dst); |
|
913 } |
|
914 |
|
915 |
|
916 void Assembler::addl(Register dst, int imm32) { |
|
917 emit_arith(0x81, 0xC0, dst, imm32); |
|
918 } |
|
919 |
|
920 |
|
921 void Assembler::addl(Register dst, Address src) { |
|
922 InstructionMark im(this); |
|
923 emit_byte(0x03); |
|
924 emit_operand(dst, src); |
|
925 } |
|
926 |
|
927 |
|
928 void Assembler::addl(Register dst, Register src) { |
|
929 emit_arith(0x03, 0xC0, dst, src); |
|
930 } |
|
931 |
|
932 |
|
933 void Assembler::andl(Register dst, int imm32) { |
|
934 emit_arith(0x81, 0xE0, dst, imm32); |
|
935 } |
|
936 |
|
937 |
|
938 void Assembler::andl(Register dst, Address src) { |
|
939 InstructionMark im(this); |
|
940 emit_byte(0x23); |
|
941 emit_operand(dst, src); |
|
942 } |
|
943 |
|
944 |
|
945 void Assembler::andl(Register dst, Register src) { |
|
946 emit_arith(0x23, 0xC0, dst, src); |
|
947 } |
|
948 |
|
949 |
|
950 void Assembler::cmpb(Address dst, int imm8) { |
|
951 InstructionMark im(this); |
|
952 emit_byte(0x80); |
|
953 emit_operand(rdi, dst); |
|
954 emit_byte(imm8); |
|
955 } |
|
956 |
|
957 void Assembler::cmpw(Address dst, int imm16) { |
|
958 InstructionMark im(this); |
|
959 emit_byte(0x66); |
|
960 emit_byte(0x81); |
|
961 emit_operand(rdi, dst); |
|
962 emit_word(imm16); |
|
963 } |
|
964 |
|
965 void Assembler::cmpl(Address dst, int imm32) { |
|
966 InstructionMark im(this); |
|
967 emit_byte(0x81); |
|
968 emit_operand(rdi, dst); |
|
969 emit_long(imm32); |
|
970 } |
|
971 |
|
972 #ifndef _LP64 |
|
973 void Assembler::cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec) { |
|
974 InstructionMark im(this); |
|
975 emit_byte(0x81); |
|
976 emit_byte(0xF8 | src1->encoding()); |
|
977 emit_data(imm32, rspec, 0); |
|
978 } |
|
979 |
|
980 void Assembler::cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec) { |
|
981 InstructionMark im(this); |
|
982 emit_byte(0x81); |
|
983 emit_operand(rdi, src1); |
|
984 emit_data(imm32, rspec, 0); |
|
985 } |
|
986 #endif // _LP64 |
|
987 |
|
988 |
|
989 void Assembler::cmpl(Register dst, int imm32) { |
|
990 emit_arith(0x81, 0xF8, dst, imm32); |
|
991 } |
|
992 |
|
993 |
|
994 void Assembler::cmpl(Register dst, Register src) { |
|
995 emit_arith(0x3B, 0xC0, dst, src); |
|
996 } |
|
997 |
|
998 |
|
999 void Assembler::cmpl(Register dst, Address src) { |
|
1000 InstructionMark im(this); |
|
1001 emit_byte(0x3B); |
|
1002 emit_operand(dst, src); |
|
1003 } |
|
1004 |
|
1005 |
|
1006 void Assembler::decl(Register dst) { |
|
1007 // Don't use it directly. Use MacroAssembler::decrement() instead. |
|
1008 emit_byte(0x48 | dst->encoding()); |
|
1009 } |
|
1010 |
|
1011 |
|
1012 void Assembler::decl(Address dst) { |
|
1013 // Don't use it directly. Use MacroAssembler::decrement() instead. |
|
1014 InstructionMark im(this); |
|
1015 emit_byte(0xFF); |
|
1016 emit_operand(rcx, dst); |
|
1017 } |
|
1018 |
|
1019 |
|
1020 void Assembler::idivl(Register src) { |
|
1021 emit_byte(0xF7); |
|
1022 emit_byte(0xF8 | src->encoding()); |
|
1023 } |
|
1024 |
|
1025 |
|
1026 void Assembler::cdql() { |
|
1027 emit_byte(0x99); |
|
1028 } |
|
1029 |
|
1030 |
|
1031 void Assembler::imull(Register dst, Register src) { |
|
1032 emit_byte(0x0F); |
|
1033 emit_byte(0xAF); |
|
1034 emit_byte(0xC0 | dst->encoding() << 3 | src->encoding()); |
|
1035 } |
|
1036 |
|
1037 |
|
1038 void Assembler::imull(Register dst, Register src, int value) { |
|
1039 if (is8bit(value)) { |
|
1040 emit_byte(0x6B); |
|
1041 emit_byte(0xC0 | dst->encoding() << 3 | src->encoding()); |
|
1042 emit_byte(value); |
|
1043 } else { |
|
1044 emit_byte(0x69); |
|
1045 emit_byte(0xC0 | dst->encoding() << 3 | src->encoding()); |
|
1046 emit_long(value); |
|
1047 } |
|
1048 } |
|
1049 |
|
1050 |
|
1051 void Assembler::incl(Register dst) { |
|
1052 // Don't use it directly. Use MacroAssembler::increment() instead. |
|
1053 emit_byte(0x40 | dst->encoding()); |
|
1054 } |
|
1055 |
|
1056 |
|
1057 void Assembler::incl(Address dst) { |
|
1058 // Don't use it directly. Use MacroAssembler::increment() instead. |
|
1059 InstructionMark im(this); |
|
1060 emit_byte(0xFF); |
|
1061 emit_operand(rax, dst); |
|
1062 } |
|
1063 |
|
1064 |
|
1065 void Assembler::leal(Register dst, Address src) { |
|
1066 InstructionMark im(this); |
|
1067 emit_byte(0x8D); |
|
1068 emit_operand(dst, src); |
|
1069 } |
|
1070 |
|
1071 void Assembler::mull(Address src) { |
|
1072 InstructionMark im(this); |
|
1073 emit_byte(0xF7); |
|
1074 emit_operand(rsp, src); |
|
1075 } |
|
1076 |
|
1077 |
|
1078 void Assembler::mull(Register src) { |
|
1079 emit_byte(0xF7); |
|
1080 emit_byte(0xE0 | src->encoding()); |
|
1081 } |
|
1082 |
|
1083 |
|
1084 void Assembler::negl(Register dst) { |
|
1085 emit_byte(0xF7); |
|
1086 emit_byte(0xD8 | dst->encoding()); |
|
1087 } |
|
1088 |
|
1089 |
|
1090 void Assembler::notl(Register dst) { |
|
1091 emit_byte(0xF7); |
|
1092 emit_byte(0xD0 | dst->encoding()); |
|
1093 } |
|
1094 |
|
1095 |
|
1096 void Assembler::orl(Address dst, int imm32) { |
|
1097 InstructionMark im(this); |
|
1098 emit_byte(0x81); |
|
1099 emit_operand(rcx, dst); |
|
1100 emit_long(imm32); |
|
1101 } |
|
1102 |
|
1103 void Assembler::orl(Register dst, int imm32) { |
|
1104 emit_arith(0x81, 0xC8, dst, imm32); |
|
1105 } |
|
1106 |
|
1107 |
|
1108 void Assembler::orl(Register dst, Address src) { |
|
1109 InstructionMark im(this); |
|
1110 emit_byte(0x0B); |
|
1111 emit_operand(dst, src); |
|
1112 } |
|
1113 |
|
1114 |
|
1115 void Assembler::orl(Register dst, Register src) { |
|
1116 emit_arith(0x0B, 0xC0, dst, src); |
|
1117 } |
|
1118 |
|
1119 |
|
1120 void Assembler::rcll(Register dst, int imm8) { |
|
1121 assert(isShiftCount(imm8), "illegal shift count"); |
|
1122 if (imm8 == 1) { |
|
1123 emit_byte(0xD1); |
|
1124 emit_byte(0xD0 | dst->encoding()); |
|
1125 } else { |
|
1126 emit_byte(0xC1); |
|
1127 emit_byte(0xD0 | dst->encoding()); |
|
1128 emit_byte(imm8); |
|
1129 } |
|
1130 } |
|
1131 |
|
1132 |
|
1133 void Assembler::sarl(Register dst, int imm8) { |
|
1134 assert(isShiftCount(imm8), "illegal shift count"); |
|
1135 if (imm8 == 1) { |
|
1136 emit_byte(0xD1); |
|
1137 emit_byte(0xF8 | dst->encoding()); |
|
1138 } else { |
|
1139 emit_byte(0xC1); |
|
1140 emit_byte(0xF8 | dst->encoding()); |
|
1141 emit_byte(imm8); |
|
1142 } |
|
1143 } |
|
1144 |
|
1145 |
|
1146 void Assembler::sarl(Register dst) { |
|
1147 emit_byte(0xD3); |
|
1148 emit_byte(0xF8 | dst->encoding()); |
|
1149 } |
|
1150 |
|
1151 |
|
1152 void Assembler::sbbl(Address dst, int imm32) { |
|
1153 InstructionMark im(this); |
|
1154 emit_arith_operand(0x81,rbx,dst,imm32); |
|
1155 } |
|
1156 |
|
1157 |
|
1158 void Assembler::sbbl(Register dst, int imm32) { |
|
1159 emit_arith(0x81, 0xD8, dst, imm32); |
|
1160 } |
|
1161 |
|
1162 |
|
1163 void Assembler::sbbl(Register dst, Address src) { |
|
1164 InstructionMark im(this); |
|
1165 emit_byte(0x1B); |
|
1166 emit_operand(dst, src); |
|
1167 } |
|
1168 |
|
1169 |
|
1170 void Assembler::sbbl(Register dst, Register src) { |
|
1171 emit_arith(0x1B, 0xC0, dst, src); |
|
1172 } |
|
1173 |
|
1174 |
|
1175 void Assembler::shldl(Register dst, Register src) { |
|
1176 emit_byte(0x0F); |
|
1177 emit_byte(0xA5); |
|
1178 emit_byte(0xC0 | src->encoding() << 3 | dst->encoding()); |
|
1179 } |
|
1180 |
|
1181 |
|
1182 void Assembler::shll(Register dst, int imm8) { |
|
1183 assert(isShiftCount(imm8), "illegal shift count"); |
|
1184 if (imm8 == 1 ) { |
|
1185 emit_byte(0xD1); |
|
1186 emit_byte(0xE0 | dst->encoding()); |
|
1187 } else { |
|
1188 emit_byte(0xC1); |
|
1189 emit_byte(0xE0 | dst->encoding()); |
|
1190 emit_byte(imm8); |
|
1191 } |
|
1192 } |
|
1193 |
|
1194 |
|
1195 void Assembler::shll(Register dst) { |
|
1196 emit_byte(0xD3); |
|
1197 emit_byte(0xE0 | dst->encoding()); |
|
1198 } |
|
1199 |
|
1200 |
|
1201 void Assembler::shrdl(Register dst, Register src) { |
|
1202 emit_byte(0x0F); |
|
1203 emit_byte(0xAD); |
|
1204 emit_byte(0xC0 | src->encoding() << 3 | dst->encoding()); |
|
1205 } |
|
1206 |
|
1207 |
|
1208 void Assembler::shrl(Register dst, int imm8) { |
|
1209 assert(isShiftCount(imm8), "illegal shift count"); |
|
1210 emit_byte(0xC1); |
|
1211 emit_byte(0xE8 | dst->encoding()); |
|
1212 emit_byte(imm8); |
|
1213 } |
|
1214 |
|
1215 |
|
1216 void Assembler::shrl(Register dst) { |
|
1217 emit_byte(0xD3); |
|
1218 emit_byte(0xE8 | dst->encoding()); |
|
1219 } |
|
1220 |
|
1221 |
|
1222 void Assembler::subl(Address dst, int imm32) { |
|
1223 if (is8bit(imm32)) { |
|
1224 InstructionMark im(this); |
|
1225 emit_byte(0x83); |
|
1226 emit_operand(rbp, dst); |
|
1227 emit_byte(imm32 & 0xFF); |
|
1228 } else { |
|
1229 InstructionMark im(this); |
|
1230 emit_byte(0x81); |
|
1231 emit_operand(rbp, dst); |
|
1232 emit_long(imm32); |
|
1233 } |
|
1234 } |
|
1235 |
|
1236 |
|
1237 void Assembler::subl(Register dst, int imm32) { |
|
1238 emit_arith(0x81, 0xE8, dst, imm32); |
|
1239 } |
|
1240 |
|
1241 |
|
1242 void Assembler::subl(Address dst, Register src) { |
|
1243 InstructionMark im(this); |
|
1244 emit_byte(0x29); |
|
1245 emit_operand(src, dst); |
|
1246 } |
|
1247 |
|
1248 |
|
1249 void Assembler::subl(Register dst, Address src) { |
|
1250 InstructionMark im(this); |
|
1251 emit_byte(0x2B); |
|
1252 emit_operand(dst, src); |
|
1253 } |
|
1254 |
|
1255 |
|
1256 void Assembler::subl(Register dst, Register src) { |
|
1257 emit_arith(0x2B, 0xC0, dst, src); |
|
1258 } |
|
1259 |
|
1260 |
|
1261 void Assembler::testb(Register dst, int imm8) { |
|
1262 assert(dst->has_byte_register(), "must have byte register"); |
|
1263 emit_arith_b(0xF6, 0xC0, dst, imm8); |
|
1264 } |
|
1265 |
|
1266 |
|
1267 void Assembler::testl(Register dst, int imm32) { |
|
1268 // not using emit_arith because test |
|
1269 // doesn't support sign-extension of |
|
1270 // 8bit operands |
|
1271 if (dst->encoding() == 0) { |
|
1272 emit_byte(0xA9); |
|
1273 } else { |
|
1274 emit_byte(0xF7); |
|
1275 emit_byte(0xC0 | dst->encoding()); |
|
1276 } |
|
1277 emit_long(imm32); |
|
1278 } |
|
1279 |
|
1280 |
|
1281 void Assembler::testl(Register dst, Register src) { |
|
1282 emit_arith(0x85, 0xC0, dst, src); |
|
1283 } |
|
1284 |
|
1285 void Assembler::testl(Register dst, Address src) { |
|
1286 InstructionMark im(this); |
|
1287 emit_byte(0x85); |
|
1288 emit_operand(dst, src); |
|
1289 } |
|
1290 |
|
1291 void Assembler::xaddl(Address dst, Register src) { |
|
1292 InstructionMark im(this); |
|
1293 emit_byte(0x0F); |
|
1294 emit_byte(0xC1); |
|
1295 emit_operand(src, dst); |
|
1296 } |
|
1297 |
|
1298 void Assembler::xorl(Register dst, int imm32) { |
|
1299 emit_arith(0x81, 0xF0, dst, imm32); |
|
1300 } |
|
1301 |
|
1302 |
|
1303 void Assembler::xorl(Register dst, Address src) { |
|
1304 InstructionMark im(this); |
|
1305 emit_byte(0x33); |
|
1306 emit_operand(dst, src); |
|
1307 } |
|
1308 |
|
1309 |
|
1310 void Assembler::xorl(Register dst, Register src) { |
|
1311 emit_arith(0x33, 0xC0, dst, src); |
|
1312 } |
|
1313 |
|
1314 |
|
1315 void Assembler::bswap(Register reg) { |
|
1316 emit_byte(0x0F); |
|
1317 emit_byte(0xC8 | reg->encoding()); |
|
1318 } |
|
1319 |
|
1320 |
|
1321 void Assembler::lock() { |
|
1322 if (Atomics & 1) { |
|
1323 // Emit either nothing, a NOP, or a NOP: prefix |
|
1324 emit_byte(0x90) ; |
|
1325 } else { |
|
1326 emit_byte(0xF0); |
|
1327 } |
|
1328 } |
|
1329 |
|
1330 |
|
1331 void Assembler::xchg(Register reg, Address adr) { |
|
1332 InstructionMark im(this); |
|
1333 emit_byte(0x87); |
|
1334 emit_operand(reg, adr); |
|
1335 } |
|
1336 |
|
1337 |
|
1338 void Assembler::xchgl(Register dst, Register src) { |
|
1339 emit_byte(0x87); |
|
1340 emit_byte(0xc0 | dst->encoding() << 3 | src->encoding()); |
|
1341 } |
|
1342 |
|
1343 |
|
1344 // The 32-bit cmpxchg compares the value at adr with the contents of rax, |
|
1345 // and stores reg into adr if so; otherwise, the value at adr is loaded into rax,. |
|
1346 // The ZF is set if the compared values were equal, and cleared otherwise. |
|
1347 void Assembler::cmpxchg(Register reg, Address adr) { |
|
1348 if (Atomics & 2) { |
|
1349 // caveat: no instructionmark, so this isn't relocatable. |
|
1350 // Emit a synthetic, non-atomic, CAS equivalent. |
|
1351 // Beware. The synthetic form sets all ICCs, not just ZF. |
|
1352 // cmpxchg r,[m] is equivalent to rax, = CAS (m, rax, r) |
|
1353 cmpl (rax, adr) ; |
|
1354 movl (rax, adr) ; |
|
1355 if (reg != rax) { |
|
1356 Label L ; |
|
1357 jcc (Assembler::notEqual, L) ; |
|
1358 movl (adr, reg) ; |
|
1359 bind (L) ; |
|
1360 } |
|
1361 } else { |
|
1362 InstructionMark im(this); |
|
1363 emit_byte(0x0F); |
|
1364 emit_byte(0xB1); |
|
1365 emit_operand(reg, adr); |
|
1366 } |
|
1367 } |
|
1368 |
|
1369 // The 64-bit cmpxchg compares the value at adr with the contents of rdx:rax, |
|
1370 // and stores rcx:rbx into adr if so; otherwise, the value at adr is loaded |
|
1371 // into rdx:rax. The ZF is set if the compared values were equal, and cleared otherwise. |
|
1372 void Assembler::cmpxchg8(Address adr) { |
|
1373 InstructionMark im(this); |
|
1374 emit_byte(0x0F); |
|
1375 emit_byte(0xc7); |
|
1376 emit_operand(rcx, adr); |
|
1377 } |
|
1378 |
|
1379 void Assembler::hlt() { |
|
1380 emit_byte(0xF4); |
|
1381 } |
|
1382 |
|
1383 |
|
1384 void Assembler::addr_nop_4() { |
|
1385 // 4 bytes: NOP DWORD PTR [EAX+0] |
|
1386 emit_byte(0x0F); |
|
1387 emit_byte(0x1F); |
|
1388 emit_byte(0x40); // emit_rm(cbuf, 0x1, EAX_enc, EAX_enc); |
|
1389 emit_byte(0); // 8-bits offset (1 byte) |
|
1390 } |
|
1391 |
|
1392 void Assembler::addr_nop_5() { |
|
1393 // 5 bytes: NOP DWORD PTR [EAX+EAX*0+0] 8-bits offset |
|
1394 emit_byte(0x0F); |
|
1395 emit_byte(0x1F); |
|
1396 emit_byte(0x44); // emit_rm(cbuf, 0x1, EAX_enc, 0x4); |
|
1397 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc); |
|
1398 emit_byte(0); // 8-bits offset (1 byte) |
|
1399 } |
|
1400 |
|
1401 void Assembler::addr_nop_7() { |
|
1402 // 7 bytes: NOP DWORD PTR [EAX+0] 32-bits offset |
|
1403 emit_byte(0x0F); |
|
1404 emit_byte(0x1F); |
|
1405 emit_byte(0x80); // emit_rm(cbuf, 0x2, EAX_enc, EAX_enc); |
|
1406 emit_long(0); // 32-bits offset (4 bytes) |
|
1407 } |
|
1408 |
|
1409 void Assembler::addr_nop_8() { |
|
1410 // 8 bytes: NOP DWORD PTR [EAX+EAX*0+0] 32-bits offset |
|
1411 emit_byte(0x0F); |
|
1412 emit_byte(0x1F); |
|
1413 emit_byte(0x84); // emit_rm(cbuf, 0x2, EAX_enc, 0x4); |
|
1414 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc); |
|
1415 emit_long(0); // 32-bits offset (4 bytes) |
|
1416 } |
|
1417 |
|
1418 void Assembler::nop(int i) { |
|
1419 assert(i > 0, " "); |
|
1420 if (UseAddressNop && VM_Version::is_intel()) { |
|
1421 // |
|
1422 // Using multi-bytes nops "0x0F 0x1F [address]" for Intel |
|
1423 // 1: 0x90 |
|
1424 // 2: 0x66 0x90 |
|
1425 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding) |
|
1426 // 4: 0x0F 0x1F 0x40 0x00 |
|
1427 // 5: 0x0F 0x1F 0x44 0x00 0x00 |
|
1428 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00 |
|
1429 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 |
|
1430 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
1431 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
1432 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
1433 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
1434 |
|
1435 // The rest coding is Intel specific - don't use consecutive address nops |
|
1436 |
|
1437 // 12: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 |
|
1438 // 13: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 |
|
1439 // 14: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 |
|
1440 // 15: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 |
|
1441 |
|
1442 while(i >= 15) { |
|
1443 // For Intel don't generate consecutive addess nops (mix with regular nops) |
|
1444 i -= 15; |
|
1445 emit_byte(0x66); // size prefix |
|
1446 emit_byte(0x66); // size prefix |
|
1447 emit_byte(0x66); // size prefix |
|
1448 addr_nop_8(); |
|
1449 emit_byte(0x66); // size prefix |
|
1450 emit_byte(0x66); // size prefix |
|
1451 emit_byte(0x66); // size prefix |
|
1452 emit_byte(0x90); // nop |
|
1453 } |
|
1454 switch (i) { |
|
1455 case 14: |
|
1456 emit_byte(0x66); // size prefix |
|
1457 case 13: |
|
1458 emit_byte(0x66); // size prefix |
|
1459 case 12: |
|
1460 addr_nop_8(); |
|
1461 emit_byte(0x66); // size prefix |
|
1462 emit_byte(0x66); // size prefix |
|
1463 emit_byte(0x66); // size prefix |
|
1464 emit_byte(0x90); // nop |
|
1465 break; |
|
1466 case 11: |
|
1467 emit_byte(0x66); // size prefix |
|
1468 case 10: |
|
1469 emit_byte(0x66); // size prefix |
|
1470 case 9: |
|
1471 emit_byte(0x66); // size prefix |
|
1472 case 8: |
|
