Mercurial > hg > truffle
annotate src/cpu/x86/vm/assembler_x86.cpp @ 1972:f95d63e2154a
6989984: Use standard include model for Hospot
Summary: Replaced MakeDeps and the includeDB files with more standardized solutions.
Reviewed-by: coleenp, kvn, kamg
author | stefank |
---|---|
date | Tue, 23 Nov 2010 13:22:55 -0800 |
parents | 2fe998383789 |
children | 0fc262af204f 2f644f85485d |
rev | line source |
---|---|
0 | 1 /* |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1513
diff
changeset
|
2 * Copyright (c) 1997, 2010, Oracle and/or its affiliates. All rights reserved. |
0 | 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 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1513
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1513
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1513
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1972 | 25 #include "precompiled.hpp" |
26 #include "assembler_x86.inline.hpp" | |
27 #include "gc_interface/collectedHeap.inline.hpp" | |
28 #include "interpreter/interpreter.hpp" | |
29 #include "memory/cardTableModRefBS.hpp" | |
30 #include "memory/resourceArea.hpp" | |
31 #include "prims/methodHandles.hpp" | |
32 #include "runtime/biasedLocking.hpp" | |
33 #include "runtime/interfaceSupport.hpp" | |
34 #include "runtime/objectMonitor.hpp" | |
35 #include "runtime/os.hpp" | |
36 #include "runtime/sharedRuntime.hpp" | |
37 #include "runtime/stubRoutines.hpp" | |
38 #ifndef SERIALGC | |
39 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp" | |
40 #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp" | |
41 #include "gc_implementation/g1/heapRegion.hpp" | |
42 #endif | |
0 | 43 |
44 // Implementation of AddressLiteral | |
45 | |
46 AddressLiteral::AddressLiteral(address target, relocInfo::relocType rtype) { | |
47 _is_lval = false; | |
48 _target = target; | |
49 switch (rtype) { | |
50 case relocInfo::oop_type: | |
51 // Oops are a special case. Normally they would be their own section | |
52 // but in cases like icBuffer they are literals in the code stream that | |
53 // we don't have a section for. We use none so that we get a literal address | |
54 // which is always patchable. | |
55 break; | |
56 case relocInfo::external_word_type: | |
57 _rspec = external_word_Relocation::spec(target); | |
58 break; | |
59 case relocInfo::internal_word_type: | |
60 _rspec = internal_word_Relocation::spec(target); | |
61 break; | |
62 case relocInfo::opt_virtual_call_type: | |
63 _rspec = opt_virtual_call_Relocation::spec(); | |
64 break; | |
65 case relocInfo::static_call_type: | |
66 _rspec = static_call_Relocation::spec(); | |
67 break; | |
68 case relocInfo::runtime_call_type: | |
69 _rspec = runtime_call_Relocation::spec(); | |
70 break; | |
71 case relocInfo::poll_type: | |
72 case relocInfo::poll_return_type: | |
73 _rspec = Relocation::spec_simple(rtype); | |
74 break; | |
75 case relocInfo::none: | |
76 break; | |
77 default: | |
78 ShouldNotReachHere(); | |
79 break; | |
80 } | |
81 } | |
82 | |
83 // Implementation of Address | |
84 | |
304 | 85 #ifdef _LP64 |
86 | |
0 | 87 Address Address::make_array(ArrayAddress adr) { |
88 // Not implementable on 64bit machines | |
89 // Should have been handled higher up the call chain. | |
90 ShouldNotReachHere(); | |
304 | 91 return Address(); |
92 } | |
93 | |
94 // exceedingly dangerous constructor | |
95 Address::Address(int disp, address loc, relocInfo::relocType rtype) { | |
96 _base = noreg; | |
97 _index = noreg; | |
98 _scale = no_scale; | |
99 _disp = disp; | |
100 switch (rtype) { | |
101 case relocInfo::external_word_type: | |
102 _rspec = external_word_Relocation::spec(loc); | |
103 break; | |
104 case relocInfo::internal_word_type: | |
105 _rspec = internal_word_Relocation::spec(loc); | |
106 break; | |
107 case relocInfo::runtime_call_type: | |
108 // HMM | |
109 _rspec = runtime_call_Relocation::spec(); | |
110 break; | |
111 case relocInfo::poll_type: | |
112 case relocInfo::poll_return_type: | |
113 _rspec = Relocation::spec_simple(rtype); | |
114 break; | |
115 case relocInfo::none: | |
116 break; | |
117 default: | |
118 ShouldNotReachHere(); | |
119 } | |
120 } | |
121 #else // LP64 | |
122 | |
123 Address Address::make_array(ArrayAddress adr) { | |
0 | 124 AddressLiteral base = adr.base(); |
125 Address index = adr.index(); | |
126 assert(index._disp == 0, "must not have disp"); // maybe it can? | |
127 Address array(index._base, index._index, index._scale, (intptr_t) base.target()); | |
128 array._rspec = base._rspec; | |
129 return array; | |
304 | 130 } |
0 | 131 |
132 // exceedingly dangerous constructor | |
133 Address::Address(address loc, RelocationHolder spec) { | |
134 _base = noreg; | |
135 _index = noreg; | |
136 _scale = no_scale; | |
137 _disp = (intptr_t) loc; | |
138 _rspec = spec; | |
139 } | |
304 | 140 |
0 | 141 #endif // _LP64 |
142 | |
304 | 143 |
144 | |
0 | 145 // Convert the raw encoding form into the form expected by the constructor for |
146 // Address. An index of 4 (rsp) corresponds to having no index, so convert | |
147 // that to noreg for the Address constructor. | |
624 | 148 Address Address::make_raw(int base, int index, int scale, int disp, bool disp_is_oop) { |
149 RelocationHolder rspec; | |
150 if (disp_is_oop) { | |
151 rspec = Relocation::spec_simple(relocInfo::oop_type); | |
152 } | |
0 | 153 bool valid_index = index != rsp->encoding(); |
154 if (valid_index) { | |
155 Address madr(as_Register(base), as_Register(index), (Address::ScaleFactor)scale, in_ByteSize(disp)); | |
624 | 156 madr._rspec = rspec; |
0 | 157 return madr; |
158 } else { | |
159 Address madr(as_Register(base), noreg, Address::no_scale, in_ByteSize(disp)); | |
624 | 160 madr._rspec = rspec; |
0 | 161 return madr; |
162 } | |
163 } | |
164 | |
165 // Implementation of Assembler | |
166 | |
167 int AbstractAssembler::code_fill_byte() { | |
168 return (u_char)'\xF4'; // hlt | |
169 } | |
170 | |
171 // make this go away someday | |
172 void Assembler::emit_data(jint data, relocInfo::relocType rtype, int format) { | |
173 if (rtype == relocInfo::none) | |
174 emit_long(data); | |
175 else emit_data(data, Relocation::spec_simple(rtype), format); | |
176 } | |
177 | |
178 void Assembler::emit_data(jint data, RelocationHolder const& rspec, int format) { | |
304 | 179 assert(imm_operand == 0, "default format must be immediate in this file"); |
0 | 180 assert(inst_mark() != NULL, "must be inside InstructionMark"); |
181 if (rspec.type() != relocInfo::none) { | |
182 #ifdef ASSERT | |
183 check_relocation(rspec, format); | |
184 #endif | |
185 // Do not use AbstractAssembler::relocate, which is not intended for | |
186 // embedded words. Instead, relocate to the enclosing instruction. | |
187 | |
188 // hack. call32 is too wide for mask so use disp32 | |
189 if (format == call32_operand) | |
190 code_section()->relocate(inst_mark(), rspec, disp32_operand); | |
191 else | |
192 code_section()->relocate(inst_mark(), rspec, format); | |
193 } | |
194 emit_long(data); | |
195 } | |
196 | |
304 | 197 static int encode(Register r) { |
198 int enc = r->encoding(); | |
199 if (enc >= 8) { | |
200 enc -= 8; | |
201 } | |
202 return enc; | |
203 } | |
204 | |
205 static int encode(XMMRegister r) { | |
206 int enc = r->encoding(); | |
207 if (enc >= 8) { | |
208 enc -= 8; | |
209 } | |
210 return enc; | |
211 } | |
0 | 212 |
213 void Assembler::emit_arith_b(int op1, int op2, Register dst, int imm8) { | |
214 assert(dst->has_byte_register(), "must have byte register"); | |
215 assert(isByte(op1) && isByte(op2), "wrong opcode"); | |
216 assert(isByte(imm8), "not a byte"); | |
217 assert((op1 & 0x01) == 0, "should be 8bit operation"); | |
218 emit_byte(op1); | |
304 | 219 emit_byte(op2 | encode(dst)); |
0 | 220 emit_byte(imm8); |
221 } | |
222 | |
223 | |
304 | 224 void Assembler::emit_arith(int op1, int op2, Register dst, int32_t imm32) { |
0 | 225 assert(isByte(op1) && isByte(op2), "wrong opcode"); |
226 assert((op1 & 0x01) == 1, "should be 32bit operation"); | |
227 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); | |
228 if (is8bit(imm32)) { | |
229 emit_byte(op1 | 0x02); // set sign bit | |
304 | 230 emit_byte(op2 | encode(dst)); |
0 | 231 emit_byte(imm32 & 0xFF); |
232 } else { | |
233 emit_byte(op1); | |
304 | 234 emit_byte(op2 | encode(dst)); |
0 | 235 emit_long(imm32); |
236 } | |
237 } | |
238 | |
239 // immediate-to-memory forms | |
304 | 240 void Assembler::emit_arith_operand(int op1, Register rm, Address adr, int32_t imm32) { |
0 | 241 assert((op1 & 0x01) == 1, "should be 32bit operation"); |
242 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); | |
243 if (is8bit(imm32)) { | |
244 emit_byte(op1 | 0x02); // set sign bit | |
304 | 245 emit_operand(rm, adr, 1); |
0 | 246 emit_byte(imm32 & 0xFF); |
247 } else { | |
248 emit_byte(op1); | |
304 | 249 emit_operand(rm, adr, 4); |
0 | 250 emit_long(imm32); |
251 } | |
252 } | |
253 | |
254 void Assembler::emit_arith(int op1, int op2, Register dst, jobject obj) { | |
304 | 255 LP64_ONLY(ShouldNotReachHere()); |
0 | 256 assert(isByte(op1) && isByte(op2), "wrong opcode"); |
257 assert((op1 & 0x01) == 1, "should be 32bit operation"); | |
258 assert((op1 & 0x02) == 0, "sign-extension bit should not be set"); | |
259 InstructionMark im(this); | |
260 emit_byte(op1); | |
304 | 261 emit_byte(op2 | encode(dst)); |
262 emit_data((intptr_t)obj, relocInfo::oop_type, 0); | |
0 | 263 } |
264 | |
265 | |
266 void Assembler::emit_arith(int op1, int op2, Register dst, Register src) { | |
267 assert(isByte(op1) && isByte(op2), "wrong opcode"); | |
268 emit_byte(op1); | |
304 | 269 emit_byte(op2 | encode(dst) << 3 | encode(src)); |
270 } | |
271 | |
272 | |
273 void Assembler::emit_operand(Register reg, Register base, Register index, | |
274 Address::ScaleFactor scale, int disp, | |
275 RelocationHolder const& rspec, | |
276 int rip_relative_correction) { | |
0 | 277 relocInfo::relocType rtype = (relocInfo::relocType) rspec.type(); |
304 | 278 |
279 // Encode the registers as needed in the fields they are used in | |
280 | |
281 int regenc = encode(reg) << 3; | |
282 int indexenc = index->is_valid() ? encode(index) << 3 : 0; | |
283 int baseenc = base->is_valid() ? encode(base) : 0; | |
284 | |
0 | 285 if (base->is_valid()) { |
286 if (index->is_valid()) { | |
287 assert(scale != Address::no_scale, "inconsistent address"); | |
288 // [base + index*scale + disp] | |
304 | 289 if (disp == 0 && rtype == relocInfo::none && |
290 base != rbp LP64_ONLY(&& base != r13)) { | |
0 | 291 // [base + index*scale] |
292 // [00 reg 100][ss index base] | |
293 assert(index != rsp, "illegal addressing mode"); | |
304 | 294 emit_byte(0x04 | regenc); |
295 emit_byte(scale << 6 | indexenc | baseenc); | |
0 | 296 } else if (is8bit(disp) && rtype == relocInfo::none) { |
297 // [base + index*scale + imm8] | |
298 // [01 reg 100][ss index base] imm8 | |
299 assert(index != rsp, "illegal addressing mode"); | |
304 | 300 emit_byte(0x44 | regenc); |
301 emit_byte(scale << 6 | indexenc | baseenc); | |
0 | 302 emit_byte(disp & 0xFF); |
303 } else { | |
304 | 304 // [base + index*scale + disp32] |
305 // [10 reg 100][ss index base] disp32 | |
0 | 306 assert(index != rsp, "illegal addressing mode"); |
304 | 307 emit_byte(0x84 | regenc); |
308 emit_byte(scale << 6 | indexenc | baseenc); | |
0 | 309 emit_data(disp, rspec, disp32_operand); |
310 } | |
304 | 311 } else if (base == rsp LP64_ONLY(|| base == r12)) { |
312 // [rsp + disp] | |
0 | 313 if (disp == 0 && rtype == relocInfo::none) { |
304 | 314 // [rsp] |
0 | 315 // [00 reg 100][00 100 100] |
304 | 316 emit_byte(0x04 | regenc); |
0 | 317 emit_byte(0x24); |
318 } else if (is8bit(disp) && rtype == relocInfo::none) { | |
304 | 319 // [rsp + imm8] |
320 // [01 reg 100][00 100 100] disp8 | |
321 emit_byte(0x44 | regenc); | |
0 | 322 emit_byte(0x24); |
323 emit_byte(disp & 0xFF); | |
324 } else { | |
304 | 325 // [rsp + imm32] |
326 // [10 reg 100][00 100 100] disp32 | |
327 emit_byte(0x84 | regenc); | |
0 | 328 emit_byte(0x24); |
329 emit_data(disp, rspec, disp32_operand); | |
330 } | |
331 } else { | |
332 // [base + disp] | |
304 | 333 assert(base != rsp LP64_ONLY(&& base != r12), "illegal addressing mode"); |
334 if (disp == 0 && rtype == relocInfo::none && | |
335 base != rbp LP64_ONLY(&& base != r13)) { | |
0 | 336 // [base] |
337 // [00 reg base] | |
304 | 338 emit_byte(0x00 | regenc | baseenc); |
0 | 339 } else if (is8bit(disp) && rtype == relocInfo::none) { |
304 | 340 // [base + disp8] |
341 // [01 reg base] disp8 | |
342 emit_byte(0x40 | regenc | baseenc); | |
0 | 343 emit_byte(disp & 0xFF); |
344 } else { | |
304 | 345 // [base + disp32] |
346 // [10 reg base] disp32 | |
347 emit_byte(0x80 | regenc | baseenc); | |
0 | 348 emit_data(disp, rspec, disp32_operand); |
349 } | |
350 } | |
351 } else { | |
352 if (index->is_valid()) { | |
353 assert(scale != Address::no_scale, "inconsistent address"); | |
354 // [index*scale + disp] | |
304 | 355 // [00 reg 100][ss index 101] disp32 |
0 | 356 assert(index != rsp, "illegal addressing mode"); |
304 | 357 emit_byte(0x04 | regenc); |
358 emit_byte(scale << 6 | indexenc | 0x05); | |
0 | 359 emit_data(disp, rspec, disp32_operand); |
304 | 360 } else if (rtype != relocInfo::none ) { |
361 // [disp] (64bit) RIP-RELATIVE (32bit) abs | |
362 // [00 000 101] disp32 | |
363 | |
364 emit_byte(0x05 | regenc); | |
365 // Note that the RIP-rel. correction applies to the generated | |
366 // disp field, but _not_ to the target address in the rspec. | |
367 | |
368 // disp was created by converting the target address minus the pc | |
369 // at the start of the instruction. That needs more correction here. | |
370 // intptr_t disp = target - next_ip; | |
371 assert(inst_mark() != NULL, "must be inside InstructionMark"); | |
372 address next_ip = pc() + sizeof(int32_t) + rip_relative_correction; | |
373 int64_t adjusted = disp; | |
374 // Do rip-rel adjustment for 64bit | |
375 LP64_ONLY(adjusted -= (next_ip - inst_mark())); | |
376 assert(is_simm32(adjusted), | |
377 "must be 32bit offset (RIP relative address)"); | |
378 emit_data((int32_t) adjusted, rspec, disp32_operand); | |
379 | |
0 | 380 } else { |
304 | 381 // 32bit never did this, did everything as the rip-rel/disp code above |
382 // [disp] ABSOLUTE | |
383 // [00 reg 100][00 100 101] disp32 | |
384 emit_byte(0x04 | regenc); | |
385 emit_byte(0x25); | |
0 | 386 emit_data(disp, rspec, disp32_operand); |
387 } | |
388 } | |
389 } | |
390 | |
304 | 391 void Assembler::emit_operand(XMMRegister reg, Register base, Register index, |
392 Address::ScaleFactor scale, int disp, | |
393 RelocationHolder const& rspec) { | |
394 emit_operand((Register)reg, base, index, scale, disp, rspec); | |
395 } | |
396 | |
0 | 397 // Secret local extension to Assembler::WhichOperand: |
398 #define end_pc_operand (_WhichOperand_limit) | |
399 | |
400 address Assembler::locate_operand(address inst, WhichOperand which) { | |
401 // Decode the given instruction, and return the address of | |
402 // an embedded 32-bit operand word. | |
403 | |
404 // If "which" is disp32_operand, selects the displacement portion | |
405 // of an effective address specifier. | |
304 | 406 // If "which" is imm64_operand, selects the trailing immediate constant. |
0 | 407 // If "which" is call32_operand, selects the displacement of a call or jump. |
408 // Caller is responsible for ensuring that there is such an operand, | |
304 | 409 // and that it is 32/64 bits wide. |
0 | 410 |
411 // If "which" is end_pc_operand, find the end of the instruction. | |
412 | |
413 address ip = inst; | |
304 | 414 bool is_64bit = false; |
415 | |
416 debug_only(bool has_disp32 = false); | |
417 int tail_size = 0; // other random bytes (#32, #16, etc.) at end of insn | |
418 | |
419 again_after_prefix: | |
0 | 420 switch (0xFF & *ip++) { |
421 | |
422 // These convenience macros generate groups of "case" labels for the switch. | |
304 | 423 #define REP4(x) (x)+0: case (x)+1: case (x)+2: case (x)+3 |
424 #define REP8(x) (x)+0: case (x)+1: case (x)+2: case (x)+3: \ | |
0 | 425 case (x)+4: case (x)+5: case (x)+6: case (x)+7 |
304 | 426 #define REP16(x) REP8((x)+0): \ |
0 | 427 case REP8((x)+8) |
428 | |
429 case CS_segment: | |
430 case SS_segment: | |
431 case DS_segment: | |
432 case ES_segment: | |
433 case FS_segment: | |
434 case GS_segment: | |
304 | 435 // Seems dubious |
436 LP64_ONLY(assert(false, "shouldn't have that prefix")); | |
0 | 437 assert(ip == inst+1, "only one prefix allowed"); |
438 goto again_after_prefix; | |
439 | |
304 | 440 case 0x67: |
441 case REX: | |
442 case REX_B: | |
443 case REX_X: | |
444 case REX_XB: | |
445 case REX_R: | |
446 case REX_RB: | |
447 case REX_RX: | |
448 case REX_RXB: | |
449 NOT_LP64(assert(false, "64bit prefixes")); | |
450 goto again_after_prefix; | |
451 | |
452 case REX_W: | |
453 case REX_WB: | |
454 case REX_WX: | |
455 case REX_WXB: | |
456 case REX_WR: | |
457 case REX_WRB: | |
458 case REX_WRX: | |
459 case REX_WRXB: | |
460 NOT_LP64(assert(false, "64bit prefixes")); | |
461 is_64bit = true; | |
462 goto again_after_prefix; | |
463 | |
464 case 0xFF: // pushq a; decl a; incl a; call a; jmp a | |
0 | 465 case 0x88: // movb a, r |
466 case 0x89: // movl a, r | |
467 case 0x8A: // movb r, a | |
468 case 0x8B: // movl r, a | |
469 case 0x8F: // popl a | |
304 | 470 debug_only(has_disp32 = true); |
0 | 471 break; |
472 | |
304 | 473 case 0x68: // pushq #32 |
474 if (which == end_pc_operand) { | |
475 return ip + 4; | |
476 } | |
477 assert(which == imm_operand && !is_64bit, "pushl has no disp32 or 64bit immediate"); | |
0 | 478 return ip; // not produced by emit_operand |
479 | |
480 case 0x66: // movw ... (size prefix) | |
304 | 481 again_after_size_prefix2: |
0 | 482 switch (0xFF & *ip++) { |
304 | 483 case REX: |
484 case REX_B: | |
485 case REX_X: | |
486 case REX_XB: | |
487 case REX_R: | |
488 case REX_RB: | |
489 case REX_RX: | |
490 case REX_RXB: | |
491 case REX_W: | |
492 case REX_WB: | |
493 case REX_WX: | |
494 case REX_WXB: | |
495 case REX_WR: | |
496 case REX_WRB: | |
497 case REX_WRX: | |
498 case REX_WRXB: | |
499 NOT_LP64(assert(false, "64bit prefix found")); | |
500 goto again_after_size_prefix2; | |
0 | 501 case 0x8B: // movw r, a |
502 case 0x89: // movw a, r | |
304 | 503 debug_only(has_disp32 = true); |
0 | 504 break; |
505 case 0xC7: // movw a, #16 | |
304 | 506 debug_only(has_disp32 = true); |
0 | 507 tail_size = 2; // the imm16 |
508 break; | |
509 case 0x0F: // several SSE/SSE2 variants | |
510 ip--; // reparse the 0x0F | |
511 goto again_after_prefix; | |
512 default: | |
513 ShouldNotReachHere(); | |
514 } | |
515 break; | |
516 | |
304 | 517 case REP8(0xB8): // movl/q r, #32/#64(oop?) |
518 if (which == end_pc_operand) return ip + (is_64bit ? 8 : 4); | |
519 // these asserts are somewhat nonsensical | |
520 #ifndef _LP64 | |
521 assert(which == imm_operand || which == disp32_operand, ""); | |
522 #else | |
523 assert((which == call32_operand || which == imm_operand) && is_64bit || | |
524 which == narrow_oop_operand && !is_64bit, ""); | |
525 #endif // _LP64 | |
0 | 526 return ip; |
527 | |
528 case 0x69: // imul r, a, #32 | |
529 case 0xC7: // movl a, #32(oop?) | |
530 tail_size = 4; | |
304 | 531 debug_only(has_disp32 = true); // has both kinds of operands! |
0 | 532 break; |
533 | |
534 case 0x0F: // movx..., etc. | |
535 switch (0xFF & *ip++) { | |
536 case 0x12: // movlps | |
537 case 0x28: // movaps | |
538 case 0x2E: // ucomiss | |
539 case 0x2F: // comiss | |
540 case 0x54: // andps | |
541 case 0x55: // andnps | |
542 case 0x56: // orps | |
543 case 0x57: // xorps | |
544 case 0x6E: // movd | |
545 case 0x7E: // movd | |
546 case 0xAE: // ldmxcsr a | |
304 | 547 // 64bit side says it these have both operands but that doesn't |
548 // appear to be true | |
549 debug_only(has_disp32 = true); | |
0 | 550 break; |
551 | |
552 case 0xAD: // shrd r, a, %cl | |
553 case 0xAF: // imul r, a | |
304 | 554 case 0xBE: // movsbl r, a (movsxb) |
555 case 0xBF: // movswl r, a (movsxw) | |
556 case 0xB6: // movzbl r, a (movzxb) | |
557 case 0xB7: // movzwl r, a (movzxw) | |
0 | 558 case REP16(0x40): // cmovl cc, r, a |
559 case 0xB0: // cmpxchgb | |
560 case 0xB1: // cmpxchg | |
561 case 0xC1: // xaddl | |
562 case 0xC7: // cmpxchg8 | |
563 case REP16(0x90): // setcc a | |
304 | 564 debug_only(has_disp32 = true); |
0 | 565 // fall out of the switch to decode the address |
566 break; | |
304 | 567 |
0 | 568 case 0xAC: // shrd r, a, #8 |
304 | 569 debug_only(has_disp32 = true); |
0 | 570 tail_size = 1; // the imm8 |
571 break; | |
304 | 572 |
0 | 573 case REP16(0x80): // jcc rdisp32 |
574 if (which == end_pc_operand) return ip + 4; | |
304 | 575 assert(which == call32_operand, "jcc has no disp32 or imm"); |
0 | 576 return ip; |
577 default: | |
578 ShouldNotReachHere(); | |
579 } | |
580 break; | |
581 | |
582 case 0x81: // addl a, #32; addl r, #32 | |
583 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl | |
304 | 584 // on 32bit in the case of cmpl, the imm might be an oop |
0 | 585 tail_size = 4; |
304 | 586 debug_only(has_disp32 = true); // has both kinds of operands! |
0 | 587 break; |
588 | |
589 case 0x83: // addl a, #8; addl r, #8 | |
590 // also: orl, adcl, sbbl, andl, subl, xorl, cmpl | |
304 | 591 debug_only(has_disp32 = true); // has both kinds of operands! |
0 | 592 tail_size = 1; |
593 break; | |
594 | |
595 case 0x9B: | |
596 switch (0xFF & *ip++) { | |
597 case 0xD9: // fnstcw a | |
304 | 598 debug_only(has_disp32 = true); |
0 | 599 break; |
600 default: | |
601 ShouldNotReachHere(); | |
602 } | |
603 break; | |
604 | |
605 case REP4(0x00): // addb a, r; addl a, r; addb r, a; addl r, a | |
606 case REP4(0x10): // adc... | |
607 case REP4(0x20): // and... | |
608 case REP4(0x30): // xor... | |
609 case REP4(0x08): // or... | |
610 case REP4(0x18): // sbb... | |
611 case REP4(0x28): // sub... | |
304 | 612 case 0xF7: // mull a |
613 case 0x8D: // lea r, a | |
614 case 0x87: // xchg r, a | |
0 | 615 case REP4(0x38): // cmp... |
304 | 616 case 0x85: // test r, a |
617 debug_only(has_disp32 = true); // has both kinds of operands! | |
0 | 618 break; |
619 | |
620 case 0xC1: // sal a, #8; sar a, #8; shl a, #8; shr a, #8 | |
621 case 0xC6: // movb a, #8 | |
622 case 0x80: // cmpb a, #8 | |
623 case 0x6B: // imul r, a, #8 | |
304 | 624 debug_only(has_disp32 = true); // has both kinds of operands! |
0 | 625 tail_size = 1; // the imm8 |
626 break; | |
627 | |
628 case 0xE8: // call rdisp32 | |
629 case 0xE9: // jmp rdisp32 | |
630 if (which == end_pc_operand) return ip + 4; | |
304 | 631 assert(which == call32_operand, "call has no disp32 or imm"); |
0 | 632 return ip; |
633 | |
634 case 0xD1: // sal a, 1; sar a, 1; shl a, 1; shr a, 1 | |
635 case 0xD3: // sal a, %cl; sar a, %cl; shl a, %cl; shr a, %cl | |
636 case 0xD9: // fld_s a; fst_s a; fstp_s a; fldcw a | |
637 case 0xDD: // fld_d a; fst_d a; fstp_d a | |
638 case 0xDB: // fild_s a; fistp_s a; fld_x a; fstp_x a | |
639 case 0xDF: // fild_d a; fistp_d a | |
640 case 0xD8: // fadd_s a; fsubr_s a; fmul_s a; fdivr_s a; fcomp_s a | |
641 case 0xDC: // fadd_d a; fsubr_d a; fmul_d a; fdivr_d a; fcomp_d a | |
642 case 0xDE: // faddp_d a; fsubrp_d a; fmulp_d a; fdivrp_d a; fcompp_d a | |
304 | 643 debug_only(has_disp32 = true); |
0 | 644 break; |
645 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
405
diff
changeset
|
646 case 0xF0: // Lock |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
405
diff
changeset
|
647 assert(os::is_MP(), "only on MP"); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
405
diff
changeset
|
648 goto again_after_prefix; |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
405
diff
changeset
|
649 |
0 | 650 case 0xF3: // For SSE |
651 case 0xF2: // For SSE2 | |
304 | 652 switch (0xFF & *ip++) { |
653 case REX: | |
654 case REX_B: | |
655 case REX_X: | |
656 case REX_XB: | |
657 case REX_R: | |
658 case REX_RB: | |
659 case REX_RX: | |
660 case REX_RXB: | |
661 case REX_W: | |
662 case REX_WB: | |
663 case REX_WX: | |
664 case REX_WXB: | |
665 case REX_WR: | |
666 case REX_WRB: | |
667 case REX_WRX: | |
668 case REX_WRXB: | |
669 NOT_LP64(assert(false, "found 64bit prefix")); | |
670 ip++; | |
671 default: | |
672 ip++; | |
673 } | |
674 debug_only(has_disp32 = true); // has both kinds of operands! | |
0 | 675 break; |
676 | |
677 default: | |
678 ShouldNotReachHere(); | |
679 | |
304 | 680 #undef REP8 |
681 #undef REP16 | |
0 | 682 } |
683 | |
684 assert(which != call32_operand, "instruction is not a call, jmp, or jcc"); | |
304 | 685 #ifdef _LP64 |
686 assert(which != imm_operand, "instruction is not a movq reg, imm64"); | |
687 #else | |
688 // assert(which != imm_operand || has_imm32, "instruction has no imm32 field"); | |
689 assert(which != imm_operand || has_disp32, "instruction has no imm32 field"); | |
690 #endif // LP64 | |
691 assert(which != disp32_operand || has_disp32, "instruction has no disp32 field"); | |
0 | 692 |
693 // parse the output of emit_operand | |
694 int op2 = 0xFF & *ip++; | |
695 int base = op2 & 0x07; | |
696 int op3 = -1; | |
697 const int b100 = 4; | |
698 const int b101 = 5; | |
699 if (base == b100 && (op2 >> 6) != 3) { | |
700 op3 = 0xFF & *ip++; | |
701 base = op3 & 0x07; // refetch the base | |
702 } | |
703 // now ip points at the disp (if any) | |
704 | |
705 switch (op2 >> 6) { | |
706 case 0: | |
707 // [00 reg 100][ss index base] | |
304 | 708 // [00 reg 100][00 100 esp] |
0 | 709 // [00 reg base] |
710 // [00 reg 100][ss index 101][disp32] | |
711 // [00 reg 101] [disp32] | |
712 | |
713 if (base == b101) { | |
714 if (which == disp32_operand) | |
715 return ip; // caller wants the disp32 | |
716 ip += 4; // skip the disp32 | |
717 } | |
718 break; | |
719 | |
720 case 1: | |
721 // [01 reg 100][ss index base][disp8] | |
304 | 722 // [01 reg 100][00 100 esp][disp8] |
0 | 723 // [01 reg base] [disp8] |
724 ip += 1; // skip the disp8 | |
725 break; | |
726 | |
727 case 2: | |
728 // [10 reg 100][ss index base][disp32] | |
304 | 729 // [10 reg 100][00 100 esp][disp32] |
0 | 730 // [10 reg base] [disp32] |
731 if (which == disp32_operand) | |
732 return ip; // caller wants the disp32 | |
733 ip += 4; // skip the disp32 | |
734 break; | |
735 | |
736 case 3: | |
737 // [11 reg base] (not a memory addressing mode) | |
738 break; | |
739 } | |
740 | |
741 if (which == end_pc_operand) { | |
742 return ip + tail_size; | |
743 } | |
744 | |
304 | 745 #ifdef _LP64 |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
746 assert(which == narrow_oop_operand && !is_64bit, "instruction is not a movl adr, imm32"); |
304 | 747 #else |
748 assert(which == imm_operand, "instruction has only an imm field"); | |
749 #endif // LP64 | |
0 | 750 return ip; |
751 } | |
752 | |
753 address Assembler::locate_next_instruction(address inst) { | |
754 // Secretly share code with locate_operand: | |
755 return locate_operand(inst, end_pc_operand); | |
756 } | |
757 | |
758 | |
759 #ifdef ASSERT | |
760 void Assembler::check_relocation(RelocationHolder const& rspec, int format) { | |
761 address inst = inst_mark(); | |
762 assert(inst != NULL && inst < pc(), "must point to beginning of instruction"); | |
763 address opnd; | |
764 | |
765 Relocation* r = rspec.reloc(); | |
766 if (r->type() == relocInfo::none) { | |
767 return; | |
768 } else if (r->is_call() || format == call32_operand) { | |
769 // assert(format == imm32_operand, "cannot specify a nonzero format"); | |
770 opnd = locate_operand(inst, call32_operand); | |
771 } else if (r->is_data()) { | |
304 | 772 assert(format == imm_operand || format == disp32_operand |
773 LP64_ONLY(|| format == narrow_oop_operand), "format ok"); | |
0 | 774 opnd = locate_operand(inst, (WhichOperand)format); |
775 } else { | |
304 | 776 assert(format == imm_operand, "cannot specify a format"); |
0 | 777 return; |
778 } | |
779 assert(opnd == pc(), "must put operand where relocs can find it"); | |
780 } | |
304 | 781 #endif // ASSERT |
782 | |
783 void Assembler::emit_operand32(Register reg, Address adr) { | |
784 assert(reg->encoding() < 8, "no extended registers"); | |
785 assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers"); | |
786 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, | |
787 adr._rspec); | |
788 } | |
789 | |
790 void Assembler::emit_operand(Register reg, Address adr, | |
791 int rip_relative_correction) { | |
792 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, | |
793 adr._rspec, | |
794 rip_relative_correction); | |
795 } | |
796 | |
797 void Assembler::emit_operand(XMMRegister reg, Address adr) { | |
798 emit_operand(reg, adr._base, adr._index, adr._scale, adr._disp, | |
799 adr._rspec); | |
800 } | |
801 | |
802 // MMX operations | |
803 void Assembler::emit_operand(MMXRegister reg, Address adr) { | |
804 assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers"); | |
805 emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec); | |
806 } | |
807 | |
808 // work around gcc (3.2.1-7a) bug | |
809 void Assembler::emit_operand(Address adr, MMXRegister reg) { | |
810 assert(!adr.base_needs_rex() && !adr.index_needs_rex(), "no extended registers"); | |
811 emit_operand((Register)reg, adr._base, adr._index, adr._scale, adr._disp, adr._rspec); | |
0 | 812 } |
813 | |
814 | |
815 void Assembler::emit_farith(int b1, int b2, int i) { | |
816 assert(isByte(b1) && isByte(b2), "wrong opcode"); | |
817 assert(0 <= i && i < 8, "illegal stack offset"); | |
818 emit_byte(b1); | |
819 emit_byte(b2 + i); | |
820 } | |
821 | |
822 | |
304 | 823 // Now the Assembler instruction (identical for 32/64 bits) |
824 | |
825 void Assembler::adcl(Register dst, int32_t imm32) { | |
826 prefix(dst); | |
0 | 827 emit_arith(0x81, 0xD0, dst, imm32); |
828 } | |
829 | |
830 void Assembler::adcl(Register dst, Address src) { | |
831 InstructionMark im(this); | |
304 | 832 prefix(src, dst); |
0 | 833 emit_byte(0x13); |
834 emit_operand(dst, src); | |
835 } | |
836 | |
837 void Assembler::adcl(Register dst, Register src) { | |
304 | 838 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
0 | 839 emit_arith(0x13, 0xC0, dst, src); |
840 } | |
841 | |
304 | 842 void Assembler::addl(Address dst, int32_t imm32) { |
843 InstructionMark im(this); | |
844 prefix(dst); | |
845 emit_arith_operand(0x81, rax, dst, imm32); | |
846 } | |
0 | 847 |
848 void Assembler::addl(Address dst, Register src) { | |
849 InstructionMark im(this); | |
304 | 850 prefix(dst, src); |
0 | 851 emit_byte(0x01); |
852 emit_operand(src, dst); | |
853 } | |
854 | |
304 | 855 void Assembler::addl(Register dst, int32_t imm32) { |
856 prefix(dst); | |
0 | 857 emit_arith(0x81, 0xC0, dst, imm32); |
858 } | |
859 | |
860 void Assembler::addl(Register dst, Address src) { | |
861 InstructionMark im(this); | |
304 | 862 prefix(src, dst); |
0 | 863 emit_byte(0x03); |
864 emit_operand(dst, src); | |
865 } | |
866 | |
867 void Assembler::addl(Register dst, Register src) { | |
304 | 868 (void) prefix_and_encode(dst->encoding(), src->encoding()); |
0 | 869 emit_arith(0x03, 0xC0, dst, src); |
870 } | |
871 | |
872 void Assembler::addr_nop_4() { | |
873 // 4 bytes: NOP DWORD PTR [EAX+0] | |
874 emit_byte(0x0F); | |
875 emit_byte(0x1F); | |
876 emit_byte(0x40); // emit_rm(cbuf, 0x1, EAX_enc, EAX_enc); | |
877 emit_byte(0); // 8-bits offset (1 byte) | |
878 } | |
879 | |
880 void Assembler::addr_nop_5() { | |
881 // 5 bytes: NOP DWORD PTR [EAX+EAX*0+0] 8-bits offset | |
882 emit_byte(0x0F); | |
883 emit_byte(0x1F); | |
884 emit_byte(0x44); // emit_rm(cbuf, 0x1, EAX_enc, 0x4); | |
885 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc); | |
886 emit_byte(0); // 8-bits offset (1 byte) | |
887 } | |
888 | |
889 void Assembler::addr_nop_7() { | |
890 // 7 bytes: NOP DWORD PTR [EAX+0] 32-bits offset | |
891 emit_byte(0x0F); | |
892 emit_byte(0x1F); | |
893 emit_byte(0x80); // emit_rm(cbuf, 0x2, EAX_enc, EAX_enc); | |
894 emit_long(0); // 32-bits offset (4 bytes) | |
895 } | |
896 | |
897 void Assembler::addr_nop_8() { | |
898 // 8 bytes: NOP DWORD PTR [EAX+EAX*0+0] 32-bits offset | |
899 emit_byte(0x0F); | |
900 emit_byte(0x1F); | |
901 emit_byte(0x84); // emit_rm(cbuf, 0x2, EAX_enc, 0x4); | |
902 emit_byte(0x00); // emit_rm(cbuf, 0x0, EAX_enc, EAX_enc); | |
903 emit_long(0); // 32-bits offset (4 bytes) | |
904 } | |
905 | |
304 | 906 void Assembler::addsd(XMMRegister dst, XMMRegister src) { |
907 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
908 emit_byte(0xF2); | |
909 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
910 emit_byte(0x0F); | |
911 emit_byte(0x58); | |
912 emit_byte(0xC0 | encode); | |
913 } | |
914 | |
915 void Assembler::addsd(XMMRegister dst, Address src) { | |
916 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
917 InstructionMark im(this); | |
918 emit_byte(0xF2); | |
919 prefix(src, dst); | |
920 emit_byte(0x0F); | |
921 emit_byte(0x58); | |
922 emit_operand(dst, src); | |
923 } | |
924 | |
925 void Assembler::addss(XMMRegister dst, XMMRegister src) { | |
926 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
927 emit_byte(0xF3); | |
928 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
929 emit_byte(0x0F); | |
930 emit_byte(0x58); | |
931 emit_byte(0xC0 | encode); | |
932 } | |
933 | |
934 void Assembler::addss(XMMRegister dst, Address src) { | |
935 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
936 InstructionMark im(this); | |
937 emit_byte(0xF3); | |
938 prefix(src, dst); | |
939 emit_byte(0x0F); | |
940 emit_byte(0x58); | |
941 emit_operand(dst, src); | |
942 } | |
943 | |
944 void Assembler::andl(Register dst, int32_t imm32) { | |
945 prefix(dst); | |
946 emit_arith(0x81, 0xE0, dst, imm32); | |
947 } | |
948 | |
949 void Assembler::andl(Register dst, Address src) { | |
950 InstructionMark im(this); | |
951 prefix(src, dst); | |
952 emit_byte(0x23); | |
953 emit_operand(dst, src); | |
954 } | |
955 | |
956 void Assembler::andl(Register dst, Register src) { | |
957 (void) prefix_and_encode(dst->encoding(), src->encoding()); | |
958 emit_arith(0x23, 0xC0, dst, src); | |
959 } | |
960 | |
961 void Assembler::andpd(XMMRegister dst, Address src) { | |
962 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
963 InstructionMark im(this); | |
964 emit_byte(0x66); | |
965 prefix(src, dst); | |
966 emit_byte(0x0F); | |
967 emit_byte(0x54); | |
968 emit_operand(dst, src); | |
969 } | |
970 | |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
971 void Assembler::bsfl(Register dst, Register src) { |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
972 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
973 emit_byte(0x0F); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
974 emit_byte(0xBC); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
975 emit_byte(0xC0 | encode); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
976 } |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
977 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
978 void Assembler::bsrl(Register dst, Register src) { |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
979 assert(!VM_Version::supports_lzcnt(), "encoding is treated as LZCNT"); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
980 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
981 emit_byte(0x0F); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
982 emit_byte(0xBD); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
983 emit_byte(0xC0 | encode); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
984 } |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
985 |
304 | 986 void Assembler::bswapl(Register reg) { // bswap |
987 int encode = prefix_and_encode(reg->encoding()); | |
988 emit_byte(0x0F); | |
989 emit_byte(0xC8 | encode); | |
990 } | |
991 | |
992 void Assembler::call(Label& L, relocInfo::relocType rtype) { | |
993 // suspect disp32 is always good | |
994 int operand = LP64_ONLY(disp32_operand) NOT_LP64(imm_operand); | |
995 | |
996 if (L.is_bound()) { | |
997 const int long_size = 5; | |
998 int offs = (int)( target(L) - pc() ); | |
999 assert(offs <= 0, "assembler error"); | |
1000 InstructionMark im(this); | |
1001 // 1110 1000 #32-bit disp | |
1002 emit_byte(0xE8); | |
1003 emit_data(offs - long_size, rtype, operand); | |
1004 } else { | |
1005 InstructionMark im(this); | |
1006 // 1110 1000 #32-bit disp | |
1007 L.add_patch_at(code(), locator()); | |
1008 | |
1009 emit_byte(0xE8); | |
1010 emit_data(int(0), rtype, operand); | |
1011 } | |
1012 } | |
1013 | |
1014 void Assembler::call(Register dst) { | |
1015 // This was originally using a 32bit register encoding | |
1016 // and surely we want 64bit! | |
1017 // this is a 32bit encoding but in 64bit mode the default | |
1018 // operand size is 64bit so there is no need for the | |
1019 // wide prefix. So prefix only happens if we use the | |
1020 // new registers. Much like push/pop. | |
1021 int x = offset(); | |
1022 // this may be true but dbx disassembles it as if it | |
1023 // were 32bits... | |
1024 // int encode = prefix_and_encode(dst->encoding()); | |
1025 // if (offset() != x) assert(dst->encoding() >= 8, "what?"); | |
1026 int encode = prefixq_and_encode(dst->encoding()); | |
1027 | |
1028 emit_byte(0xFF); | |
1029 emit_byte(0xD0 | encode); | |
1030 } | |
1031 | |
1032 | |
1033 void Assembler::call(Address adr) { | |
1034 InstructionMark im(this); | |
1035 prefix(adr); | |
1036 emit_byte(0xFF); | |
1037 emit_operand(rdx, adr); | |
1038 } | |
1039 | |
1040 void Assembler::call_literal(address entry, RelocationHolder const& rspec) { | |
1041 assert(entry != NULL, "call most probably wrong"); | |
1042 InstructionMark im(this); | |
1043 emit_byte(0xE8); | |
1044 intptr_t disp = entry - (_code_pos + sizeof(int32_t)); | |
1045 assert(is_simm32(disp), "must be 32bit offset (call2)"); | |
1046 // Technically, should use call32_operand, but this format is | |
1047 // implied by the fact that we're emitting a call instruction. | |
1048 | |
1049 int operand = LP64_ONLY(disp32_operand) NOT_LP64(call32_operand); | |
1050 emit_data((int) disp, rspec, operand); | |
1051 } | |
1052 | |
1053 void Assembler::cdql() { | |
1054 emit_byte(0x99); | |
1055 } | |
1056 | |
1057 void Assembler::cmovl(Condition cc, Register dst, Register src) { | |
1058 NOT_LP64(guarantee(VM_Version::supports_cmov(), "illegal instruction")); | |
1059 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1060 emit_byte(0x0F); | |
1061 emit_byte(0x40 | cc); | |
1062 emit_byte(0xC0 | encode); | |
1063 } | |
1064 | |
1065 | |
1066 void Assembler::cmovl(Condition cc, Register dst, Address src) { | |
1067 NOT_LP64(guarantee(VM_Version::supports_cmov(), "illegal instruction")); | |
1068 prefix(src, dst); | |
1069 emit_byte(0x0F); | |
1070 emit_byte(0x40 | cc); | |
1071 emit_operand(dst, src); | |
1072 } | |
1073 | |
1074 void Assembler::cmpb(Address dst, int imm8) { | |
1075 InstructionMark im(this); | |
1076 prefix(dst); | |
1077 emit_byte(0x80); | |
1078 emit_operand(rdi, dst, 1); | |
1079 emit_byte(imm8); | |
1080 } | |
1081 | |
1082 void Assembler::cmpl(Address dst, int32_t imm32) { | |
1083 InstructionMark im(this); | |
1084 prefix(dst); | |
1085 emit_byte(0x81); | |
1086 emit_operand(rdi, dst, 4); | |
1087 emit_long(imm32); | |
1088 } | |
1089 | |
1090 void Assembler::cmpl(Register dst, int32_t imm32) { | |
1091 prefix(dst); | |
1092 emit_arith(0x81, 0xF8, dst, imm32); | |
1093 } | |
1094 | |
1095 void Assembler::cmpl(Register dst, Register src) { | |
1096 (void) prefix_and_encode(dst->encoding(), src->encoding()); | |
1097 emit_arith(0x3B, 0xC0, dst, src); | |
1098 } | |
1099 | |
1100 | |
1101 void Assembler::cmpl(Register dst, Address src) { | |
1102 InstructionMark im(this); | |
1103 prefix(src, dst); | |
1104 emit_byte(0x3B); | |
1105 emit_operand(dst, src); | |
1106 } | |
1107 | |
1108 void Assembler::cmpw(Address dst, int imm16) { | |
1109 InstructionMark im(this); | |
1110 assert(!dst.base_needs_rex() && !dst.index_needs_rex(), "no extended registers"); | |
1111 emit_byte(0x66); | |
1112 emit_byte(0x81); | |
1113 emit_operand(rdi, dst, 2); | |
1114 emit_word(imm16); | |
1115 } | |
1116 | |
1117 // The 32-bit cmpxchg compares the value at adr with the contents of rax, | |
1118 // and stores reg into adr if so; otherwise, the value at adr is loaded into rax,. | |
1119 // The ZF is set if the compared values were equal, and cleared otherwise. | |
1120 void Assembler::cmpxchgl(Register reg, Address adr) { // cmpxchg | |
1121 if (Atomics & 2) { | |
1122 // caveat: no instructionmark, so this isn't relocatable. | |
1123 // Emit a synthetic, non-atomic, CAS equivalent. | |
1124 // Beware. The synthetic form sets all ICCs, not just ZF. | |
1125 // cmpxchg r,[m] is equivalent to rax, = CAS (m, rax, r) | |
1126 cmpl(rax, adr); | |
1127 movl(rax, adr); | |
1128 if (reg != rax) { | |
1129 Label L ; | |
1130 jcc(Assembler::notEqual, L); | |
1131 movl(adr, reg); | |
1132 bind(L); | |
1133 } | |
1134 } else { | |
1135 InstructionMark im(this); | |
1136 prefix(adr, reg); | |
1137 emit_byte(0x0F); | |
1138 emit_byte(0xB1); | |
1139 emit_operand(reg, adr); | |
1140 } | |
1141 } | |
1142 | |
1143 void Assembler::comisd(XMMRegister dst, Address src) { | |
1144 // NOTE: dbx seems to decode this as comiss even though the | |
1145 // 0x66 is there. Strangly ucomisd comes out correct | |
1146 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1147 emit_byte(0x66); | |
1148 comiss(dst, src); | |
1149 } | |
1150 | |
1151 void Assembler::comiss(XMMRegister dst, Address src) { | |
1152 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1153 | |
1154 InstructionMark im(this); | |
1155 prefix(src, dst); | |
1156 emit_byte(0x0F); | |
1157 emit_byte(0x2F); | |
1158 emit_operand(dst, src); | |
1159 } | |
1160 | |
1161 void Assembler::cvtdq2pd(XMMRegister dst, XMMRegister src) { | |
1162 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1163 emit_byte(0xF3); | |
1164 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1165 emit_byte(0x0F); | |
1166 emit_byte(0xE6); | |
1167 emit_byte(0xC0 | encode); | |
1168 } | |
1169 | |
1170 void Assembler::cvtdq2ps(XMMRegister dst, XMMRegister src) { | |
1171 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1172 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1173 emit_byte(0x0F); | |
1174 emit_byte(0x5B); | |
1175 emit_byte(0xC0 | encode); | |
1176 } | |
1177 | |
1178 void Assembler::cvtsd2ss(XMMRegister dst, XMMRegister src) { | |
1179 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1180 emit_byte(0xF2); | |
1181 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1182 emit_byte(0x0F); | |
1183 emit_byte(0x5A); | |
1184 emit_byte(0xC0 | encode); | |
1185 } | |
1186 | |
1187 void Assembler::cvtsi2sdl(XMMRegister dst, Register src) { | |
1188 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1189 emit_byte(0xF2); | |
1190 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1191 emit_byte(0x0F); | |
1192 emit_byte(0x2A); | |
1193 emit_byte(0xC0 | encode); | |
1194 } | |
1195 | |
1196 void Assembler::cvtsi2ssl(XMMRegister dst, Register src) { | |
1197 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1198 emit_byte(0xF3); | |
1199 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1200 emit_byte(0x0F); | |
1201 emit_byte(0x2A); | |
1202 emit_byte(0xC0 | encode); | |
1203 } | |
1204 | |
1205 void Assembler::cvtss2sd(XMMRegister dst, XMMRegister src) { | |
1206 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1207 emit_byte(0xF3); | |
1208 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1209 emit_byte(0x0F); | |
1210 emit_byte(0x5A); | |
1211 emit_byte(0xC0 | encode); | |
1212 } | |
1213 | |
1214 void Assembler::cvttsd2sil(Register dst, XMMRegister src) { | |