1473 addr_nop_8(); |
|
1474 break; |
|
1475 case 7: |
|
1476 addr_nop_7(); |
|
1477 break; |
|
1478 case 6: |
|
1479 emit_byte(0x66); // size prefix |
|
1480 case 5: |
|
1481 addr_nop_5(); |
|
1482 break; |
|
1483 case 4: |
|
1484 addr_nop_4(); |
|
1485 break; |
|
1486 case 3: |
|
1487 // Don't use "0x0F 0x1F 0x00" - need patching safe padding |
|
1488 emit_byte(0x66); // size prefix |
|
1489 case 2: |
|
1490 emit_byte(0x66); // size prefix |
|
1491 case 1: |
|
1492 emit_byte(0x90); // nop |
|
1493 break; |
|
1494 default: |
|
1495 assert(i == 0, " "); |
|
1496 } |
|
1497 return; |
|
1498 } |
|
1499 if (UseAddressNop && VM_Version::is_amd()) { |
|
1500 // |
|
1501 // Using multi-bytes nops "0x0F 0x1F [address]" for AMD. |
|
1502 // 1: 0x90 |
|
1503 // 2: 0x66 0x90 |
|
1504 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding) |
|
1505 // 4: 0x0F 0x1F 0x40 0x00 |
|
1506 // 5: 0x0F 0x1F 0x44 0x00 0x00 |
|
1507 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00 |
|
1508 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 |
|
1509 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
1510 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
1511 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
1512 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
1513 |
|
1514 // The rest coding is AMD specific - use consecutive address nops |
|
1515 |
|
1516 // 12: 0x66 0x0F 0x1F 0x44 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00 |
|
1517 // 13: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00 |
|
1518 // 14: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 |
|
1519 // 15: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 |
|
1520 // 16: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 |
|
1521 // Size prefixes (0x66) are added for larger sizes |
|
1522 |
|
1523 while(i >= 22) { |
|
1524 i -= 11; |
|
1525 emit_byte(0x66); // size prefix |
|
1526 emit_byte(0x66); // size prefix |
|
1527 emit_byte(0x66); // size prefix |
|
1528 addr_nop_8(); |
|
1529 } |
|
1530 // Generate first nop for size between 21-12 |
|
1531 switch (i) { |
|
1532 case 21: |
|
1533 i -= 1; |
|
1534 emit_byte(0x66); // size prefix |
|
1535 case 20: |
|
1536 case 19: |
|
1537 i -= 1; |
|
1538 emit_byte(0x66); // size prefix |
|
1539 case 18: |
|
1540 case 17: |
|
1541 i -= 1; |
|
1542 emit_byte(0x66); // size prefix |
|
1543 case 16: |
|
1544 case 15: |
|
1545 i -= 8; |
|
1546 addr_nop_8(); |
|
1547 break; |
|
1548 case 14: |
|
1549 case 13: |
|
1550 i -= 7; |
|
1551 addr_nop_7(); |
|
1552 break; |
|
1553 case 12: |
|
1554 i -= 6; |
|
1555 emit_byte(0x66); // size prefix |
|
1556 addr_nop_5(); |
|
1557 break; |
|
1558 default: |
|
1559 assert(i < 12, " "); |
|
1560 } |
|
1561 |
|
1562 // Generate second nop for size between 11-1 |
|
1563 switch (i) { |
|
1564 case 11: |
|
1565 emit_byte(0x66); // size prefix |
|
1566 case 10: |
|
1567 emit_byte(0x66); // size prefix |
|
1568 case 9: |
|
1569 emit_byte(0x66); // size prefix |
|
1570 case 8: |
|
1571 addr_nop_8(); |
|
1572 break; |
|
1573 case 7: |
|
1574 addr_nop_7(); |
|
1575 break; |
|
1576 case 6: |
|
1577 emit_byte(0x66); // size prefix |
|
1578 case 5: |
|
1579 addr_nop_5(); |
|
1580 break; |
|
1581 case 4: |
|
1582 addr_nop_4(); |
|
1583 break; |
|
1584 case 3: |
|
1585 // Don't use "0x0F 0x1F 0x00" - need patching safe padding |
|
1586 emit_byte(0x66); // size prefix |
|
1587 case 2: |
|
1588 emit_byte(0x66); // size prefix |
|
1589 case 1: |
|
1590 emit_byte(0x90); // nop |
|
1591 break; |
|
1592 default: |
|
1593 assert(i == 0, " "); |
|
1594 } |
|
1595 return; |
|
1596 } |
|
1597 |
|
1598 // Using nops with size prefixes "0x66 0x90". |
|
1599 // From AMD Optimization Guide: |
|
1600 // 1: 0x90 |
|
1601 // 2: 0x66 0x90 |
|
1602 // 3: 0x66 0x66 0x90 |
|
1603 // 4: 0x66 0x66 0x66 0x90 |
|
1604 // 5: 0x66 0x66 0x90 0x66 0x90 |
|
1605 // 6: 0x66 0x66 0x90 0x66 0x66 0x90 |
|
1606 // 7: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 |
|
1607 // 8: 0x66 0x66 0x66 0x90 0x66 0x66 0x66 0x90 |
|
1608 // 9: 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90 |
|
1609 // 10: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90 |
|
1610 // |
|
1611 while(i > 12) { |
|
1612 i -= 4; |
|
1613 emit_byte(0x66); // size prefix |
|
1614 emit_byte(0x66); |
|
1615 emit_byte(0x66); |
|
1616 emit_byte(0x90); // nop |
|
1617 } |
|
1618 // 1 - 12 nops |
|
1619 if(i > 8) { |
|
1620 if(i > 9) { |
|
1621 i -= 1; |
|
1622 emit_byte(0x66); |
|
1623 } |
|
1624 i -= 3; |
|
1625 emit_byte(0x66); |
|
1626 emit_byte(0x66); |
|
1627 emit_byte(0x90); |
|
1628 } |
|
1629 // 1 - 8 nops |
|
1630 if(i > 4) { |
|
1631 if(i > 6) { |
|
1632 i -= 1; |
|
1633 emit_byte(0x66); |
|
1634 } |
|
1635 i -= 3; |
|
1636 emit_byte(0x66); |
|
1637 emit_byte(0x66); |
|
1638 emit_byte(0x90); |
|
1639 } |
|
1640 switch (i) { |
|
1641 case 4: |
|
1642 emit_byte(0x66); |
|
1643 case 3: |
|
1644 emit_byte(0x66); |
|
1645 case 2: |
|
1646 emit_byte(0x66); |
|
1647 case 1: |
|
1648 emit_byte(0x90); |
|
1649 break; |
|
1650 default: |
|
1651 assert(i == 0, " "); |
|
1652 } |
|
1653 } |
|
1654 |
|
1655 void Assembler::ret(int imm16) { |
|
1656 if (imm16 == 0) { |
|
1657 emit_byte(0xC3); |
|
1658 } else { |
|
1659 emit_byte(0xC2); |
|
1660 emit_word(imm16); |
|
1661 } |
|
1662 } |
|
1663 |
|
1664 |
|
1665 void Assembler::set_byte_if_not_zero(Register dst) { |
|
1666 emit_byte(0x0F); |
|
1667 emit_byte(0x95); |
|
1668 emit_byte(0xE0 | dst->encoding()); |
|
1669 } |
|
1670 |
|
1671 |
|
1672 // copies a single word from [esi] to [edi] |
|
1673 void Assembler::smovl() { |
|
1674 emit_byte(0xA5); |
|
1675 } |
|
1676 |
|
1677 // copies data from [esi] to [edi] using rcx double words (m32) |
|
1678 void Assembler::rep_movl() { |
|
1679 emit_byte(0xF3); |
|
1680 emit_byte(0xA5); |
|
1681 } |
|
1682 |
|
1683 |
|
1684 // sets rcx double words (m32) with rax, value at [edi] |
|
1685 void Assembler::rep_set() { |
|
1686 emit_byte(0xF3); |
|
1687 emit_byte(0xAB); |
|
1688 } |
|
1689 |
|
1690 // scans rcx double words (m32) at [edi] for occurance of rax, |
|
1691 void Assembler::repne_scan() { |
|
1692 emit_byte(0xF2); |
|
1693 emit_byte(0xAF); |
|
1694 } |
|
1695 |
|
1696 |
|
1697 void Assembler::setb(Condition cc, Register dst) { |
|
1698 assert(0 <= cc && cc < 16, "illegal cc"); |
|
1699 emit_byte(0x0F); |
|
1700 emit_byte(0x90 | cc); |
|
1701 emit_byte(0xC0 | dst->encoding()); |
|
1702 } |
|
1703 |
|
1704 void Assembler::cld() { |
|
1705 emit_byte(0xfc); |
|
1706 } |
|
1707 |
|
1708 void Assembler::std() { |
|
1709 emit_byte(0xfd); |
|
1710 } |
|
1711 |
|
1712 void Assembler::emit_raw (unsigned char b) { |
|
1713 emit_byte (b) ; |
|
1714 } |
|
1715 |
|
1716 // Serializes memory. |
|
1717 void Assembler::membar() { |
|
1718 // Memory barriers are only needed on multiprocessors |
|
1719 if (os::is_MP()) { |
|
1720 if( VM_Version::supports_sse2() ) { |
|
1721 emit_byte( 0x0F ); // MFENCE; faster blows no regs |
|
1722 emit_byte( 0xAE ); |
|
1723 emit_byte( 0xF0 ); |
|
1724 } else { |
|
1725 // All usable chips support "locked" instructions which suffice |
|
1726 // as barriers, and are much faster than the alternative of |
|
1727 // using cpuid instruction. We use here a locked add [esp],0. |
|
1728 // This is conveniently otherwise a no-op except for blowing |
|
1729 // flags (which we save and restore.) |
|
1730 pushfd(); // Save eflags register |
|
1731 lock(); |
|
1732 addl(Address(rsp, 0), 0);// Assert the lock# signal here |
|
1733 popfd(); // Restore eflags register |
|
1734 } |
|
1735 } |
|
1736 } |
|
1737 |
|
1738 // Identify processor type and features |
|
1739 void Assembler::cpuid() { |
|
1740 // Note: we can't assert VM_Version::supports_cpuid() here |
|
1741 // because this instruction is used in the processor |
|
1742 // identification code. |
|
1743 emit_byte( 0x0F ); |
|
1744 emit_byte( 0xA2 ); |
|
1745 } |
|
1746 |
|
1747 void Assembler::call(Label& L, relocInfo::relocType rtype) { |
|
1748 if (L.is_bound()) { |
|
1749 const int long_size = 5; |
|
1750 int offs = target(L) - pc(); |
|
1751 assert(offs <= 0, "assembler error"); |
|
1752 InstructionMark im(this); |
|
1753 // 1110 1000 #32-bit disp |
|
1754 emit_byte(0xE8); |
|
1755 emit_data(offs - long_size, rtype, 0); |
|
1756 } else { |
|
1757 InstructionMark im(this); |
|
1758 // 1110 1000 #32-bit disp |
|
1759 L.add_patch_at(code(), locator()); |
|
1760 emit_byte(0xE8); |
|
1761 emit_data(int(0), rtype, 0); |
|
1762 } |
|
1763 } |
|
1764 |
|
1765 void Assembler::call(Register dst) { |
|
1766 emit_byte(0xFF); |
|
1767 emit_byte(0xD0 | dst->encoding()); |
|
1768 } |
|
1769 |
|
1770 |
|
1771 void Assembler::call(Address adr) { |
|
1772 InstructionMark im(this); |
|
1773 relocInfo::relocType rtype = adr.reloc(); |
|
1774 if (rtype != relocInfo::runtime_call_type) { |
|
1775 emit_byte(0xFF); |
|
1776 emit_operand(rdx, adr); |
|
1777 } else { |
|
1778 assert(false, "ack"); |
|
1779 } |
|
1780 |
|
1781 } |
|
1782 |
|
1783 void Assembler::call_literal(address dest, RelocationHolder const& rspec) { |
|
1784 InstructionMark im(this); |
|
1785 emit_byte(0xE8); |
|
1786 intptr_t disp = dest - (_code_pos + sizeof(int32_t)); |
|
1787 assert(dest != NULL, "must have a target"); |
|
1788 emit_data(disp, rspec, call32_operand); |
|
1789 |
|
1790 } |
|
1791 |
|
1792 void Assembler::jmp(Register entry) { |
|
1793 emit_byte(0xFF); |
|
1794 emit_byte(0xE0 | entry->encoding()); |
|
1795 } |
|
1796 |
|
1797 |
|
1798 void Assembler::jmp(Address adr) { |
|
1799 InstructionMark im(this); |
|
1800 emit_byte(0xFF); |
|
1801 emit_operand(rsp, adr); |
|
1802 } |
|
1803 |
|
1804 void Assembler::jmp_literal(address dest, RelocationHolder const& rspec) { |
|
1805 InstructionMark im(this); |
|
1806 emit_byte(0xE9); |
|
1807 assert(dest != NULL, "must have a target"); |
|
1808 intptr_t disp = dest - (_code_pos + sizeof(int32_t)); |
|
1809 emit_data(disp, rspec.reloc(), call32_operand); |
|
1810 } |
|
1811 |
|
1812 void Assembler::jmp(Label& L, relocInfo::relocType rtype) { |
|
1813 if (L.is_bound()) { |
|
1814 address entry = target(L); |
|
1815 assert(entry != NULL, "jmp most probably wrong"); |
|
1816 InstructionMark im(this); |
|
1817 const int short_size = 2; |
|
1818 const int long_size = 5; |
|
1819 intptr_t offs = entry - _code_pos; |
|
1820 if (rtype == relocInfo::none && is8bit(offs - short_size)) { |
|
1821 emit_byte(0xEB); |
|
1822 emit_byte((offs - short_size) & 0xFF); |
|
1823 } else { |
|
1824 emit_byte(0xE9); |
|
1825 emit_long(offs - long_size); |
|
1826 } |
|
1827 } else { |
|
1828 // By default, forward jumps are always 32-bit displacements, since |
|
1829 // we can't yet know where the label will be bound. If you're sure that |
|
1830 // the forward jump will not run beyond 256 bytes, use jmpb to |
|
1831 // force an 8-bit displacement. |
|
1832 InstructionMark im(this); |
|
1833 relocate(rtype); |
|
1834 L.add_patch_at(code(), locator()); |
|
1835 emit_byte(0xE9); |
|
1836 emit_long(0); |
|
1837 } |
|
1838 } |
|
1839 |
|
1840 void Assembler::jmpb(Label& L) { |
|
1841 if (L.is_bound()) { |
|
1842 const int short_size = 2; |
|
1843 address entry = target(L); |
|
1844 assert(is8bit((entry - _code_pos) + short_size), |
|
1845 "Dispacement too large for a short jmp"); |
|
1846 assert(entry != NULL, "jmp most probably wrong"); |
|
1847 intptr_t offs = entry - _code_pos; |
|
1848 emit_byte(0xEB); |
|
1849 emit_byte((offs - short_size) & 0xFF); |
|
1850 } else { |
|
1851 InstructionMark im(this); |
|
1852 L.add_patch_at(code(), locator()); |
|
1853 emit_byte(0xEB); |
|
1854 emit_byte(0); |
|
1855 } |
|
1856 } |
|
1857 |
|
1858 void Assembler::jcc(Condition cc, Label& L, relocInfo::relocType rtype) { |
|
1859 InstructionMark im(this); |
|
1860 relocate(rtype); |
|
1861 assert((0 <= cc) && (cc < 16), "illegal cc"); |
|
1862 if (L.is_bound()) { |
|
1863 address dst = target(L); |
|
1864 assert(dst != NULL, "jcc most probably wrong"); |
|
1865 |
|
1866 const int short_size = 2; |
|
1867 const int long_size = 6; |
|
1868 int offs = (int)dst - ((int)_code_pos); |
|
1869 if (rtype == relocInfo::none && is8bit(offs - short_size)) { |
|
1870 // 0111 tttn #8-bit disp |
|
1871 emit_byte(0x70 | cc); |
|
1872 emit_byte((offs - short_size) & 0xFF); |
|
1873 } else { |
|
1874 // 0000 1111 1000 tttn #32-bit disp |
|
1875 emit_byte(0x0F); |
|
1876 emit_byte(0x80 | cc); |
|
1877 emit_long(offs - long_size); |
|
1878 } |
|
1879 } else { |
|
1880 // Note: could eliminate cond. jumps to this jump if condition |
|
1881 // is the same however, seems to be rather unlikely case. |
|
1882 // Note: use jccb() if label to be bound is very close to get |
|
1883 // an 8-bit displacement |
|
1884 L.add_patch_at(code(), locator()); |
|
1885 emit_byte(0x0F); |
|
1886 emit_byte(0x80 | cc); |
|
1887 emit_long(0); |
|
1888 } |
|
1889 } |
|
1890 |
|
1891 void Assembler::jccb(Condition cc, Label& L) { |
|
1892 if (L.is_bound()) { |
|
1893 const int short_size = 2; |
|
1894 address entry = target(L); |
|
1895 assert(is8bit((intptr_t)entry - ((intptr_t)_code_pos + short_size)), |
|
1896 "Dispacement too large for a short jmp"); |
|
1897 intptr_t offs = (intptr_t)entry - (intptr_t)_code_pos; |
|
1898 // 0111 tttn #8-bit disp |
|
1899 emit_byte(0x70 | cc); |
|
1900 emit_byte((offs - short_size) & 0xFF); |
|
1901 jcc(cc, L); |
|
1902 } else { |
|
1903 InstructionMark im(this); |
|
1904 L.add_patch_at(code(), locator()); |
|
1905 emit_byte(0x70 | cc); |
|
1906 emit_byte(0); |
|
1907 } |
|
1908 } |
|
1909 |
|
1910 // FPU instructions |
|
1911 |
|
1912 void Assembler::fld1() { |
|
1913 emit_byte(0xD9); |
|
1914 emit_byte(0xE8); |
|
1915 } |
|
1916 |
|
1917 |
|
1918 void Assembler::fldz() { |
|
1919 emit_byte(0xD9); |
|
1920 emit_byte(0xEE); |
|
1921 } |
|
1922 |
|
1923 |
|
1924 void Assembler::fld_s(Address adr) { |
|
1925 InstructionMark im(this); |
|
1926 emit_byte(0xD9); |
|
1927 emit_operand(rax, adr); |
|
1928 } |
|
1929 |
|
1930 |
|
1931 void Assembler::fld_s (int index) { |
|
1932 emit_farith(0xD9, 0xC0, index); |
|
1933 } |
|
1934 |
|
1935 |
|
1936 void Assembler::fld_d(Address adr) { |
|
1937 InstructionMark im(this); |
|
1938 emit_byte(0xDD); |
|
1939 emit_operand(rax, adr); |
|
1940 } |
|
1941 |
|
1942 |
|
1943 void Assembler::fld_x(Address adr) { |
|
1944 InstructionMark im(this); |
|
1945 emit_byte(0xDB); |
|
1946 emit_operand(rbp, adr); |
|
1947 } |
|
1948 |
|
1949 |
|
1950 void Assembler::fst_s(Address adr) { |
|
1951 InstructionMark im(this); |
|
1952 emit_byte(0xD9); |
|
1953 emit_operand(rdx, adr); |
|
1954 } |
|
1955 |
|
1956 |
|
1957 void Assembler::fst_d(Address adr) { |
|
1958 InstructionMark im(this); |
|
1959 emit_byte(0xDD); |
|
1960 emit_operand(rdx, adr); |
|
1961 } |
|
1962 |
|
1963 |
|
1964 void Assembler::fstp_s(Address adr) { |
|
1965 InstructionMark im(this); |
|
1966 emit_byte(0xD9); |
|
1967 emit_operand(rbx, adr); |
|
1968 } |
|
1969 |
|
1970 |
|
1971 void Assembler::fstp_d(Address adr) { |
|
1972 InstructionMark im(this); |
|
1973 emit_byte(0xDD); |
|
1974 emit_operand(rbx, adr); |
|
1975 } |
|
1976 |
|
1977 |
|
1978 void Assembler::fstp_x(Address adr) { |
|
1979 InstructionMark im(this); |
|
1980 emit_byte(0xDB); |
|
1981 emit_operand(rdi, adr); |
|
1982 } |
|
1983 |
|
1984 |
|
1985 void Assembler::fstp_d(int index) { |
|
1986 emit_farith(0xDD, 0xD8, index); |
|
1987 } |
|
1988 |
|
1989 |
|
1990 void Assembler::fild_s(Address adr) { |
|
1991 InstructionMark im(this); |
|
1992 emit_byte(0xDB); |
|
1993 emit_operand(rax, adr); |
|
1994 } |
|
1995 |
|
1996 |
|
1997 void Assembler::fild_d(Address adr) { |
|
1998 InstructionMark im(this); |
|
1999 emit_byte(0xDF); |
|
2000 emit_operand(rbp, adr); |
|
2001 } |
|
2002 |
|
2003 |
|
2004 void Assembler::fistp_s(Address adr) { |
|
2005 InstructionMark im(this); |
|
2006 emit_byte(0xDB); |
|
2007 emit_operand(rbx, adr); |
|
2008 } |
|
2009 |
|
2010 |
|
2011 void Assembler::fistp_d(Address adr) { |
|
2012 InstructionMark im(this); |
|
2013 emit_byte(0xDF); |
|
2014 emit_operand(rdi, adr); |
|
2015 } |
|
2016 |
|
2017 |
|
2018 void Assembler::fist_s(Address adr) { |
|
2019 InstructionMark im(this); |
|
2020 emit_byte(0xDB); |
|
2021 emit_operand(rdx, adr); |
|
2022 } |
|
2023 |
|
2024 |
|
2025 void Assembler::fabs() { |
|
2026 emit_byte(0xD9); |
|
2027 emit_byte(0xE1); |
|
2028 } |
|
2029 |
|
2030 |
|
2031 void Assembler::fldln2() { |
|
2032 emit_byte(0xD9); |
|
2033 emit_byte(0xED); |
|
2034 } |
|
2035 |
|
2036 void Assembler::fyl2x() { |
|
2037 emit_byte(0xD9); |
|
2038 emit_byte(0xF1); |
|
2039 } |
|
2040 |
|
2041 |
|
2042 void Assembler::fldlg2() { |
|
2043 emit_byte(0xD9); |
|
2044 emit_byte(0xEC); |
|
2045 } |
|
2046 |
|
2047 |
|
2048 void Assembler::flog() { |
|
2049 fldln2(); |
|
2050 fxch(); |
|
2051 fyl2x(); |
|
2052 } |
|
2053 |
|
2054 |
|
2055 void Assembler::flog10() { |
|
2056 fldlg2(); |
|
2057 fxch(); |
|
2058 fyl2x(); |
|
2059 } |
|
2060 |
|
2061 |
|
2062 void Assembler::fsin() { |
|
2063 emit_byte(0xD9); |
|
2064 emit_byte(0xFE); |
|
2065 } |
|
2066 |
|
2067 |
|
2068 void Assembler::fcos() { |
|
2069 emit_byte(0xD9); |
|
2070 emit_byte(0xFF); |
|
2071 } |
|
2072 |
|
2073 void Assembler::ftan() { |
|
2074 emit_byte(0xD9); |
|
2075 emit_byte(0xF2); |
|
2076 emit_byte(0xDD); |
|
2077 emit_byte(0xD8); |
|
2078 } |
|
2079 |
|
2080 void Assembler::fsqrt() { |
|
2081 emit_byte(0xD9); |
|
2082 emit_byte(0xFA); |
|
2083 } |
|
2084 |
|
2085 |
|
2086 void Assembler::fchs() { |
|
2087 emit_byte(0xD9); |
|
2088 emit_byte(0xE0); |
|
2089 } |
|
2090 |
|
2091 |
|
2092 void Assembler::fadd_s(Address src) { |
|
2093 InstructionMark im(this); |
|
2094 emit_byte(0xD8); |
|
2095 emit_operand(rax, src); |
|
2096 } |
|
2097 |
|
2098 |
|
2099 void Assembler::fadd_d(Address src) { |
|
2100 InstructionMark im(this); |
|
2101 emit_byte(0xDC); |
|
2102 emit_operand(rax, src); |
|
2103 } |
|
2104 |
|
2105 |
|
2106 void Assembler::fadd(int i) { |
|
2107 emit_farith(0xD8, 0xC0, i); |
|
2108 } |
|
2109 |
|
2110 |
|
2111 void Assembler::fadda(int i) { |
|
2112 emit_farith(0xDC, 0xC0, i); |
|
2113 } |
|
2114 |
|
2115 |
|
2116 void Assembler::fsub_d(Address src) { |
|
2117 InstructionMark im(this); |
|
2118 emit_byte(0xDC); |
|
2119 emit_operand(rsp, src); |
|
2120 } |
|
2121 |
|
2122 |
|
2123 void Assembler::fsub_s(Address src) { |
|
2124 InstructionMark im(this); |
|
2125 emit_byte(0xD8); |
|
2126 emit_operand(rsp, src); |
|
2127 } |
|
2128 |
|
2129 |
|
2130 void Assembler::fsubr_s(Address src) { |
|
2131 InstructionMark im(this); |
|
2132 emit_byte(0xD8); |
|
2133 emit_operand(rbp, src); |
|
2134 } |
|
2135 |
|
2136 |
|
2137 void Assembler::fsubr_d(Address src) { |
|
2138 InstructionMark im(this); |
|
2139 emit_byte(0xDC); |
|
2140 emit_operand(rbp, src); |
|
2141 } |
|
2142 |
|
2143 |
|
2144 void Assembler::fmul_s(Address src) { |
|
2145 InstructionMark im(this); |
|
2146 emit_byte(0xD8); |
|
2147 emit_operand(rcx, src); |
|
2148 } |
|
2149 |
|
2150 |
|
2151 void Assembler::fmul_d(Address src) { |
|
2152 InstructionMark im(this); |
|
2153 emit_byte(0xDC); |
|
2154 emit_operand(rcx, src); |
|
2155 } |
|
2156 |
|
2157 |
|
2158 void Assembler::fmul(int i) { |
|
2159 emit_farith(0xD8, 0xC8, i); |
|
2160 } |
|
2161 |
|
2162 |
|
2163 void Assembler::fmula(int i) { |
|
2164 emit_farith(0xDC, 0xC8, i); |
|
2165 } |
|
2166 |
|
2167 |
|
2168 void Assembler::fdiv_s(Address src) { |
|
2169 InstructionMark im(this); |
|
2170 emit_byte(0xD8); |
|
2171 emit_operand(rsi, src); |
|
2172 } |
|
2173 |
|
2174 |
|
2175 void Assembler::fdiv_d(Address src) { |
|
2176 InstructionMark im(this); |
|
2177 emit_byte(0xDC); |
|
2178 emit_operand(rsi, src); |
|
2179 } |
|
2180 |
|
2181 |
|
2182 void Assembler::fdivr_s(Address src) { |
|
2183 InstructionMark im(this); |
|
2184 emit_byte(0xD8); |
|
2185 emit_operand(rdi, src); |
|
2186 } |
|
2187 |
|
2188 |
|
2189 void Assembler::fdivr_d(Address src) { |
|
2190 InstructionMark im(this); |
|
2191 emit_byte(0xDC); |
|
2192 emit_operand(rdi, src); |
|
2193 } |
|
2194 |
|
2195 |
|
2196 void Assembler::fsub(int i) { |
|
2197 emit_farith(0xD8, 0xE0, i); |
|
2198 } |
|
2199 |
|
2200 |
|
2201 void Assembler::fsuba(int i) { |
|
2202 emit_farith(0xDC, 0xE8, i); |
|
2203 } |
|
2204 |
|
2205 |
|
2206 void Assembler::fsubr(int i) { |
|
2207 emit_farith(0xD8, 0xE8, i); |
|
2208 } |
|
2209 |
|
2210 |
|
2211 void Assembler::fsubra(int i) { |
|
2212 emit_farith(0xDC, 0xE0, i); |
|
2213 } |
|
2214 |
|
2215 |
|
2216 void Assembler::fdiv(int i) { |
|
2217 emit_farith(0xD8, 0xF0, i); |
|
2218 } |
|
2219 |
|
2220 |
|
2221 void Assembler::fdiva(int i) { |
|
2222 emit_farith(0xDC, 0xF8, i); |
|
2223 } |
|
2224 |
|
2225 |
|
2226 void Assembler::fdivr(int i) { |
|
2227 emit_farith(0xD8, 0xF8, i); |
|
2228 } |
|
2229 |
|
2230 |
|
2231 void Assembler::fdivra(int i) { |
|
2232 emit_farith(0xDC, 0xF0, i); |
|
2233 } |
|
2234 |
|
2235 |
|
2236 // Note: The Intel manual (Pentium Processor User's Manual, Vol.3, 1994) |
|
2237 // is erroneous for some of the floating-point instructions below. |
|
2238 |
|
2239 void Assembler::fdivp(int i) { |
|
2240 emit_farith(0xDE, 0xF8, i); // ST(0) <- ST(0) / ST(1) and pop (Intel manual wrong) |
|
2241 } |
|
2242 |
|
2243 |
|
2244 void Assembler::fdivrp(int i) { |
|
2245 emit_farith(0xDE, 0xF0, i); // ST(0) <- ST(1) / ST(0) and pop (Intel manual wrong) |
|
2246 } |
|
2247 |
|
2248 |
|
2249 void Assembler::fsubp(int i) { |
|
2250 emit_farith(0xDE, 0xE8, i); // ST(0) <- ST(0) - ST(1) and pop (Intel manual wrong) |
|
2251 } |
|
2252 |
|
2253 |
|
2254 void Assembler::fsubrp(int i) { |
|
2255 emit_farith(0xDE, 0xE0, i); // ST(0) <- ST(1) - ST(0) and pop (Intel manual wrong) |
|
2256 } |
|
2257 |
|
2258 |
|
2259 void Assembler::faddp(int i) { |
|
2260 emit_farith(0xDE, 0xC0, i); |
|
2261 } |
|
2262 |
|
2263 |
|
2264 void Assembler::fmulp(int i) { |
|
2265 emit_farith(0xDE, 0xC8, i); |
|
2266 } |
|
2267 |
|
2268 |
|
2269 void Assembler::fprem() { |
|
2270 emit_byte(0xD9); |
|
2271 emit_byte(0xF8); |
|
2272 } |
|
2273 |
|
2274 |
|
2275 void Assembler::fprem1() { |
|
2276 emit_byte(0xD9); |
|
2277 emit_byte(0xF5); |
|
2278 } |
|
2279 |
|
2280 |
|
2281 void Assembler::fxch(int i) { |
|
2282 emit_farith(0xD9, 0xC8, i); |
|
2283 } |
|
2284 |
|
2285 |
|
2286 void Assembler::fincstp() { |
|
2287 emit_byte(0xD9); |
|
2288 emit_byte(0xF7); |
|
2289 } |
|
2290 |
|
2291 |
|
2292 void Assembler::fdecstp() { |
|
2293 emit_byte(0xD9); |
|
2294 emit_byte(0xF6); |
|
2295 } |
|
2296 |
|
2297 |
|
2298 void Assembler::ffree(int i) { |
|
2299 emit_farith(0xDD, 0xC0, i); |
|
2300 } |
|
2301 |
|
2302 |
|
2303 void Assembler::fcomp_s(Address src) { |
|
2304 InstructionMark im(this); |
|
2305 emit_byte(0xD8); |
|
2306 emit_operand(rbx, src); |
|
2307 } |
|
2308 |
|
2309 |
|
2310 void Assembler::fcomp_d(Address src) { |
|
2311 InstructionMark im(this); |
|
2312 emit_byte(0xDC); |
|
2313 emit_operand(rbx, src); |
|
2314 } |
|
2315 |
|
2316 |
|
2317 void Assembler::fcom(int i) { |
|
2318 emit_farith(0xD8, 0xD0, i); |
|
2319 } |
|
2320 |
|
2321 |
|
2322 void Assembler::fcomp(int i) { |
|
2323 emit_farith(0xD8, 0xD8, i); |
|
2324 } |
|
2325 |
|
2326 |
|
2327 void Assembler::fcompp() { |
|
2328 emit_byte(0xDE); |
|
2329 emit_byte(0xD9); |
|
2330 } |
|
2331 |
|
2332 |
|
2333 void Assembler::fucomi(int i) { |
|
2334 // make sure the instruction is supported (introduced for P6, together with cmov) |
|
2335 guarantee(VM_Version::supports_cmov(), "illegal instruction"); |
|
2336 emit_farith(0xDB, 0xE8, i); |
|
2337 } |
|
2338 |
|
2339 |
|
2340 void Assembler::fucomip(int i) { |
|
2341 // make sure the instruction is supported (introduced for P6, together with cmov) |
|
2342 guarantee(VM_Version::supports_cmov(), "illegal instruction"); |
|
2343 emit_farith(0xDF, 0xE8, i); |
|
2344 } |
|
2345 |
|
2346 |
|
2347 void Assembler::ftst() { |
|
2348 emit_byte(0xD9); |
|
2349 emit_byte(0xE4); |
|
2350 } |
|
2351 |
|
2352 |
|
2353 void Assembler::fnstsw_ax() { |
|
2354 emit_byte(0xdF); |
|
2355 emit_byte(0xE0); |
|
2356 } |
|
2357 |
|
2358 |
|
2359 void Assembler::fwait() { |
|
2360 emit_byte(0x9B); |
|
2361 } |
|
2362 |
|
2363 |
|
2364 void Assembler::finit() { |
|
2365 emit_byte(0x9B); |
|
2366 emit_byte(0xDB); |
|
2367 emit_byte(0xE3); |
|
2368 } |
|
2369 |
|
2370 |
|
2371 void Assembler::fldcw(Address src) { |
|
2372 InstructionMark im(this); |
|
2373 emit_byte(0xd9); |
|
2374 emit_operand(rbp, src); |
|
2375 } |
|
2376 |
|
2377 |
|
2378 void Assembler::fnstcw(Address src) { |
|
2379 InstructionMark im(this); |
|
2380 emit_byte(0x9B); |
|
2381 emit_byte(0xD9); |
|
2382 emit_operand(rdi, src); |
|
2383 } |
|
2384 |
|
2385 void Assembler::fnsave(Address dst) { |
|
2386 InstructionMark im(this); |
|
2387 emit_byte(0xDD); |
|
2388 emit_operand(rsi, dst); |
|
2389 } |
|
2390 |
|
2391 |
|
2392 void Assembler::frstor(Address src) { |
|
2393 InstructionMark im(this); |
|
2394 emit_byte(0xDD); |
|
2395 emit_operand(rsp, src); |
|
2396 } |
|
2397 |
|
2398 |
|
2399 void Assembler::fldenv(Address src) { |
|
2400 InstructionMark im(this); |
|
2401 emit_byte(0xD9); |
|
2402 emit_operand(rsp, src); |
|
2403 } |
|
2404 |
|
2405 |
|
2406 void Assembler::sahf() { |
|
2407 emit_byte(0x9E); |
|
2408 } |
|
2409 |
|
2410 // MMX operations |
|
2411 void Assembler::emit_operand(MMXRegister reg, Address adr) { |
|
2412 emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec); |
|
2413 } |
|
2414 |
|
2415 void Assembler::movq( MMXRegister dst, Address src ) { |
|
2416 assert( VM_Version::supports_mmx(), "" ); |
|
2417 emit_byte(0x0F); |
|
2418 emit_byte(0x6F); |
|
2419 emit_operand(dst,src); |
|
2420 } |
|
2421 |
|
2422 void Assembler::movq( Address dst, MMXRegister src ) { |
|
2423 assert( VM_Version::supports_mmx(), "" ); |
|
2424 emit_byte(0x0F); |
|
2425 emit_byte(0x7F); |
|
2426 emit_operand(src,dst); |
|
2427 } |
|
2428 |
|
2429 void Assembler::emms() { |
|
2430 emit_byte(0x0F); |
|
2431 emit_byte(0x77); |
|
2432 } |
|
2433 |
|
2434 |
|
2435 |
|
2436 |
|
2437 // SSE and SSE2 instructions |
|
2438 inline void Assembler::emit_sse_operand(XMMRegister reg, Address adr) { |
|
2439 assert(((Register)reg)->encoding() == reg->encoding(), "otherwise typecast is invalid"); |
|
2440 emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec); |
|
2441 } |
|
2442 inline void Assembler::emit_sse_operand(Register reg, Address adr) { |
|
2443 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec); |
|
2444 } |
|
2445 |
|
2446 inline void Assembler::emit_sse_operand(XMMRegister dst, XMMRegister src) { |
|
2447 emit_byte(0xC0 | dst->encoding() << 3 | src->encoding()); |
|
2448 } |
|
2449 inline void Assembler::emit_sse_operand(XMMRegister dst, Register src) { |
|
2450 emit_byte(0xC0 | dst->encoding() << 3 | src->encoding()); |
|
2451 } |
|
2452 inline void Assembler::emit_sse_operand(Register dst, XMMRegister src) { |
|
2453 emit_byte(0xC0 | dst->encoding() << 3 | src->encoding()); |
|
2454 } |
|
2455 |
|
2456 |
|
2457 // Macro for creation of SSE2 instructions |
|
2458 // The SSE2 instricution set is highly regular, so this macro saves |
|
2459 // a lot of cut&paste |
|
2460 // Each macro expansion creates two methods (same name with different |
|
2461 // parameter list) |
|
2462 // |
|
2463 // Macro parameters: |
|
2464 // * name: name of the created methods |
|
2465 // * sse_version: either sse or sse2 for the assertion if instruction supported by processor |
|
2466 // * prefix: first opcode byte of the instruction (or 0 if no prefix byte) |
|
2467 // * opcode: last opcode byte of the instruction |
|
2468 // * conversion instruction have parameters of type Register instead of XMMRegister, |
|
2469 // so this can also configured with macro parameters |
|
2470 #define emit_sse_instruction(name, sse_version, prefix, opcode, dst_register_type, src_register_type) \ |
|
2471 \ |
|
2472 void Assembler:: name (dst_register_type dst, Address src) { \ |
|
2473 assert(VM_Version::supports_##sse_version(), ""); \ |
|
2474 \ |
|
2475 InstructionMark im(this); \ |
|
2476 if (prefix != 0) emit_byte(prefix); \ |
|
2477 emit_byte(0x0F); \ |
|
2478 emit_byte(opcode); \ |
|
2479 emit_sse_operand(dst, src); \ |
|
2480 } \ |
|
2481 \ |
|
2482 void Assembler:: name (dst_register_type dst, src_register_type src) { \ |
|
2483 assert(VM_Version::supports_##sse_version(), ""); \ |
|
2484 \ |
|
2485 if (prefix != 0) emit_byte(prefix); \ |
|
2486 emit_byte(0x0F); \ |
|
2487 emit_byte(opcode); \ |
|
2488 emit_sse_operand(dst, src); \ |
|
2489 } \ |
|
2490 |
|
2491 emit_sse_instruction(addss, sse, 0xF3, 0x58, XMMRegister, XMMRegister); |
|
2492 emit_sse_instruction(addsd, sse2, 0xF2, 0x58, XMMRegister, XMMRegister) |
|
2493 emit_sse_instruction(subss, sse, 0xF3, 0x5C, XMMRegister, XMMRegister) |
|
2494 emit_sse_instruction(subsd, sse2, 0xF2, 0x5C, XMMRegister, XMMRegister) |
|
2495 emit_sse_instruction(mulss, sse, 0xF3, 0x59, XMMRegister, XMMRegister) |
|
2496 emit_sse_instruction(mulsd, sse2, 0xF2, 0x59, XMMRegister, XMMRegister) |
|
2497 emit_sse_instruction(divss, sse, 0xF3, 0x5E, XMMRegister, XMMRegister) |
|
2498 emit_sse_instruction(divsd, sse2, 0xF2, 0x5E, XMMRegister, XMMRegister) |
|
2499 emit_sse_instruction(sqrtss, sse, 0xF3, 0x51, XMMRegister, XMMRegister) |
|
2500 emit_sse_instruction(sqrtsd, sse2, 0xF2, 0x51, XMMRegister, XMMRegister) |
|
2501 |
|
2502 emit_sse_instruction(pxor, sse2, 0x66, 0xEF, XMMRegister, XMMRegister) |
|
2503 |
|
2504 emit_sse_instruction(comiss, sse, 0, 0x2F, XMMRegister, XMMRegister) |
|
2505 emit_sse_instruction(comisd, sse2, 0x66, 0x2F, XMMRegister, XMMRegister) |
|
2506 emit_sse_instruction(ucomiss, sse, 0, 0x2E, XMMRegister, XMMRegister) |
|
2507 emit_sse_instruction(ucomisd, sse2, 0x66, 0x2E, XMMRegister, XMMRegister) |
|
2508 |
|
2509 emit_sse_instruction(cvtss2sd, sse2, 0xF3, 0x5A, XMMRegister, XMMRegister); |
|
2510 emit_sse_instruction(cvtsd2ss, sse2, 0xF2, 0x5A, XMMRegister, XMMRegister) |
|
2511 emit_sse_instruction(cvtsi2ss, sse, 0xF3, 0x2A, XMMRegister, Register); |
|
2512 emit_sse_instruction(cvtsi2sd, sse2, 0xF2, 0x2A, XMMRegister, Register) |
|
2513 emit_sse_instruction(cvtss2si, sse, 0xF3, 0x2D, Register, XMMRegister); |
|
2514 emit_sse_instruction(cvtsd2si, sse2, 0xF2, 0x2D, Register, XMMRegister) |
|
2515 emit_sse_instruction(cvttss2si, sse, 0xF3, 0x2C, Register, XMMRegister); |
|
2516 emit_sse_instruction(cvttsd2si, sse2, 0xF2, 0x2C, Register, XMMRegister) |
|
2517 |
|
2518 emit_sse_instruction(movss, sse, 0xF3, 0x10, XMMRegister, XMMRegister) |
|
2519 emit_sse_instruction(movsd, sse2, 0xF2, 0x10, XMMRegister, XMMRegister) |
|
2520 |
|
2521 emit_sse_instruction(movq, sse2, 0xF3, 0x7E, XMMRegister, XMMRegister); |
|
2522 emit_sse_instruction(movd, sse2, 0x66, 0x6E, XMMRegister, Register); |
|
2523 emit_sse_instruction(movdqa, sse2, 0x66, 0x6F, XMMRegister, XMMRegister); |
|
2524 |
|
2525 emit_sse_instruction(punpcklbw, sse2, 0x66, 0x60, XMMRegister, XMMRegister); |
|
2526 |
|
2527 |
|
2528 // Instruction not covered by macro |
|
2529 void Assembler::movq(Address dst, XMMRegister src) { |
|
2530 assert(VM_Version::supports_sse2(), ""); |
|
2531 |
|
2532 InstructionMark im(this); |
|
2533 emit_byte(0x66); |
|
2534 emit_byte(0x0F); |
|
2535 emit_byte(0xD6); |
|
2536 emit_sse_operand(src, dst); |
|
2537 } |
|
2538 |
|
2539 void Assembler::movd(Address dst, XMMRegister src) { |
|
2540 assert(VM_Version::supports_sse2(), ""); |
|
2541 |
|
2542 InstructionMark im(this); |
|
2543 emit_byte(0x66); |
|
2544 emit_byte(0x0F); |
|
2545 emit_byte(0x7E); |
|
2546 emit_sse_operand(src, dst); |
|
2547 } |
|
2548 |
|
2549 void Assembler::movd(Register dst, XMMRegister src) { |
|
2550 assert(VM_Version::supports_sse2(), ""); |
|
2551 |
|
2552 emit_byte(0x66); |
|
2553 emit_byte(0x0F); |
|
2554 emit_byte(0x7E); |
|
2555 emit_sse_operand(src, dst); |
|
2556 } |
|
2557 |
|
2558 void Assembler::movdqa(Address dst, XMMRegister src) { |
|
2559 assert(VM_Version::supports_sse2(), ""); |
|
2560 |
|
2561 InstructionMark im(this); |
|
2562 emit_byte(0x66); |
|
2563 emit_byte(0x0F); |
|
2564 emit_byte(0x7F); |
|
2565 emit_sse_operand(src, dst); |
|
2566 } |
|
2567 |
|
2568 void Assembler::pshufd(XMMRegister dst, XMMRegister src, int mode) { |
|
2569 assert(isByte(mode), "invalid value"); |
|
2570 assert(VM_Version::supports_sse2(), ""); |
|
2571 |
|
2572 emit_byte(0x66); |
|
2573 emit_byte(0x0F); |
|
2574 emit_byte(0x70); |
|
2575 emit_sse_operand(dst, src); |
|
2576 emit_byte(mode & 0xFF); |
|
2577 } |
|
2578 |
|
2579 void Assembler::pshufd(XMMRegister dst, Address src, int mode) { |
|
2580 assert(isByte(mode), "invalid value"); |
|
2581 assert(VM_Version::supports_sse2(), ""); |
|
2582 |
|
2583 InstructionMark im(this); |
|
2584 emit_byte(0x66); |
|
2585 emit_byte(0x0F); |
|
2586 emit_byte(0x70); |
|
2587 emit_sse_operand(dst, src); |
|
2588 emit_byte(mode & 0xFF); |
|
2589 } |
|
2590 |
|
2591 void Assembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) { |
|
2592 assert(isByte(mode), "invalid value"); |
|
2593 assert(VM_Version::supports_sse2(), ""); |
|
2594 |
|
2595 emit_byte(0xF2); |
|
2596 emit_byte(0x0F); |
|
2597 emit_byte(0x70); |
|
2598 emit_sse_operand(dst, src); |
|
2599 emit_byte(mode & 0xFF); |
|
2600 } |
|
2601 |
|
2602 void Assembler::pshuflw(XMMRegister dst, Address src, int mode) { |
|
2603 assert(isByte(mode), "invalid value"); |
|
2604 assert(VM_Version::supports_sse2(), ""); |