1215 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1216 emit_byte(0xF2); | |
1217 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1218 emit_byte(0x0F); | |
1219 emit_byte(0x2C); | |
1220 emit_byte(0xC0 | encode); | |
1221 } | |
1222 | |
1223 void Assembler::cvttss2sil(Register dst, XMMRegister src) { | |
1224 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1225 emit_byte(0xF3); | |
1226 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1227 emit_byte(0x0F); | |
1228 emit_byte(0x2C); | |
1229 emit_byte(0xC0 | encode); | |
1230 } | |
1231 | |
1232 void Assembler::decl(Address dst) { | |
1233 // Don't use it directly. Use MacroAssembler::decrement() instead. | |
1234 InstructionMark im(this); | |
1235 prefix(dst); | |
1236 emit_byte(0xFF); | |
1237 emit_operand(rcx, dst); | |
1238 } | |
1239 | |
1240 void Assembler::divsd(XMMRegister dst, Address src) { | |
1241 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1242 InstructionMark im(this); | |
1243 emit_byte(0xF2); | |
1244 prefix(src, dst); | |
1245 emit_byte(0x0F); | |
1246 emit_byte(0x5E); | |
1247 emit_operand(dst, src); | |
1248 } | |
1249 | |
1250 void Assembler::divsd(XMMRegister dst, XMMRegister src) { | |
1251 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1252 emit_byte(0xF2); | |
1253 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1254 emit_byte(0x0F); | |
1255 emit_byte(0x5E); | |
1256 emit_byte(0xC0 | encode); | |
1257 } | |
1258 | |
1259 void Assembler::divss(XMMRegister dst, Address src) { | |
1260 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1261 InstructionMark im(this); | |
1262 emit_byte(0xF3); | |
1263 prefix(src, dst); | |
1264 emit_byte(0x0F); | |
1265 emit_byte(0x5E); | |
1266 emit_operand(dst, src); | |
1267 } | |
1268 | |
1269 void Assembler::divss(XMMRegister dst, XMMRegister src) { | |
1270 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1271 emit_byte(0xF3); | |
1272 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1273 emit_byte(0x0F); | |
1274 emit_byte(0x5E); | |
1275 emit_byte(0xC0 | encode); | |
1276 } | |
1277 | |
1278 void Assembler::emms() { | |
1279 NOT_LP64(assert(VM_Version::supports_mmx(), "")); | |
1280 emit_byte(0x0F); | |
1281 emit_byte(0x77); | |
1282 } | |
1283 | |
1284 void Assembler::hlt() { | |
1285 emit_byte(0xF4); | |
1286 } | |
1287 | |
1288 void Assembler::idivl(Register src) { | |
1289 int encode = prefix_and_encode(src->encoding()); | |
1290 emit_byte(0xF7); | |
1291 emit_byte(0xF8 | encode); | |
1292 } | |
1293 | |
1920 | 1294 void Assembler::divl(Register src) { // Unsigned |
1295 int encode = prefix_and_encode(src->encoding()); | |
1296 emit_byte(0xF7); | |
1297 emit_byte(0xF0 | encode); | |
1298 } | |
1299 | |
304 | 1300 void Assembler::imull(Register dst, Register src) { |
1301 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1302 emit_byte(0x0F); | |
1303 emit_byte(0xAF); | |
1304 emit_byte(0xC0 | encode); | |
1305 } | |
1306 | |
1307 | |
1308 void Assembler::imull(Register dst, Register src, int value) { | |
1309 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1310 if (is8bit(value)) { | |
1311 emit_byte(0x6B); | |
1312 emit_byte(0xC0 | encode); | |
1914
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1846
diff
changeset
|
1313 emit_byte(value & 0xFF); |
304 | 1314 } else { |
1315 emit_byte(0x69); | |
1316 emit_byte(0xC0 | encode); | |
1317 emit_long(value); | |
1318 } | |
1319 } | |
1320 | |
1321 void Assembler::incl(Address dst) { | |
1322 // Don't use it directly. Use MacroAssembler::increment() instead. | |
1323 InstructionMark im(this); | |
1324 prefix(dst); | |
1325 emit_byte(0xFF); | |
1326 emit_operand(rax, dst); | |
1327 } | |
1328 | |
1329 void Assembler::jcc(Condition cc, Label& L, relocInfo::relocType rtype) { | |
1330 InstructionMark im(this); | |
1331 relocate(rtype); | |
1332 assert((0 <= cc) && (cc < 16), "illegal cc"); | |
1333 if (L.is_bound()) { | |
1334 address dst = target(L); | |
1335 assert(dst != NULL, "jcc most probably wrong"); | |
1336 | |
1337 const int short_size = 2; | |
1338 const int long_size = 6; | |
1339 intptr_t offs = (intptr_t)dst - (intptr_t)_code_pos; | |
1340 if (rtype == relocInfo::none && is8bit(offs - short_size)) { | |
1341 // 0111 tttn #8-bit disp | |
1342 emit_byte(0x70 | cc); | |
1343 emit_byte((offs - short_size) & 0xFF); | |
1344 } else { | |
1345 // 0000 1111 1000 tttn #32-bit disp | |
1346 assert(is_simm32(offs - long_size), | |
1347 "must be 32bit offset (call4)"); | |
1348 emit_byte(0x0F); | |
1349 emit_byte(0x80 | cc); | |
1350 emit_long(offs - long_size); | |
1351 } | |
1352 } else { | |
1353 // Note: could eliminate cond. jumps to this jump if condition | |
1354 // is the same however, seems to be rather unlikely case. | |
1355 // Note: use jccb() if label to be bound is very close to get | |
1356 // an 8-bit displacement | |
1357 L.add_patch_at(code(), locator()); | |
1358 emit_byte(0x0F); | |
1359 emit_byte(0x80 | cc); | |
1360 emit_long(0); | |
1361 } | |
1362 } | |
1363 | |
1364 void Assembler::jccb(Condition cc, Label& L) { | |
1365 if (L.is_bound()) { | |
1366 const int short_size = 2; | |
1367 address entry = target(L); | |
1368 assert(is8bit((intptr_t)entry - ((intptr_t)_code_pos + short_size)), | |
1369 "Dispacement too large for a short jmp"); | |
1370 intptr_t offs = (intptr_t)entry - (intptr_t)_code_pos; | |
1371 // 0111 tttn #8-bit disp | |
1372 emit_byte(0x70 | cc); | |
1373 emit_byte((offs - short_size) & 0xFF); | |
1374 } else { | |
1375 InstructionMark im(this); | |
1376 L.add_patch_at(code(), locator()); | |
1377 emit_byte(0x70 | cc); | |
1378 emit_byte(0); | |
1379 } | |
1380 } | |
1381 | |
1382 void Assembler::jmp(Address adr) { | |
1383 InstructionMark im(this); | |
1384 prefix(adr); | |
1385 emit_byte(0xFF); | |
1386 emit_operand(rsp, adr); | |
1387 } | |
1388 | |
1389 void Assembler::jmp(Label& L, relocInfo::relocType rtype) { | |
1390 if (L.is_bound()) { | |
1391 address entry = target(L); | |
1392 assert(entry != NULL, "jmp most probably wrong"); | |
1393 InstructionMark im(this); | |
1394 const int short_size = 2; | |
1395 const int long_size = 5; | |
1396 intptr_t offs = entry - _code_pos; | |
1397 if (rtype == relocInfo::none && is8bit(offs - short_size)) { | |
1398 emit_byte(0xEB); | |
1399 emit_byte((offs - short_size) & 0xFF); | |
1400 } else { | |
1401 emit_byte(0xE9); | |
1402 emit_long(offs - long_size); | |
1403 } | |
1404 } else { | |
1405 // By default, forward jumps are always 32-bit displacements, since | |
1406 // we can't yet know where the label will be bound. If you're sure that | |
1407 // the forward jump will not run beyond 256 bytes, use jmpb to | |
1408 // force an 8-bit displacement. | |
1409 InstructionMark im(this); | |
1410 relocate(rtype); | |
1411 L.add_patch_at(code(), locator()); | |
1412 emit_byte(0xE9); | |
1413 emit_long(0); | |
1414 } | |
1415 } | |
1416 | |
1417 void Assembler::jmp(Register entry) { | |
1418 int encode = prefix_and_encode(entry->encoding()); | |
1419 emit_byte(0xFF); | |
1420 emit_byte(0xE0 | encode); | |
1421 } | |
1422 | |
1423 void Assembler::jmp_literal(address dest, RelocationHolder const& rspec) { | |
1424 InstructionMark im(this); | |
1425 emit_byte(0xE9); | |
1426 assert(dest != NULL, "must have a target"); | |
1427 intptr_t disp = dest - (_code_pos + sizeof(int32_t)); | |
1428 assert(is_simm32(disp), "must be 32bit offset (jmp)"); | |
1429 emit_data(disp, rspec.reloc(), call32_operand); | |
1430 } | |
1431 | |
1432 void Assembler::jmpb(Label& L) { | |
1433 if (L.is_bound()) { | |
1434 const int short_size = 2; | |
1435 address entry = target(L); | |
1436 assert(is8bit((entry - _code_pos) + short_size), | |
1437 "Dispacement too large for a short jmp"); | |
1438 assert(entry != NULL, "jmp most probably wrong"); | |
1439 intptr_t offs = entry - _code_pos; | |
1440 emit_byte(0xEB); | |
1441 emit_byte((offs - short_size) & 0xFF); | |
1442 } else { | |
1443 InstructionMark im(this); | |
1444 L.add_patch_at(code(), locator()); | |
1445 emit_byte(0xEB); | |
1446 emit_byte(0); | |
1447 } | |
1448 } | |
1449 | |
1450 void Assembler::ldmxcsr( Address src) { | |
1451 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1452 InstructionMark im(this); | |
1453 prefix(src); | |
1454 emit_byte(0x0F); | |
1455 emit_byte(0xAE); | |
1456 emit_operand(as_Register(2), src); | |
1457 } | |
1458 | |
1459 void Assembler::leal(Register dst, Address src) { | |
1460 InstructionMark im(this); | |
1461 #ifdef _LP64 | |
1462 emit_byte(0x67); // addr32 | |
1463 prefix(src, dst); | |
1464 #endif // LP64 | |
1465 emit_byte(0x8D); | |
1466 emit_operand(dst, src); | |
1467 } | |
1468 | |
1469 void Assembler::lock() { | |
1470 if (Atomics & 1) { | |
1471 // Emit either nothing, a NOP, or a NOP: prefix | |
1472 emit_byte(0x90) ; | |
1473 } else { | |
1474 emit_byte(0xF0); | |
1475 } | |
1476 } | |
1477 | |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1478 void Assembler::lzcntl(Register dst, Register src) { |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1479 assert(VM_Version::supports_lzcnt(), "encoding is treated as BSR"); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1480 emit_byte(0xF3); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1481 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1482 emit_byte(0x0F); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1483 emit_byte(0xBD); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1484 emit_byte(0xC0 | encode); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1485 } |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
1486 |
671
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1487 // Emit mfence instruction |
304 | 1488 void Assembler::mfence() { |
671
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1489 NOT_LP64(assert(VM_Version::supports_sse2(), "unsupported");) |
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1490 emit_byte( 0x0F ); |
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1491 emit_byte( 0xAE ); |
d0994e5bebce
6822204: volatile fences should prefer lock:addl to actual mfence instructions
never
parents:
665
diff
changeset
|
1492 emit_byte( 0xF0 ); |
304 | 1493 } |
1494 | |
1495 void Assembler::mov(Register dst, Register src) { | |
1496 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src)); | |
1497 } | |
1498 | |
1499 void Assembler::movapd(XMMRegister dst, XMMRegister src) { | |
1500 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1501 int dstenc = dst->encoding(); | |
1502 int srcenc = src->encoding(); | |
1503 emit_byte(0x66); | |
1504 if (dstenc < 8) { | |
1505 if (srcenc >= 8) { | |
1506 prefix(REX_B); | |
1507 srcenc -= 8; | |
1508 } | |
1509 } else { | |
1510 if (srcenc < 8) { | |
1511 prefix(REX_R); | |
1512 } else { | |
1513 prefix(REX_RB); | |
1514 srcenc -= 8; | |
1515 } | |
1516 dstenc -= 8; | |
1517 } | |
1518 emit_byte(0x0F); | |
1519 emit_byte(0x28); | |
1520 emit_byte(0xC0 | dstenc << 3 | srcenc); | |
1521 } | |
1522 | |
1523 void Assembler::movaps(XMMRegister dst, XMMRegister src) { | |
1524 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1525 int dstenc = dst->encoding(); | |
1526 int srcenc = src->encoding(); | |
1527 if (dstenc < 8) { | |
1528 if (srcenc >= 8) { | |
1529 prefix(REX_B); | |
1530 srcenc -= 8; | |
1531 } | |
1532 } else { | |
1533 if (srcenc < 8) { | |
1534 prefix(REX_R); | |
1535 } else { | |
1536 prefix(REX_RB); | |
1537 srcenc -= 8; | |
1538 } | |
1539 dstenc -= 8; | |
1540 } | |
1541 emit_byte(0x0F); | |
1542 emit_byte(0x28); | |
1543 emit_byte(0xC0 | dstenc << 3 | srcenc); | |
1544 } | |
1545 | |
1546 void Assembler::movb(Register dst, Address src) { | |
1547 NOT_LP64(assert(dst->has_byte_register(), "must have byte register")); | |
1548 InstructionMark im(this); | |
1549 prefix(src, dst, true); | |
1550 emit_byte(0x8A); | |
1551 emit_operand(dst, src); | |
1552 } | |
1553 | |
1554 | |
1555 void Assembler::movb(Address dst, int imm8) { | |
1556 InstructionMark im(this); | |
1557 prefix(dst); | |
1558 emit_byte(0xC6); | |
1559 emit_operand(rax, dst, 1); | |
1560 emit_byte(imm8); | |
1561 } | |
1562 | |
1563 | |
1564 void Assembler::movb(Address dst, Register src) { | |
1565 assert(src->has_byte_register(), "must have byte register"); | |
1566 InstructionMark im(this); | |
1567 prefix(dst, src, true); | |
1568 emit_byte(0x88); | |
1569 emit_operand(src, dst); | |
1570 } | |
1571 | |
1572 void Assembler::movdl(XMMRegister dst, Register src) { | |
1573 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1574 emit_byte(0x66); | |
1575 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1576 emit_byte(0x0F); | |
1577 emit_byte(0x6E); | |
1578 emit_byte(0xC0 | encode); | |
1579 } | |
1580 | |
1581 void Assembler::movdl(Register dst, XMMRegister src) { | |
1582 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1583 emit_byte(0x66); | |
1584 // swap src/dst to get correct prefix | |
1585 int encode = prefix_and_encode(src->encoding(), dst->encoding()); | |
1586 emit_byte(0x0F); | |
1587 emit_byte(0x7E); | |
1588 emit_byte(0xC0 | encode); | |
1589 } | |
1590 | |
1591 void Assembler::movdqa(XMMRegister dst, Address src) { | |
1592 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1593 InstructionMark im(this); | |
1594 emit_byte(0x66); | |
1595 prefix(src, dst); | |
1596 emit_byte(0x0F); | |
1597 emit_byte(0x6F); | |
1598 emit_operand(dst, src); | |
1599 } | |
1600 | |
1601 void Assembler::movdqa(XMMRegister dst, XMMRegister src) { | |
1602 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1603 emit_byte(0x66); | |
1604 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
1605 emit_byte(0x0F); | |
1606 emit_byte(0x6F); | |
1607 emit_byte(0xC0 | encode); | |
1608 } | |
1609 | |
1610 void Assembler::movdqa(Address dst, XMMRegister src) { | |
1611 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1612 InstructionMark im(this); | |
1613 emit_byte(0x66); | |
1614 prefix(dst, src); | |
1615 emit_byte(0x0F); | |
1616 emit_byte(0x7F); | |
1617 emit_operand(src, dst); | |
1618 } | |
1619 | |
405 | 1620 void Assembler::movdqu(XMMRegister dst, Address src) { |
1621 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1622 InstructionMark im(this); | |
1623 emit_byte(0xF3); | |
1624 prefix(src, dst); | |
1625 emit_byte(0x0F); | |
1626 emit_byte(0x6F); | |
1627 emit_operand(dst, src); | |
1628 } | |
1629 | |
1630 void Assembler::movdqu(XMMRegister dst, XMMRegister src) { | |
1631 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1632 emit_byte(0xF3); | |
1633 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
1634 emit_byte(0x0F); | |
1635 emit_byte(0x6F); | |
1636 emit_byte(0xC0 | encode); | |
1637 } | |
1638 | |
1639 void Assembler::movdqu(Address dst, XMMRegister src) { | |
1640 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1641 InstructionMark im(this); | |
1642 emit_byte(0xF3); | |
1643 prefix(dst, src); | |
1644 emit_byte(0x0F); | |
1645 emit_byte(0x7F); | |
1646 emit_operand(src, dst); | |
1647 } | |
1648 | |
304 | 1649 // Uses zero extension on 64bit |
1650 | |
1651 void Assembler::movl(Register dst, int32_t imm32) { | |
1652 int encode = prefix_and_encode(dst->encoding()); | |
1653 emit_byte(0xB8 | encode); | |
1654 emit_long(imm32); | |
1655 } | |
1656 | |
1657 void Assembler::movl(Register dst, Register src) { | |
1658 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1659 emit_byte(0x8B); | |
1660 emit_byte(0xC0 | encode); | |
1661 } | |
1662 | |
1663 void Assembler::movl(Register dst, Address src) { | |
1664 InstructionMark im(this); | |
1665 prefix(src, dst); | |
1666 emit_byte(0x8B); | |
1667 emit_operand(dst, src); | |
1668 } | |
1669 | |
1670 void Assembler::movl(Address dst, int32_t imm32) { | |
1671 InstructionMark im(this); | |
1672 prefix(dst); | |
1673 emit_byte(0xC7); | |
1674 emit_operand(rax, dst, 4); | |
1675 emit_long(imm32); | |
1676 } | |
1677 | |
1678 void Assembler::movl(Address dst, Register src) { | |
1679 InstructionMark im(this); | |
1680 prefix(dst, src); | |
1681 emit_byte(0x89); | |
1682 emit_operand(src, dst); | |
1683 } | |
1684 | |
1685 // New cpus require to use movsd and movss to avoid partial register stall | |
1686 // when loading from memory. But for old Opteron use movlpd instead of movsd. | |
1687 // The selection is done in MacroAssembler::movdbl() and movflt(). | |
1688 void Assembler::movlpd(XMMRegister dst, Address src) { | |
1689 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1690 InstructionMark im(this); | |
1691 emit_byte(0x66); | |
1692 prefix(src, dst); | |
1693 emit_byte(0x0F); | |
1694 emit_byte(0x12); | |
1695 emit_operand(dst, src); | |
1696 } | |
1697 | |
1698 void Assembler::movq( MMXRegister dst, Address src ) { | |
1699 assert( VM_Version::supports_mmx(), "" ); | |
1700 emit_byte(0x0F); | |
1701 emit_byte(0x6F); | |
1702 emit_operand(dst, src); | |
1703 } | |
1704 | |
1705 void Assembler::movq( Address dst, MMXRegister src ) { | |
1706 assert( VM_Version::supports_mmx(), "" ); | |
1707 emit_byte(0x0F); | |
1708 emit_byte(0x7F); | |
1709 // workaround gcc (3.2.1-7a) bug | |
1710 // In that version of gcc with only an emit_operand(MMX, Address) | |
1711 // gcc will tail jump and try and reverse the parameters completely | |
1712 // obliterating dst in the process. By having a version available | |
1713 // that doesn't need to swap the args at the tail jump the bug is | |
1714 // avoided. | |
1715 emit_operand(dst, src); | |
1716 } | |
1717 | |
1718 void Assembler::movq(XMMRegister dst, Address src) { | |
1719 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1720 InstructionMark im(this); | |
1721 emit_byte(0xF3); | |
1722 prefix(src, dst); | |
1723 emit_byte(0x0F); | |
1724 emit_byte(0x7E); | |
1725 emit_operand(dst, src); | |
1726 } | |
1727 | |
1728 void Assembler::movq(Address dst, XMMRegister src) { | |
1729 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1730 InstructionMark im(this); | |
1731 emit_byte(0x66); | |
1732 prefix(dst, src); | |
1733 emit_byte(0x0F); | |
1734 emit_byte(0xD6); | |
1735 emit_operand(src, dst); | |
1736 } | |
1737 | |
1738 void Assembler::movsbl(Register dst, Address src) { // movsxb | |
1739 InstructionMark im(this); | |
1740 prefix(src, dst); | |
1741 emit_byte(0x0F); | |
1742 emit_byte(0xBE); | |
1743 emit_operand(dst, src); | |
1744 } | |
1745 | |
1746 void Assembler::movsbl(Register dst, Register src) { // movsxb | |
1747 NOT_LP64(assert(src->has_byte_register(), "must have byte register")); | |
1748 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true); | |
1749 emit_byte(0x0F); | |
1750 emit_byte(0xBE); | |
1751 emit_byte(0xC0 | encode); | |
1752 } | |
1753 | |
1754 void Assembler::movsd(XMMRegister dst, XMMRegister src) { | |
1755 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1756 emit_byte(0xF2); | |
1757 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1758 emit_byte(0x0F); | |
1759 emit_byte(0x10); | |
1760 emit_byte(0xC0 | encode); | |
1761 } | |
1762 | |
1763 void Assembler::movsd(XMMRegister dst, Address src) { | |
1764 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1765 InstructionMark im(this); | |
1766 emit_byte(0xF2); | |
1767 prefix(src, dst); | |
1768 emit_byte(0x0F); | |
1769 emit_byte(0x10); | |
1770 emit_operand(dst, src); | |
1771 } | |
1772 | |
1773 void Assembler::movsd(Address dst, XMMRegister src) { | |
1774 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1775 InstructionMark im(this); | |
1776 emit_byte(0xF2); | |
1777 prefix(dst, src); | |
1778 emit_byte(0x0F); | |
1779 emit_byte(0x11); | |
1780 emit_operand(src, dst); | |
1781 } | |
1782 | |
1783 void Assembler::movss(XMMRegister dst, XMMRegister src) { | |
1784 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1785 emit_byte(0xF3); | |
1786 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1787 emit_byte(0x0F); | |
1788 emit_byte(0x10); | |
1789 emit_byte(0xC0 | encode); | |
1790 } | |
1791 | |
1792 void Assembler::movss(XMMRegister dst, Address src) { | |
1793 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1794 InstructionMark im(this); | |
1795 emit_byte(0xF3); | |
1796 prefix(src, dst); | |
1797 emit_byte(0x0F); | |
1798 emit_byte(0x10); | |
1799 emit_operand(dst, src); | |
1800 } | |
1801 | |
1802 void Assembler::movss(Address dst, XMMRegister src) { | |
1803 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1804 InstructionMark im(this); | |
1805 emit_byte(0xF3); | |
1806 prefix(dst, src); | |
1807 emit_byte(0x0F); | |
1808 emit_byte(0x11); | |
1809 emit_operand(src, dst); | |
1810 } | |
1811 | |
1812 void Assembler::movswl(Register dst, Address src) { // movsxw | |
1813 InstructionMark im(this); | |
1814 prefix(src, dst); | |
1815 emit_byte(0x0F); | |
1816 emit_byte(0xBF); | |
1817 emit_operand(dst, src); | |
1818 } | |
1819 | |
1820 void Assembler::movswl(Register dst, Register src) { // movsxw | |
1821 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1822 emit_byte(0x0F); | |
1823 emit_byte(0xBF); | |
1824 emit_byte(0xC0 | encode); | |
1825 } | |
1826 | |
1827 void Assembler::movw(Address dst, int imm16) { | |
1828 InstructionMark im(this); | |
1829 | |
1830 emit_byte(0x66); // switch to 16-bit mode | |
1831 prefix(dst); | |
1832 emit_byte(0xC7); | |
1833 emit_operand(rax, dst, 2); | |
1834 emit_word(imm16); | |
1835 } | |
1836 | |
1837 void Assembler::movw(Register dst, Address src) { | |
1838 InstructionMark im(this); | |
1839 emit_byte(0x66); | |
1840 prefix(src, dst); | |
1841 emit_byte(0x8B); | |
1842 emit_operand(dst, src); | |
1843 } | |
1844 | |
1845 void Assembler::movw(Address dst, Register src) { | |
1846 InstructionMark im(this); | |
1847 emit_byte(0x66); | |
1848 prefix(dst, src); | |
1849 emit_byte(0x89); | |
1850 emit_operand(src, dst); | |
1851 } | |
1852 | |
1853 void Assembler::movzbl(Register dst, Address src) { // movzxb | |
1854 InstructionMark im(this); | |
1855 prefix(src, dst); | |
1856 emit_byte(0x0F); | |
1857 emit_byte(0xB6); | |
1858 emit_operand(dst, src); | |
1859 } | |
1860 | |
1861 void Assembler::movzbl(Register dst, Register src) { // movzxb | |
1862 NOT_LP64(assert(src->has_byte_register(), "must have byte register")); | |
1863 int encode = prefix_and_encode(dst->encoding(), src->encoding(), true); | |
1864 emit_byte(0x0F); | |
1865 emit_byte(0xB6); | |
1866 emit_byte(0xC0 | encode); | |
1867 } | |
1868 | |
1869 void Assembler::movzwl(Register dst, Address src) { // movzxw | |
1870 InstructionMark im(this); | |
1871 prefix(src, dst); | |
1872 emit_byte(0x0F); | |
1873 emit_byte(0xB7); | |
1874 emit_operand(dst, src); | |
1875 } | |
1876 | |
1877 void Assembler::movzwl(Register dst, Register src) { // movzxw | |
1878 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1879 emit_byte(0x0F); | |
1880 emit_byte(0xB7); | |
1881 emit_byte(0xC0 | encode); | |
1882 } | |
1883 | |
1884 void Assembler::mull(Address src) { | |
1885 InstructionMark im(this); | |
1886 prefix(src); | |
1887 emit_byte(0xF7); | |
1888 emit_operand(rsp, src); | |
1889 } | |
1890 | |
1891 void Assembler::mull(Register src) { | |
1892 int encode = prefix_and_encode(src->encoding()); | |
1893 emit_byte(0xF7); | |
1894 emit_byte(0xE0 | encode); | |
1895 } | |
1896 | |
1897 void Assembler::mulsd(XMMRegister dst, Address src) { | |
1898 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1899 InstructionMark im(this); | |
1900 emit_byte(0xF2); | |
1901 prefix(src, dst); | |
1902 emit_byte(0x0F); | |
1903 emit_byte(0x59); | |
1904 emit_operand(dst, src); | |
1905 } | |
1906 | |
1907 void Assembler::mulsd(XMMRegister dst, XMMRegister src) { | |
1908 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
1909 emit_byte(0xF2); | |
1910 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1911 emit_byte(0x0F); | |
1912 emit_byte(0x59); | |
1913 emit_byte(0xC0 | encode); | |
1914 } | |
1915 | |
1916 void Assembler::mulss(XMMRegister dst, Address src) { | |
1917 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1918 InstructionMark im(this); | |
1919 emit_byte(0xF3); | |
1920 prefix(src, dst); | |
1921 emit_byte(0x0F); | |
1922 emit_byte(0x59); | |
1923 emit_operand(dst, src); | |
1924 } | |
1925 | |
1926 void Assembler::mulss(XMMRegister dst, XMMRegister src) { | |
1927 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
1928 emit_byte(0xF3); | |
1929 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
1930 emit_byte(0x0F); | |
1931 emit_byte(0x59); | |
1932 emit_byte(0xC0 | encode); | |
1933 } | |
1934 | |
1935 void Assembler::negl(Register dst) { | |
1936 int encode = prefix_and_encode(dst->encoding()); | |
1937 emit_byte(0xF7); | |
1938 emit_byte(0xD8 | encode); | |
1939 } | |
1940 | |
0 | 1941 void Assembler::nop(int i) { |
304 | 1942 #ifdef ASSERT |
0 | 1943 assert(i > 0, " "); |
304 | 1944 // The fancy nops aren't currently recognized by debuggers making it a |
1945 // pain to disassemble code while debugging. If asserts are on clearly | |
1946 // speed is not an issue so simply use the single byte traditional nop | |
1947 // to do alignment. | |
1948 | |
1949 for (; i > 0 ; i--) emit_byte(0x90); | |
1950 return; | |
1951 | |
1952 #endif // ASSERT | |
1953 | |
0 | 1954 if (UseAddressNop && VM_Version::is_intel()) { |
1955 // | |
1956 // Using multi-bytes nops "0x0F 0x1F [address]" for Intel | |
1957 // 1: 0x90 | |
1958 // 2: 0x66 0x90 | |
1959 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding) | |
1960 // 4: 0x0F 0x1F 0x40 0x00 | |
1961 // 5: 0x0F 0x1F 0x44 0x00 0x00 | |
1962 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00 | |
1963 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 | |
1964 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 | |
1965 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 | |
1966 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 | |
1967 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 | |
1968 | |
1969 // The rest coding is Intel specific - don't use consecutive address nops | |
1970 | |
1971 // 12: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 | |
1972 // 13: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 | |
1973 // 14: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 | |
1974 // 15: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x66 0x66 0x66 0x90 | |
1975 | |
1976 while(i >= 15) { | |
1977 // For Intel don't generate consecutive addess nops (mix with regular nops) | |
1978 i -= 15; | |
1979 emit_byte(0x66); // size prefix | |
1980 emit_byte(0x66); // size prefix | |
1981 emit_byte(0x66); // size prefix | |
1982 addr_nop_8(); | |
1983 emit_byte(0x66); // size prefix | |
1984 emit_byte(0x66); // size prefix | |
1985 emit_byte(0x66); // size prefix | |
1986 emit_byte(0x90); // nop | |
1987 } | |
1988 switch (i) { | |
1989 case 14: | |
1990 emit_byte(0x66); // size prefix | |
1991 case 13: | |
1992 emit_byte(0x66); // size prefix | |
1993 case 12: | |
1994 addr_nop_8(); | |
1995 emit_byte(0x66); // size prefix | |
1996 emit_byte(0x66); // size prefix | |
1997 emit_byte(0x66); // size prefix | |
1998 emit_byte(0x90); // nop | |
1999 break; | |
2000 case 11: | |
2001 emit_byte(0x66); // size prefix | |
2002 case 10: | |
2003 emit_byte(0x66); // size prefix | |
2004 case 9: | |
2005 emit_byte(0x66); // size prefix | |
2006 case 8: | |
2007 addr_nop_8(); | |
2008 break; | |
2009 case 7: | |
2010 addr_nop_7(); | |
2011 break; | |
2012 case 6: | |
2013 emit_byte(0x66); // size prefix | |
2014 case 5: | |
2015 addr_nop_5(); | |
2016 break; | |
2017 case 4: | |
2018 addr_nop_4(); | |
2019 break; | |
2020 case 3: | |
2021 // Don't use "0x0F 0x1F 0x00" - need patching safe padding | |
2022 emit_byte(0x66); // size prefix | |
2023 case 2: | |
2024 emit_byte(0x66); // size prefix | |
2025 case 1: | |
2026 emit_byte(0x90); // nop | |
2027 break; | |
2028 default: | |
2029 assert(i == 0, " "); | |
2030 } | |
2031 return; | |
2032 } | |
2033 if (UseAddressNop && VM_Version::is_amd()) { | |
2034 // | |
2035 // Using multi-bytes nops "0x0F 0x1F [address]" for AMD. | |
2036 // 1: 0x90 | |
2037 // 2: 0x66 0x90 | |
2038 // 3: 0x66 0x66 0x90 (don't use "0x0F 0x1F 0x00" - need patching safe padding) | |
2039 // 4: 0x0F 0x1F 0x40 0x00 | |
2040 // 5: 0x0F 0x1F 0x44 0x00 0x00 | |
2041 // 6: 0x66 0x0F 0x1F 0x44 0x00 0x00 | |
2042 // 7: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 | |
2043 // 8: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 | |
2044 // 9: 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 | |
2045 // 10: 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 | |
2046 // 11: 0x66 0x66 0x66 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 | |
2047 | |
2048 // The rest coding is AMD specific - use consecutive address nops | |
2049 | |
2050 // 12: 0x66 0x0F 0x1F 0x44 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00 | |
2051 // 13: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x66 0x0F 0x1F 0x44 0x00 0x00 | |
2052 // 14: 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 | |
2053 // 15: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x80 0x00 0x00 0x00 0x00 | |
2054 // 16: 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 0x0F 0x1F 0x84 0x00 0x00 0x00 0x00 0x00 | |
2055 // Size prefixes (0x66) are added for larger sizes | |
2056 | |
2057 while(i >= 22) { | |
2058 i -= 11; | |
2059 emit_byte(0x66); // size prefix | |
2060 emit_byte(0x66); // size prefix | |
2061 emit_byte(0x66); // size prefix | |
2062 addr_nop_8(); | |
2063 } | |
2064 // Generate first nop for size between 21-12 | |
2065 switch (i) { | |
2066 case 21: | |
2067 i -= 1; | |
2068 emit_byte(0x66); // size prefix | |
2069 case 20: | |
2070 case 19: | |
2071 i -= 1; | |
2072 emit_byte(0x66); // size prefix | |
2073 case 18: | |
2074 case 17: | |
2075 i -= 1; | |
2076 emit_byte(0x66); // size prefix | |
2077 case 16: | |
2078 case 15: | |
2079 i -= 8; | |
2080 addr_nop_8(); | |
2081 break; | |
2082 case 14: | |
2083 case 13: | |
2084 i -= 7; | |
2085 addr_nop_7(); | |
2086 break; | |
2087 case 12: | |
2088 i -= 6; | |
2089 emit_byte(0x66); // size prefix | |
2090 addr_nop_5(); | |
2091 break; | |
2092 default: | |
2093 assert(i < 12, " "); | |
2094 } | |
2095 | |
2096 // Generate second nop for size between 11-1 | |
2097 switch (i) { | |
2098 case 11: | |
2099 emit_byte(0x66); // size prefix | |
2100 case 10: | |
2101 emit_byte(0x66); // size prefix | |
2102 case 9: | |
2103 emit_byte(0x66); // size prefix | |
2104 case 8: | |
2105 addr_nop_8(); | |
2106 break; | |
2107 case 7: | |
2108 addr_nop_7(); | |
2109 break; | |
2110 case 6: | |
2111 emit_byte(0x66); // size prefix | |
2112 case 5: | |
2113 addr_nop_5(); | |
2114 break; | |
2115 case 4: | |
2116 addr_nop_4(); | |
2117 break; | |
2118 case 3: | |
2119 // Don't use "0x0F 0x1F 0x00" - need patching safe padding | |
2120 emit_byte(0x66); // size prefix | |
2121 case 2: | |
2122 emit_byte(0x66); // size prefix | |
2123 case 1: | |
2124 emit_byte(0x90); // nop | |
2125 break; | |
2126 default: | |
2127 assert(i == 0, " "); | |
2128 } | |
2129 return; | |
2130 } | |
2131 | |
2132 // Using nops with size prefixes "0x66 0x90". | |
2133 // From AMD Optimization Guide: | |
2134 // 1: 0x90 | |
2135 // 2: 0x66 0x90 | |
2136 // 3: 0x66 0x66 0x90 | |
2137 // 4: 0x66 0x66 0x66 0x90 | |
2138 // 5: 0x66 0x66 0x90 0x66 0x90 | |
2139 // 6: 0x66 0x66 0x90 0x66 0x66 0x90 | |
2140 // 7: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 | |
2141 // 8: 0x66 0x66 0x66 0x90 0x66 0x66 0x66 0x90 | |
2142 // 9: 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90 | |
2143 // 10: 0x66 0x66 0x66 0x90 0x66 0x66 0x90 0x66 0x66 0x90 | |
2144 // | |
2145 while(i > 12) { | |
2146 i -= 4; | |
2147 emit_byte(0x66); // size prefix | |
2148 emit_byte(0x66); | |
2149 emit_byte(0x66); | |
2150 emit_byte(0x90); // nop | |
2151 } | |
2152 // 1 - 12 nops | |
2153 if(i > 8) { | |
2154 if(i > 9) { | |
2155 i -= 1; | |
2156 emit_byte(0x66); | |
2157 } | |
2158 i -= 3; | |
2159 emit_byte(0x66); | |
2160 emit_byte(0x66); | |
2161 emit_byte(0x90); | |
2162 } | |
2163 // 1 - 8 nops | |
2164 if(i > 4) { | |
2165 if(i > 6) { | |
2166 i -= 1; | |
2167 emit_byte(0x66); | |
2168 } | |
2169 i -= 3; | |
2170 emit_byte(0x66); | |
2171 emit_byte(0x66); | |
2172 emit_byte(0x90); | |
2173 } | |
2174 switch (i) { | |
2175 case 4: | |
2176 emit_byte(0x66); | |
2177 case 3: | |
2178 emit_byte(0x66); | |
2179 case 2: | |
2180 emit_byte(0x66); | |
2181 case 1: | |
2182 emit_byte(0x90); | |
2183 break; | |
2184 default: | |
2185 assert(i == 0, " "); | |
2186 } | |
2187 } | |
2188 | |
304 | 2189 void Assembler::notl(Register dst) { |
2190 int encode = prefix_and_encode(dst->encoding()); | |
2191 emit_byte(0xF7); | |
2192 emit_byte(0xD0 | encode ); | |
2193 } | |
2194 | |
2195 void Assembler::orl(Address dst, int32_t imm32) { | |
2196 InstructionMark im(this); | |
2197 prefix(dst); | |
2198 emit_byte(0x81); | |
2199 emit_operand(rcx, dst, 4); | |
2200 emit_long(imm32); | |
2201 } | |
2202 | |
2203 void Assembler::orl(Register dst, int32_t imm32) { | |
2204 prefix(dst); | |
2205 emit_arith(0x81, 0xC8, dst, imm32); | |
2206 } | |
2207 | |
2208 | |
2209 void Assembler::orl(Register dst, Address src) { | |
2210 InstructionMark im(this); | |
2211 prefix(src, dst); | |
2212 emit_byte(0x0B); | |
2213 emit_operand(dst, src); | |
2214 } | |
2215 | |
2216 | |
2217 void Assembler::orl(Register dst, Register src) { | |
2218 (void) prefix_and_encode(dst->encoding(), src->encoding()); | |
2219 emit_arith(0x0B, 0xC0, dst, src); | |
2220 } | |
2221 | |
681 | 2222 void Assembler::pcmpestri(XMMRegister dst, Address src, int imm8) { |
2223 assert(VM_Version::supports_sse4_2(), ""); | |
2224 | |
2225 InstructionMark im(this); | |
2226 emit_byte(0x66); | |
2227 prefix(src, dst); | |
2228 emit_byte(0x0F); | |
2229 emit_byte(0x3A); | |
2230 emit_byte(0x61); | |
2231 emit_operand(dst, src); | |
2232 emit_byte(imm8); | |
2233 } | |
2234 | |
2235 void Assembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) { | |
2236 assert(VM_Version::supports_sse4_2(), ""); | |
2237 | |
2238 emit_byte(0x66); | |
2239 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
2240 emit_byte(0x0F); | |
2241 emit_byte(0x3A); | |
2242 emit_byte(0x61); | |
2243 emit_byte(0xC0 | encode); | |
2244 emit_byte(imm8); | |
2245 } | |
2246 | |
304 | 2247 // generic |
2248 void Assembler::pop(Register dst) { | |
2249 int encode = prefix_and_encode(dst->encoding()); | |
2250 emit_byte(0x58 | encode); | |
2251 } | |
2252 | |
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2253 void Assembler::popcntl(Register dst, Address src) { |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2254 assert(VM_Version::supports_popcnt(), "must support"); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2255 InstructionMark im(this); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2256 emit_byte(0xF3); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2257 prefix(src, dst); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2258 emit_byte(0x0F); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2259 emit_byte(0xB8); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2260 emit_operand(dst, src); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2261 } |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2262 |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2263 void Assembler::popcntl(Register dst, Register src) { |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2264 assert(VM_Version::supports_popcnt(), "must support"); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2265 emit_byte(0xF3); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2266 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2267 emit_byte(0x0F); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2268 emit_byte(0xB8); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2269 emit_byte(0xC0 | encode); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2270 } |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
2271 |
304 | 2272 void Assembler::popf() { |
2273 emit_byte(0x9D); | |
2274 } | |
2275 | |
1060 | 2276 #ifndef _LP64 // no 32bit push/pop on amd64 |
304 | 2277 void Assembler::popl(Address dst) { |
2278 // NOTE: this will adjust stack by 8byte on 64bits | |
2279 InstructionMark im(this); | |
2280 prefix(dst); | |
2281 emit_byte(0x8F); | |
2282 emit_operand(rax, dst); | |
2283 } | |
1060 | 2284 #endif |
304 | 2285 |
2286 void Assembler::prefetch_prefix(Address src) { | |
2287 prefix(src); | |
2288 emit_byte(0x0F); | |
2289 } | |
2290 | |
2291 void Assembler::prefetchnta(Address src) { | |
2292 NOT_LP64(assert(VM_Version::supports_sse2(), "must support")); | |
2293 InstructionMark im(this); | |
2294 prefetch_prefix(src); | |
2295 emit_byte(0x18); | |
2296 emit_operand(rax, src); // 0, src | |
2297 } | |
2298 | |
2299 void Assembler::prefetchr(Address src) { | |
2300 NOT_LP64(assert(VM_Version::supports_3dnow(), "must support")); | |
2301 InstructionMark im(this); | |
2302 prefetch_prefix(src); | |
2303 emit_byte(0x0D); | |
2304 emit_operand(rax, src); // 0, src | |
2305 } | |
2306 | |
2307 void Assembler::prefetcht0(Address src) { | |
2308 NOT_LP64(assert(VM_Version::supports_sse(), "must support")); | |
2309 InstructionMark im(this); | |
2310 prefetch_prefix(src); | |
2311 emit_byte(0x18); | |
2312 emit_operand(rcx, src); // 1, src | |
2313 } | |
2314 | |
2315 void Assembler::prefetcht1(Address src) { | |
2316 NOT_LP64(assert(VM_Version::supports_sse(), "must support")); | |
2317 InstructionMark im(this); | |
2318 prefetch_prefix(src); | |
2319 emit_byte(0x18); | |
2320 emit_operand(rdx, src); // 2, src | |
2321 } | |
2322 | |
2323 void Assembler::prefetcht2(Address src) { | |
2324 NOT_LP64(assert(VM_Version::supports_sse(), "must support")); | |
2325 InstructionMark im(this); | |
2326 prefetch_prefix(src); | |
2327 emit_byte(0x18); | |
2328 emit_operand(rbx, src); // 3, src | |
2329 } | |
2330 | |
2331 void Assembler::prefetchw(Address src) { | |
2332 NOT_LP64(assert(VM_Version::supports_3dnow(), "must support")); | |
2333 InstructionMark im(this); | |
2334 prefetch_prefix(src); | |
2335 emit_byte(0x0D); | |
2336 emit_operand(rcx, src); // 1, src | |
2337 } | |
2338 | |
2339 void Assembler::prefix(Prefix p) { | |
2340 a_byte(p); | |
2341 } | |
2342 | |
2343 void Assembler::pshufd(XMMRegister dst, XMMRegister src, int mode) { | |
2344 assert(isByte(mode), "invalid value"); | |
2345 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2346 | |
2347 emit_byte(0x66); | |
2348 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
2349 emit_byte(0x0F); | |
2350 emit_byte(0x70); | |
2351 emit_byte(0xC0 | encode); | |
2352 emit_byte(mode & 0xFF); | |
2353 | |
2354 } | |
2355 | |
2356 void Assembler::pshufd(XMMRegister dst, Address src, int mode) { | |
2357 assert(isByte(mode), "invalid value"); | |
2358 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2359 | |
2360 InstructionMark im(this); | |
2361 emit_byte(0x66); | |
2362 prefix(src, dst); | |
2363 emit_byte(0x0F); | |
2364 emit_byte(0x70); | |
2365 emit_operand(dst, src); | |
2366 emit_byte(mode & 0xFF); | |
2367 } | |
2368 | |
2369 void Assembler::pshuflw(XMMRegister dst, XMMRegister src, int mode) { | |
2370 assert(isByte(mode), "invalid value"); | |
2371 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2372 | |
2373 emit_byte(0xF2); | |
2374 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
2375 emit_byte(0x0F); | |
2376 emit_byte(0x70); | |
2377 emit_byte(0xC0 | encode); | |
2378 emit_byte(mode & 0xFF); | |
2379 } | |
2380 | |
2381 void Assembler::pshuflw(XMMRegister dst, Address src, int mode) { | |
2382 assert(isByte(mode), "invalid value"); | |
2383 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2384 | |
2385 InstructionMark im(this); | |
2386 emit_byte(0xF2); | |
2387 prefix(src, dst); // QQ new | |
2388 emit_byte(0x0F); | |
2389 emit_byte(0x70); | |
2390 emit_operand(dst, src); | |
2391 emit_byte(mode & 0xFF); | |
2392 } | |
2393 | |
2394 void Assembler::psrlq(XMMRegister dst, int shift) { | |
2395 // HMM Table D-1 says sse2 or mmx | |
2396 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
2397 | |
2398 int encode = prefixq_and_encode(xmm2->encoding(), dst->encoding()); | |
2399 emit_byte(0x66); | |
2400 emit_byte(0x0F); | |
2401 emit_byte(0x73); | |
2402 emit_byte(0xC0 | encode); | |
2403 emit_byte(shift); | |
2404 } | |
2405 | |
681 | 2406 void Assembler::ptest(XMMRegister dst, Address src) { |
2407 assert(VM_Version::supports_sse4_1(), ""); | |
2408 | |
2409 InstructionMark im(this); | |
2410 emit_byte(0x66); | |
2411 prefix(src, dst); | |
2412 emit_byte(0x0F); | |
2413 emit_byte(0x38); | |
2414 emit_byte(0x17); | |
2415 emit_operand(dst, src); | |
2416 } | |
2417 | |
2418 void Assembler::ptest(XMMRegister dst, XMMRegister src) { | |
2419 assert(VM_Version::supports_sse4_1(), ""); | |
2420 | |
2421 emit_byte(0x66); | |
2422 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
2423 emit_byte(0x0F); | |
2424 emit_byte(0x38); | |
2425 emit_byte(0x17); | |
2426 emit_byte(0xC0 | encode); | |
2427 } | |
2428 | |
304 | 2429 void Assembler::punpcklbw(XMMRegister dst, XMMRegister src) { |
2430 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2431 emit_byte(0x66); | |
2432 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
2433 emit_byte(0x0F); | |
2434 emit_byte(0x60); | |
2435 emit_byte(0xC0 | encode); | |
2436 } | |
2437 | |
2438 void Assembler::push(int32_t imm32) { | |
2439 // in 64bits we push 64bits onto the stack but only | |
2440 // take a 32bit immediate | |
2441 emit_byte(0x68); | |
2442 emit_long(imm32); | |
2443 } | |
2444 | |
2445 void Assembler::push(Register src) { | |
2446 int encode = prefix_and_encode(src->encoding()); | |
2447 | |
2448 emit_byte(0x50 | encode); | |
2449 } | |
2450 | |
2451 void Assembler::pushf() { | |
2452 emit_byte(0x9C); | |
2453 } | |
2454 | |
1060 | 2455 #ifndef _LP64 // no 32bit push/pop on amd64 |
304 | 2456 void Assembler::pushl(Address src) { |
2457 // Note this will push 64bit on 64bit | |
2458 InstructionMark im(this); | |
2459 prefix(src); | |
2460 emit_byte(0xFF); | |
2461 emit_operand(rsi, src); | |
2462 } | |
1060 | 2463 #endif |
304 | 2464 |
2465 void Assembler::pxor(XMMRegister dst, Address src) { | |
2466 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2467 InstructionMark im(this); | |
2468 emit_byte(0x66); | |
2469 prefix(src, dst); | |
2470 emit_byte(0x0F); | |
2471 emit_byte(0xEF); | |
2472 emit_operand(dst, src); | |
2473 } | |
2474 | |
2475 void Assembler::pxor(XMMRegister dst, XMMRegister src) { | |
2476 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2477 InstructionMark im(this); | |
2478 emit_byte(0x66); | |
2479 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
2480 emit_byte(0x0F); | |
2481 emit_byte(0xEF); | |
2482 emit_byte(0xC0 | encode); | |
2483 } | |
2484 | |
2485 void Assembler::rcll(Register dst, int imm8) { | |
2486 assert(isShiftCount(imm8), "illegal shift count"); | |
2487 int encode = prefix_and_encode(dst->encoding()); | |
2488 if (imm8 == 1) { | |
2489 emit_byte(0xD1); | |
2490 emit_byte(0xD0 | encode); | |
2491 } else { | |
2492 emit_byte(0xC1); | |
2493 emit_byte(0xD0 | encode); | |
2494 emit_byte(imm8); | |
2495 } | |
2496 } | |
2497 | |
2498 // copies data from [esi] to [edi] using rcx pointer sized words | |
2499 // generic | |
2500 void Assembler::rep_mov() { | |
2501 emit_byte(0xF3); | |
2502 // MOVSQ | |
2503 LP64_ONLY(prefix(REX_W)); | |
2504 emit_byte(0xA5); | |
2505 } | |
2506 | |
2507 // sets rcx pointer sized words with rax, value at [edi] | |
2508 // generic | |
2509 void Assembler::rep_set() { // rep_set | |
2510 emit_byte(0xF3); | |
2511 // STOSQ | |
2512 LP64_ONLY(prefix(REX_W)); | |
2513 emit_byte(0xAB); | |
2514 } | |
2515 | |
2516 // scans rcx pointer sized words at [edi] for occurance of rax, | |
2517 // generic | |
2518 void Assembler::repne_scan() { // repne_scan | |