|
2605 |
|
2606 InstructionMark im(this); |
|
2607 emit_byte(0xF2); |
|
2608 emit_byte(0x0F); |
|
2609 emit_byte(0x70); |
|
2610 emit_sse_operand(dst, src); |
|
2611 emit_byte(mode & 0xFF); |
|
2612 } |
|
2613 |
|
2614 void Assembler::psrlq(XMMRegister dst, int shift) { |
|
2615 assert(VM_Version::supports_sse2(), ""); |
|
2616 |
|
2617 emit_byte(0x66); |
|
2618 emit_byte(0x0F); |
|
2619 emit_byte(0x73); |
|
2620 emit_sse_operand(xmm2, dst); |
|
2621 emit_byte(shift); |
|
2622 } |
|
2623 |
|
2624 void Assembler::movss( Address dst, XMMRegister src ) { |
|
2625 assert(VM_Version::supports_sse(), ""); |
|
2626 |
|
2627 InstructionMark im(this); |
|
2628 emit_byte(0xF3); // single |
|
2629 emit_byte(0x0F); |
|
2630 emit_byte(0x11); // store |
|
2631 emit_sse_operand(src, dst); |
|
2632 } |
|
2633 |
|
2634 void Assembler::movsd( Address dst, XMMRegister src ) { |
|
2635 assert(VM_Version::supports_sse2(), ""); |
|
2636 |
|
2637 InstructionMark im(this); |
|
2638 emit_byte(0xF2); // double |
|
2639 emit_byte(0x0F); |
|
2640 emit_byte(0x11); // store |
|
2641 emit_sse_operand(src,dst); |
|
2642 } |
|
2643 |
|
2644 // New cpus require to use movaps and movapd to avoid partial register stall |
|
2645 // when moving between registers. |
|
2646 void Assembler::movaps(XMMRegister dst, XMMRegister src) { |
|
2647 assert(VM_Version::supports_sse(), ""); |
|
2648 |
|
2649 emit_byte(0x0F); |
|
2650 emit_byte(0x28); |
|
2651 emit_sse_operand(dst, src); |
|
2652 } |
|
2653 void Assembler::movapd(XMMRegister dst, XMMRegister src) { |
|
2654 assert(VM_Version::supports_sse2(), ""); |
|
2655 |
|
2656 emit_byte(0x66); |
|
2657 emit_byte(0x0F); |
|
2658 emit_byte(0x28); |
|
2659 emit_sse_operand(dst, src); |
|
2660 } |
|
2661 |
|
2662 // New cpus require to use movsd and movss to avoid partial register stall |
|
2663 // when loading from memory. But for old Opteron use movlpd instead of movsd. |
|
2664 // The selection is done in MacroAssembler::movdbl() and movflt(). |
|
2665 void Assembler::movlpd(XMMRegister dst, Address src) { |
|
2666 assert(VM_Version::supports_sse(), ""); |
|
2667 |
|
2668 InstructionMark im(this); |
|
2669 emit_byte(0x66); |
|
2670 emit_byte(0x0F); |
|
2671 emit_byte(0x12); |
|
2672 emit_sse_operand(dst, src); |
|
2673 } |
|
2674 |
|
2675 |
|
2676 emit_sse_instruction(andps, sse, 0, 0x54, XMMRegister, XMMRegister); |
|
2677 emit_sse_instruction(andpd, sse2, 0x66, 0x54, XMMRegister, XMMRegister); |
|
2678 emit_sse_instruction(andnps, sse, 0, 0x55, XMMRegister, XMMRegister); |
|
2679 emit_sse_instruction(andnpd, sse2, 0x66, 0x55, XMMRegister, XMMRegister); |
|
2680 emit_sse_instruction(orps, sse, 0, 0x56, XMMRegister, XMMRegister); |
|
2681 emit_sse_instruction(orpd, sse2, 0x66, 0x56, XMMRegister, XMMRegister); |
|
2682 emit_sse_instruction(xorps, sse, 0, 0x57, XMMRegister, XMMRegister); |
|
2683 emit_sse_instruction(xorpd, sse2, 0x66, 0x57, XMMRegister, XMMRegister); |
|
2684 |
|
2685 |
|
2686 void Assembler::ldmxcsr( Address src) { |
|
2687 InstructionMark im(this); |
|
2688 emit_byte(0x0F); |
|
2689 emit_byte(0xAE); |
|
2690 emit_operand(rdx /* 2 */, src); |
|
2691 } |
|
2692 |
|
2693 void Assembler::stmxcsr( Address dst) { |
|
2694 InstructionMark im(this); |
|
2695 emit_byte(0x0F); |
|
2696 emit_byte(0xAE); |
|
2697 emit_operand(rbx /* 3 */, dst); |
|
2698 } |
|
2699 |
|
2700 // Implementation of MacroAssembler |
|
2701 |
|
2702 Address MacroAssembler::as_Address(AddressLiteral adr) { |
|
2703 // amd64 always does this as a pc-rel |
|
2704 // we can be absolute or disp based on the instruction type |
|
2705 // jmp/call are displacements others are absolute |
|
2706 assert(!adr.is_lval(), "must be rval"); |
|
2707 |
|
2708 return Address(adr.target(), adr.rspec()); |
|
2709 } |
|
2710 |
|
2711 Address MacroAssembler::as_Address(ArrayAddress adr) { |
|
2712 return Address::make_array(adr); |
|
2713 } |
|
2714 |
|
2715 void MacroAssembler::fat_nop() { |
|
2716 // A 5 byte nop that is safe for patching (see patch_verified_entry) |
|
2717 emit_byte(0x26); // es: |
|
2718 emit_byte(0x2e); // cs: |
|
2719 emit_byte(0x64); // fs: |
|
2720 emit_byte(0x65); // gs: |
|
2721 emit_byte(0x90); |
|
2722 } |
|
2723 |
|
2724 // 32bit can do a case table jump in one instruction but we no longer allow the base |
|
2725 // to be installed in the Address class |
|
2726 void MacroAssembler::jump(ArrayAddress entry) { |
|
2727 jmp(as_Address(entry)); |
|
2728 } |
|
2729 |
|
2730 void MacroAssembler::jump(AddressLiteral dst) { |
|
2731 jmp_literal(dst.target(), dst.rspec()); |
|
2732 } |
|
2733 |
|
2734 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst) { |
|
2735 assert((0 <= cc) && (cc < 16), "illegal cc"); |
|
2736 |
|
2737 InstructionMark im(this); |
|
2738 |
|
2739 relocInfo::relocType rtype = dst.reloc(); |
|
2740 relocate(rtype); |
|
2741 const int short_size = 2; |
|
2742 const int long_size = 6; |
|
2743 int offs = (int)dst.target() - ((int)_code_pos); |
|
2744 if (rtype == relocInfo::none && is8bit(offs - short_size)) { |
|
2745 // 0111 tttn #8-bit disp |
|
2746 emit_byte(0x70 | cc); |
|
2747 emit_byte((offs - short_size) & 0xFF); |
|
2748 } else { |
|
2749 // 0000 1111 1000 tttn #32-bit disp |
|
2750 emit_byte(0x0F); |
|
2751 emit_byte(0x80 | cc); |
|
2752 emit_long(offs - long_size); |
|
2753 } |
|
2754 } |
|
2755 |
|
2756 // Calls |
|
2757 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) { |
|
2758 Assembler::call(L, rtype); |
|
2759 } |
|
2760 |
|
2761 void MacroAssembler::call(Register entry) { |
|
2762 Assembler::call(entry); |
|
2763 } |
|
2764 |
|
2765 void MacroAssembler::call(AddressLiteral entry) { |
|
2766 Assembler::call_literal(entry.target(), entry.rspec()); |
|
2767 } |
|
2768 |
|
2769 |
|
2770 void MacroAssembler::cmp8(AddressLiteral src1, int8_t imm) { |
|
2771 Assembler::cmpb(as_Address(src1), imm); |
|
2772 } |
|
2773 |
|
2774 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm) { |
|
2775 Assembler::cmpl(as_Address(src1), imm); |
|
2776 } |
|
2777 |
|
2778 void MacroAssembler::cmp32(Register src1, AddressLiteral src2) { |
|
2779 if (src2.is_lval()) { |
|
2780 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec()); |
|
2781 } else { |
|
2782 Assembler::cmpl(src1, as_Address(src2)); |
|
2783 } |
|
2784 } |
|
2785 |
|
2786 void MacroAssembler::cmp32(Register src1, int32_t imm) { |
|
2787 Assembler::cmpl(src1, imm); |
|
2788 } |
|
2789 |
|
2790 void MacroAssembler::cmp32(Register src1, Address src2) { |
|
2791 Assembler::cmpl(src1, src2); |
|
2792 } |
|
2793 |
|
2794 void MacroAssembler::cmpoop(Address src1, jobject obj) { |
|
2795 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate()); |
|
2796 } |
|
2797 |
|
2798 void MacroAssembler::cmpoop(Register src1, jobject obj) { |
|
2799 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate()); |
|
2800 } |
|
2801 |
|
2802 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2) { |
|
2803 if (src2.is_lval()) { |
|
2804 // compare the effect address of src2 to src1 |
|
2805 cmp_literal32(src1, (int32_t)src2.target(), src2.rspec()); |
|
2806 } else { |
|
2807 Assembler::cmpl(src1, as_Address(src2)); |
|
2808 } |
|
2809 } |
|
2810 |
|
2811 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2) { |
|
2812 assert(src2.is_lval(), "not a mem-mem compare"); |
|
2813 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec()); |
|
2814 } |
|
2815 |
|
2816 |
|
2817 void MacroAssembler::cmpxchgptr(Register reg, AddressLiteral adr) { |
|
2818 cmpxchg(reg, as_Address(adr)); |
|
2819 } |
|
2820 |
|
2821 void MacroAssembler::increment(AddressLiteral dst) { |
|
2822 increment(as_Address(dst)); |
|
2823 } |
|
2824 |
|
2825 void MacroAssembler::increment(ArrayAddress dst) { |
|
2826 increment(as_Address(dst)); |
|
2827 } |
|
2828 |
|
2829 void MacroAssembler::lea(Register dst, AddressLiteral adr) { |
|
2830 // leal(dst, as_Address(adr)); |
|
2831 // see note in movl as to why we musr use a move |
|
2832 mov_literal32(dst, (int32_t) adr.target(), adr.rspec()); |
|
2833 } |
|
2834 |
|
2835 void MacroAssembler::lea(Address dst, AddressLiteral adr) { |
|
2836 // leal(dst, as_Address(adr)); |
|
2837 // see note in movl as to why we musr use a move |
|
2838 mov_literal32(dst, (int32_t) adr.target(), adr.rspec()); |
|
2839 } |
|
2840 |
|
2841 void MacroAssembler::mov32(AddressLiteral dst, Register src) { |
|
2842 Assembler::movl(as_Address(dst), src); |
|
2843 } |
|
2844 |
|
2845 void MacroAssembler::mov32(Register dst, AddressLiteral src) { |
|
2846 Assembler::movl(dst, as_Address(src)); |
|
2847 } |
|
2848 |
|
2849 void MacroAssembler::movbyte(ArrayAddress dst, int src) { |
|
2850 movb(as_Address(dst), src); |
|
2851 } |
|
2852 |
|
2853 void MacroAssembler::movoop(Address dst, jobject obj) { |
|
2854 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate()); |
|
2855 } |
|
2856 |
|
2857 void MacroAssembler::movoop(Register dst, jobject obj) { |
|
2858 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate()); |
|
2859 } |
|
2860 |
|
2861 void MacroAssembler::movptr(Register dst, AddressLiteral src) { |
|
2862 if (src.is_lval()) { |
|
2863 // essentially an lea |
|
2864 mov_literal32(dst, (int32_t) src.target(), src.rspec()); |
|
2865 } else { |
|
2866 // mov 32bits from an absolute address |
|
2867 movl(dst, as_Address(src)); |
|
2868 } |
|
2869 } |
|
2870 |
|
2871 void MacroAssembler::movptr(ArrayAddress dst, Register src) { |
|
2872 movl(as_Address(dst), src); |
|
2873 } |
|
2874 |
|
2875 void MacroAssembler::movptr(Register dst, ArrayAddress src) { |
|
2876 movl(dst, as_Address(src)); |
|
2877 } |
|
2878 |
|
2879 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src) { |
|
2880 movss(dst, as_Address(src)); |
|
2881 } |
|
2882 |
|
2883 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src) { |
|
2884 if (UseXmmLoadAndClearUpper) { movsd (dst, as_Address(src)); return; } |
|
2885 else { movlpd(dst, as_Address(src)); return; } |
|
2886 } |
|
2887 |
|
2888 void Assembler::pushoop(jobject obj) { |
|
2889 push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate()); |
|
2890 } |
|
2891 |
|
2892 |
|
2893 void MacroAssembler::pushptr(AddressLiteral src) { |
|
2894 if (src.is_lval()) { |
|
2895 push_literal32((int32_t)src.target(), src.rspec()); |
|
2896 } else { |
|
2897 pushl(as_Address(src)); |
|
2898 } |
|
2899 } |
|
2900 |
|
2901 void MacroAssembler::test32(Register src1, AddressLiteral src2) { |
|
2902 // src2 must be rval |
|
2903 testl(src1, as_Address(src2)); |
|
2904 } |
|
2905 |
|
2906 // FPU |
|
2907 |
|
2908 void MacroAssembler::fld_x(AddressLiteral src) { |
|
2909 Assembler::fld_x(as_Address(src)); |
|
2910 } |
|
2911 |
|
2912 void MacroAssembler::fld_d(AddressLiteral src) { |
|
2913 fld_d(as_Address(src)); |
|
2914 } |
|
2915 |
|
2916 void MacroAssembler::fld_s(AddressLiteral src) { |
|
2917 fld_s(as_Address(src)); |
|
2918 } |
|
2919 |
|
2920 void MacroAssembler::fldcw(AddressLiteral src) { |
|
2921 Assembler::fldcw(as_Address(src)); |
|
2922 } |
|
2923 |
|
2924 void MacroAssembler::ldmxcsr(AddressLiteral src) { |
|
2925 Assembler::ldmxcsr(as_Address(src)); |
|
2926 } |
|
2927 |
|
2928 // SSE |
|
2929 |
|
2930 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src) { |
|
2931 andpd(dst, as_Address(src)); |
|
2932 } |
|
2933 |
|
2934 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src) { |
|
2935 comisd(dst, as_Address(src)); |
|
2936 } |
|
2937 |
|
2938 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src) { |
|
2939 comiss(dst, as_Address(src)); |
|
2940 } |
|
2941 |
|
2942 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src) { |
|
2943 movsd(dst, as_Address(src)); |
|
2944 } |
|
2945 |
|
2946 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src) { |
|
2947 movss(dst, as_Address(src)); |
|
2948 } |
|
2949 |
|
2950 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src) { |
|
2951 xorpd(dst, as_Address(src)); |
|
2952 } |
|
2953 |
|
2954 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src) { |
|
2955 xorps(dst, as_Address(src)); |
|
2956 } |
|
2957 |
|
2958 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src) { |
|
2959 ucomisd(dst, as_Address(src)); |
|
2960 } |
|
2961 |
|
2962 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src) { |
|
2963 ucomiss(dst, as_Address(src)); |
|
2964 } |
|
2965 |
|
2966 void MacroAssembler::null_check(Register reg, int offset) { |
|
2967 if (needs_explicit_null_check(offset)) { |
|
2968 // provoke OS NULL exception if reg = NULL by |
|
2969 // accessing M[reg] w/o changing any (non-CC) registers |
|
2970 cmpl(rax, Address(reg, 0)); |
|
2971 // Note: should probably use testl(rax, Address(reg, 0)); |
|
2972 // may be shorter code (however, this version of |
|
2973 // testl needs to be implemented first) |
|
2974 } else { |
|
2975 // nothing to do, (later) access of M[reg + offset] |
|
2976 // will provoke OS NULL exception if reg = NULL |
|
2977 } |
|
2978 } |
|
2979 |
|
2980 |
|
2981 int MacroAssembler::load_unsigned_byte(Register dst, Address src) { |
|
2982 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16, |
|
2983 // and "3.9 Partial Register Penalties", p. 22). |
|
2984 int off; |
|
2985 if (VM_Version::is_P6() || src.uses(dst)) { |
|
2986 off = offset(); |
|
2987 movzxb(dst, src); |
|
2988 } else { |
|
2989 xorl(dst, dst); |
|
2990 off = offset(); |
|
2991 movb(dst, src); |
|
2992 } |
|
2993 return off; |
|
2994 } |
|
2995 |
|
2996 |
|
2997 int MacroAssembler::load_unsigned_word(Register dst, Address src) { |
|
2998 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16, |
|
2999 // and "3.9 Partial Register Penalties", p. 22). |
|
3000 int off; |
|
3001 if (VM_Version::is_P6() || src.uses(dst)) { |
|
3002 off = offset(); |
|
3003 movzxw(dst, src); |
|
3004 } else { |
|
3005 xorl(dst, dst); |
|
3006 off = offset(); |
|
3007 movw(dst, src); |
|
3008 } |
|
3009 return off; |
|
3010 } |
|
3011 |
|
3012 |
|
3013 int MacroAssembler::load_signed_byte(Register dst, Address src) { |
|
3014 int off; |
|
3015 if (VM_Version::is_P6()) { |
|
3016 off = offset(); |
|
3017 movsxb(dst, src); |
|
3018 } else { |
|
3019 off = load_unsigned_byte(dst, src); |
|
3020 shll(dst, 24); |
|
3021 sarl(dst, 24); |
|
3022 } |
|
3023 return off; |
|
3024 } |
|
3025 |
|
3026 |
|
3027 int MacroAssembler::load_signed_word(Register dst, Address src) { |
|
3028 int off; |
|
3029 if (VM_Version::is_P6()) { |
|
3030 off = offset(); |
|
3031 movsxw(dst, src); |
|
3032 } else { |
|
3033 off = load_unsigned_word(dst, src); |
|
3034 shll(dst, 16); |
|
3035 sarl(dst, 16); |
|
3036 } |
|
3037 return off; |
|
3038 } |
|
3039 |
|
3040 |
|
3041 void MacroAssembler::extend_sign(Register hi, Register lo) { |
|
3042 // According to Intel Doc. AP-526, "Integer Divide", p.18. |
|
3043 if (VM_Version::is_P6() && hi == rdx && lo == rax) { |
|
3044 cdql(); |
|
3045 } else { |
|
3046 movl(hi, lo); |
|
3047 sarl(hi, 31); |
|
3048 } |
|
3049 } |
|
3050 |
|
3051 |
|
3052 void MacroAssembler::increment(Register reg, int value) { |
|
3053 if (value == min_jint) {addl(reg, value); return; } |
|
3054 if (value < 0) { decrement(reg, -value); return; } |
|
3055 if (value == 0) { ; return; } |
|
3056 if (value == 1 && UseIncDec) { incl(reg); return; } |
|
3057 /* else */ { addl(reg, value) ; return; } |
|
3058 } |
|
3059 |
|
3060 void MacroAssembler::increment(Address dst, int value) { |
|
3061 if (value == min_jint) {addl(dst, value); return; } |
|
3062 if (value < 0) { decrement(dst, -value); return; } |
|
3063 if (value == 0) { ; return; } |
|
3064 if (value == 1 && UseIncDec) { incl(dst); return; } |
|
3065 /* else */ { addl(dst, value) ; return; } |
|
3066 } |
|
3067 |
|
3068 void MacroAssembler::decrement(Register reg, int value) { |
|
3069 if (value == min_jint) {subl(reg, value); return; } |
|
3070 if (value < 0) { increment(reg, -value); return; } |
|
3071 if (value == 0) { ; return; } |
|
3072 if (value == 1 && UseIncDec) { decl(reg); return; } |
|
3073 /* else */ { subl(reg, value) ; return; } |
|
3074 } |
|
3075 |
|
3076 void MacroAssembler::decrement(Address dst, int value) { |
|
3077 if (value == min_jint) {subl(dst, value); return; } |
|
3078 if (value < 0) { increment(dst, -value); return; } |
|
3079 if (value == 0) { ; return; } |
|
3080 if (value == 1 && UseIncDec) { decl(dst); return; } |
|
3081 /* else */ { subl(dst, value) ; return; } |
|
3082 } |
|
3083 |
|
3084 void MacroAssembler::align(int modulus) { |
|
3085 if (offset() % modulus != 0) nop(modulus - (offset() % modulus)); |
|
3086 } |
|
3087 |
|
3088 |
|
3089 void MacroAssembler::enter() { |
|
3090 pushl(rbp); |
|
3091 movl(rbp, rsp); |
|
3092 } |
|
3093 |
|
3094 |
|
3095 void MacroAssembler::leave() { |
|
3096 movl(rsp, rbp); |
|
3097 popl(rbp); |
|
3098 } |
|
3099 |
|
3100 void MacroAssembler::set_last_Java_frame(Register java_thread, |
|
3101 Register last_java_sp, |
|
3102 Register last_java_fp, |
|
3103 address last_java_pc) { |
|
3104 // determine java_thread register |
|
3105 if (!java_thread->is_valid()) { |
|
3106 java_thread = rdi; |
|
3107 get_thread(java_thread); |
|
3108 } |
|
3109 // determine last_java_sp register |
|
3110 if (!last_java_sp->is_valid()) { |
|
3111 last_java_sp = rsp; |
|
3112 } |
|
3113 |
|
3114 // last_java_fp is optional |
|
3115 |
|
3116 if (last_java_fp->is_valid()) { |
|
3117 movl(Address(java_thread, JavaThread::last_Java_fp_offset()), last_java_fp); |
|
3118 } |
|
3119 |
|
3120 // last_java_pc is optional |
|
3121 |
|
3122 if (last_java_pc != NULL) { |
|
3123 lea(Address(java_thread, |
|
3124 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()), |
|
3125 InternalAddress(last_java_pc)); |
|
3126 |
|
3127 } |
|
3128 movl(Address(java_thread, JavaThread::last_Java_sp_offset()), last_java_sp); |
|
3129 } |
|
3130 |
|
3131 void MacroAssembler::reset_last_Java_frame(Register java_thread, bool clear_fp, bool clear_pc) { |
|
3132 // determine java_thread register |
|
3133 if (!java_thread->is_valid()) { |
|
3134 java_thread = rdi; |
|
3135 get_thread(java_thread); |
|
3136 } |
|
3137 // we must set sp to zero to clear frame |
|
3138 movl(Address(java_thread, JavaThread::last_Java_sp_offset()), 0); |
|
3139 if (clear_fp) { |
|
3140 movl(Address(java_thread, JavaThread::last_Java_fp_offset()), 0); |
|
3141 } |
|
3142 |
|
3143 if (clear_pc) |
|
3144 movl(Address(java_thread, JavaThread::last_Java_pc_offset()), 0); |
|
3145 |
|
3146 } |
|
3147 |
|
3148 |
|
3149 |
|
3150 // Implementation of call_VM versions |
|
3151 |
|
3152 void MacroAssembler::call_VM_leaf_base( |
|
3153 address entry_point, |
|
3154 int number_of_arguments |
|
3155 ) { |
|
3156 call(RuntimeAddress(entry_point)); |
|
3157 increment(rsp, number_of_arguments * wordSize); |
|
3158 } |
|
3159 |
|
3160 |
|
3161 void MacroAssembler::call_VM_base( |
|
3162 Register oop_result, |
|
3163 Register java_thread, |
|
3164 Register last_java_sp, |
|
3165 address entry_point, |
|
3166 int number_of_arguments, |
|
3167 bool check_exceptions |
|
3168 ) { |
|
3169 // determine java_thread register |
|
3170 if (!java_thread->is_valid()) { |
|
3171 java_thread = rdi; |
|
3172 get_thread(java_thread); |
|
3173 } |
|
3174 // determine last_java_sp register |
|
3175 if (!last_java_sp->is_valid()) { |
|
3176 last_java_sp = rsp; |
|
3177 } |
|
3178 // debugging support |
|
3179 assert(number_of_arguments >= 0 , "cannot have negative number of arguments"); |
|
3180 assert(java_thread != oop_result , "cannot use the same register for java_thread & oop_result"); |
|
3181 assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp"); |
|
3182 // push java thread (becomes first argument of C function) |
|
3183 pushl(java_thread); |
|
3184 // set last Java frame before call |
|
3185 assert(last_java_sp != rbp, "this code doesn't work for last_java_sp == rbp, which currently can't portably work anyway since C2 doesn't save rbp,"); |
|
3186 // Only interpreter should have to set fp |
|
3187 set_last_Java_frame(java_thread, last_java_sp, rbp, NULL); |
|
3188 // do the call |
|
3189 call(RuntimeAddress(entry_point)); |
|
3190 // restore the thread (cannot use the pushed argument since arguments |
|
3191 // may be overwritten by C code generated by an optimizing compiler); |
|
3192 // however can use the register value directly if it is callee saved. |
|
3193 if (java_thread == rdi || java_thread == rsi) { |
|
3194 // rdi & rsi are callee saved -> nothing to do |
|
3195 #ifdef ASSERT |
|
3196 guarantee(java_thread != rax, "change this code"); |
|
3197 pushl(rax); |
|
3198 { Label L; |
|
3199 get_thread(rax); |
|
3200 cmpl(java_thread, rax); |
|
3201 jcc(Assembler::equal, L); |
|
3202 stop("MacroAssembler::call_VM_base: rdi not callee saved?"); |
|
3203 bind(L); |
|
3204 } |
|
3205 popl(rax); |
|
3206 #endif |
|
3207 } else { |
|
3208 get_thread(java_thread); |
|
3209 } |
|
3210 // reset last Java frame |
|
3211 // Only interpreter should have to clear fp |
|
3212 reset_last_Java_frame(java_thread, true, false); |
|
3213 // discard thread and arguments |
|
3214 addl(rsp, (1 + number_of_arguments)*wordSize); |
|
3215 |
|
3216 #ifndef CC_INTERP |
|
3217 // C++ interp handles this in the interpreter |
|
3218 check_and_handle_popframe(java_thread); |
|
3219 check_and_handle_earlyret(java_thread); |
|
3220 #endif /* CC_INTERP */ |
|
3221 |
|
3222 if (check_exceptions) { |
|
3223 // check for pending exceptions (java_thread is set upon return) |
|
3224 cmpl(Address(java_thread, Thread::pending_exception_offset()), NULL_WORD); |
|
3225 jump_cc(Assembler::notEqual, |
|
3226 RuntimeAddress(StubRoutines::forward_exception_entry())); |
|
3227 } |
|
3228 |
|
3229 // get oop result if there is one and reset the value in the thread |
|
3230 if (oop_result->is_valid()) { |
|
3231 movl(oop_result, Address(java_thread, JavaThread::vm_result_offset())); |
|
3232 movl(Address(java_thread, JavaThread::vm_result_offset()), NULL_WORD); |
|
3233 verify_oop(oop_result); |
|
3234 } |
|
3235 } |
|
3236 |
|
3237 |
|
3238 void MacroAssembler::check_and_handle_popframe(Register java_thread) { |
|
3239 } |
|
3240 |
|
3241 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { |
|
3242 } |
|
3243 |
|
3244 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) { |
|
3245 leal(rax, Address(rsp, (1 + number_of_arguments) * wordSize)); |
|
3246 call_VM_base(oop_result, noreg, rax, entry_point, number_of_arguments, check_exceptions); |
|
3247 } |
|
3248 |
|
3249 |
|
3250 void MacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { |
|
3251 Label C, E; |
|
3252 call(C, relocInfo::none); |
|
3253 jmp(E); |
|
3254 |
|
3255 bind(C); |
|
3256 call_VM_helper(oop_result, entry_point, 0, check_exceptions); |
|
3257 ret(0); |
|
3258 |
|
3259 bind(E); |
|
3260 } |
|
3261 |
|
3262 |
|
3263 void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) { |
|
3264 Label C, E; |
|
3265 call(C, relocInfo::none); |
|
3266 jmp(E); |
|
3267 |
|
3268 bind(C); |
|
3269 pushl(arg_1); |
|
3270 call_VM_helper(oop_result, entry_point, 1, check_exceptions); |
|
3271 ret(0); |
|
3272 |
|
3273 bind(E); |
|
3274 } |
|
3275 |
|
3276 |
|
3277 void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) { |
|
3278 Label C, E; |
|
3279 call(C, relocInfo::none); |
|
3280 jmp(E); |
|
3281 |
|
3282 bind(C); |
|
3283 pushl(arg_2); |
|
3284 pushl(arg_1); |
|
3285 call_VM_helper(oop_result, entry_point, 2, check_exceptions); |
|
3286 ret(0); |
|
3287 |
|
3288 bind(E); |
|
3289 } |
|
3290 |
|
3291 |
|
3292 void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) { |
|
3293 Label C, E; |
|
3294 call(C, relocInfo::none); |
|
3295 jmp(E); |
|
3296 |
|
3297 bind(C); |
|
3298 pushl(arg_3); |
|
3299 pushl(arg_2); |
|
3300 pushl(arg_1); |
|
3301 call_VM_helper(oop_result, entry_point, 3, check_exceptions); |
|
3302 ret(0); |
|
3303 |
|
3304 bind(E); |
|
3305 } |
|
3306 |
|
3307 |
|
3308 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments, bool check_exceptions) { |
|
3309 call_VM_base(oop_result, noreg, last_java_sp, entry_point, number_of_arguments, check_exceptions); |
|
3310 } |
|
3311 |
|
3312 |
|
3313 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions) { |
|
3314 pushl(arg_1); |
|
3315 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions); |
|
3316 } |
|
3317 |
|
3318 |
|
3319 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) { |
|
3320 pushl(arg_2); |
|
3321 pushl(arg_1); |
|
3322 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions); |
|
3323 } |
|
3324 |
|
3325 |
|
3326 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) { |
|
3327 pushl(arg_3); |
|
3328 pushl(arg_2); |
|
3329 pushl(arg_1); |
|
3330 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions); |
|
3331 } |
|
3332 |
|
3333 |
|
3334 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) { |
|
3335 call_VM_leaf_base(entry_point, number_of_arguments); |
|
3336 } |
|
3337 |
|
3338 |
|
3339 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_1) { |
|
3340 pushl(arg_1); |
|
3341 call_VM_leaf(entry_point, 1); |
|
3342 } |
|
3343 |
|
3344 |
|
3345 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_1, Register arg_2) { |
|
3346 pushl(arg_2); |
|
3347 pushl(arg_1); |
|
3348 call_VM_leaf(entry_point, 2); |
|
3349 } |
|
3350 |
|
3351 |
|
3352 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3) { |
|
3353 pushl(arg_3); |
|
3354 pushl(arg_2); |
|
3355 pushl(arg_1); |
|
3356 call_VM_leaf(entry_point, 3); |
|
3357 } |
|
3358 |
|
3359 |
|
3360 // Calls to C land |
|
3361 // |
|
3362 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded |
|
3363 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp |
|
3364 // has to be reset to 0. This is required to allow proper stack traversal. |
|
3365 |
|
3366 void MacroAssembler::store_check(Register obj) { |
|
3367 // Does a store check for the oop in register obj. The content of |
|
3368 // register obj is destroyed afterwards. |
|
3369 store_check_part_1(obj); |
|
3370 store_check_part_2(obj); |
|
3371 } |
|
3372 |
|
3373 |
|
3374 void MacroAssembler::store_check(Register obj, Address dst) { |
|
3375 store_check(obj); |
|
3376 } |
|
3377 |
|
3378 |
|
3379 // split the store check operation so that other instructions can be scheduled inbetween |
|
3380 void MacroAssembler::store_check_part_1(Register obj) { |
|
3381 BarrierSet* bs = Universe::heap()->barrier_set(); |
|
3382 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind"); |
|
3383 shrl(obj, CardTableModRefBS::card_shift); |
|
3384 } |
|
3385 |
|
3386 |
|
3387 void MacroAssembler::store_check_part_2(Register obj) { |
|
3388 BarrierSet* bs = Universe::heap()->barrier_set(); |
|
3389 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind"); |
|
3390 CardTableModRefBS* ct = (CardTableModRefBS*)bs; |
|
3391 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); |
|
3392 ExternalAddress cardtable((address)ct->byte_map_base); |
|
3393 Address index(noreg, obj, Address::times_1); |
|
3394 |
|
3395 movb(as_Address(ArrayAddress(cardtable, index)), 0); |
|
3396 } |
|
3397 |
|
3398 |
|
3399 void MacroAssembler::c2bool(Register x) { |
|
3400 // implements x == 0 ? 0 : 1 |
|
3401 // note: must only look at least-significant byte of x |
|
3402 // since C-style booleans are stored in one byte |
|
3403 // only! (was bug) |
|
3404 andl(x, 0xFF); |
|
3405 setb(Assembler::notZero, x); |
|
3406 } |
|
3407 |
|
3408 |
|
3409 int MacroAssembler::corrected_idivl(Register reg) { |
|
3410 // Full implementation of Java idiv and irem; checks for |
|
3411 // special case as described in JVM spec., p.243 & p.271. |
|
3412 // The function returns the (pc) offset of the idivl |
|
3413 // instruction - may be needed for implicit exceptions. |
|
3414 // |
|
3415 // normal case special case |
|
3416 // |
|
3417 // input : rax,: dividend min_int |
|
3418 // reg: divisor (may not be rax,/rdx) -1 |
|
3419 // |
|
3420 // output: rax,: quotient (= rax, idiv reg) min_int |
|
3421 // rdx: remainder (= rax, irem reg) 0 |
|
3422 assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register"); |
|
3423 const int min_int = 0x80000000; |
|
3424 Label normal_case, special_case; |
|
3425 |
|
3426 // check for special case |
|
3427 cmpl(rax, min_int); |
|
3428 jcc(Assembler::notEqual, normal_case); |
|
3429 xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0) |
|
3430 cmpl(reg, -1); |
|
3431 jcc(Assembler::equal, special_case); |
|
3432 |
|
3433 // handle normal case |
|
3434 bind(normal_case); |
|
3435 cdql(); |
|
3436 int idivl_offset = offset(); |
|
3437 idivl(reg); |
|
3438 |
|
3439 // normal and special case exit |
|
3440 bind(special_case); |
|
3441 |
|
3442 return idivl_offset; |
|
3443 } |
|
3444 |
|
3445 |
|
3446 void MacroAssembler::lneg(Register hi, Register lo) { |
|
3447 negl(lo); |
|
3448 adcl(hi, 0); |
|
3449 negl(hi); |
|
3450 } |
|
3451 |
|
3452 |
|
3453 void MacroAssembler::lmul(int x_rsp_offset, int y_rsp_offset) { |
|
3454 // Multiplication of two Java long values stored on the stack |
|
3455 // as illustrated below. Result is in rdx:rax. |
|
3456 // |
|
3457 // rsp ---> [ ?? ] \ \ |
|
3458 // .... | y_rsp_offset | |
|
3459 // [ y_lo ] / (in bytes) | x_rsp_offset |
|
3460 // [ y_hi ] | (in bytes) |
|
3461 // .... | |
|
3462 // [ x_lo ] / |
|
3463 // [ x_hi ] |
|
3464 // .... |
|
3465 // |
|
3466 // Basic idea: lo(result) = lo(x_lo * y_lo) |
|
3467 // hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi) |
|
3468 Address x_hi(rsp, x_rsp_offset + wordSize); Address x_lo(rsp, x_rsp_offset); |
|
3469 Address y_hi(rsp, y_rsp_offset + wordSize); Address y_lo(rsp, y_rsp_offset); |
|
3470 Label quick; |
|
3471 // load x_hi, y_hi and check if quick |
|
3472 // multiplication is possible |
|
3473 movl(rbx, x_hi); |
|
3474 movl(rcx, y_hi); |
|
3475 movl(rax, rbx); |
|
3476 orl(rbx, rcx); // rbx, = 0 <=> x_hi = 0 and y_hi = 0 |
|
3477 jcc(Assembler::zero, quick); // if rbx, = 0 do quick multiply |
|
3478 // do full multiplication |
|
3479 // 1st step |
|
3480 mull(y_lo); // x_hi * y_lo |
|
3481 movl(rbx, rax); // save lo(x_hi * y_lo) in rbx, |
|
3482 // 2nd step |
|
3483 movl(rax, x_lo); |
|
3484 mull(rcx); // x_lo * y_hi |
|
3485 addl(rbx, rax); // add lo(x_lo * y_hi) to rbx, |
|
3486 // 3rd step |
|
3487 bind(quick); // note: rbx, = 0 if quick multiply! |
|
3488 movl(rax, x_lo); |
|
3489 mull(y_lo); // x_lo * y_lo |
|
3490 addl(rdx, rbx); // correct hi(x_lo * y_lo) |
|
3491 } |
|
3492 |
|
3493 |
|
3494 void MacroAssembler::lshl(Register hi, Register lo) { |
|
3495 // Java shift left long support (semantics as described in JVM spec., p.305) |
|
3496 // (basic idea for shift counts s >= n: x << s == (x << n) << (s - n)) |
|
3497 // shift value is in rcx ! |
|
3498 assert(hi != rcx, "must not use rcx"); |
|
3499 assert(lo != rcx, "must not use rcx"); |
|
3500 const Register s = rcx; // shift count |
|
3501 const int n = BitsPerWord; |
|
3502 Label L; |
|
3503 andl(s, 0x3f); // s := s & 0x3f (s < 0x40) |
|
3504 cmpl(s, n); // if (s < n) |
|
3505 jcc(Assembler::less, L); // else (s >= n) |
|
3506 movl(hi, lo); // x := x << n |
|
3507 xorl(lo, lo); |
|
3508 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n! |
|
3509 bind(L); // s (mod n) < n |
|
3510 shldl(hi, lo); // x := x << s |
|
3511 shll(lo); |
|
3512 } |
|
3513 |
|
3514 |
|
3515 void MacroAssembler::lshr(Register hi, Register lo, bool sign_extension) { |
|
3516 // Java shift right long support (semantics as described in JVM spec., p.306 & p.310) |
|
3517 // (basic idea for shift counts s >= n: x >> s == (x >> n) >> (s - n)) |
|
3518 assert(hi != rcx, "must not use rcx"); |
|
3519 assert(lo != rcx, "must not use rcx"); |
|
3520 const Register s = rcx; // shift count |
|
3521 const int n = BitsPerWord; |
|
3522 Label L; |
|
3523 andl(s, 0x3f); // s := s & 0x3f (s < 0x40) |
|
3524 cmpl(s, n); // if (s < n) |
|
3525 jcc(Assembler::less, L); // else (s >= n) |
|
3526 movl(lo, hi); // x := x >> n |
|
3527 if (sign_extension) sarl(hi, 31); |
|
3528 else xorl(hi, hi); |
|
3529 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n! |
|
3530 bind(L); // s (mod n) < n |
|
3531 shrdl(lo, hi); // x := x >> s |
|
3532 if (sign_extension) sarl(hi); |
|
3533 else shrl(hi); |
|
3534 } |
|
3535 |
|
3536 |
|
3537 // Note: y_lo will be destroyed |
|
3538 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) { |
|
3539 // Long compare for Java (semantics as described in JVM spec.) |
|
3540 Label high, low, done; |
|
3541 |
|
3542 cmpl(x_hi, y_hi); |
|
3543 jcc(Assembler::less, low); |
|
3544 jcc(Assembler::greater, high); |
|
3545 // x_hi is the return register |
|
3546 xorl(x_hi, x_hi); |
|
3547 cmpl(x_lo, y_lo); |
|
3548 jcc(Assembler::below, low); |
|
3549 jcc(Assembler::equal, done); |
|
3550 |
|
3551 bind(high); |
|
3552 xorl(x_hi, x_hi); |
|
3553 increment(x_hi); |
|
3554 jmp(done); |
|
3555 |
|
3556 bind(low); |
|
3557 xorl(x_hi, x_hi); |
|
3558 decrement(x_hi); |
|
3559 |
|
3560 bind(done); |
|
3561 } |
|
3562 |
|
3563 |
|
3564 void MacroAssembler::save_rax(Register tmp) { |
|
3565 if (tmp == noreg) pushl(rax); |
|
3566 else if (tmp != rax) movl(tmp, rax); |
|
3567 } |
|
3568 |
|
3569 |
|
3570 void MacroAssembler::restore_rax(Register tmp) { |
|
3571 if (tmp == noreg) popl(rax); |
|
3572 else if (tmp != rax) movl(rax, tmp); |
|
3573 } |
|
3574 |
|
3575 |
|
3576 void MacroAssembler::fremr(Register tmp) { |
|
3577 save_rax(tmp); |
|
3578 { Label L; |
|
3579 bind(L); |
|
3580 fprem(); |
|