2519 emit_byte(0xF2); | |
2520 // SCASQ | |
2521 LP64_ONLY(prefix(REX_W)); | |
2522 emit_byte(0xAF); | |
2523 } | |
2524 | |
2525 #ifdef _LP64 | |
2526 // scans rcx 4 byte words at [edi] for occurance of rax, | |
2527 // generic | |
2528 void Assembler::repne_scanl() { // repne_scan | |
2529 emit_byte(0xF2); | |
2530 // SCASL | |
2531 emit_byte(0xAF); | |
2532 } | |
2533 #endif | |
2534 | |
0 | 2535 void Assembler::ret(int imm16) { |
2536 if (imm16 == 0) { | |
2537 emit_byte(0xC3); | |
2538 } else { | |
2539 emit_byte(0xC2); | |
2540 emit_word(imm16); | |
2541 } | |
2542 } | |
2543 | |
304 | 2544 void Assembler::sahf() { |
2545 #ifdef _LP64 | |
2546 // Not supported in 64bit mode | |
2547 ShouldNotReachHere(); | |
2548 #endif | |
2549 emit_byte(0x9E); | |
2550 } | |
2551 | |
2552 void Assembler::sarl(Register dst, int imm8) { | |
2553 int encode = prefix_and_encode(dst->encoding()); | |
2554 assert(isShiftCount(imm8), "illegal shift count"); | |
2555 if (imm8 == 1) { | |
2556 emit_byte(0xD1); | |
2557 emit_byte(0xF8 | encode); | |
2558 } else { | |
2559 emit_byte(0xC1); | |
2560 emit_byte(0xF8 | encode); | |
2561 emit_byte(imm8); | |
2562 } | |
2563 } | |
2564 | |
2565 void Assembler::sarl(Register dst) { | |
2566 int encode = prefix_and_encode(dst->encoding()); | |
2567 emit_byte(0xD3); | |
2568 emit_byte(0xF8 | encode); | |
2569 } | |
2570 | |
2571 void Assembler::sbbl(Address dst, int32_t imm32) { | |
2572 InstructionMark im(this); | |
2573 prefix(dst); | |
2574 emit_arith_operand(0x81, rbx, dst, imm32); | |
2575 } | |
2576 | |
2577 void Assembler::sbbl(Register dst, int32_t imm32) { | |
2578 prefix(dst); | |
2579 emit_arith(0x81, 0xD8, dst, imm32); | |
2580 } | |
2581 | |
2582 | |
2583 void Assembler::sbbl(Register dst, Address src) { | |
2584 InstructionMark im(this); | |
2585 prefix(src, dst); | |
2586 emit_byte(0x1B); | |
2587 emit_operand(dst, src); | |
2588 } | |
2589 | |
2590 void Assembler::sbbl(Register dst, Register src) { | |
2591 (void) prefix_and_encode(dst->encoding(), src->encoding()); | |
2592 emit_arith(0x1B, 0xC0, dst, src); | |
2593 } | |
2594 | |
2595 void Assembler::setb(Condition cc, Register dst) { | |
2596 assert(0 <= cc && cc < 16, "illegal cc"); | |
2597 int encode = prefix_and_encode(dst->encoding(), true); | |
0 | 2598 emit_byte(0x0F); |
304 | 2599 emit_byte(0x90 | cc); |
2600 emit_byte(0xC0 | encode); | |
2601 } | |
2602 | |
2603 void Assembler::shll(Register dst, int imm8) { | |
2604 assert(isShiftCount(imm8), "illegal shift count"); | |
2605 int encode = prefix_and_encode(dst->encoding()); | |
2606 if (imm8 == 1 ) { | |
2607 emit_byte(0xD1); | |
2608 emit_byte(0xE0 | encode); | |
2609 } else { | |
2610 emit_byte(0xC1); | |
2611 emit_byte(0xE0 | encode); | |
2612 emit_byte(imm8); | |
2613 } | |
2614 } | |
2615 | |
2616 void Assembler::shll(Register dst) { | |
2617 int encode = prefix_and_encode(dst->encoding()); | |
2618 emit_byte(0xD3); | |
2619 emit_byte(0xE0 | encode); | |
2620 } | |
2621 | |
2622 void Assembler::shrl(Register dst, int imm8) { | |
2623 assert(isShiftCount(imm8), "illegal shift count"); | |
2624 int encode = prefix_and_encode(dst->encoding()); | |
2625 emit_byte(0xC1); | |
2626 emit_byte(0xE8 | encode); | |
2627 emit_byte(imm8); | |
2628 } | |
2629 | |
2630 void Assembler::shrl(Register dst) { | |
2631 int encode = prefix_and_encode(dst->encoding()); | |
2632 emit_byte(0xD3); | |
2633 emit_byte(0xE8 | encode); | |
2634 } | |
0 | 2635 |
2636 // copies a single word from [esi] to [edi] | |
2637 void Assembler::smovl() { | |
2638 emit_byte(0xA5); | |
2639 } | |
2640 | |
304 | 2641 void Assembler::sqrtsd(XMMRegister dst, XMMRegister src) { |
2642 // HMM Table D-1 says sse2 | |
2643 // NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
2644 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2645 emit_byte(0xF2); | |
2646 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
2647 emit_byte(0x0F); | |
2648 emit_byte(0x51); | |
2649 emit_byte(0xC0 | encode); | |
2650 } | |
2651 | |
2652 void Assembler::stmxcsr( Address dst) { | |
2653 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
2654 InstructionMark im(this); | |
2655 prefix(dst); | |
2656 emit_byte(0x0F); | |
2657 emit_byte(0xAE); | |
2658 emit_operand(as_Register(3), dst); | |
2659 } | |
2660 | |
2661 void Assembler::subl(Address dst, int32_t imm32) { | |
2662 InstructionMark im(this); | |
2663 prefix(dst); | |
2664 if (is8bit(imm32)) { | |
2665 emit_byte(0x83); | |
2666 emit_operand(rbp, dst, 1); | |
2667 emit_byte(imm32 & 0xFF); | |
2668 } else { | |
2669 emit_byte(0x81); | |
2670 emit_operand(rbp, dst, 4); | |
2671 emit_long(imm32); | |
2672 } | |
2673 } | |
2674 | |
2675 void Assembler::subl(Register dst, int32_t imm32) { | |
2676 prefix(dst); | |
2677 emit_arith(0x81, 0xE8, dst, imm32); | |
2678 } | |
2679 | |
2680 void Assembler::subl(Address dst, Register src) { | |
2681 InstructionMark im(this); | |
2682 prefix(dst, src); | |
2683 emit_byte(0x29); | |
2684 emit_operand(src, dst); | |
2685 } | |
2686 | |
2687 void Assembler::subl(Register dst, Address src) { | |
2688 InstructionMark im(this); | |
2689 prefix(src, dst); | |
2690 emit_byte(0x2B); | |
2691 emit_operand(dst, src); | |
2692 } | |
2693 | |
2694 void Assembler::subl(Register dst, Register src) { | |
2695 (void) prefix_and_encode(dst->encoding(), src->encoding()); | |
2696 emit_arith(0x2B, 0xC0, dst, src); | |
2697 } | |
2698 | |
2699 void Assembler::subsd(XMMRegister dst, XMMRegister src) { | |
2700 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2701 emit_byte(0xF2); | |
2702 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
2703 emit_byte(0x0F); | |
2704 emit_byte(0x5C); | |
2705 emit_byte(0xC0 | encode); | |
2706 } | |
2707 | |
2708 void Assembler::subsd(XMMRegister dst, Address src) { | |
2709 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2710 InstructionMark im(this); | |
2711 emit_byte(0xF2); | |
2712 prefix(src, dst); | |
2713 emit_byte(0x0F); | |
2714 emit_byte(0x5C); | |
2715 emit_operand(dst, src); | |
2716 } | |
2717 | |
2718 void Assembler::subss(XMMRegister dst, XMMRegister src) { | |
2719 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
0 | 2720 emit_byte(0xF3); |
304 | 2721 int encode = prefix_and_encode(dst->encoding(), src->encoding()); |
2722 emit_byte(0x0F); | |
2723 emit_byte(0x5C); | |
2724 emit_byte(0xC0 | encode); | |
2725 } | |
2726 | |
2727 void Assembler::subss(XMMRegister dst, Address src) { | |
2728 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
2729 InstructionMark im(this); | |
2730 emit_byte(0xF3); | |
2731 prefix(src, dst); | |
2732 emit_byte(0x0F); | |
2733 emit_byte(0x5C); | |
2734 emit_operand(dst, src); | |
2735 } | |
2736 | |
2737 void Assembler::testb(Register dst, int imm8) { | |
2738 NOT_LP64(assert(dst->has_byte_register(), "must have byte register")); | |
2739 (void) prefix_and_encode(dst->encoding(), true); | |
2740 emit_arith_b(0xF6, 0xC0, dst, imm8); | |
2741 } | |
2742 | |
2743 void Assembler::testl(Register dst, int32_t imm32) { | |
2744 // not using emit_arith because test | |
2745 // doesn't support sign-extension of | |
2746 // 8bit operands | |
2747 int encode = dst->encoding(); | |
2748 if (encode == 0) { | |
2749 emit_byte(0xA9); | |
2750 } else { | |
2751 encode = prefix_and_encode(encode); | |
2752 emit_byte(0xF7); | |
2753 emit_byte(0xC0 | encode); | |
2754 } | |
2755 emit_long(imm32); | |
2756 } | |
2757 | |
2758 void Assembler::testl(Register dst, Register src) { | |
2759 (void) prefix_and_encode(dst->encoding(), src->encoding()); | |
2760 emit_arith(0x85, 0xC0, dst, src); | |
2761 } | |
2762 | |
2763 void Assembler::testl(Register dst, Address src) { | |
2764 InstructionMark im(this); | |
2765 prefix(src, dst); | |
2766 emit_byte(0x85); | |
2767 emit_operand(dst, src); | |
2768 } | |
2769 | |
2770 void Assembler::ucomisd(XMMRegister dst, Address src) { | |
2771 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2772 emit_byte(0x66); | |
2773 ucomiss(dst, src); | |
2774 } | |
2775 | |
2776 void Assembler::ucomisd(XMMRegister dst, XMMRegister src) { | |
2777 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2778 emit_byte(0x66); | |
2779 ucomiss(dst, src); | |
2780 } | |
2781 | |
2782 void Assembler::ucomiss(XMMRegister dst, Address src) { | |
2783 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
2784 | |
2785 InstructionMark im(this); | |
2786 prefix(src, dst); | |
2787 emit_byte(0x0F); | |
2788 emit_byte(0x2E); | |
2789 emit_operand(dst, src); | |
2790 } | |
2791 | |
2792 void Assembler::ucomiss(XMMRegister dst, XMMRegister src) { | |
2793 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
2794 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
2795 emit_byte(0x0F); | |
2796 emit_byte(0x2E); | |
2797 emit_byte(0xC0 | encode); | |
2798 } | |
2799 | |
2800 | |
2801 void Assembler::xaddl(Address dst, Register src) { | |
2802 InstructionMark im(this); | |
2803 prefix(dst, src); | |
0 | 2804 emit_byte(0x0F); |
304 | 2805 emit_byte(0xC1); |
2806 emit_operand(src, dst); | |
2807 } | |
2808 | |
2809 void Assembler::xchgl(Register dst, Address src) { // xchg | |
2810 InstructionMark im(this); | |
2811 prefix(src, dst); | |
2812 emit_byte(0x87); | |
2813 emit_operand(dst, src); | |
2814 } | |
2815 | |
2816 void Assembler::xchgl(Register dst, Register src) { | |
2817 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
2818 emit_byte(0x87); | |
2819 emit_byte(0xc0 | encode); | |
2820 } | |
2821 | |
2822 void Assembler::xorl(Register dst, int32_t imm32) { | |
2823 prefix(dst); | |
2824 emit_arith(0x81, 0xF0, dst, imm32); | |
2825 } | |
2826 | |
2827 void Assembler::xorl(Register dst, Address src) { | |
2828 InstructionMark im(this); | |
2829 prefix(src, dst); | |
2830 emit_byte(0x33); | |
2831 emit_operand(dst, src); | |
2832 } | |
2833 | |
2834 void Assembler::xorl(Register dst, Register src) { | |
2835 (void) prefix_and_encode(dst->encoding(), src->encoding()); | |
2836 emit_arith(0x33, 0xC0, dst, src); | |
2837 } | |
2838 | |
2839 void Assembler::xorpd(XMMRegister dst, XMMRegister src) { | |
2840 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2841 emit_byte(0x66); | |
2842 xorps(dst, src); | |
2843 } | |
2844 | |
2845 void Assembler::xorpd(XMMRegister dst, Address src) { | |
2846 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
2847 InstructionMark im(this); | |
2848 emit_byte(0x66); | |
2849 prefix(src, dst); | |
2850 emit_byte(0x0F); | |
2851 emit_byte(0x57); | |
2852 emit_operand(dst, src); | |
2853 } | |
2854 | |
2855 | |
2856 void Assembler::xorps(XMMRegister dst, XMMRegister src) { | |
2857 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
2858 int encode = prefix_and_encode(dst->encoding(), src->encoding()); | |
2859 emit_byte(0x0F); | |
2860 emit_byte(0x57); | |
2861 emit_byte(0xC0 | encode); | |
2862 } | |
2863 | |
2864 void Assembler::xorps(XMMRegister dst, Address src) { | |
2865 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
2866 InstructionMark im(this); | |
2867 prefix(src, dst); | |
2868 emit_byte(0x0F); | |
2869 emit_byte(0x57); | |
2870 emit_operand(dst, src); | |
2871 } | |
2872 | |
2873 #ifndef _LP64 | |
2874 // 32bit only pieces of the assembler | |
2875 | |
2876 void Assembler::cmp_literal32(Register src1, int32_t imm32, RelocationHolder const& rspec) { | |
2877 // NO PREFIX AS NEVER 64BIT | |
2878 InstructionMark im(this); | |
2879 emit_byte(0x81); | |
2880 emit_byte(0xF8 | src1->encoding()); | |
2881 emit_data(imm32, rspec, 0); | |
2882 } | |
2883 | |
2884 void Assembler::cmp_literal32(Address src1, int32_t imm32, RelocationHolder const& rspec) { | |
2885 // NO PREFIX AS NEVER 64BIT (not even 32bit versions of 64bit regs | |
2886 InstructionMark im(this); | |
2887 emit_byte(0x81); | |
2888 emit_operand(rdi, src1); | |
2889 emit_data(imm32, rspec, 0); | |
2890 } | |
2891 | |
2892 // The 64-bit (32bit platform) cmpxchg compares the value at adr with the contents of rdx:rax, | |
2893 // and stores rcx:rbx into adr if so; otherwise, the value at adr is loaded | |
2894 // into rdx:rax. The ZF is set if the compared values were equal, and cleared otherwise. | |
2895 void Assembler::cmpxchg8(Address adr) { | |
2896 InstructionMark im(this); | |
2897 emit_byte(0x0F); | |
2898 emit_byte(0xc7); | |
2899 emit_operand(rcx, adr); | |
2900 } | |
2901 | |
2902 void Assembler::decl(Register dst) { | |
2903 // Don't use it directly. Use MacroAssembler::decrementl() instead. | |
2904 emit_byte(0x48 | dst->encoding()); | |
2905 } | |
2906 | |
2907 #endif // _LP64 | |
2908 | |
2909 // 64bit typically doesn't use the x87 but needs to for the trig funcs | |
2910 | |
2911 void Assembler::fabs() { | |
2912 emit_byte(0xD9); | |
2913 emit_byte(0xE1); | |
2914 } | |
2915 | |
2916 void Assembler::fadd(int i) { | |
2917 emit_farith(0xD8, 0xC0, i); | |
2918 } | |
2919 | |
2920 void Assembler::fadd_d(Address src) { | |
2921 InstructionMark im(this); | |
2922 emit_byte(0xDC); | |
2923 emit_operand32(rax, src); | |
2924 } | |
2925 | |
2926 void Assembler::fadd_s(Address src) { | |
2927 InstructionMark im(this); | |
2928 emit_byte(0xD8); | |
2929 emit_operand32(rax, src); | |
2930 } | |
2931 | |
2932 void Assembler::fadda(int i) { | |
2933 emit_farith(0xDC, 0xC0, i); | |
2934 } | |
2935 | |
2936 void Assembler::faddp(int i) { | |
2937 emit_farith(0xDE, 0xC0, i); | |
2938 } | |
2939 | |
2940 void Assembler::fchs() { | |
2941 emit_byte(0xD9); | |
2942 emit_byte(0xE0); | |
2943 } | |
2944 | |
2945 void Assembler::fcom(int i) { | |
2946 emit_farith(0xD8, 0xD0, i); | |
2947 } | |
2948 | |
2949 void Assembler::fcomp(int i) { | |
2950 emit_farith(0xD8, 0xD8, i); | |
2951 } | |
2952 | |
2953 void Assembler::fcomp_d(Address src) { | |
2954 InstructionMark im(this); | |
2955 emit_byte(0xDC); | |
2956 emit_operand32(rbx, src); | |
2957 } | |
2958 | |
2959 void Assembler::fcomp_s(Address src) { | |
2960 InstructionMark im(this); | |
2961 emit_byte(0xD8); | |
2962 emit_operand32(rbx, src); | |
2963 } | |
2964 | |
2965 void Assembler::fcompp() { | |
2966 emit_byte(0xDE); | |
2967 emit_byte(0xD9); | |
2968 } | |
2969 | |
2970 void Assembler::fcos() { | |
2971 emit_byte(0xD9); | |
0 | 2972 emit_byte(0xFF); |
304 | 2973 } |
2974 | |
2975 void Assembler::fdecstp() { | |
2976 emit_byte(0xD9); | |
2977 emit_byte(0xF6); | |
2978 } | |
2979 | |
2980 void Assembler::fdiv(int i) { | |
2981 emit_farith(0xD8, 0xF0, i); | |
2982 } | |
2983 | |
2984 void Assembler::fdiv_d(Address src) { | |
2985 InstructionMark im(this); | |
2986 emit_byte(0xDC); | |
2987 emit_operand32(rsi, src); | |
2988 } | |
2989 | |
2990 void Assembler::fdiv_s(Address src) { | |
2991 InstructionMark im(this); | |
2992 emit_byte(0xD8); | |
2993 emit_operand32(rsi, src); | |
2994 } | |
2995 | |
2996 void Assembler::fdiva(int i) { | |
2997 emit_farith(0xDC, 0xF8, i); | |
2998 } | |
2999 | |
3000 // Note: The Intel manual (Pentium Processor User's Manual, Vol.3, 1994) | |
3001 // is erroneous for some of the floating-point instructions below. | |
3002 | |
3003 void Assembler::fdivp(int i) { | |
3004 emit_farith(0xDE, 0xF8, i); // ST(0) <- ST(0) / ST(1) and pop (Intel manual wrong) | |
3005 } | |
3006 | |
3007 void Assembler::fdivr(int i) { | |
3008 emit_farith(0xD8, 0xF8, i); | |
3009 } | |
3010 | |
3011 void Assembler::fdivr_d(Address src) { | |
3012 InstructionMark im(this); | |
3013 emit_byte(0xDC); | |
3014 emit_operand32(rdi, src); | |
3015 } | |
3016 | |
3017 void Assembler::fdivr_s(Address src) { | |
3018 InstructionMark im(this); | |
3019 emit_byte(0xD8); | |
3020 emit_operand32(rdi, src); | |
3021 } | |
3022 | |
3023 void Assembler::fdivra(int i) { | |
3024 emit_farith(0xDC, 0xF0, i); | |
3025 } | |
3026 | |
3027 void Assembler::fdivrp(int i) { | |
3028 emit_farith(0xDE, 0xF0, i); // ST(0) <- ST(1) / ST(0) and pop (Intel manual wrong) | |
3029 } | |
3030 | |
3031 void Assembler::ffree(int i) { | |
3032 emit_farith(0xDD, 0xC0, i); | |
3033 } | |
3034 | |
3035 void Assembler::fild_d(Address adr) { | |
3036 InstructionMark im(this); | |
3037 emit_byte(0xDF); | |
3038 emit_operand32(rbp, adr); | |
3039 } | |
3040 | |
3041 void Assembler::fild_s(Address adr) { | |
3042 InstructionMark im(this); | |
3043 emit_byte(0xDB); | |
3044 emit_operand32(rax, adr); | |
3045 } | |
3046 | |
3047 void Assembler::fincstp() { | |
3048 emit_byte(0xD9); | |
3049 emit_byte(0xF7); | |
3050 } | |
3051 | |
3052 void Assembler::finit() { | |
3053 emit_byte(0x9B); | |
3054 emit_byte(0xDB); | |
3055 emit_byte(0xE3); | |
3056 } | |
3057 | |
3058 void Assembler::fist_s(Address adr) { | |
3059 InstructionMark im(this); | |
3060 emit_byte(0xDB); | |
3061 emit_operand32(rdx, adr); | |
3062 } | |
3063 | |
3064 void Assembler::fistp_d(Address adr) { | |
3065 InstructionMark im(this); | |
3066 emit_byte(0xDF); | |
3067 emit_operand32(rdi, adr); | |
3068 } | |
3069 | |
3070 void Assembler::fistp_s(Address adr) { | |
3071 InstructionMark im(this); | |
3072 emit_byte(0xDB); | |
3073 emit_operand32(rbx, adr); | |
3074 } | |
0 | 3075 |
3076 void Assembler::fld1() { | |
3077 emit_byte(0xD9); | |
3078 emit_byte(0xE8); | |
3079 } | |
3080 | |
304 | 3081 void Assembler::fld_d(Address adr) { |
3082 InstructionMark im(this); | |
3083 emit_byte(0xDD); | |
3084 emit_operand32(rax, adr); | |
3085 } | |
0 | 3086 |
3087 void Assembler::fld_s(Address adr) { | |
3088 InstructionMark im(this); | |
3089 emit_byte(0xD9); | |
304 | 3090 emit_operand32(rax, adr); |
3091 } | |
3092 | |
3093 | |
3094 void Assembler::fld_s(int index) { | |
0 | 3095 emit_farith(0xD9, 0xC0, index); |
3096 } | |
3097 | |
3098 void Assembler::fld_x(Address adr) { | |
3099 InstructionMark im(this); | |
3100 emit_byte(0xDB); | |
304 | 3101 emit_operand32(rbp, adr); |
3102 } | |
3103 | |
3104 void Assembler::fldcw(Address src) { | |
3105 InstructionMark im(this); | |
3106 emit_byte(0xd9); | |
3107 emit_operand32(rbp, src); | |
3108 } | |
3109 | |
3110 void Assembler::fldenv(Address src) { | |
0 | 3111 InstructionMark im(this); |
3112 emit_byte(0xD9); | |
304 | 3113 emit_operand32(rsp, src); |
3114 } | |
3115 | |
3116 void Assembler::fldlg2() { | |
0 | 3117 emit_byte(0xD9); |
304 | 3118 emit_byte(0xEC); |
3119 } | |
0 | 3120 |
3121 void Assembler::fldln2() { | |
3122 emit_byte(0xD9); | |
3123 emit_byte(0xED); | |
3124 } | |
3125 | |
304 | 3126 void Assembler::fldz() { |
0 | 3127 emit_byte(0xD9); |
304 | 3128 emit_byte(0xEE); |
3129 } | |
0 | 3130 |
3131 void Assembler::flog() { | |
3132 fldln2(); | |
3133 fxch(); | |
3134 fyl2x(); | |
3135 } | |
3136 | |
3137 void Assembler::flog10() { | |
3138 fldlg2(); | |
3139 fxch(); | |
3140 fyl2x(); | |
3141 } | |
3142 | |
304 | 3143 void Assembler::fmul(int i) { |
3144 emit_farith(0xD8, 0xC8, i); | |
3145 } | |
3146 | |
3147 void Assembler::fmul_d(Address src) { | |
3148 InstructionMark im(this); | |
3149 emit_byte(0xDC); | |
3150 emit_operand32(rcx, src); | |
3151 } | |
3152 | |
3153 void Assembler::fmul_s(Address src) { | |
3154 InstructionMark im(this); | |
3155 emit_byte(0xD8); | |
3156 emit_operand32(rcx, src); | |
3157 } | |
3158 | |
3159 void Assembler::fmula(int i) { | |
3160 emit_farith(0xDC, 0xC8, i); | |
3161 } | |
3162 | |
3163 void Assembler::fmulp(int i) { | |
3164 emit_farith(0xDE, 0xC8, i); | |
3165 } | |
3166 | |
3167 void Assembler::fnsave(Address dst) { | |
3168 InstructionMark im(this); | |
3169 emit_byte(0xDD); | |
3170 emit_operand32(rsi, dst); | |
3171 } | |
3172 | |
3173 void Assembler::fnstcw(Address src) { | |
3174 InstructionMark im(this); | |
3175 emit_byte(0x9B); | |
3176 emit_byte(0xD9); | |
3177 emit_operand32(rdi, src); | |
3178 } | |
3179 | |
3180 void Assembler::fnstsw_ax() { | |
3181 emit_byte(0xdF); | |
3182 emit_byte(0xE0); | |
3183 } | |
3184 | |
3185 void Assembler::fprem() { | |
3186 emit_byte(0xD9); | |
3187 emit_byte(0xF8); | |
3188 } | |
3189 | |
3190 void Assembler::fprem1() { | |
3191 emit_byte(0xD9); | |
3192 emit_byte(0xF5); | |
3193 } | |
3194 | |
3195 void Assembler::frstor(Address src) { | |
3196 InstructionMark im(this); | |
3197 emit_byte(0xDD); | |
3198 emit_operand32(rsp, src); | |
3199 } | |
0 | 3200 |
3201 void Assembler::fsin() { | |
3202 emit_byte(0xD9); | |
3203 emit_byte(0xFE); | |
3204 } | |
3205 | |
304 | 3206 void Assembler::fsqrt() { |
3207 emit_byte(0xD9); | |
3208 emit_byte(0xFA); | |
3209 } | |
3210 | |
3211 void Assembler::fst_d(Address adr) { | |
3212 InstructionMark im(this); | |
3213 emit_byte(0xDD); | |
3214 emit_operand32(rdx, adr); | |
3215 } | |
3216 | |
3217 void Assembler::fst_s(Address adr) { | |
3218 InstructionMark im(this); | |
3219 emit_byte(0xD9); | |
3220 emit_operand32(rdx, adr); | |
3221 } | |
3222 | |
3223 void Assembler::fstp_d(Address adr) { | |
3224 InstructionMark im(this); | |
3225 emit_byte(0xDD); | |
3226 emit_operand32(rbx, adr); | |
3227 } | |
3228 | |
3229 void Assembler::fstp_d(int index) { | |
3230 emit_farith(0xDD, 0xD8, index); | |
3231 } | |
3232 | |
3233 void Assembler::fstp_s(Address adr) { | |
3234 InstructionMark im(this); | |
0 | 3235 emit_byte(0xD9); |
304 | 3236 emit_operand32(rbx, adr); |
3237 } | |
3238 | |
3239 void Assembler::fstp_x(Address adr) { | |
3240 InstructionMark im(this); | |
3241 emit_byte(0xDB); | |
3242 emit_operand32(rdi, adr); | |
3243 } | |
3244 | |
3245 void Assembler::fsub(int i) { | |
3246 emit_farith(0xD8, 0xE0, i); | |
3247 } | |
3248 | |
3249 void Assembler::fsub_d(Address src) { | |
3250 InstructionMark im(this); | |
3251 emit_byte(0xDC); | |
3252 emit_operand32(rsp, src); | |
3253 } | |
3254 | |
3255 void Assembler::fsub_s(Address src) { | |
3256 InstructionMark im(this); | |
3257 emit_byte(0xD8); | |
3258 emit_operand32(rsp, src); | |
3259 } | |
3260 | |
3261 void Assembler::fsuba(int i) { | |
3262 emit_farith(0xDC, 0xE8, i); | |
3263 } | |
3264 | |
3265 void Assembler::fsubp(int i) { | |
3266 emit_farith(0xDE, 0xE8, i); // ST(0) <- ST(0) - ST(1) and pop (Intel manual wrong) | |
3267 } | |
3268 | |
3269 void Assembler::fsubr(int i) { | |