3581 fwait(); fnstsw_ax(); |
|
3582 sahf(); |
|
3583 jcc(Assembler::parity, L); |
|
3584 } |
|
3585 restore_rax(tmp); |
|
3586 // Result is in ST0. |
|
3587 // Note: fxch & fpop to get rid of ST1 |
|
3588 // (otherwise FPU stack could overflow eventually) |
|
3589 fxch(1); |
|
3590 fpop(); |
|
3591 } |
|
3592 |
|
3593 |
|
3594 static const double pi_4 = 0.7853981633974483; |
|
3595 |
|
3596 void MacroAssembler::trigfunc(char trig, int num_fpu_regs_in_use) { |
|
3597 // A hand-coded argument reduction for values in fabs(pi/4, pi/2) |
|
3598 // was attempted in this code; unfortunately it appears that the |
|
3599 // switch to 80-bit precision and back causes this to be |
|
3600 // unprofitable compared with simply performing a runtime call if |
|
3601 // the argument is out of the (-pi/4, pi/4) range. |
|
3602 |
|
3603 Register tmp = noreg; |
|
3604 if (!VM_Version::supports_cmov()) { |
|
3605 // fcmp needs a temporary so preserve rbx, |
|
3606 tmp = rbx; |
|
3607 pushl(tmp); |
|
3608 } |
|
3609 |
|
3610 Label slow_case, done; |
|
3611 |
|
3612 // x ?<= pi/4 |
|
3613 fld_d(ExternalAddress((address)&pi_4)); |
|
3614 fld_s(1); // Stack: X PI/4 X |
|
3615 fabs(); // Stack: |X| PI/4 X |
|
3616 fcmp(tmp); |
|
3617 jcc(Assembler::above, slow_case); |
|
3618 |
|
3619 // fastest case: -pi/4 <= x <= pi/4 |
|
3620 switch(trig) { |
|
3621 case 's': |
|
3622 fsin(); |
|
3623 break; |
|
3624 case 'c': |
|
3625 fcos(); |
|
3626 break; |
|
3627 case 't': |
|
3628 ftan(); |
|
3629 break; |
|
3630 default: |
|
3631 assert(false, "bad intrinsic"); |
|
3632 break; |
|
3633 } |
|
3634 jmp(done); |
|
3635 |
|
3636 // slow case: runtime call |
|
3637 bind(slow_case); |
|
3638 // Preserve registers across runtime call |
|
3639 pushad(); |
|
3640 int incoming_argument_and_return_value_offset = -1; |
|
3641 if (num_fpu_regs_in_use > 1) { |
|
3642 // Must preserve all other FPU regs (could alternatively convert |
|
3643 // SharedRuntime::dsin and dcos into assembly routines known not to trash |
|
3644 // FPU state, but can not trust C compiler) |
|
3645 NEEDS_CLEANUP; |
|
3646 // NOTE that in this case we also push the incoming argument to |
|
3647 // the stack and restore it later; we also use this stack slot to |
|
3648 // hold the return value from dsin or dcos. |
|
3649 for (int i = 0; i < num_fpu_regs_in_use; i++) { |
|
3650 subl(rsp, wordSize*2); |
|
3651 fstp_d(Address(rsp, 0)); |
|
3652 } |
|
3653 incoming_argument_and_return_value_offset = 2*wordSize*(num_fpu_regs_in_use-1); |
|
3654 fld_d(Address(rsp, incoming_argument_and_return_value_offset)); |
|
3655 } |
|
3656 subl(rsp, wordSize*2); |
|
3657 fstp_d(Address(rsp, 0)); |
|
3658 // NOTE: we must not use call_VM_leaf here because that requires a |
|
3659 // complete interpreter frame in debug mode -- same bug as 4387334 |
|
3660 NEEDS_CLEANUP; |
|
3661 // Need to add stack banging before this runtime call if it needs to |
|
3662 // be taken; however, there is no generic stack banging routine at |
|
3663 // the MacroAssembler level |
|
3664 switch(trig) { |
|
3665 case 's': |
|
3666 { |
|
3667 call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dsin))); |
|
3668 } |
|
3669 break; |
|
3670 case 'c': |
|
3671 { |
|
3672 call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dcos))); |
|
3673 } |
|
3674 break; |
|
3675 case 't': |
|
3676 { |
|
3677 call(RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::dtan))); |
|
3678 } |
|
3679 break; |
|
3680 default: |
|
3681 assert(false, "bad intrinsic"); |
|
3682 break; |
|
3683 } |
|
3684 addl(rsp, wordSize * 2); |
|
3685 if (num_fpu_regs_in_use > 1) { |
|
3686 // Must save return value to stack and then restore entire FPU stack |
|
3687 fstp_d(Address(rsp, incoming_argument_and_return_value_offset)); |
|
3688 for (int i = 0; i < num_fpu_regs_in_use; i++) { |
|
3689 fld_d(Address(rsp, 0)); |
|
3690 addl(rsp, wordSize*2); |
|
3691 } |
|
3692 } |
|
3693 popad(); |
|
3694 |
|
3695 // Come here with result in F-TOS |
|
3696 bind(done); |
|
3697 |
|
3698 if (tmp != noreg) { |
|
3699 popl(tmp); |
|
3700 } |
|
3701 } |
|
3702 |
|
3703 void MacroAssembler::jC2(Register tmp, Label& L) { |
|
3704 // set parity bit if FPU flag C2 is set (via rax) |
|
3705 save_rax(tmp); |
|
3706 fwait(); fnstsw_ax(); |
|
3707 sahf(); |
|
3708 restore_rax(tmp); |
|
3709 // branch |
|
3710 jcc(Assembler::parity, L); |
|
3711 } |
|
3712 |
|
3713 |
|
3714 void MacroAssembler::jnC2(Register tmp, Label& L) { |
|
3715 // set parity bit if FPU flag C2 is set (via rax) |
|
3716 save_rax(tmp); |
|
3717 fwait(); fnstsw_ax(); |
|
3718 sahf(); |
|
3719 restore_rax(tmp); |
|
3720 // branch |
|
3721 jcc(Assembler::noParity, L); |
|
3722 } |
|
3723 |
|
3724 |
|
3725 void MacroAssembler::fcmp(Register tmp) { |
|
3726 fcmp(tmp, 1, true, true); |
|
3727 } |
|
3728 |
|
3729 |
|
3730 void MacroAssembler::fcmp(Register tmp, int index, bool pop_left, bool pop_right) { |
|
3731 assert(!pop_right || pop_left, "usage error"); |
|
3732 if (VM_Version::supports_cmov()) { |
|
3733 assert(tmp == noreg, "unneeded temp"); |
|
3734 if (pop_left) { |
|
3735 fucomip(index); |
|
3736 } else { |
|
3737 fucomi(index); |
|
3738 } |
|
3739 if (pop_right) { |
|
3740 fpop(); |
|
3741 } |
|
3742 } else { |
|
3743 assert(tmp != noreg, "need temp"); |
|
3744 if (pop_left) { |
|
3745 if (pop_right) { |
|
3746 fcompp(); |
|
3747 } else { |
|
3748 fcomp(index); |
|
3749 } |
|
3750 } else { |
|
3751 fcom(index); |
|
3752 } |
|
3753 // convert FPU condition into eflags condition via rax, |
|
3754 save_rax(tmp); |
|
3755 fwait(); fnstsw_ax(); |
|
3756 sahf(); |
|
3757 restore_rax(tmp); |
|
3758 } |
|
3759 // condition codes set as follows: |
|
3760 // |
|
3761 // CF (corresponds to C0) if x < y |
|
3762 // PF (corresponds to C2) if unordered |
|
3763 // ZF (corresponds to C3) if x = y |
|
3764 } |
|
3765 |
|
3766 |
|
3767 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less) { |
|
3768 fcmp2int(dst, unordered_is_less, 1, true, true); |
|
3769 } |
|
3770 |
|
3771 |
|
3772 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right) { |
|
3773 fcmp(VM_Version::supports_cmov() ? noreg : dst, index, pop_left, pop_right); |
|
3774 Label L; |
|
3775 if (unordered_is_less) { |
|
3776 movl(dst, -1); |
|
3777 jcc(Assembler::parity, L); |
|
3778 jcc(Assembler::below , L); |
|
3779 movl(dst, 0); |
|
3780 jcc(Assembler::equal , L); |
|
3781 increment(dst); |
|
3782 } else { // unordered is greater |
|
3783 movl(dst, 1); |
|
3784 jcc(Assembler::parity, L); |
|
3785 jcc(Assembler::above , L); |
|
3786 movl(dst, 0); |
|
3787 jcc(Assembler::equal , L); |
|
3788 decrement(dst); |
|
3789 } |
|
3790 bind(L); |
|
3791 } |
|
3792 |
|
3793 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) { |
|
3794 ucomiss(opr1, opr2); |
|
3795 |
|
3796 Label L; |
|
3797 if (unordered_is_less) { |
|
3798 movl(dst, -1); |
|
3799 jcc(Assembler::parity, L); |
|
3800 jcc(Assembler::below , L); |
|
3801 movl(dst, 0); |
|
3802 jcc(Assembler::equal , L); |
|
3803 increment(dst); |
|
3804 } else { // unordered is greater |
|
3805 movl(dst, 1); |
|
3806 jcc(Assembler::parity, L); |
|
3807 jcc(Assembler::above , L); |
|
3808 movl(dst, 0); |
|
3809 jcc(Assembler::equal , L); |
|
3810 decrement(dst); |
|
3811 } |
|
3812 bind(L); |
|
3813 } |
|
3814 |
|
3815 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) { |
|
3816 ucomisd(opr1, opr2); |
|
3817 |
|
3818 Label L; |
|
3819 if (unordered_is_less) { |
|
3820 movl(dst, -1); |
|
3821 jcc(Assembler::parity, L); |
|
3822 jcc(Assembler::below , L); |
|
3823 movl(dst, 0); |
|
3824 jcc(Assembler::equal , L); |
|
3825 increment(dst); |
|
3826 } else { // unordered is greater |
|
3827 movl(dst, 1); |
|
3828 jcc(Assembler::parity, L); |
|
3829 jcc(Assembler::above , L); |
|
3830 movl(dst, 0); |
|
3831 jcc(Assembler::equal , L); |
|
3832 decrement(dst); |
|
3833 } |
|
3834 bind(L); |
|
3835 } |
|
3836 |
|
3837 |
|
3838 |
|
3839 void MacroAssembler::fpop() { |
|
3840 ffree(); |
|
3841 fincstp(); |
|
3842 } |
|
3843 |
|
3844 |
|
3845 void MacroAssembler::sign_extend_short(Register reg) { |
|
3846 if (VM_Version::is_P6()) { |
|
3847 movsxw(reg, reg); |
|
3848 } else { |
|
3849 shll(reg, 16); |
|
3850 sarl(reg, 16); |
|
3851 } |
|
3852 } |
|
3853 |
|
3854 |
|
3855 void MacroAssembler::sign_extend_byte(Register reg) { |
|
3856 if (VM_Version::is_P6() && reg->has_byte_register()) { |
|
3857 movsxb(reg, reg); |
|
3858 } else { |
|
3859 shll(reg, 24); |
|
3860 sarl(reg, 24); |
|
3861 } |
|
3862 } |
|
3863 |
|
3864 |
|
3865 void MacroAssembler::division_with_shift (Register reg, int shift_value) { |
|
3866 assert (shift_value > 0, "illegal shift value"); |
|
3867 Label _is_positive; |
|
3868 testl (reg, reg); |
|
3869 jcc (Assembler::positive, _is_positive); |
|
3870 int offset = (1 << shift_value) - 1 ; |
|
3871 |
|
3872 increment(reg, offset); |
|
3873 |
|
3874 bind (_is_positive); |
|
3875 sarl(reg, shift_value); |
|
3876 } |
|
3877 |
|
3878 |
|
3879 void MacroAssembler::round_to(Register reg, int modulus) { |
|
3880 addl(reg, modulus - 1); |
|
3881 andl(reg, -modulus); |
|
3882 } |
|
3883 |
|
3884 // C++ bool manipulation |
|
3885 |
|
3886 void MacroAssembler::movbool(Register dst, Address src) { |
|
3887 if(sizeof(bool) == 1) |
|
3888 movb(dst, src); |
|
3889 else if(sizeof(bool) == 2) |
|
3890 movw(dst, src); |
|
3891 else if(sizeof(bool) == 4) |
|
3892 movl(dst, src); |
|
3893 else |
|
3894 // unsupported |
|
3895 ShouldNotReachHere(); |
|
3896 } |
|
3897 |
|
3898 void MacroAssembler::movbool(Address dst, bool boolconst) { |
|
3899 if(sizeof(bool) == 1) |
|
3900 movb(dst, (int) boolconst); |
|
3901 else if(sizeof(bool) == 2) |
|
3902 movw(dst, (int) boolconst); |
|
3903 else if(sizeof(bool) == 4) |
|
3904 movl(dst, (int) boolconst); |
|
3905 else |
|
3906 // unsupported |
|
3907 ShouldNotReachHere(); |
|
3908 } |
|
3909 |
|
3910 void MacroAssembler::movbool(Address dst, Register src) { |
|
3911 if(sizeof(bool) == 1) |
|
3912 movb(dst, src); |
|
3913 else if(sizeof(bool) == 2) |
|
3914 movw(dst, src); |
|
3915 else if(sizeof(bool) == 4) |
|
3916 movl(dst, src); |
|
3917 else |
|
3918 // unsupported |
|
3919 ShouldNotReachHere(); |
|
3920 } |
|
3921 |
|
3922 void MacroAssembler::testbool(Register dst) { |
|
3923 if(sizeof(bool) == 1) |
|
3924 testb(dst, (int) 0xff); |
|
3925 else if(sizeof(bool) == 2) { |
|
3926 // testw implementation needed for two byte bools |
|
3927 ShouldNotReachHere(); |
|
3928 } else if(sizeof(bool) == 4) |
|
3929 testl(dst, dst); |
|
3930 else |
|
3931 // unsupported |
|
3932 ShouldNotReachHere(); |
|
3933 } |
|
3934 |
|
3935 void MacroAssembler::verify_oop(Register reg, const char* s) { |
|
3936 if (!VerifyOops) return; |
|
3937 // Pass register number to verify_oop_subroutine |
|
3938 char* b = new char[strlen(s) + 50]; |
|
3939 sprintf(b, "verify_oop: %s: %s", reg->name(), s); |
|
3940 pushl(rax); // save rax, |
|
3941 pushl(reg); // pass register argument |
|
3942 ExternalAddress buffer((address) b); |
|
3943 pushptr(buffer.addr()); |
|
3944 // call indirectly to solve generation ordering problem |
|
3945 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address())); |
|
3946 call(rax); |
|
3947 } |
|
3948 |
|
3949 |
|
3950 void MacroAssembler::verify_oop_addr(Address addr, const char* s) { |
|
3951 if (!VerifyOops) return; |
|
3952 // QQQ fix this |
|
3953 // Address adjust(addr.base(), addr.index(), addr.scale(), addr.disp() + BytesPerWord); |
|
3954 // Pass register number to verify_oop_subroutine |
|
3955 char* b = new char[strlen(s) + 50]; |
|
3956 sprintf(b, "verify_oop_addr: %s", s); |
|
3957 pushl(rax); // save rax, |
|
3958 // addr may contain rsp so we will have to adjust it based on the push |
|
3959 // we just did |
|
3960 if (addr.uses(rsp)) { |
|
3961 leal(rax, addr); |
|
3962 pushl(Address(rax, BytesPerWord)); |
|
3963 } else { |
|
3964 pushl(addr); |
|
3965 } |
|
3966 ExternalAddress buffer((address) b); |
|
3967 // pass msg argument |
|
3968 pushptr(buffer.addr()); |
|
3969 // call indirectly to solve generation ordering problem |
|
3970 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address())); |
|
3971 call(rax); |
|
3972 // Caller pops the arguments and restores rax, from the stack |
|
3973 } |
|
3974 |
|
3975 |
|
3976 void MacroAssembler::stop(const char* msg) { |
|
3977 ExternalAddress message((address)msg); |
|
3978 // push address of message |
|
3979 pushptr(message.addr()); |
|
3980 { Label L; call(L, relocInfo::none); bind(L); } // push eip |
|
3981 pushad(); // push registers |
|
3982 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug))); |
|
3983 hlt(); |
|
3984 } |
|
3985 |
|
3986 |
|
3987 void MacroAssembler::warn(const char* msg) { |
|
3988 push_CPU_state(); |
|
3989 |
|
3990 ExternalAddress message((address) msg); |
|
3991 // push address of message |
|
3992 pushptr(message.addr()); |
|
3993 |
|
3994 call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning))); |
|
3995 addl(rsp, wordSize); // discard argument |
|
3996 pop_CPU_state(); |
|
3997 } |
|
3998 |
|
3999 |
|
4000 void MacroAssembler::debug(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg) { |
|
4001 // In order to get locks to work, we need to fake a in_VM state |
|
4002 JavaThread* thread = JavaThread::current(); |
|
4003 JavaThreadState saved_state = thread->thread_state(); |
|
4004 thread->set_thread_state(_thread_in_vm); |
|
4005 if (ShowMessageBoxOnError) { |
|
4006 JavaThread* thread = JavaThread::current(); |
|
4007 JavaThreadState saved_state = thread->thread_state(); |
|
4008 thread->set_thread_state(_thread_in_vm); |
|
4009 ttyLocker ttyl; |
|
4010 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) { |
|
4011 BytecodeCounter::print(); |
|
4012 } |
|
4013 // To see where a verify_oop failed, get $ebx+40/X for this frame. |
|
4014 // This is the value of eip which points to where verify_oop will return. |
|
4015 if (os::message_box(msg, "Execution stopped, print registers?")) { |
|
4016 tty->print_cr("eip = 0x%08x", eip); |
|
4017 tty->print_cr("rax, = 0x%08x", rax); |
|
4018 tty->print_cr("rbx, = 0x%08x", rbx); |
|
4019 tty->print_cr("rcx = 0x%08x", rcx); |
|
4020 tty->print_cr("rdx = 0x%08x", rdx); |
|
4021 tty->print_cr("rdi = 0x%08x", rdi); |
|
4022 tty->print_cr("rsi = 0x%08x", rsi); |
|
4023 tty->print_cr("rbp, = 0x%08x", rbp); |
|
4024 tty->print_cr("rsp = 0x%08x", rsp); |
|
4025 BREAKPOINT; |
|
4026 } |
|
4027 } else { |
|
4028 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg); |
|
4029 assert(false, "DEBUG MESSAGE"); |
|
4030 } |
|
4031 ThreadStateTransition::transition(thread, _thread_in_vm, saved_state); |
|
4032 } |
|
4033 |
|
4034 |
|
4035 |
|
4036 void MacroAssembler::os_breakpoint() { |
|
4037 // instead of directly emitting a breakpoint, call os:breakpoint for better debugability |
|
4038 // (e.g., MSVC can't call ps() otherwise) |
|
4039 call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint))); |
|
4040 } |
|
4041 |
|
4042 |
|
4043 void MacroAssembler::push_fTOS() { |
|
4044 subl(rsp, 2 * wordSize); |
|
4045 fstp_d(Address(rsp, 0)); |
|
4046 } |
|
4047 |
|
4048 |
|
4049 void MacroAssembler::pop_fTOS() { |
|
4050 fld_d(Address(rsp, 0)); |
|
4051 addl(rsp, 2 * wordSize); |
|
4052 } |
|
4053 |
|
4054 |
|
4055 void MacroAssembler::empty_FPU_stack() { |
|
4056 if (VM_Version::supports_mmx()) { |
|
4057 emms(); |
|
4058 } else { |
|
4059 for (int i = 8; i-- > 0; ) ffree(i); |
|
4060 } |
|
4061 } |
|
4062 |
|
4063 |
|
4064 class ControlWord { |
|
4065 public: |
|
4066 int32_t _value; |
|
4067 |
|
4068 int rounding_control() const { return (_value >> 10) & 3 ; } |
|
4069 int precision_control() const { return (_value >> 8) & 3 ; } |
|
4070 bool precision() const { return ((_value >> 5) & 1) != 0; } |
|
4071 bool underflow() const { return ((_value >> 4) & 1) != 0; } |
|
4072 bool overflow() const { return ((_value >> 3) & 1) != 0; } |
|
4073 bool zero_divide() const { return ((_value >> 2) & 1) != 0; } |
|
4074 bool denormalized() const { return ((_value >> 1) & 1) != 0; } |
|
4075 bool invalid() const { return ((_value >> 0) & 1) != 0; } |
|
4076 |
|
4077 void print() const { |
|
4078 // rounding control |
|
4079 const char* rc; |
|
4080 switch (rounding_control()) { |
|
4081 case 0: rc = "round near"; break; |
|
4082 case 1: rc = "round down"; break; |
|
4083 case 2: rc = "round up "; break; |
|
4084 case 3: rc = "chop "; break; |
|
4085 }; |
|
4086 // precision control |
|
4087 const char* pc; |
|
4088 switch (precision_control()) { |
|
4089 case 0: pc = "24 bits "; break; |
|
4090 case 1: pc = "reserved"; break; |
|
4091 case 2: pc = "53 bits "; break; |
|
4092 case 3: pc = "64 bits "; break; |
|
4093 }; |
|
4094 // flags |
|
4095 char f[9]; |
|
4096 f[0] = ' '; |
|
4097 f[1] = ' '; |
|
4098 f[2] = (precision ()) ? 'P' : 'p'; |
|
4099 f[3] = (underflow ()) ? 'U' : 'u'; |
|
4100 f[4] = (overflow ()) ? 'O' : 'o'; |
|
4101 f[5] = (zero_divide ()) ? 'Z' : 'z'; |
|
4102 f[6] = (denormalized()) ? 'D' : 'd'; |
|
4103 f[7] = (invalid ()) ? 'I' : 'i'; |
|
4104 f[8] = '\x0'; |
|
4105 // output |
|
4106 printf("%04x masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc); |
|
4107 } |
|
4108 |
|
4109 }; |
|
4110 |
|
4111 |
|
4112 class StatusWord { |
|
4113 public: |
|
4114 int32_t _value; |
|
4115 |
|
4116 bool busy() const { return ((_value >> 15) & 1) != 0; } |
|
4117 bool C3() const { return ((_value >> 14) & 1) != 0; } |
|
4118 bool C2() const { return ((_value >> 10) & 1) != 0; } |
|
4119 bool C1() const { return ((_value >> 9) & 1) != 0; } |
|
4120 bool C0() const { return ((_value >> 8) & 1) != 0; } |
|
4121 int top() const { return (_value >> 11) & 7 ; } |
|
4122 bool error_status() const { return ((_value >> 7) & 1) != 0; } |
|
4123 bool stack_fault() const { return ((_value >> 6) & 1) != 0; } |
|
4124 bool precision() const { return ((_value >> 5) & 1) != 0; } |
|
4125 bool underflow() const { return ((_value >> 4) & 1) != 0; } |
|
4126 bool overflow() const { return ((_value >> 3) & 1) != 0; } |
|
4127 bool zero_divide() const { return ((_value >> 2) & 1) != 0; } |
|
4128 bool denormalized() const { return ((_value >> 1) & 1) != 0; } |
|
4129 bool invalid() const { return ((_value >> 0) & 1) != 0; } |
|
4130 |
|
4131 void print() const { |
|
4132 // condition codes |
|
4133 char c[5]; |
|
4134 c[0] = (C3()) ? '3' : '-'; |
|
4135 c[1] = (C2()) ? '2' : '-'; |
|
4136 c[2] = (C1()) ? '1' : '-'; |
|
4137 c[3] = (C0()) ? '0' : '-'; |
|
4138 c[4] = '\x0'; |
|
4139 // flags |
|
4140 char f[9]; |
|
4141 f[0] = (error_status()) ? 'E' : '-'; |
|
4142 f[1] = (stack_fault ()) ? 'S' : '-'; |
|
4143 f[2] = (precision ()) ? 'P' : '-'; |
|
4144 f[3] = (underflow ()) ? 'U' : '-'; |
|
4145 f[4] = (overflow ()) ? 'O' : '-'; |
|
4146 f[5] = (zero_divide ()) ? 'Z' : '-'; |
|
4147 f[6] = (denormalized()) ? 'D' : '-'; |
|
4148 f[7] = (invalid ()) ? 'I' : '-'; |
|
4149 f[8] = '\x0'; |
|
4150 // output |
|
4151 printf("%04x flags = %s, cc = %s, top = %d", _value & 0xFFFF, f, c, top()); |
|
4152 } |
|
4153 |
|
4154 }; |
|
4155 |
|
4156 |
|
4157 class TagWord { |
|
4158 public: |
|
4159 int32_t _value; |
|
4160 |
|
4161 int tag_at(int i) const { return (_value >> (i*2)) & 3; } |
|
4162 |
|
4163 void print() const { |
|
4164 printf("%04x", _value & 0xFFFF); |
|
4165 } |
|
4166 |
|
4167 }; |
|
4168 |
|
4169 |
|
4170 class FPU_Register { |
|
4171 public: |
|
4172 int32_t _m0; |
|
4173 int32_t _m1; |
|
4174 int16_t _ex; |
|
4175 |
|
4176 bool is_indefinite() const { |
|
4177 return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0; |
|
4178 } |
|
4179 |
|
4180 void print() const { |
|
4181 char sign = (_ex < 0) ? '-' : '+'; |
|
4182 const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : " "; |
|
4183 printf("%c%04hx.%08x%08x %s", sign, _ex, _m1, _m0, kind); |
|
4184 }; |
|
4185 |
|
4186 }; |
|
4187 |
|
4188 |
|
4189 class FPU_State { |
|
4190 public: |
|
4191 enum { |
|
4192 register_size = 10, |
|
4193 number_of_registers = 8, |
|
4194 register_mask = 7 |
|
4195 }; |
|
4196 |
|
4197 ControlWord _control_word; |
|
4198 StatusWord _status_word; |
|
4199 TagWord _tag_word; |
|
4200 int32_t _error_offset; |
|
4201 int32_t _error_selector; |
|
4202 int32_t _data_offset; |
|
4203 int32_t _data_selector; |
|
4204 int8_t _register[register_size * number_of_registers]; |
|
4205 |
|
4206 int tag_for_st(int i) const { return _tag_word.tag_at((_status_word.top() + i) & register_mask); } |
|
4207 FPU_Register* st(int i) const { return (FPU_Register*)&_register[register_size * i]; } |
|
4208 |
|
4209 const char* tag_as_string(int tag) const { |
|
4210 switch (tag) { |
|
4211 case 0: return "valid"; |
|
4212 case 1: return "zero"; |
|
4213 case 2: return "special"; |
|
4214 case 3: return "empty"; |
|
4215 } |
|
4216 ShouldNotReachHere() |
|
4217 return NULL; |
|
4218 } |
|
4219 |
|
4220 void print() const { |
|
4221 // print computation registers |
|
4222 { int t = _status_word.top(); |
|
4223 for (int i = 0; i < number_of_registers; i++) { |
|
4224 int j = (i - t) & register_mask; |
|
4225 printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j); |
|
4226 st(j)->print(); |
|
4227 printf(" %s\n", tag_as_string(_tag_word.tag_at(i))); |
|
4228 } |
|
4229 } |
|
4230 printf("\n"); |
|
4231 // print control registers |
|
4232 printf("ctrl = "); _control_word.print(); printf("\n"); |
|
4233 printf("stat = "); _status_word .print(); printf("\n"); |
|
4234 printf("tags = "); _tag_word .print(); printf("\n"); |
|
4235 } |
|
4236 |
|
4237 }; |
|
4238 |
|
4239 |
|
4240 class Flag_Register { |
|
4241 public: |
|
4242 int32_t _value; |
|
4243 |
|
4244 bool overflow() const { return ((_value >> 11) & 1) != 0; } |
|
4245 bool direction() const { return ((_value >> 10) & 1) != 0; } |
|
4246 bool sign() const { return ((_value >> 7) & 1) != 0; } |
|
4247 bool zero() const { return ((_value >> 6) & 1) != 0; } |
|
4248 bool auxiliary_carry() const { return ((_value >> 4) & 1) != 0; } |
|
4249 bool parity() const { return ((_value >> 2) & 1) != 0; } |
|
4250 bool carry() const { return ((_value >> 0) & 1) != 0; } |
|
4251 |
|
4252 void print() const { |
|
4253 // flags |
|
4254 char f[8]; |
|
4255 f[0] = (overflow ()) ? 'O' : '-'; |
|
4256 f[1] = (direction ()) ? 'D' : '-'; |
|
4257 f[2] = (sign ()) ? 'S' : '-'; |
|
4258 f[3] = (zero ()) ? 'Z' : '-'; |
|
4259 f[4] = (auxiliary_carry()) ? 'A' : '-'; |
|
4260 f[5] = (parity ()) ? 'P' : '-'; |
|
4261 f[6] = (carry ()) ? 'C' : '-'; |
|
4262 f[7] = '\x0'; |
|
4263 // output |
|
4264 printf("%08x flags = %s", _value, f); |
|
4265 } |
|
4266 |
|
4267 }; |
|
4268 |
|
4269 |
|
4270 class IU_Register { |
|
4271 public: |
|
4272 int32_t _value; |
|
4273 |
|
4274 void print() const { |
|
4275 printf("%08x %11d", _value, _value); |
|
4276 } |
|
4277 |
|
4278 }; |
|
4279 |
|
4280 |
|
4281 class IU_State { |
|
4282 public: |
|
4283 Flag_Register _eflags; |
|
4284 IU_Register _rdi; |
|
4285 IU_Register _rsi; |
|
4286 IU_Register _rbp; |
|
4287 IU_Register _rsp; |
|
4288 IU_Register _rbx; |
|
4289 IU_Register _rdx; |
|
4290 IU_Register _rcx; |
|
4291 IU_Register _rax; |
|
4292 |
|
4293 void print() const { |
|
4294 // computation registers |
|
4295 printf("rax, = "); _rax.print(); printf("\n"); |
|
4296 printf("rbx, = "); _rbx.print(); printf("\n"); |
|
4297 printf("rcx = "); _rcx.print(); printf("\n"); |
|
4298 printf("rdx = "); _rdx.print(); printf("\n"); |
|
4299 printf("rdi = "); _rdi.print(); printf("\n"); |
|
4300 printf("rsi = "); _rsi.print(); printf("\n"); |
|
4301 printf("rbp, = "); _rbp.print(); printf("\n"); |
|
4302 printf("rsp = "); _rsp.print(); printf("\n"); |
|
4303 printf("\n"); |
|
4304 // control registers |
|
4305 printf("flgs = "); _eflags.print(); printf("\n"); |
|
4306 } |
|
4307 }; |
|
4308 |
|
4309 |
|
4310 class CPU_State { |
|
4311 public: |
|
4312 FPU_State _fpu_state; |
|
4313 IU_State _iu_state; |
|
4314 |
|
4315 void print() const { |
|
4316 printf("--------------------------------------------------\n"); |
|
4317 _iu_state .print(); |
|
4318 printf("\n"); |
|
4319 _fpu_state.print(); |
|
4320 printf("--------------------------------------------------\n"); |
|
4321 } |
|
4322 |
|
4323 }; |
|
4324 |
|
4325 |
|
4326 static void _print_CPU_state(CPU_State* state) { |
|
4327 state->print(); |
|
4328 }; |
|
4329 |
|
4330 |
|
4331 void MacroAssembler::print_CPU_state() { |
|
4332 push_CPU_state(); |
|
4333 pushl(rsp); // pass CPU state |
|
4334 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state))); |
|
4335 addl(rsp, wordSize); // discard argument |
|
4336 pop_CPU_state(); |
|
4337 } |
|
4338 |
|
4339 |
|
4340 static bool _verify_FPU(int stack_depth, char* s, CPU_State* state) { |
|
4341 static int counter = 0; |
|
4342 FPU_State* fs = &state->_fpu_state; |
|
4343 counter++; |
|
4344 // For leaf calls, only verify that the top few elements remain empty. |
|
4345 // We only need 1 empty at the top for C2 code. |
|
4346 if( stack_depth < 0 ) { |
|
4347 if( fs->tag_for_st(7) != 3 ) { |
|
4348 printf("FPR7 not empty\n"); |
|
4349 state->print(); |
|
4350 assert(false, "error"); |
|
4351 return false; |
|
4352 } |
|
4353 return true; // All other stack states do not matter |
|
4354 } |
|
4355 |
|
4356 assert((fs->_control_word._value & 0xffff) == StubRoutines::_fpu_cntrl_wrd_std, |
|
4357 "bad FPU control word"); |
|
4358 |
|
4359 // compute stack depth |
|
4360 int i = 0; |
|
4361 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) < 3) i++; |
|
4362 int d = i; |
|
4363 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) == 3) i++; |
|
4364 // verify findings |
|
4365 if (i != FPU_State::number_of_registers) { |
|
4366 // stack not contiguous |
|
4367 printf("%s: stack not contiguous at ST%d\n", s, i); |
|
4368 state->print(); |
|
4369 assert(false, "error"); |
|
4370 return false; |
|
4371 } |
|
4372 // check if computed stack depth corresponds to expected stack depth |
|
4373 if (stack_depth < 0) { |
|
4374 // expected stack depth is -stack_depth or less |
|
4375 if (d > -stack_depth) { |
|
4376 // too many elements on the stack |
|
4377 printf("%s: <= %d stack elements expected but found %d\n", s, -stack_depth, d); |
|
4378 state->print(); |
|
4379 assert(false, "error"); |
|
4380 return false; |
|
4381 } |
|
4382 } else { |
|
4383 // expected stack depth is stack_depth |
|
4384 if (d != stack_depth) { |
|
4385 // wrong stack depth |
|
4386 printf("%s: %d stack elements expected but found %d\n", s, stack_depth, d); |
|
4387 state->print(); |
|
4388 assert(false, "error"); |
|
4389 return false; |
|
4390 } |
|
4391 } |
|
4392 // everything is cool |
|
4393 return true; |
|
4394 } |
|
4395 |
|
4396 |
|
4397 void MacroAssembler::verify_FPU(int stack_depth, const char* s) { |
|
4398 if (!VerifyFPU) return; |
|
4399 push_CPU_state(); |
|
4400 pushl(rsp); // pass CPU state |
|
4401 ExternalAddress msg((address) s); |
|
4402 // pass message string s |
|
4403 pushptr(msg.addr()); |
|
4404 pushl(stack_depth); // pass stack depth |
|
4405 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _verify_FPU))); |
|
4406 addl(rsp, 3 * wordSize); // discard arguments |
|
4407 // check for error |
|
4408 { Label L; |
|
4409 testl(rax, rax); |
|
4410 jcc(Assembler::notZero, L); |
|
4411 int3(); // break if error condition |
|
4412 bind(L); |
|
4413 } |
|
4414 pop_CPU_state(); |
|
4415 } |
|
4416 |
|
4417 |
|
4418 void MacroAssembler::push_IU_state() { |
|
4419 pushad(); |
|
4420 pushfd(); |
|
4421 } |
|
4422 |
|
4423 |
|
4424 void MacroAssembler::pop_IU_state() { |
|
4425 popfd(); |
|
4426 popad(); |
|
4427 } |
|
4428 |
|
4429 |
|
4430 void MacroAssembler::push_FPU_state() { |
|
4431 subl(rsp, FPUStateSizeInWords * wordSize); |
|
4432 fnsave(Address(rsp, 0)); |
|
4433 fwait(); |
|
4434 } |
|
4435 |
|
4436 |
|
4437 void MacroAssembler::pop_FPU_state() { |
|
4438 frstor(Address(rsp, 0)); |
|
4439 addl(rsp, FPUStateSizeInWords * wordSize); |
|
4440 } |
|
4441 |
|
4442 |
|
4443 void MacroAssembler::push_CPU_state() { |
|
4444 push_IU_state(); |
|
4445 push_FPU_state(); |
|
4446 } |
|
4447 |
|
4448 |
|
4449 void MacroAssembler::pop_CPU_state() { |
|
4450 pop_FPU_state(); |
|
4451 pop_IU_state(); |
|
4452 } |
|
4453 |
|
4454 |
|
4455 void MacroAssembler::push_callee_saved_registers() { |
|
4456 pushl(rsi); |
|
4457 pushl(rdi); |
|
4458 pushl(rdx); |
|
4459 pushl(rcx); |
|
4460 } |
|
4461 |
|
4462 |
|
4463 void MacroAssembler::pop_callee_saved_registers() { |
|
4464 popl(rcx); |
|
4465 popl(rdx); |
|
4466 popl(rdi); |
|
4467 popl(rsi); |
|
4468 } |
|
4469 |
|
4470 |
|
4471 void MacroAssembler::set_word_if_not_zero(Register dst) { |
|
4472 xorl(dst, dst); |
|
4473 set_byte_if_not_zero(dst); |
|
4474 } |
|
4475 |
|
4476 // Write serialization page so VM thread can do a pseudo remote membar. |
|
4477 // We use the current thread pointer to calculate a thread specific |
|
4478 // offset to write to within the page. This minimizes bus traffic |
|
4479 // due to cache line collision. |
|
4480 void MacroAssembler::serialize_memory(Register thread, Register tmp) { |
|
4481 movl(tmp, thread); |
|
4482 shrl(tmp, os::get_serialize_page_shift_count()); |
|
4483 andl(tmp, (os::vm_page_size() - sizeof(int))); |
|
4484 |
|
4485 Address index(noreg, tmp, Address::times_1); |
|
4486 ExternalAddress page(os::get_memory_serialize_page()); |
|
4487 |
|
4488 movptr(ArrayAddress(page, index), tmp); |
|
4489 } |
|
4490 |
|
4491 |
|
4492 void MacroAssembler::verify_tlab() { |
|
4493 #ifdef ASSERT |
|
4494 if (UseTLAB && VerifyOops) { |
|
4495 Label next, ok; |
|
4496 Register t1 = rsi; |
|
4497 Register thread_reg = rbx; |
|
4498 |
|
4499 pushl(t1); |
|
4500 pushl(thread_reg); |
|
4501 get_thread(thread_reg); |
|
4502 |
|
4503 movl(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); |
|
4504 cmpl(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset()))); |
|
4505 jcc(Assembler::aboveEqual, next); |
|
4506 stop("assert(top >= start)"); |
|
4507 should_not_reach_here(); |
|
4508 |
|
4509 bind(next); |
|
4510 movl(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset()))); |
|
4511 cmpl(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); |
|
4512 jcc(Assembler::aboveEqual, ok); |
|
4513 stop("assert(top <= end)"); |
|
4514 should_not_reach_here(); |
|
4515 |
|
4516 bind(ok); |
|
4517 popl(thread_reg); |
|
4518 popl(t1); |
|
4519 } |
|
4520 #endif |
|
4521 } |
|
4522 |
|
4523 |
|
4524 // Defines obj, preserves var_size_in_bytes |
|
4525 void MacroAssembler::eden_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, |
|
4526 Register t1, Label& slow_case) { |
|
4527 assert(obj == rax, "obj must be in rax, for cmpxchg"); |
|
4528 assert_different_registers(obj, var_size_in_bytes, t1); |
|
4529 Register end = t1; |
|
4530 Label retry; |
|
4531 bind(retry); |
|
4532 ExternalAddress heap_top((address) Universe::heap()->top_addr()); |
|
4533 movptr(obj, heap_top); |
|
4534 if (var_size_in_bytes == noreg) { |
|
4535 leal(end, Address(obj, con_size_in_bytes)); |
|
4536 } else { |
|
4537 leal(end, Address(obj, var_size_in_bytes, Address::times_1)); |
|
4538 } |
|
4539 // if end < obj then we wrapped around => object too long => slow case |
|
4540 cmpl(end, obj); |
|
4541 jcc(Assembler::below, slow_case); |
|
4542 cmpptr(end, ExternalAddress((address) Universe::heap()->end_addr())); |
|
4543 jcc(Assembler::above, slow_case); |
|
4544 // Compare obj with the top addr, and if still equal, store the new top addr in |
|
4545 // end at the address of the top addr pointer. Sets ZF if was equal, and clears |
|
4546 // it otherwise. Use lock prefix for atomicity on MPs. |
|
4547 if (os::is_MP()) { |
|
4548 lock(); |
|
4549 } |
|
4550 cmpxchgptr(end, heap_top); |
|
4551 jcc(Assembler::notEqual, retry); |
|
4552 } |
|
4553 |
|
4554 |
|
4555 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes. |
|
4556 void MacroAssembler::tlab_allocate(Register obj, Register var_size_in_bytes, int con_size_in_bytes, |
|
4557 Register t1, Register t2, Label& slow_case) { |
|
4558 assert_different_registers(obj, t1, t2); |
|
4559 assert_different_registers(obj, var_size_in_bytes, t1); |
|
4560 Register end = t2; |
|
4561 Register thread = t1; |
|
4562 |
|
4563 verify_tlab(); |
|
4564 |
|
4565 get_thread(thread); |
|
4566 |
|
4567 movl(obj, Address(thread, JavaThread::tlab_top_offset())); |
|
4568 if (var_size_in_bytes == noreg) { |
|
4569 leal(end, Address(obj, con_size_in_bytes)); |
|
4570 } else { |
|
4571 leal(end, Address(obj, var_size_in_bytes, Address::times_1)); |
|
4572 } |
|
4573 cmpl(end, Address(thread, JavaThread::tlab_end_offset())); |
|
4574 jcc(Assembler::above, slow_case); |
|
4575 |
|
4576 // update the tlab top pointer |
|
4577 movl(Address(thread, JavaThread::tlab_top_offset()), end); |
|
4578 |
|
4579 // recover var_size_in_bytes if necessary |
|
4580 if (var_size_in_bytes == end) { |
|
4581 subl(var_size_in_bytes, obj); |
|
4582 } |
|
4583 verify_tlab(); |
|
4584 } |
|
4585 |
|
4586 // Preserves rbx, and rdx. |
|
4587 void MacroAssembler::tlab_refill(Label& retry, Label& try_eden, Label& slow_case) { |
|
4588 Register top = rax; |
|
4589 Register t1 = rcx; |
|
4590 Register t2 = rsi; |
|
4591 Register thread_reg = rdi; |
|
4592 assert_different_registers(top, thread_reg, t1, t2, /* preserve: */ rbx, rdx); |
|
4593 Label do_refill, discard_tlab; |
|
4594 |
|
4595 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) { |