3270 emit_farith(0xD8, 0xE8, i); | |
3271 } | |
3272 | |
3273 void Assembler::fsubr_d(Address src) { | |
3274 InstructionMark im(this); | |
3275 emit_byte(0xDC); | |
3276 emit_operand32(rbp, src); | |
3277 } | |
3278 | |
3279 void Assembler::fsubr_s(Address src) { | |
3280 InstructionMark im(this); | |
3281 emit_byte(0xD8); | |
3282 emit_operand32(rbp, src); | |
3283 } | |
3284 | |
3285 void Assembler::fsubra(int i) { | |
3286 emit_farith(0xDC, 0xE0, i); | |
3287 } | |
3288 | |
3289 void Assembler::fsubrp(int i) { | |
3290 emit_farith(0xDE, 0xE0, i); // ST(0) <- ST(1) - ST(0) and pop (Intel manual wrong) | |
0 | 3291 } |
3292 | |
3293 void Assembler::ftan() { | |
3294 emit_byte(0xD9); | |
3295 emit_byte(0xF2); | |
3296 emit_byte(0xDD); | |
3297 emit_byte(0xD8); | |
3298 } | |
3299 | |
304 | 3300 void Assembler::ftst() { |
0 | 3301 emit_byte(0xD9); |
304 | 3302 emit_byte(0xE4); |
3303 } | |
0 | 3304 |
3305 void Assembler::fucomi(int i) { | |
3306 // make sure the instruction is supported (introduced for P6, together with cmov) | |
3307 guarantee(VM_Version::supports_cmov(), "illegal instruction"); | |
3308 emit_farith(0xDB, 0xE8, i); | |
3309 } | |
3310 | |
3311 void Assembler::fucomip(int i) { | |
3312 // make sure the instruction is supported (introduced for P6, together with cmov) | |
3313 guarantee(VM_Version::supports_cmov(), "illegal instruction"); | |
3314 emit_farith(0xDF, 0xE8, i); | |
3315 } | |
3316 | |
3317 void Assembler::fwait() { | |
3318 emit_byte(0x9B); | |
3319 } | |
3320 | |
304 | 3321 void Assembler::fxch(int i) { |
3322 emit_farith(0xD9, 0xC8, i); | |
3323 } | |
3324 | |
3325 void Assembler::fyl2x() { | |
0 | 3326 emit_byte(0xD9); |
304 | 3327 emit_byte(0xF1); |
3328 } | |
3329 | |
3330 | |
3331 #ifndef _LP64 | |
3332 | |
3333 void Assembler::incl(Register dst) { | |
3334 // Don't use it directly. Use MacroAssembler::incrementl() instead. | |
3335 emit_byte(0x40 | dst->encoding()); | |
3336 } | |
3337 | |
3338 void Assembler::lea(Register dst, Address src) { | |
3339 leal(dst, src); | |
3340 } | |
3341 | |
3342 void Assembler::mov_literal32(Address dst, int32_t imm32, RelocationHolder const& rspec) { | |
3343 InstructionMark im(this); | |
3344 emit_byte(0xC7); | |
3345 emit_operand(rax, dst); | |
3346 emit_data((int)imm32, rspec, 0); | |
3347 } | |
3348 | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3349 void Assembler::mov_literal32(Register dst, int32_t imm32, RelocationHolder const& rspec) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3350 InstructionMark im(this); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3351 int encode = prefix_and_encode(dst->encoding()); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3352 emit_byte(0xB8 | encode); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3353 emit_data((int)imm32, rspec, 0); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3354 } |
304 | 3355 |
3356 void Assembler::popa() { // 32bit | |
3357 emit_byte(0x61); | |
3358 } | |
3359 | |
3360 void Assembler::push_literal32(int32_t imm32, RelocationHolder const& rspec) { | |
3361 InstructionMark im(this); | |
3362 emit_byte(0x68); | |
3363 emit_data(imm32, rspec, 0); | |
3364 } | |
3365 | |
3366 void Assembler::pusha() { // 32bit | |
3367 emit_byte(0x60); | |
3368 } | |
3369 | |
3370 void Assembler::set_byte_if_not_zero(Register dst) { | |
0 | 3371 emit_byte(0x0F); |
304 | 3372 emit_byte(0x95); |
3373 emit_byte(0xE0 | dst->encoding()); | |
3374 } | |
3375 | |
3376 void Assembler::shldl(Register dst, Register src) { | |
0 | 3377 emit_byte(0x0F); |
304 | 3378 emit_byte(0xA5); |
3379 emit_byte(0xC0 | src->encoding() << 3 | dst->encoding()); | |
3380 } | |
3381 | |
3382 void Assembler::shrdl(Register dst, Register src) { | |
0 | 3383 emit_byte(0x0F); |
304 | 3384 emit_byte(0xAD); |
3385 emit_byte(0xC0 | src->encoding() << 3 | dst->encoding()); | |
3386 } | |
3387 | |
3388 #else // LP64 | |
3389 | |
1369 | 3390 void Assembler::set_byte_if_not_zero(Register dst) { |
3391 int enc = prefix_and_encode(dst->encoding(), true); | |
3392 emit_byte(0x0F); | |
3393 emit_byte(0x95); | |
3394 emit_byte(0xE0 | enc); | |
3395 } | |
3396 | |
304 | 3397 // 64bit only pieces of the assembler |
3398 // This should only be used by 64bit instructions that can use rip-relative | |
3399 // it cannot be used by instructions that want an immediate value. | |
3400 | |
3401 bool Assembler::reachable(AddressLiteral adr) { | |
3402 int64_t disp; | |
3403 // None will force a 64bit literal to the code stream. Likely a placeholder | |
3404 // for something that will be patched later and we need to certain it will | |
3405 // always be reachable. | |
3406 if (adr.reloc() == relocInfo::none) { | |
3407 return false; | |
3408 } | |
3409 if (adr.reloc() == relocInfo::internal_word_type) { | |
3410 // This should be rip relative and easily reachable. | |
3411 return true; | |
3412 } | |
3413 if (adr.reloc() == relocInfo::virtual_call_type || | |
3414 adr.reloc() == relocInfo::opt_virtual_call_type || | |
3415 adr.reloc() == relocInfo::static_call_type || | |
3416 adr.reloc() == relocInfo::static_stub_type ) { | |
3417 // This should be rip relative within the code cache and easily | |
3418 // reachable until we get huge code caches. (At which point | |
3419 // ic code is going to have issues). | |
3420 return true; | |
3421 } | |
3422 if (adr.reloc() != relocInfo::external_word_type && | |
3423 adr.reloc() != relocInfo::poll_return_type && // these are really external_word but need special | |
3424 adr.reloc() != relocInfo::poll_type && // relocs to identify them | |
3425 adr.reloc() != relocInfo::runtime_call_type ) { | |
3426 return false; | |
3427 } | |
3428 | |
3429 // Stress the correction code | |
3430 if (ForceUnreachable) { | |
3431 // Must be runtimecall reloc, see if it is in the codecache | |
3432 // Flipping stuff in the codecache to be unreachable causes issues | |
3433 // with things like inline caches where the additional instructions | |
3434 // are not handled. | |
3435 if (CodeCache::find_blob(adr._target) == NULL) { | |
3436 return false; | |
3437 } | |
3438 } | |
3439 // For external_word_type/runtime_call_type if it is reachable from where we | |
3440 // are now (possibly a temp buffer) and where we might end up | |
3441 // anywhere in the codeCache then we are always reachable. | |
3442 // This would have to change if we ever save/restore shared code | |
3443 // to be more pessimistic. | |
3444 | |
3445 disp = (int64_t)adr._target - ((int64_t)CodeCache::low_bound() + sizeof(int)); | |
3446 if (!is_simm32(disp)) return false; | |
3447 disp = (int64_t)adr._target - ((int64_t)CodeCache::high_bound() + sizeof(int)); | |
3448 if (!is_simm32(disp)) return false; | |
3449 | |
3450 disp = (int64_t)adr._target - ((int64_t)_code_pos + sizeof(int)); | |
3451 | |
3452 // Because rip relative is a disp + address_of_next_instruction and we | |
3453 // don't know the value of address_of_next_instruction we apply a fudge factor | |
3454 // to make sure we will be ok no matter the size of the instruction we get placed into. | |
3455 // We don't have to fudge the checks above here because they are already worst case. | |
3456 | |
3457 // 12 == override/rex byte, opcode byte, rm byte, sib byte, a 4-byte disp , 4-byte literal | |
3458 // + 4 because better safe than sorry. | |
3459 const int fudge = 12 + 4; | |
3460 if (disp < 0) { | |
3461 disp -= fudge; | |
3462 } else { | |
3463 disp += fudge; | |
3464 } | |
3465 return is_simm32(disp); | |
3466 } | |
3467 | |
3468 void Assembler::emit_data64(jlong data, | |
3469 relocInfo::relocType rtype, | |
3470 int format) { | |
3471 if (rtype == relocInfo::none) { | |
3472 emit_long64(data); | |
3473 } else { | |
3474 emit_data64(data, Relocation::spec_simple(rtype), format); | |
3475 } | |
3476 } | |
3477 | |
3478 void Assembler::emit_data64(jlong data, | |
3479 RelocationHolder const& rspec, | |
3480 int format) { | |
3481 assert(imm_operand == 0, "default format must be immediate in this file"); | |
3482 assert(imm_operand == format, "must be immediate"); | |
3483 assert(inst_mark() != NULL, "must be inside InstructionMark"); | |
3484 // Do not use AbstractAssembler::relocate, which is not intended for | |
3485 // embedded words. Instead, relocate to the enclosing instruction. | |
3486 code_section()->relocate(inst_mark(), rspec, format); | |
3487 #ifdef ASSERT | |
3488 check_relocation(rspec, format); | |
3489 #endif | |
3490 emit_long64(data); | |
3491 } | |
3492 | |
3493 int Assembler::prefix_and_encode(int reg_enc, bool byteinst) { | |
3494 if (reg_enc >= 8) { | |
3495 prefix(REX_B); | |
3496 reg_enc -= 8; | |
3497 } else if (byteinst && reg_enc >= 4) { | |
3498 prefix(REX); | |
3499 } | |
3500 return reg_enc; | |
3501 } | |
3502 | |
3503 int Assembler::prefixq_and_encode(int reg_enc) { | |
3504 if (reg_enc < 8) { | |
3505 prefix(REX_W); | |
3506 } else { | |
3507 prefix(REX_WB); | |
3508 reg_enc -= 8; | |
3509 } | |
3510 return reg_enc; | |
3511 } | |
3512 | |
3513 int Assembler::prefix_and_encode(int dst_enc, int src_enc, bool byteinst) { | |
3514 if (dst_enc < 8) { | |
3515 if (src_enc >= 8) { | |
3516 prefix(REX_B); | |
3517 src_enc -= 8; | |
3518 } else if (byteinst && src_enc >= 4) { | |
3519 prefix(REX); | |
3520 } | |
3521 } else { | |
3522 if (src_enc < 8) { | |
3523 prefix(REX_R); | |
3524 } else { | |
3525 prefix(REX_RB); | |
3526 src_enc -= 8; | |
3527 } | |
3528 dst_enc -= 8; | |
3529 } | |
3530 return dst_enc << 3 | src_enc; | |
3531 } | |
3532 | |
3533 int Assembler::prefixq_and_encode(int dst_enc, int src_enc) { | |
3534 if (dst_enc < 8) { | |
3535 if (src_enc < 8) { | |
3536 prefix(REX_W); | |
3537 } else { | |
3538 prefix(REX_WB); | |
3539 src_enc -= 8; | |
3540 } | |
3541 } else { | |
3542 if (src_enc < 8) { | |
3543 prefix(REX_WR); | |
3544 } else { | |
3545 prefix(REX_WRB); | |
3546 src_enc -= 8; | |
3547 } | |
3548 dst_enc -= 8; | |
3549 } | |
3550 return dst_enc << 3 | src_enc; | |
3551 } | |
3552 | |
3553 void Assembler::prefix(Register reg) { | |
3554 if (reg->encoding() >= 8) { | |
3555 prefix(REX_B); | |
3556 } | |
3557 } | |
3558 | |
3559 void Assembler::prefix(Address adr) { | |
3560 if (adr.base_needs_rex()) { | |
3561 if (adr.index_needs_rex()) { | |
3562 prefix(REX_XB); | |
3563 } else { | |
3564 prefix(REX_B); | |
3565 } | |
3566 } else { | |
3567 if (adr.index_needs_rex()) { | |
3568 prefix(REX_X); | |
3569 } | |
3570 } | |
3571 } | |
3572 | |
3573 void Assembler::prefixq(Address adr) { | |
3574 if (adr.base_needs_rex()) { | |
3575 if (adr.index_needs_rex()) { | |
3576 prefix(REX_WXB); | |
3577 } else { | |
3578 prefix(REX_WB); | |
3579 } | |
3580 } else { | |
3581 if (adr.index_needs_rex()) { | |
3582 prefix(REX_WX); | |
3583 } else { | |
3584 prefix(REX_W); | |
3585 } | |
3586 } | |
3587 } | |
3588 | |
3589 | |
3590 void Assembler::prefix(Address adr, Register reg, bool byteinst) { | |
3591 if (reg->encoding() < 8) { | |
3592 if (adr.base_needs_rex()) { | |
3593 if (adr.index_needs_rex()) { | |
3594 prefix(REX_XB); | |
3595 } else { | |
3596 prefix(REX_B); | |
3597 } | |
3598 } else { | |
3599 if (adr.index_needs_rex()) { | |
3600 prefix(REX_X); | |
3601 } else if (reg->encoding() >= 4 ) { | |
3602 prefix(REX); | |
3603 } | |
3604 } | |
3605 } else { | |
3606 if (adr.base_needs_rex()) { | |
3607 if (adr.index_needs_rex()) { | |
3608 prefix(REX_RXB); | |
3609 } else { | |
3610 prefix(REX_RB); | |
3611 } | |
3612 } else { | |
3613 if (adr.index_needs_rex()) { | |
3614 prefix(REX_RX); | |
3615 } else { | |
3616 prefix(REX_R); | |
3617 } | |
3618 } | |
3619 } | |
3620 } | |
3621 | |
3622 void Assembler::prefixq(Address adr, Register src) { | |
3623 if (src->encoding() < 8) { | |
3624 if (adr.base_needs_rex()) { | |
3625 if (adr.index_needs_rex()) { | |
3626 prefix(REX_WXB); | |
3627 } else { | |
3628 prefix(REX_WB); | |
3629 } | |
3630 } else { | |
3631 if (adr.index_needs_rex()) { | |
3632 prefix(REX_WX); | |
3633 } else { | |
3634 prefix(REX_W); | |
3635 } | |
3636 } | |
3637 } else { | |
3638 if (adr.base_needs_rex()) { | |
3639 if (adr.index_needs_rex()) { | |
3640 prefix(REX_WRXB); | |
3641 } else { | |
3642 prefix(REX_WRB); | |
3643 } | |
3644 } else { | |
3645 if (adr.index_needs_rex()) { | |
3646 prefix(REX_WRX); | |
3647 } else { | |
3648 prefix(REX_WR); | |
3649 } | |
3650 } | |
3651 } | |
3652 } | |
3653 | |
3654 void Assembler::prefix(Address adr, XMMRegister reg) { | |
3655 if (reg->encoding() < 8) { | |
3656 if (adr.base_needs_rex()) { | |
3657 if (adr.index_needs_rex()) { | |
3658 prefix(REX_XB); | |
3659 } else { | |
3660 prefix(REX_B); | |
3661 } | |
3662 } else { | |
3663 if (adr.index_needs_rex()) { | |
3664 prefix(REX_X); | |
3665 } | |
3666 } | |
3667 } else { | |
3668 if (adr.base_needs_rex()) { | |
3669 if (adr.index_needs_rex()) { | |
3670 prefix(REX_RXB); | |
3671 } else { | |
3672 prefix(REX_RB); | |
3673 } | |
3674 } else { | |
3675 if (adr.index_needs_rex()) { | |
3676 prefix(REX_RX); | |
3677 } else { | |
3678 prefix(REX_R); | |
3679 } | |
3680 } | |
3681 } | |
3682 } | |
3683 | |
3684 void Assembler::adcq(Register dst, int32_t imm32) { | |
3685 (void) prefixq_and_encode(dst->encoding()); | |
3686 emit_arith(0x81, 0xD0, dst, imm32); | |
3687 } | |
3688 | |
3689 void Assembler::adcq(Register dst, Address src) { | |
3690 InstructionMark im(this); | |
3691 prefixq(src, dst); | |
3692 emit_byte(0x13); | |
3693 emit_operand(dst, src); | |
3694 } | |
3695 | |
3696 void Assembler::adcq(Register dst, Register src) { | |
3697 (int) prefixq_and_encode(dst->encoding(), src->encoding()); | |
3698 emit_arith(0x13, 0xC0, dst, src); | |
3699 } | |
3700 | |
3701 void Assembler::addq(Address dst, int32_t imm32) { | |
3702 InstructionMark im(this); | |
3703 prefixq(dst); | |
3704 emit_arith_operand(0x81, rax, dst,imm32); | |
3705 } | |
3706 | |
3707 void Assembler::addq(Address dst, Register src) { | |
3708 InstructionMark im(this); | |
3709 prefixq(dst, src); | |
3710 emit_byte(0x01); | |
3711 emit_operand(src, dst); | |
3712 } | |
3713 | |
3714 void Assembler::addq(Register dst, int32_t imm32) { | |
3715 (void) prefixq_and_encode(dst->encoding()); | |
3716 emit_arith(0x81, 0xC0, dst, imm32); | |
3717 } | |
3718 | |
3719 void Assembler::addq(Register dst, Address src) { | |
3720 InstructionMark im(this); | |
3721 prefixq(src, dst); | |
3722 emit_byte(0x03); | |
3723 emit_operand(dst, src); | |
3724 } | |
3725 | |
3726 void Assembler::addq(Register dst, Register src) { | |
3727 (void) prefixq_and_encode(dst->encoding(), src->encoding()); | |
3728 emit_arith(0x03, 0xC0, dst, src); | |
3729 } | |
3730 | |
3731 void Assembler::andq(Register dst, int32_t imm32) { | |
3732 (void) prefixq_and_encode(dst->encoding()); | |
3733 emit_arith(0x81, 0xE0, dst, imm32); | |
3734 } | |
3735 | |
3736 void Assembler::andq(Register dst, Address src) { | |
3737 InstructionMark im(this); | |
3738 prefixq(src, dst); | |
3739 emit_byte(0x23); | |
3740 emit_operand(dst, src); | |
3741 } | |
3742 | |
3743 void Assembler::andq(Register dst, Register src) { | |
3744 (int) prefixq_and_encode(dst->encoding(), src->encoding()); | |
3745 emit_arith(0x23, 0xC0, dst, src); | |
3746 } | |
3747 | |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3748 void Assembler::bsfq(Register dst, Register src) { |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3749 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3750 emit_byte(0x0F); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3751 emit_byte(0xBC); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3752 emit_byte(0xC0 | encode); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3753 } |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3754 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3755 void Assembler::bsrq(Register dst, Register src) { |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3756 assert(!VM_Version::supports_lzcnt(), "encoding is treated as LZCNT"); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3757 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3758 emit_byte(0x0F); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3759 emit_byte(0xBD); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3760 emit_byte(0xC0 | encode); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3761 } |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
3762 |
304 | 3763 void Assembler::bswapq(Register reg) { |
3764 int encode = prefixq_and_encode(reg->encoding()); | |
3765 emit_byte(0x0F); | |
3766 emit_byte(0xC8 | encode); | |
3767 } | |
3768 | |
3769 void Assembler::cdqq() { | |
3770 prefix(REX_W); | |
3771 emit_byte(0x99); | |
3772 } | |
3773 | |
3774 void Assembler::clflush(Address adr) { | |
3775 prefix(adr); | |
3776 emit_byte(0x0F); | |
3777 emit_byte(0xAE); | |
3778 emit_operand(rdi, adr); | |
3779 } | |
3780 | |
3781 void Assembler::cmovq(Condition cc, Register dst, Register src) { | |
3782 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
3783 emit_byte(0x0F); | |
3784 emit_byte(0x40 | cc); | |
3785 emit_byte(0xC0 | encode); | |
3786 } | |
3787 | |
3788 void Assembler::cmovq(Condition cc, Register dst, Address src) { | |
3789 InstructionMark im(this); | |
3790 prefixq(src, dst); | |
3791 emit_byte(0x0F); | |
3792 emit_byte(0x40 | cc); | |
3793 emit_operand(dst, src); | |
3794 } | |
3795 | |
3796 void Assembler::cmpq(Address dst, int32_t imm32) { | |
3797 InstructionMark im(this); | |
3798 prefixq(dst); | |
3799 emit_byte(0x81); | |
3800 emit_operand(rdi, dst, 4); | |
3801 emit_long(imm32); | |
3802 } | |
3803 | |
3804 void Assembler::cmpq(Register dst, int32_t imm32) { | |
3805 (void) prefixq_and_encode(dst->encoding()); | |
3806 emit_arith(0x81, 0xF8, dst, imm32); | |
3807 } | |
3808 | |
3809 void Assembler::cmpq(Address dst, Register src) { | |
3810 InstructionMark im(this); | |
3811 prefixq(dst, src); | |
3812 emit_byte(0x3B); | |
3813 emit_operand(src, dst); | |
3814 } | |
3815 | |
3816 void Assembler::cmpq(Register dst, Register src) { | |
3817 (void) prefixq_and_encode(dst->encoding(), src->encoding()); | |
3818 emit_arith(0x3B, 0xC0, dst, src); | |
3819 } | |
3820 | |
3821 void Assembler::cmpq(Register dst, Address src) { | |
3822 InstructionMark im(this); | |
3823 prefixq(src, dst); | |
3824 emit_byte(0x3B); | |
3825 emit_operand(dst, src); | |
3826 } | |
3827 | |
3828 void Assembler::cmpxchgq(Register reg, Address adr) { | |
3829 InstructionMark im(this); | |
3830 prefixq(adr, reg); | |
3831 emit_byte(0x0F); | |
3832 emit_byte(0xB1); | |
3833 emit_operand(reg, adr); | |
3834 } | |
3835 | |
3836 void Assembler::cvtsi2sdq(XMMRegister dst, Register src) { | |
3837 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
3838 emit_byte(0xF2); | |
3839 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
3840 emit_byte(0x0F); | |
3841 emit_byte(0x2A); | |
3842 emit_byte(0xC0 | encode); | |
3843 } | |
3844 | |
3845 void Assembler::cvtsi2ssq(XMMRegister dst, Register src) { | |
3846 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
3847 emit_byte(0xF3); | |
3848 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
3849 emit_byte(0x0F); | |
3850 emit_byte(0x2A); | |
3851 emit_byte(0xC0 | encode); | |
3852 } | |
3853 | |
3854 void Assembler::cvttsd2siq(Register dst, XMMRegister src) { | |
3855 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
3856 emit_byte(0xF2); | |
3857 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
3858 emit_byte(0x0F); | |
3859 emit_byte(0x2C); | |
3860 emit_byte(0xC0 | encode); | |
3861 } | |
3862 | |
3863 void Assembler::cvttss2siq(Register dst, XMMRegister src) { | |
3864 NOT_LP64(assert(VM_Version::supports_sse(), "")); | |
3865 emit_byte(0xF3); | |
3866 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
3867 emit_byte(0x0F); | |
3868 emit_byte(0x2C); | |
3869 emit_byte(0xC0 | encode); | |
3870 } | |
3871 | |
3872 void Assembler::decl(Register dst) { | |
3873 // Don't use it directly. Use MacroAssembler::decrementl() instead. | |
3874 // Use two-byte form (one-byte form is a REX prefix in 64-bit mode) | |
3875 int encode = prefix_and_encode(dst->encoding()); | |
3876 emit_byte(0xFF); | |
3877 emit_byte(0xC8 | encode); | |
3878 } | |
3879 | |
3880 void Assembler::decq(Register dst) { | |
3881 // Don't use it directly. Use MacroAssembler::decrementq() instead. | |
3882 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) | |
3883 int encode = prefixq_and_encode(dst->encoding()); | |
3884 emit_byte(0xFF); | |
3885 emit_byte(0xC8 | encode); | |
3886 } | |
3887 | |
3888 void Assembler::decq(Address dst) { | |
3889 // Don't use it directly. Use MacroAssembler::decrementq() instead. | |
3890 InstructionMark im(this); | |
3891 prefixq(dst); | |
3892 emit_byte(0xFF); | |
3893 emit_operand(rcx, dst); | |
3894 } | |
3895 | |
3896 void Assembler::fxrstor(Address src) { | |
3897 prefixq(src); | |
3898 emit_byte(0x0F); | |
3899 emit_byte(0xAE); | |
3900 emit_operand(as_Register(1), src); | |
3901 } | |
3902 | |
3903 void Assembler::fxsave(Address dst) { | |
3904 prefixq(dst); | |
3905 emit_byte(0x0F); | |
3906 emit_byte(0xAE); | |
3907 emit_operand(as_Register(0), dst); | |
3908 } | |
3909 | |
3910 void Assembler::idivq(Register src) { | |
3911 int encode = prefixq_and_encode(src->encoding()); | |
3912 emit_byte(0xF7); | |
3913 emit_byte(0xF8 | encode); | |
3914 } | |
3915 | |
3916 void Assembler::imulq(Register dst, Register src) { | |
3917 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
3918 emit_byte(0x0F); | |
3919 emit_byte(0xAF); | |
3920 emit_byte(0xC0 | encode); | |
3921 } | |
3922 | |
3923 void Assembler::imulq(Register dst, Register src, int value) { | |
3924 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
3925 if (is8bit(value)) { | |
3926 emit_byte(0x6B); | |
3927 emit_byte(0xC0 | encode); | |
1914
ae065c367d93
6987135: Performance regression on Intel platform with 32-bits edition between 6u13 and 6u14.