|
4596 // No allocation in the shared eden. |
|
4597 jmp(slow_case); |
|
4598 } |
|
4599 |
|
4600 get_thread(thread_reg); |
|
4601 |
|
4602 movl(top, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); |
|
4603 movl(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset()))); |
|
4604 |
|
4605 // calculate amount of free space |
|
4606 subl(t1, top); |
|
4607 shrl(t1, LogHeapWordSize); |
|
4608 |
|
4609 // Retain tlab and allocate object in shared space if |
|
4610 // the amount free in the tlab is too large to discard. |
|
4611 cmpl(t1, Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset()))); |
|
4612 jcc(Assembler::lessEqual, discard_tlab); |
|
4613 |
|
4614 // Retain |
|
4615 movl(t2, ThreadLocalAllocBuffer::refill_waste_limit_increment()); |
|
4616 addl(Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset())), t2); |
|
4617 if (TLABStats) { |
|
4618 // increment number of slow_allocations |
|
4619 addl(Address(thread_reg, in_bytes(JavaThread::tlab_slow_allocations_offset())), 1); |
|
4620 } |
|
4621 jmp(try_eden); |
|
4622 |
|
4623 bind(discard_tlab); |
|
4624 if (TLABStats) { |
|
4625 // increment number of refills |
|
4626 addl(Address(thread_reg, in_bytes(JavaThread::tlab_number_of_refills_offset())), 1); |
|
4627 // accumulate wastage -- t1 is amount free in tlab |
|
4628 addl(Address(thread_reg, in_bytes(JavaThread::tlab_fast_refill_waste_offset())), t1); |
|
4629 } |
|
4630 |
|
4631 // if tlab is currently allocated (top or end != null) then |
|
4632 // fill [top, end + alignment_reserve) with array object |
|
4633 testl (top, top); |
|
4634 jcc(Assembler::zero, do_refill); |
|
4635 |
|
4636 // set up the mark word |
|
4637 movl(Address(top, oopDesc::mark_offset_in_bytes()), (int)markOopDesc::prototype()->copy_set_hash(0x2)); |
|
4638 // set the length to the remaining space |
|
4639 subl(t1, typeArrayOopDesc::header_size(T_INT)); |
|
4640 addl(t1, ThreadLocalAllocBuffer::alignment_reserve()); |
|
4641 shll(t1, log2_intptr(HeapWordSize/sizeof(jint))); |
|
4642 movl(Address(top, arrayOopDesc::length_offset_in_bytes()), t1); |
|
4643 // set klass to intArrayKlass |
|
4644 // dubious reloc why not an oop reloc? |
|
4645 movptr(t1, ExternalAddress((address) Universe::intArrayKlassObj_addr())); |
|
4646 movl(Address(top, oopDesc::klass_offset_in_bytes()), t1); |
|
4647 |
|
4648 // refill the tlab with an eden allocation |
|
4649 bind(do_refill); |
|
4650 movl(t1, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset()))); |
|
4651 shll(t1, LogHeapWordSize); |
|
4652 // add object_size ?? |
|
4653 eden_allocate(top, t1, 0, t2, slow_case); |
|
4654 |
|
4655 // Check that t1 was preserved in eden_allocate. |
|
4656 #ifdef ASSERT |
|
4657 if (UseTLAB) { |
|
4658 Label ok; |
|
4659 Register tsize = rsi; |
|
4660 assert_different_registers(tsize, thread_reg, t1); |
|
4661 pushl(tsize); |
|
4662 movl(tsize, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset()))); |
|
4663 shll(tsize, LogHeapWordSize); |
|
4664 cmpl(t1, tsize); |
|
4665 jcc(Assembler::equal, ok); |
|
4666 stop("assert(t1 != tlab size)"); |
|
4667 should_not_reach_here(); |
|
4668 |
|
4669 bind(ok); |
|
4670 popl(tsize); |
|
4671 } |
|
4672 #endif |
|
4673 movl(Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())), top); |
|
4674 movl(Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())), top); |
|
4675 addl(top, t1); |
|
4676 subl(top, ThreadLocalAllocBuffer::alignment_reserve_in_bytes()); |
|
4677 movl(Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())), top); |
|
4678 verify_tlab(); |
|
4679 jmp(retry); |
|
4680 } |
|
4681 |
|
4682 |
|
4683 int MacroAssembler::biased_locking_enter(Register lock_reg, Register obj_reg, Register swap_reg, Register tmp_reg, |
|
4684 bool swap_reg_contains_mark, |
|
4685 Label& done, Label* slow_case, |
|
4686 BiasedLockingCounters* counters) { |
|
4687 assert(UseBiasedLocking, "why call this otherwise?"); |
|
4688 assert(swap_reg == rax, "swap_reg must be rax, for cmpxchg"); |
|
4689 assert_different_registers(lock_reg, obj_reg, swap_reg); |
|
4690 |
|
4691 if (PrintBiasedLockingStatistics && counters == NULL) |
|
4692 counters = BiasedLocking::counters(); |
|
4693 |
|
4694 bool need_tmp_reg = false; |
|
4695 if (tmp_reg == noreg) { |
|
4696 need_tmp_reg = true; |
|
4697 tmp_reg = lock_reg; |
|
4698 } else { |
|
4699 assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg); |
|
4700 } |
|
4701 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout"); |
|
4702 Address mark_addr (obj_reg, oopDesc::mark_offset_in_bytes()); |
|
4703 Address klass_addr (obj_reg, oopDesc::klass_offset_in_bytes()); |
|
4704 Address saved_mark_addr(lock_reg, 0); |
|
4705 |
|
4706 // Biased locking |
|
4707 // See whether the lock is currently biased toward our thread and |
|
4708 // whether the epoch is still valid |
|
4709 // Note that the runtime guarantees sufficient alignment of JavaThread |
|
4710 // pointers to allow age to be placed into low bits |
|
4711 // First check to see whether biasing is even enabled for this object |
|
4712 Label cas_label; |
|
4713 int null_check_offset = -1; |
|
4714 if (!swap_reg_contains_mark) { |
|
4715 null_check_offset = offset(); |
|
4716 movl(swap_reg, mark_addr); |
|
4717 } |
|
4718 if (need_tmp_reg) { |
|
4719 pushl(tmp_reg); |
|
4720 } |
|
4721 movl(tmp_reg, swap_reg); |
|
4722 andl(tmp_reg, markOopDesc::biased_lock_mask_in_place); |
|
4723 cmpl(tmp_reg, markOopDesc::biased_lock_pattern); |
|
4724 if (need_tmp_reg) { |
|
4725 popl(tmp_reg); |
|
4726 } |
|
4727 jcc(Assembler::notEqual, cas_label); |
|
4728 // The bias pattern is present in the object's header. Need to check |
|
4729 // whether the bias owner and the epoch are both still current. |
|
4730 // Note that because there is no current thread register on x86 we |
|
4731 // need to store off the mark word we read out of the object to |
|
4732 // avoid reloading it and needing to recheck invariants below. This |
|
4733 // store is unfortunate but it makes the overall code shorter and |
|
4734 // simpler. |
|
4735 movl(saved_mark_addr, swap_reg); |
|
4736 if (need_tmp_reg) { |
|
4737 pushl(tmp_reg); |
|
4738 } |
|
4739 get_thread(tmp_reg); |
|
4740 xorl(swap_reg, tmp_reg); |
|
4741 if (swap_reg_contains_mark) { |
|
4742 null_check_offset = offset(); |
|
4743 } |
|
4744 movl(tmp_reg, klass_addr); |
|
4745 xorl(swap_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); |
|
4746 andl(swap_reg, ~((int) markOopDesc::age_mask_in_place)); |
|
4747 if (need_tmp_reg) { |
|
4748 popl(tmp_reg); |
|
4749 } |
|
4750 if (counters != NULL) { |
|
4751 cond_inc32(Assembler::zero, |
|
4752 ExternalAddress((address)counters->biased_lock_entry_count_addr())); |
|
4753 } |
|
4754 jcc(Assembler::equal, done); |
|
4755 |
|
4756 Label try_revoke_bias; |
|
4757 Label try_rebias; |
|
4758 |
|
4759 // At this point we know that the header has the bias pattern and |
|
4760 // that we are not the bias owner in the current epoch. We need to |
|
4761 // figure out more details about the state of the header in order to |
|
4762 // know what operations can be legally performed on the object's |
|
4763 // header. |
|
4764 |
|
4765 // If the low three bits in the xor result aren't clear, that means |
|
4766 // the prototype header is no longer biased and we have to revoke |
|
4767 // the bias on this object. |
|
4768 testl(swap_reg, markOopDesc::biased_lock_mask_in_place); |
|
4769 jcc(Assembler::notZero, try_revoke_bias); |
|
4770 |
|
4771 // Biasing is still enabled for this data type. See whether the |
|
4772 // epoch of the current bias is still valid, meaning that the epoch |
|
4773 // bits of the mark word are equal to the epoch bits of the |
|
4774 // prototype header. (Note that the prototype header's epoch bits |
|
4775 // only change at a safepoint.) If not, attempt to rebias the object |
|
4776 // toward the current thread. Note that we must be absolutely sure |
|
4777 // that the current epoch is invalid in order to do this because |
|
4778 // otherwise the manipulations it performs on the mark word are |
|
4779 // illegal. |
|
4780 testl(swap_reg, markOopDesc::epoch_mask_in_place); |
|
4781 jcc(Assembler::notZero, try_rebias); |
|
4782 |
|
4783 // The epoch of the current bias is still valid but we know nothing |
|
4784 // about the owner; it might be set or it might be clear. Try to |
|
4785 // acquire the bias of the object using an atomic operation. If this |
|
4786 // fails we will go in to the runtime to revoke the object's bias. |
|
4787 // Note that we first construct the presumed unbiased header so we |
|
4788 // don't accidentally blow away another thread's valid bias. |
|
4789 movl(swap_reg, saved_mark_addr); |
|
4790 andl(swap_reg, |
|
4791 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place); |
|
4792 if (need_tmp_reg) { |
|
4793 pushl(tmp_reg); |
|
4794 } |
|
4795 get_thread(tmp_reg); |
|
4796 orl(tmp_reg, swap_reg); |
|
4797 if (os::is_MP()) { |
|
4798 lock(); |
|
4799 } |
|
4800 cmpxchg(tmp_reg, Address(obj_reg, 0)); |
|
4801 if (need_tmp_reg) { |
|
4802 popl(tmp_reg); |
|
4803 } |
|
4804 // If the biasing toward our thread failed, this means that |
|
4805 // another thread succeeded in biasing it toward itself and we |
|
4806 // need to revoke that bias. The revocation will occur in the |
|
4807 // interpreter runtime in the slow case. |
|
4808 if (counters != NULL) { |
|
4809 cond_inc32(Assembler::zero, |
|
4810 ExternalAddress((address)counters->anonymously_biased_lock_entry_count_addr())); |
|
4811 } |
|
4812 if (slow_case != NULL) { |
|
4813 jcc(Assembler::notZero, *slow_case); |
|
4814 } |
|
4815 jmp(done); |
|
4816 |
|
4817 bind(try_rebias); |
|
4818 // At this point we know the epoch has expired, meaning that the |
|
4819 // current "bias owner", if any, is actually invalid. Under these |
|
4820 // circumstances _only_, we are allowed to use the current header's |
|
4821 // value as the comparison value when doing the cas to acquire the |
|
4822 // bias in the current epoch. In other words, we allow transfer of |
|
4823 // the bias from one thread to another directly in this situation. |
|
4824 // |
|
4825 // FIXME: due to a lack of registers we currently blow away the age |
|
4826 // bits in this situation. Should attempt to preserve them. |
|
4827 if (need_tmp_reg) { |
|
4828 pushl(tmp_reg); |
|
4829 } |
|
4830 get_thread(tmp_reg); |
|
4831 movl(swap_reg, klass_addr); |
|
4832 orl(tmp_reg, Address(swap_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); |
|
4833 movl(swap_reg, saved_mark_addr); |
|
4834 if (os::is_MP()) { |
|
4835 lock(); |
|
4836 } |
|
4837 cmpxchg(tmp_reg, Address(obj_reg, 0)); |
|
4838 if (need_tmp_reg) { |
|
4839 popl(tmp_reg); |
|
4840 } |
|
4841 // If the biasing toward our thread failed, then another thread |
|
4842 // succeeded in biasing it toward itself and we need to revoke that |
|
4843 // bias. The revocation will occur in the runtime in the slow case. |
|
4844 if (counters != NULL) { |
|
4845 cond_inc32(Assembler::zero, |
|
4846 ExternalAddress((address)counters->rebiased_lock_entry_count_addr())); |
|
4847 } |
|
4848 if (slow_case != NULL) { |
|
4849 jcc(Assembler::notZero, *slow_case); |
|
4850 } |
|
4851 jmp(done); |
|
4852 |
|
4853 bind(try_revoke_bias); |
|
4854 // The prototype mark in the klass doesn't have the bias bit set any |
|
4855 // more, indicating that objects of this data type are not supposed |
|
4856 // to be biased any more. We are going to try to reset the mark of |
|
4857 // this object to the prototype value and fall through to the |
|
4858 // CAS-based locking scheme. Note that if our CAS fails, it means |
|
4859 // that another thread raced us for the privilege of revoking the |
|
4860 // bias of this particular object, so it's okay to continue in the |
|
4861 // normal locking code. |
|
4862 // |
|
4863 // FIXME: due to a lack of registers we currently blow away the age |
|
4864 // bits in this situation. Should attempt to preserve them. |
|
4865 movl(swap_reg, saved_mark_addr); |
|
4866 if (need_tmp_reg) { |
|
4867 pushl(tmp_reg); |
|
4868 } |
|
4869 movl(tmp_reg, klass_addr); |
|
4870 movl(tmp_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); |
|
4871 if (os::is_MP()) { |
|
4872 lock(); |
|
4873 } |
|
4874 cmpxchg(tmp_reg, Address(obj_reg, 0)); |
|
4875 if (need_tmp_reg) { |
|
4876 popl(tmp_reg); |
|
4877 } |
|
4878 // Fall through to the normal CAS-based lock, because no matter what |
|
4879 // the result of the above CAS, some thread must have succeeded in |
|
4880 // removing the bias bit from the object's header. |
|
4881 if (counters != NULL) { |
|
4882 cond_inc32(Assembler::zero, |
|
4883 ExternalAddress((address)counters->revoked_lock_entry_count_addr())); |
|
4884 } |
|
4885 |
|
4886 bind(cas_label); |
|
4887 |
|
4888 return null_check_offset; |
|
4889 } |
|
4890 |
|
4891 |
|
4892 void MacroAssembler::biased_locking_exit(Register obj_reg, Register temp_reg, Label& done) { |
|
4893 assert(UseBiasedLocking, "why call this otherwise?"); |
|
4894 |
|
4895 // Check for biased locking unlock case, which is a no-op |
|
4896 // Note: we do not have to check the thread ID for two reasons. |
|
4897 // First, the interpreter checks for IllegalMonitorStateException at |
|
4898 // a higher level. Second, if the bias was revoked while we held the |
|
4899 // lock, the object could not be rebiased toward another thread, so |
|
4900 // the bias bit would be clear. |
|
4901 movl(temp_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); |
|
4902 andl(temp_reg, markOopDesc::biased_lock_mask_in_place); |
|
4903 cmpl(temp_reg, markOopDesc::biased_lock_pattern); |
|
4904 jcc(Assembler::equal, done); |
|
4905 } |
|
4906 |
|
4907 |
|
4908 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) { |
|
4909 switch (cond) { |
|
4910 // Note some conditions are synonyms for others |
|
4911 case Assembler::zero: return Assembler::notZero; |
|
4912 case Assembler::notZero: return Assembler::zero; |
|
4913 case Assembler::less: return Assembler::greaterEqual; |
|
4914 case Assembler::lessEqual: return Assembler::greater; |
|
4915 case Assembler::greater: return Assembler::lessEqual; |
|
4916 case Assembler::greaterEqual: return Assembler::less; |
|
4917 case Assembler::below: return Assembler::aboveEqual; |
|
4918 case Assembler::belowEqual: return Assembler::above; |
|
4919 case Assembler::above: return Assembler::belowEqual; |
|
4920 case Assembler::aboveEqual: return Assembler::below; |
|
4921 case Assembler::overflow: return Assembler::noOverflow; |
|
4922 case Assembler::noOverflow: return Assembler::overflow; |
|
4923 case Assembler::negative: return Assembler::positive; |
|
4924 case Assembler::positive: return Assembler::negative; |
|
4925 case Assembler::parity: return Assembler::noParity; |
|
4926 case Assembler::noParity: return Assembler::parity; |
|
4927 } |
|
4928 ShouldNotReachHere(); return Assembler::overflow; |
|
4929 } |
|
4930 |
|
4931 |
|
4932 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr) { |
|
4933 Condition negated_cond = negate_condition(cond); |
|
4934 Label L; |
|
4935 jcc(negated_cond, L); |
|
4936 atomic_incl(counter_addr); |
|
4937 bind(L); |
|
4938 } |
|
4939 |
|
4940 void MacroAssembler::atomic_incl(AddressLiteral counter_addr) { |
|
4941 pushfd(); |
|
4942 if (os::is_MP()) |
|
4943 lock(); |
|
4944 increment(counter_addr); |
|
4945 popfd(); |
|
4946 } |
|
4947 |
|
4948 SkipIfEqual::SkipIfEqual( |
|
4949 MacroAssembler* masm, const bool* flag_addr, bool value) { |
|
4950 _masm = masm; |
|
4951 _masm->cmp8(ExternalAddress((address)flag_addr), value); |
|
4952 _masm->jcc(Assembler::equal, _label); |
|
4953 } |
|
4954 |
|
4955 SkipIfEqual::~SkipIfEqual() { |
|
4956 _masm->bind(_label); |
|
4957 } |
|
4958 |
|
4959 |
|
4960 // Writes to stack successive pages until offset reached to check for |
|
4961 // stack overflow + shadow pages. This clobbers tmp. |
|
4962 void MacroAssembler::bang_stack_size(Register size, Register tmp) { |
|
4963 movl(tmp, rsp); |
|
4964 // Bang stack for total size given plus shadow page size. |
|
4965 // Bang one page at a time because large size can bang beyond yellow and |
|
4966 // red zones. |
|
4967 Label loop; |
|
4968 bind(loop); |
|
4969 movl(Address(tmp, (-os::vm_page_size())), size ); |
|
4970 subl(tmp, os::vm_page_size()); |
|
4971 subl(size, os::vm_page_size()); |
|
4972 jcc(Assembler::greater, loop); |
|
4973 |
|
4974 // Bang down shadow pages too. |
|
4975 // The -1 because we already subtracted 1 page. |
|
4976 for (int i = 0; i< StackShadowPages-1; i++) { |
|
4977 movl(Address(tmp, (-i*os::vm_page_size())), size ); |
|
4978 } |
|
4979 } |