kvn
parents:
1846
diff
changeset
|
3928 emit_byte(value & 0xFF); |
304 | 3929 } else { |
3930 emit_byte(0x69); | |
3931 emit_byte(0xC0 | encode); | |
3932 emit_long(value); | |
3933 } | |
3934 } | |
3935 | |
3936 void Assembler::incl(Register dst) { | |
3937 // Don't use it directly. Use MacroAssembler::incrementl() instead. | |
3938 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) | |
3939 int encode = prefix_and_encode(dst->encoding()); | |
3940 emit_byte(0xFF); | |
3941 emit_byte(0xC0 | encode); | |
3942 } | |
3943 | |
3944 void Assembler::incq(Register dst) { | |
3945 // Don't use it directly. Use MacroAssembler::incrementq() instead. | |
3946 // Use two-byte form (one-byte from is a REX prefix in 64-bit mode) | |
3947 int encode = prefixq_and_encode(dst->encoding()); | |
3948 emit_byte(0xFF); | |
3949 emit_byte(0xC0 | encode); | |
3950 } | |
3951 | |
3952 void Assembler::incq(Address dst) { | |
3953 // Don't use it directly. Use MacroAssembler::incrementq() instead. | |
3954 InstructionMark im(this); | |
3955 prefixq(dst); | |
3956 emit_byte(0xFF); | |
3957 emit_operand(rax, dst); | |
3958 } | |
3959 | |
3960 void Assembler::lea(Register dst, Address src) { | |
3961 leaq(dst, src); | |
3962 } | |
3963 | |
3964 void Assembler::leaq(Register dst, Address src) { | |
3965 InstructionMark im(this); | |
3966 prefixq(src, dst); | |
3967 emit_byte(0x8D); | |
3968 emit_operand(dst, src); | |
3969 } | |
3970 | |
3971 void Assembler::mov64(Register dst, int64_t imm64) { | |
3972 InstructionMark im(this); | |
3973 int encode = prefixq_and_encode(dst->encoding()); | |
3974 emit_byte(0xB8 | encode); | |
3975 emit_long64(imm64); | |
3976 } | |
3977 | |
3978 void Assembler::mov_literal64(Register dst, intptr_t imm64, RelocationHolder const& rspec) { | |
3979 InstructionMark im(this); | |
3980 int encode = prefixq_and_encode(dst->encoding()); | |
3981 emit_byte(0xB8 | encode); | |
3982 emit_data64(imm64, rspec); | |
3983 } | |
3984 | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3985 void Assembler::mov_narrow_oop(Register dst, int32_t imm32, RelocationHolder const& rspec) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3986 InstructionMark im(this); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3987 int encode = prefix_and_encode(dst->encoding()); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3988 emit_byte(0xB8 | encode); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3989 emit_data((int)imm32, rspec, narrow_oop_operand); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3990 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3991 |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3992 void Assembler::mov_narrow_oop(Address dst, int32_t imm32, RelocationHolder const& rspec) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3993 InstructionMark im(this); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3994 prefix(dst); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3995 emit_byte(0xC7); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3996 emit_operand(rax, dst, 4); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3997 emit_data((int)imm32, rspec, narrow_oop_operand); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3998 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
3999 |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4000 void Assembler::cmp_narrow_oop(Register src1, int32_t imm32, RelocationHolder const& rspec) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4001 InstructionMark im(this); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4002 int encode = prefix_and_encode(src1->encoding()); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4003 emit_byte(0x81); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4004 emit_byte(0xF8 | encode); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4005 emit_data((int)imm32, rspec, narrow_oop_operand); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4006 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4007 |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4008 void Assembler::cmp_narrow_oop(Address src1, int32_t imm32, RelocationHolder const& rspec) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4009 InstructionMark im(this); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4010 prefix(src1); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4011 emit_byte(0x81); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4012 emit_operand(rax, src1, 4); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4013 emit_data((int)imm32, rspec, narrow_oop_operand); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4014 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
4015 |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
4016 void Assembler::lzcntq(Register dst, Register src) { |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
4017 assert(VM_Version::supports_lzcnt(), "encoding is treated as BSR"); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
4018 emit_byte(0xF3); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
4019 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
4020 emit_byte(0x0F); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
4021 emit_byte(0xBD); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
4022 emit_byte(0xC0 | encode); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
4023 } |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
710
diff
changeset
|
4024 |
304 | 4025 void Assembler::movdq(XMMRegister dst, Register src) { |
4026 // table D-1 says MMX/SSE2 | |
4027 NOT_LP64(assert(VM_Version::supports_sse2() || VM_Version::supports_mmx(), "")); | |
0 | 4028 emit_byte(0x66); |
304 | 4029 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
0 | 4030 emit_byte(0x0F); |
304 | 4031 emit_byte(0x6E); |
4032 emit_byte(0xC0 | encode); | |
4033 } | |
4034 | |
4035 void Assembler::movdq(Register dst, XMMRegister src) { | |
4036 // table D-1 says MMX/SSE2 | |
4037 NOT_LP64(assert(VM_Version::supports_sse2() || VM_Version::supports_mmx(), "")); | |
0 | 4038 emit_byte(0x66); |
304 | 4039 // swap src/dst to get correct prefix |
4040 int encode = prefixq_and_encode(src->encoding(), dst->encoding()); | |
0 | 4041 emit_byte(0x0F); |
4042 emit_byte(0x7E); | |
304 | 4043 emit_byte(0xC0 | encode); |
4044 } | |
4045 | |
4046 void Assembler::movq(Register dst, Register src) { | |
4047 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
4048 emit_byte(0x8B); | |
4049 emit_byte(0xC0 | encode); | |
4050 } | |
4051 | |
4052 void Assembler::movq(Register dst, Address src) { | |
4053 InstructionMark im(this); | |
4054 prefixq(src, dst); | |
4055 emit_byte(0x8B); | |
4056 emit_operand(dst, src); | |
4057 } | |
4058 | |
4059 void Assembler::movq(Address dst, Register src) { | |
4060 InstructionMark im(this); | |
4061 prefixq(dst, src); | |
4062 emit_byte(0x89); | |
4063 emit_operand(src, dst); | |
4064 } | |
4065 | |
624 | 4066 void Assembler::movsbq(Register dst, Address src) { |
4067 InstructionMark im(this); | |
4068 prefixq(src, dst); | |
4069 emit_byte(0x0F); | |
4070 emit_byte(0xBE); | |
4071 emit_operand(dst, src); | |
4072 } | |
4073 | |
4074 void Assembler::movsbq(Register dst, Register src) { | |
4075 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
4076 emit_byte(0x0F); | |
4077 emit_byte(0xBE); | |
4078 emit_byte(0xC0 | encode); | |
4079 } | |
4080 | |
304 | 4081 void Assembler::movslq(Register dst, int32_t imm32) { |
4082 // dbx shows movslq(rcx, 3) as movq $0x0000000049000000,(%rbx) | |
4083 // and movslq(r8, 3); as movl $0x0000000048000000,(%rbx) | |
4084 // as a result we shouldn't use until tested at runtime... | |
4085 ShouldNotReachHere(); | |
4086 InstructionMark im(this); | |
4087 int encode = prefixq_and_encode(dst->encoding()); | |
4088 emit_byte(0xC7 | encode); | |
4089 emit_long(imm32); | |
4090 } | |
4091 | |
4092 void Assembler::movslq(Address dst, int32_t imm32) { | |
4093 assert(is_simm32(imm32), "lost bits"); | |
4094 InstructionMark im(this); | |
4095 prefixq(dst); | |
4096 emit_byte(0xC7); | |
4097 emit_operand(rax, dst, 4); | |
4098 emit_long(imm32); | |
4099 } | |
4100 | |
4101 void Assembler::movslq(Register dst, Address src) { | |
4102 InstructionMark im(this); | |
4103 prefixq(src, dst); | |
4104 emit_byte(0x63); | |
4105 emit_operand(dst, src); | |
4106 } | |
4107 | |
4108 void Assembler::movslq(Register dst, Register src) { | |
4109 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
4110 emit_byte(0x63); | |
4111 emit_byte(0xC0 | encode); | |
4112 } | |
4113 | |
624 | 4114 void Assembler::movswq(Register dst, Address src) { |
4115 InstructionMark im(this); | |
4116 prefixq(src, dst); | |
4117 emit_byte(0x0F); | |
4118 emit_byte(0xBF); | |
4119 emit_operand(dst, src); | |
4120 } | |
4121 | |
4122 void Assembler::movswq(Register dst, Register src) { | |
4123 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
4124 emit_byte(0x0F); | |
4125 emit_byte(0xBF); | |
4126 emit_byte(0xC0 | encode); | |
4127 } | |
4128 | |
4129 void Assembler::movzbq(Register dst, Address src) { | |
4130 InstructionMark im(this); | |
4131 prefixq(src, dst); | |
4132 emit_byte(0x0F); | |
4133 emit_byte(0xB6); | |
4134 emit_operand(dst, src); | |
4135 } | |
4136 | |
4137 void Assembler::movzbq(Register dst, Register src) { | |
4138 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
4139 emit_byte(0x0F); | |
4140 emit_byte(0xB6); | |
4141 emit_byte(0xC0 | encode); | |
4142 } | |
4143 | |
4144 void Assembler::movzwq(Register dst, Address src) { | |
4145 InstructionMark im(this); | |
4146 prefixq(src, dst); | |
4147 emit_byte(0x0F); | |
4148 emit_byte(0xB7); | |
4149 emit_operand(dst, src); | |
4150 } | |
4151 | |
4152 void Assembler::movzwq(Register dst, Register src) { | |
4153 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
4154 emit_byte(0x0F); | |
4155 emit_byte(0xB7); | |
4156 emit_byte(0xC0 | encode); | |
4157 } | |
4158 | |
304 | 4159 void Assembler::negq(Register dst) { |
4160 int encode = prefixq_and_encode(dst->encoding()); | |
4161 emit_byte(0xF7); | |
4162 emit_byte(0xD8 | encode); | |
4163 } | |
4164 | |
4165 void Assembler::notq(Register dst) { | |
4166 int encode = prefixq_and_encode(dst->encoding()); | |
4167 emit_byte(0xF7); | |
4168 emit_byte(0xD0 | encode); | |
4169 } | |
4170 | |
4171 void Assembler::orq(Address dst, int32_t imm32) { | |
4172 InstructionMark im(this); | |
4173 prefixq(dst); | |
4174 emit_byte(0x81); | |
4175 emit_operand(rcx, dst, 4); | |
4176 emit_long(imm32); | |
4177 } | |
4178 | |
4179 void Assembler::orq(Register dst, int32_t imm32) { | |
4180 (void) prefixq_and_encode(dst->encoding()); | |
4181 emit_arith(0x81, 0xC8, dst, imm32); | |
4182 } | |
4183 | |
4184 void Assembler::orq(Register dst, Address src) { | |
4185 InstructionMark im(this); | |
4186 prefixq(src, dst); | |
4187 emit_byte(0x0B); | |
4188 emit_operand(dst, src); | |
4189 } | |
4190 | |
4191 void Assembler::orq(Register dst, Register src) { | |
4192 (void) prefixq_and_encode(dst->encoding(), src->encoding()); | |
4193 emit_arith(0x0B, 0xC0, dst, src); | |
4194 } | |
4195 | |
4196 void Assembler::popa() { // 64bit | |
4197 movq(r15, Address(rsp, 0)); | |
4198 movq(r14, Address(rsp, wordSize)); | |
4199 movq(r13, Address(rsp, 2 * wordSize)); | |
4200 movq(r12, Address(rsp, 3 * wordSize)); | |
4201 movq(r11, Address(rsp, 4 * wordSize)); | |
4202 movq(r10, Address(rsp, 5 * wordSize)); | |
4203 movq(r9, Address(rsp, 6 * wordSize)); | |
4204 movq(r8, Address(rsp, 7 * wordSize)); | |
4205 movq(rdi, Address(rsp, 8 * wordSize)); | |
4206 movq(rsi, Address(rsp, 9 * wordSize)); | |
4207 movq(rbp, Address(rsp, 10 * wordSize)); | |
4208 // skip rsp | |
4209 movq(rbx, Address(rsp, 12 * wordSize)); | |
4210 movq(rdx, Address(rsp, 13 * wordSize)); | |
4211 movq(rcx, Address(rsp, 14 * wordSize)); | |
4212 movq(rax, Address(rsp, 15 * wordSize)); | |
4213 | |
4214 addq(rsp, 16 * wordSize); | |
4215 } | |
4216 | |
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4217 void Assembler::popcntq(Register dst, Address src) { |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4218 assert(VM_Version::supports_popcnt(), "must support"); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4219 InstructionMark im(this); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4220 emit_byte(0xF3); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4221 prefixq(src, dst); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4222 emit_byte(0x0F); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4223 emit_byte(0xB8); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4224 emit_operand(dst, src); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4225 } |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4226 |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4227 void Assembler::popcntq(Register dst, Register src) { |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4228 assert(VM_Version::supports_popcnt(), "must support"); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4229 emit_byte(0xF3); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4230 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4231 emit_byte(0x0F); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4232 emit_byte(0xB8); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4233 emit_byte(0xC0 | encode); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4234 } |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
4235 |
304 | 4236 void Assembler::popq(Address dst) { |
4237 InstructionMark im(this); | |
4238 prefixq(dst); | |
4239 emit_byte(0x8F); | |
4240 emit_operand(rax, dst); | |
4241 } | |
4242 | |
4243 void Assembler::pusha() { // 64bit | |
4244 // we have to store original rsp. ABI says that 128 bytes | |
4245 // below rsp are local scratch. | |
4246 movq(Address(rsp, -5 * wordSize), rsp); | |
4247 | |
4248 subq(rsp, 16 * wordSize); | |
4249 | |
4250 movq(Address(rsp, 15 * wordSize), rax); | |
4251 movq(Address(rsp, 14 * wordSize), rcx); | |
4252 movq(Address(rsp, 13 * wordSize), rdx); | |
4253 movq(Address(rsp, 12 * wordSize), rbx); | |
4254 // skip rsp | |
4255 movq(Address(rsp, 10 * wordSize), rbp); | |
4256 movq(Address(rsp, 9 * wordSize), rsi); | |
4257 movq(Address(rsp, 8 * wordSize), rdi); | |
4258 movq(Address(rsp, 7 * wordSize), r8); | |
4259 movq(Address(rsp, 6 * wordSize), r9); | |
4260 movq(Address(rsp, 5 * wordSize), r10); | |
4261 movq(Address(rsp, 4 * wordSize), r11); | |
4262 movq(Address(rsp, 3 * wordSize), r12); | |
4263 movq(Address(rsp, 2 * wordSize), r13); | |
4264 movq(Address(rsp, wordSize), r14); | |
4265 movq(Address(rsp, 0), r15); | |
4266 } | |
4267 | |
4268 void Assembler::pushq(Address src) { | |
4269 InstructionMark im(this); | |
4270 prefixq(src); | |
4271 emit_byte(0xFF); | |
4272 emit_operand(rsi, src); | |
4273 } | |
4274 | |
4275 void Assembler::rclq(Register dst, int imm8) { | |
4276 assert(isShiftCount(imm8 >> 1), "illegal shift count"); | |
4277 int encode = prefixq_and_encode(dst->encoding()); | |
4278 if (imm8 == 1) { | |
4279 emit_byte(0xD1); | |
4280 emit_byte(0xD0 | encode); | |
4281 } else { | |
4282 emit_byte(0xC1); | |
4283 emit_byte(0xD0 | encode); | |
4284 emit_byte(imm8); | |
4285 } | |
4286 } | |
4287 void Assembler::sarq(Register dst, int imm8) { | |
4288 assert(isShiftCount(imm8 >> 1), "illegal shift count"); | |
4289 int encode = prefixq_and_encode(dst->encoding()); | |
4290 if (imm8 == 1) { | |
4291 emit_byte(0xD1); | |
4292 emit_byte(0xF8 | encode); | |
4293 } else { | |
4294 emit_byte(0xC1); | |
4295 emit_byte(0xF8 | encode); | |
4296 emit_byte(imm8); | |
4297 } | |
4298 } | |
4299 | |
4300 void Assembler::sarq(Register dst) { | |
4301 int encode = prefixq_and_encode(dst->encoding()); | |
4302 emit_byte(0xD3); | |
4303 emit_byte(0xF8 | encode); | |
4304 } | |
4305 void Assembler::sbbq(Address dst, int32_t imm32) { | |
4306 InstructionMark im(this); | |
4307 prefixq(dst); | |
4308 emit_arith_operand(0x81, rbx, dst, imm32); | |
4309 } | |
4310 | |
4311 void Assembler::sbbq(Register dst, int32_t imm32) { | |
4312 (void) prefixq_and_encode(dst->encoding()); | |
4313 emit_arith(0x81, 0xD8, dst, imm32); | |
4314 } | |
4315 | |
4316 void Assembler::sbbq(Register dst, Address src) { | |
4317 InstructionMark im(this); | |
4318 prefixq(src, dst); | |
4319 emit_byte(0x1B); | |
4320 emit_operand(dst, src); | |
4321 } | |
4322 | |
4323 void Assembler::sbbq(Register dst, Register src) { | |
4324 (void) prefixq_and_encode(dst->encoding(), src->encoding()); | |
4325 emit_arith(0x1B, 0xC0, dst, src); | |
4326 } | |
4327 | |
4328 void Assembler::shlq(Register dst, int imm8) { | |
4329 assert(isShiftCount(imm8 >> 1), "illegal shift count"); | |
4330 int encode = prefixq_and_encode(dst->encoding()); | |
4331 if (imm8 == 1) { | |
4332 emit_byte(0xD1); | |
4333 emit_byte(0xE0 | encode); | |
4334 } else { | |
4335 emit_byte(0xC1); | |
4336 emit_byte(0xE0 | encode); | |
4337 emit_byte(imm8); | |
4338 } | |
4339 } | |
4340 | |
4341 void Assembler::shlq(Register dst) { | |
4342 int encode = prefixq_and_encode(dst->encoding()); | |
4343 emit_byte(0xD3); | |
4344 emit_byte(0xE0 | encode); | |
4345 } | |
4346 | |
4347 void Assembler::shrq(Register dst, int imm8) { | |
4348 assert(isShiftCount(imm8 >> 1), "illegal shift count"); | |
4349 int encode = prefixq_and_encode(dst->encoding()); | |
4350 emit_byte(0xC1); | |
4351 emit_byte(0xE8 | encode); | |
4352 emit_byte(imm8); | |
4353 } | |
4354 | |
4355 void Assembler::shrq(Register dst) { | |
4356 int encode = prefixq_and_encode(dst->encoding()); | |
4357 emit_byte(0xD3); | |
4358 emit_byte(0xE8 | encode); | |
4359 } | |
4360 | |
4361 void Assembler::sqrtsd(XMMRegister dst, Address src) { | |
4362 NOT_LP64(assert(VM_Version::supports_sse2(), "")); | |
0 | 4363 InstructionMark im(this); |
4364 emit_byte(0xF2); | |
304 | 4365 prefix(src, dst); |
0 | 4366 emit_byte(0x0F); |
304 | 4367 emit_byte(0x51); |
4368 emit_operand(dst, src); | |
4369 } | |
4370 | |
4371 void Assembler::subq(Address dst, int32_t imm32) { | |
4372 InstructionMark im(this); | |
4373 prefixq(dst); | |
4374 if (is8bit(imm32)) { | |
4375 emit_byte(0x83); | |
4376 emit_operand(rbp, dst, 1); | |
4377 emit_byte(imm32 & 0xFF); | |
4378 } else { | |
4379 emit_byte(0x81); | |
4380 emit_operand(rbp, dst, 4); | |
4381 emit_long(imm32); | |
4382 } | |
4383 } | |
4384 | |
4385 void Assembler::subq(Register dst, int32_t imm32) { | |
4386 (void) prefixq_and_encode(dst->encoding()); | |
4387 emit_arith(0x81, 0xE8, dst, imm32); | |
4388 } | |
4389 | |
4390 void Assembler::subq(Address dst, Register src) { | |
4391 InstructionMark im(this); | |
4392 prefixq(dst, src); | |
4393 emit_byte(0x29); | |
4394 emit_operand(src, dst); | |
4395 } | |
4396 | |
4397 void Assembler::subq(Register dst, Address src) { | |
4398 InstructionMark im(this); | |
4399 prefixq(src, dst); | |
4400 emit_byte(0x2B); | |
4401 emit_operand(dst, src); | |
4402 } | |
4403 | |
4404 void Assembler::subq(Register dst, Register src) { | |
4405 (void) prefixq_and_encode(dst->encoding(), src->encoding()); | |
4406 emit_arith(0x2B, 0xC0, dst, src); | |
4407 } | |
4408 | |
4409 void Assembler::testq(Register dst, int32_t imm32) { | |
4410 // not using emit_arith because test | |
4411 // doesn't support sign-extension of | |
4412 // 8bit operands | |
4413 int encode = dst->encoding(); | |
4414 if (encode == 0) { | |
4415 prefix(REX_W); | |
4416 emit_byte(0xA9); | |
4417 } else { | |
4418 encode = prefixq_and_encode(encode); | |
4419 emit_byte(0xF7); | |
4420 emit_byte(0xC0 | encode); | |
4421 } | |
4422 emit_long(imm32); | |
4423 } | |
4424 | |
4425 void Assembler::testq(Register dst, Register src) { | |
4426 (void) prefixq_and_encode(dst->encoding(), src->encoding()); | |
4427 emit_arith(0x85, 0xC0, dst, src); | |
4428 } | |
4429 | |
4430 void Assembler::xaddq(Address dst, Register src) { | |
4431 InstructionMark im(this); | |
4432 prefixq(dst, src); | |
71 | 4433 emit_byte(0x0F); |
304 | 4434 emit_byte(0xC1); |
4435 emit_operand(src, dst); | |
4436 } | |
4437 | |
4438 void Assembler::xchgq(Register dst, Address src) { | |
4439 InstructionMark im(this); | |
4440 prefixq(src, dst); | |
4441 emit_byte(0x87); | |
4442 emit_operand(dst, src); | |
4443 } | |
4444 | |
4445 void Assembler::xchgq(Register dst, Register src) { | |
4446 int encode = prefixq_and_encode(dst->encoding(), src->encoding()); | |
4447 emit_byte(0x87); | |
4448 emit_byte(0xc0 | encode); | |
4449 } | |
4450 | |
4451 void Assembler::xorq(Register dst, Register src) { | |
4452 (void) prefixq_and_encode(dst->encoding(), src->encoding()); | |
4453 emit_arith(0x33, 0xC0, dst, src); | |
4454 } | |
4455 | |
4456 void Assembler::xorq(Register dst, Address src) { | |
4457 InstructionMark im(this); | |
4458 prefixq(src, dst); | |
4459 emit_byte(0x33); | |
4460 emit_operand(dst, src); | |
4461 } | |
4462 | |
4463 #endif // !LP64 | |
4464 | |
4465 static Assembler::Condition reverse[] = { | |
4466 Assembler::noOverflow /* overflow = 0x0 */ , | |
4467 Assembler::overflow /* noOverflow = 0x1 */ , | |
4468 Assembler::aboveEqual /* carrySet = 0x2, below = 0x2 */ , | |
4469 Assembler::below /* aboveEqual = 0x3, carryClear = 0x3 */ , | |
4470 Assembler::notZero /* zero = 0x4, equal = 0x4 */ , | |
4471 Assembler::zero /* notZero = 0x5, notEqual = 0x5 */ , | |
4472 Assembler::above /* belowEqual = 0x6 */ , | |
4473 Assembler::belowEqual /* above = 0x7 */ , | |
4474 Assembler::positive /* negative = 0x8 */ , | |
4475 Assembler::negative /* positive = 0x9 */ , | |
4476 Assembler::noParity /* parity = 0xa */ , | |
4477 Assembler::parity /* noParity = 0xb */ , | |
4478 Assembler::greaterEqual /* less = 0xc */ , | |
4479 Assembler::less /* greaterEqual = 0xd */ , | |
4480 Assembler::greater /* lessEqual = 0xe */ , | |
4481 Assembler::lessEqual /* greater = 0xf, */ | |
4482 | |
4483 }; | |
4484 | |
0 | 4485 |
4486 // Implementation of MacroAssembler | |
4487 | |
304 | 4488 // First all the versions that have distinct versions depending on 32/64 bit |
4489 // Unless the difference is trivial (1 line or so). | |
4490 | |
4491 #ifndef _LP64 | |
4492 | |
4493 // 32bit versions | |
4494 | |
0 | 4495 Address MacroAssembler::as_Address(AddressLiteral adr) { |
4496 return Address(adr.target(), adr.rspec()); | |
4497 } | |
4498 | |
4499 Address MacroAssembler::as_Address(ArrayAddress adr) { | |
4500 return Address::make_array(adr); | |
4501 } | |
4502 | |
304 | 4503 int MacroAssembler::biased_locking_enter(Register lock_reg, |
4504 Register obj_reg, | |
4505 Register swap_reg, | |
4506 Register tmp_reg, | |
4507 bool swap_reg_contains_mark, | |
4508 Label& done, | |
4509 Label* slow_case, | |
4510 BiasedLockingCounters* counters) { | |
4511 assert(UseBiasedLocking, "why call this otherwise?"); | |
4512 assert(swap_reg == rax, "swap_reg must be rax, for cmpxchg"); | |
4513 assert_different_registers(lock_reg, obj_reg, swap_reg); | |
4514 | |
4515 if (PrintBiasedLockingStatistics && counters == NULL) | |
4516 counters = BiasedLocking::counters(); | |
4517 | |
4518 bool need_tmp_reg = false; | |
4519 if (tmp_reg == noreg) { | |
4520 need_tmp_reg = true; | |
4521 tmp_reg = lock_reg; | |
4522 } else { | |
4523 assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg); | |
4524 } | |
4525 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout"); | |
4526 Address mark_addr (obj_reg, oopDesc::mark_offset_in_bytes()); | |
4527 Address klass_addr (obj_reg, oopDesc::klass_offset_in_bytes()); | |
4528 Address saved_mark_addr(lock_reg, 0); | |
4529 | |
4530 // Biased locking | |
4531 // See whether the lock is currently biased toward our thread and | |
4532 // whether the epoch is still valid | |
4533 // Note that the runtime guarantees sufficient alignment of JavaThread | |
4534 // pointers to allow age to be placed into low bits | |
4535 // First check to see whether biasing is even enabled for this object | |
4536 Label cas_label; | |
4537 int null_check_offset = -1; | |
4538 if (!swap_reg_contains_mark) { | |
4539 null_check_offset = offset(); | |
4540 movl(swap_reg, mark_addr); | |
4541 } | |
4542 if (need_tmp_reg) { | |
4543 push(tmp_reg); | |
4544 } | |
4545 movl(tmp_reg, swap_reg); | |
4546 andl(tmp_reg, markOopDesc::biased_lock_mask_in_place); | |
4547 cmpl(tmp_reg, markOopDesc::biased_lock_pattern); | |
4548 if (need_tmp_reg) { | |
4549 pop(tmp_reg); | |
4550 } | |
4551 jcc(Assembler::notEqual, cas_label); | |
4552 // The bias pattern is present in the object's header. Need to check | |
4553 // whether the bias owner and the epoch are both still current. | |
4554 // Note that because there is no current thread register on x86 we | |
4555 // need to store off the mark word we read out of the object to | |
4556 // avoid reloading it and needing to recheck invariants below. This | |
4557 // store is unfortunate but it makes the overall code shorter and | |
4558 // simpler. | |
4559 movl(saved_mark_addr, swap_reg); | |
4560 if (need_tmp_reg) { | |
4561 push(tmp_reg); | |
4562 } | |
4563 get_thread(tmp_reg); | |
4564 xorl(swap_reg, tmp_reg); | |
4565 if (swap_reg_contains_mark) { | |
4566 null_check_offset = offset(); | |
4567 } | |
4568 movl(tmp_reg, klass_addr); | |
4569 xorl(swap_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); | |
4570 andl(swap_reg, ~((int) markOopDesc::age_mask_in_place)); | |
4571 if (need_tmp_reg) { | |
4572 pop(tmp_reg); | |
4573 } | |
4574 if (counters != NULL) { | |
4575 cond_inc32(Assembler::zero, | |
4576 ExternalAddress((address)counters->biased_lock_entry_count_addr())); | |
4577 } | |
4578 jcc(Assembler::equal, done); | |
4579 | |
4580 Label try_revoke_bias; | |
4581 Label try_rebias; | |
4582 | |
4583 // At this point we know that the header has the bias pattern and | |
4584 // that we are not the bias owner in the current epoch. We need to | |
4585 // figure out more details about the state of the header in order to | |
4586 // know what operations can be legally performed on the object's | |
4587 // header. | |
4588 | |
4589 // If the low three bits in the xor result aren't clear, that means | |
4590 // the prototype header is no longer biased and we have to revoke | |
4591 // the bias on this object. | |
4592 testl(swap_reg, markOopDesc::biased_lock_mask_in_place); | |
4593 jcc(Assembler::notZero, try_revoke_bias); | |
4594 | |
4595 // Biasing is still enabled for this data type. See whether the | |
4596 // epoch of the current bias is still valid, meaning that the epoch | |
4597 // bits of the mark word are equal to the epoch bits of the | |
4598 // prototype header. (Note that the prototype header's epoch bits | |
4599 // only change at a safepoint.) If not, attempt to rebias the object | |
4600 // toward the current thread. Note that we must be absolutely sure | |
4601 // that the current epoch is invalid in order to do this because | |
4602 // otherwise the manipulations it performs on the mark word are | |
4603 // illegal. | |
4604 testl(swap_reg, markOopDesc::epoch_mask_in_place); | |
4605 jcc(Assembler::notZero, try_rebias); | |
4606 | |
4607 // The epoch of the current bias is still valid but we know nothing | |
4608 // about the owner; it might be set or it might be clear. Try to | |
4609 // acquire the bias of the object using an atomic operation. If this | |
4610 // fails we will go in to the runtime to revoke the object's bias. | |
4611 // Note that we first construct the presumed unbiased header so we | |
4612 // don't accidentally blow away another thread's valid bias. | |
4613 movl(swap_reg, saved_mark_addr); | |
4614 andl(swap_reg, | |
4615 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place); | |
4616 if (need_tmp_reg) { | |
4617 push(tmp_reg); | |
4618 } | |
4619 get_thread(tmp_reg); | |
4620 orl(tmp_reg, swap_reg); | |
4621 if (os::is_MP()) { | |
4622 lock(); | |
4623 } | |
4624 cmpxchgptr(tmp_reg, Address(obj_reg, 0)); | |
4625 if (need_tmp_reg) { | |
4626 pop(tmp_reg); | |
4627 } | |
4628 // If the biasing toward our thread failed, this means that | |
4629 // another thread succeeded in biasing it toward itself and we | |
4630 // need to revoke that bias. The revocation will occur in the | |
4631 // interpreter runtime in the slow case. | |
4632 if (counters != NULL) { | |
4633 cond_inc32(Assembler::zero, | |
4634 ExternalAddress((address)counters->anonymously_biased_lock_entry_count_addr())); | |
4635 } | |
4636 if (slow_case != NULL) { | |
4637 jcc(Assembler::notZero, *slow_case); | |
4638 } | |
4639 jmp(done); | |
4640 | |
4641 bind(try_rebias); | |
4642 // At this point we know the epoch has expired, meaning that the | |
4643 // current "bias owner", if any, is actually invalid. Under these | |
4644 // circumstances _only_, we are allowed to use the current header's | |
4645 // value as the comparison value when doing the cas to acquire the | |
4646 // bias in the current epoch. In other words, we allow transfer of | |
4647 // the bias from one thread to another directly in this situation. | |
4648 // | |
4649 // FIXME: due to a lack of registers we currently blow away the age | |
4650 // bits in this situation. Should attempt to preserve them. | |
4651 if (need_tmp_reg) { | |
4652 push(tmp_reg); | |
4653 } | |
4654 get_thread(tmp_reg); | |
4655 movl(swap_reg, klass_addr); | |
4656 orl(tmp_reg, Address(swap_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); | |
4657 movl(swap_reg, saved_mark_addr); | |
4658 if (os::is_MP()) { | |
4659 lock(); | |
4660 } | |
4661 cmpxchgptr(tmp_reg, Address(obj_reg, 0)); | |
4662 if (need_tmp_reg) { | |
4663 pop(tmp_reg); | |
4664 } | |
4665 // If the biasing toward our thread failed, then another thread | |
4666 // succeeded in biasing it toward itself and we need to revoke that | |
4667 // bias. The revocation will occur in the runtime in the slow case. | |
4668 if (counters != NULL) { | |
4669 cond_inc32(Assembler::zero, | |
4670 ExternalAddress((address)counters->rebiased_lock_entry_count_addr())); | |
4671 } | |
4672 if (slow_case != NULL) { | |
4673 jcc(Assembler::notZero, *slow_case); | |
4674 } | |
4675 jmp(done); | |
4676 | |
4677 bind(try_revoke_bias); | |
4678 // The prototype mark in the klass doesn't have the bias bit set any | |
4679 // more, indicating that objects of this data type are not supposed | |
4680 // to be biased any more. We are going to try to reset the mark of | |
4681 // this object to the prototype value and fall through to the | |
4682 // CAS-based locking scheme. Note that if our CAS fails, it means | |
4683 // that another thread raced us for the privilege of revoking the | |
4684 // bias of this particular object, so it's okay to continue in the | |
4685 // normal locking code. | |
4686 // | |
4687 // FIXME: due to a lack of registers we currently blow away the age | |
4688 // bits in this situation. Should attempt to preserve them. | |
4689 movl(swap_reg, saved_mark_addr); | |
4690 if (need_tmp_reg) { | |
4691 push(tmp_reg); | |
4692 } | |
4693 movl(tmp_reg, klass_addr); | |
4694 movl(tmp_reg, Address(tmp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); | |
4695 if (os::is_MP()) { | |
4696 lock(); | |
4697 } | |
4698 cmpxchgptr(tmp_reg, Address(obj_reg, 0)); | |
4699 if (need_tmp_reg) { | |
4700 pop(tmp_reg); | |
4701 } | |
4702 // Fall through to the normal CAS-based lock, because no matter what | |
4703 // the result of the above CAS, some thread must have succeeded in | |
4704 // removing the bias bit from the object's header. | |
4705 if (counters != NULL) { | |
4706 cond_inc32(Assembler::zero, | |
4707 ExternalAddress((address)counters->revoked_lock_entry_count_addr())); | |
4708 } | |
4709 | |
4710 bind(cas_label); | |
4711 | |
4712 return null_check_offset; | |
4713 } | |
4714 void MacroAssembler::call_VM_leaf_base(address entry_point, | |
4715 int number_of_arguments) { | |
4716 call(RuntimeAddress(entry_point)); | |
4717 increment(rsp, number_of_arguments * wordSize); | |
4718 } | |
4719 | |
4720 void MacroAssembler::cmpoop(Address src1, jobject obj) { | |
4721 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate()); | |
4722 } | |
4723 | |
4724 void MacroAssembler::cmpoop(Register src1, jobject obj) { | |
4725 cmp_literal32(src1, (int32_t)obj, oop_Relocation::spec_for_immediate()); | |
4726 } | |
4727 | |
4728 void MacroAssembler::extend_sign(Register hi, Register lo) { | |
4729 // According to Intel Doc. AP-526, "Integer Divide", p.18. | |
4730 if (VM_Version::is_P6() && hi == rdx && lo == rax) { | |
4731 cdql(); | |
4732 } else { | |
4733 movl(hi, lo); | |
4734 sarl(hi, 31); | |
4735 } | |
4736 } | |
4737 | |
0 | 4738 void MacroAssembler::fat_nop() { |
4739 // A 5 byte nop that is safe for patching (see patch_verified_entry) | |
4740 emit_byte(0x26); // es: | |
4741 emit_byte(0x2e); // cs: | |
4742 emit_byte(0x64); // fs: | |
4743 emit_byte(0x65); // gs: | |
4744 emit_byte(0x90); | |
4745 } | |
4746 | |
304 | 4747 void MacroAssembler::jC2(Register tmp, Label& L) { |
4748 // set parity bit if FPU flag C2 is set (via rax) | |
4749 save_rax(tmp); | |
4750 fwait(); fnstsw_ax(); | |
4751 sahf(); | |
4752 restore_rax(tmp); | |
4753 // branch | |
4754 jcc(Assembler::parity, L); | |
4755 } | |
4756 | |
4757 void MacroAssembler::jnC2(Register tmp, Label& L) { | |
4758 // set parity bit if FPU flag C2 is set (via rax) | |
4759 save_rax(tmp); | |
4760 fwait(); fnstsw_ax(); | |
4761 sahf(); | |
4762 restore_rax(tmp); | |
4763 // branch | |
4764 jcc(Assembler::noParity, L); | |
4765 } | |
4766 | |
0 | 4767 // 32bit can do a case table jump in one instruction but we no longer allow the base |
4768 // to be installed in the Address class | |
4769 void MacroAssembler::jump(ArrayAddress entry) { | |
4770 jmp(as_Address(entry)); | |
4771 } | |
4772 | |
304 | 4773 // Note: y_lo will be destroyed |
4774 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) { | |
4775 // Long compare for Java (semantics as described in JVM spec.) | |
4776 Label high, low, done; | |
4777 | |
4778 cmpl(x_hi, y_hi); | |
4779 jcc(Assembler::less, low); | |
4780 jcc(Assembler::greater, high); | |
4781 // x_hi is the return register | |
4782 xorl(x_hi, x_hi); | |
4783 cmpl(x_lo, y_lo); | |
4784 jcc(Assembler::below, low); | |
4785 jcc(Assembler::equal, done); | |
4786 | |
4787 bind(high); | |
4788 xorl(x_hi, x_hi); | |
4789 increment(x_hi); | |
4790 jmp(done); | |
4791 | |
4792 bind(low); | |
4793 xorl(x_hi, x_hi); | |
4794 decrementl(x_hi); | |
4795 | |
4796 bind(done); | |
4797 } | |
4798 | |
4799 void MacroAssembler::lea(Register dst, AddressLiteral src) { | |
4800 mov_literal32(dst, (int32_t)src.target(), src.rspec()); | |
0 | 4801 } |
4802 | |
4803 void MacroAssembler::lea(Address dst, AddressLiteral adr) { | |
4804 // leal(dst, as_Address(adr)); | |
304 | 4805 // see note in movl as to why we must use a move |
0 | 4806 mov_literal32(dst, (int32_t) adr.target(), adr.rspec()); |
4807 } | |
4808 | |
4809 void MacroAssembler::leave() { | |
304 | 4810 mov(rsp, rbp); |
4811 pop(rbp); | |
4812 } | |
0 | 4813 |
4814 void MacroAssembler::lmul(int x_rsp_offset, int y_rsp_offset) { | |
4815 // Multiplication of two Java long values stored on the stack | |
4816 // as illustrated below. Result is in rdx:rax. | |
4817 // | |
4818 // rsp ---> [ ?? ] \ \ | |
4819 // .... | y_rsp_offset | | |
4820 // [ y_lo ] / (in bytes) | x_rsp_offset | |
4821 // [ y_hi ] | (in bytes) | |
4822 // .... | | |
4823 // [ x_lo ] / | |
4824 // [ x_hi ] | |
4825 // .... | |
4826 // | |
4827 // Basic idea: lo(result) = lo(x_lo * y_lo) | |
4828 // hi(result) = hi(x_lo * y_lo) + lo(x_hi * y_lo) + lo(x_lo * y_hi) | |
4829 Address x_hi(rsp, x_rsp_offset + wordSize); Address x_lo(rsp, x_rsp_offset); | |
4830 Address y_hi(rsp, y_rsp_offset + wordSize); Address y_lo(rsp, y_rsp_offset); | |
4831 Label quick; | |
4832 // load x_hi, y_hi and check if quick | |
4833 // multiplication is possible | |
4834 movl(rbx, x_hi); | |
4835 movl(rcx, y_hi); | |
4836 movl(rax, rbx); | |
4837 orl(rbx, rcx); // rbx, = 0 <=> x_hi = 0 and y_hi = 0 | |
4838 jcc(Assembler::zero, quick); // if rbx, = 0 do quick multiply | |
4839 // do full multiplication | |
4840 // 1st step | |
4841 mull(y_lo); // x_hi * y_lo | |
4842 movl(rbx, rax); // save lo(x_hi * y_lo) in rbx, | |
4843 // 2nd step | |
4844 movl(rax, x_lo); | |
4845 mull(rcx); // x_lo * y_hi | |
4846 addl(rbx, rax); // add lo(x_lo * y_hi) to rbx, | |
4847 // 3rd step | |
4848 bind(quick); // note: rbx, = 0 if quick multiply! | |
4849 movl(rax, x_lo); | |
4850 mull(y_lo); // x_lo * y_lo | |
4851 addl(rdx, rbx); // correct hi(x_lo * y_lo) | |
4852 } | |
4853 | |
304 | 4854 void MacroAssembler::lneg(Register hi, Register lo) { |
4855 negl(lo); | |
4856 adcl(hi, 0); | |
4857 negl(hi); | |
4858 } | |
0 | 4859 |
4860 void MacroAssembler::lshl(Register hi, Register lo) { | |
4861 // Java shift left long support (semantics as described in JVM spec., p.305) | |
4862 // (basic idea for shift counts s >= n: x << s == (x << n) << (s - n)) | |
4863 // shift value is in rcx ! | |
4864 assert(hi != rcx, "must not use rcx"); | |
4865 assert(lo != rcx, "must not use rcx"); | |
4866 const Register s = rcx; // shift count | |
4867 const int n = BitsPerWord; | |
4868 Label L; | |
4869 andl(s, 0x3f); // s := s & 0x3f (s < 0x40) | |
4870 cmpl(s, n); // if (s < n) | |
4871 jcc(Assembler::less, L); // else (s >= n) | |
4872 movl(hi, lo); // x := x << n | |
4873 xorl(lo, lo); | |
4874 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n! | |
4875 bind(L); // s (mod n) < n | |
4876 shldl(hi, lo); // x := x << s | |
4877 shll(lo); | |
4878 } | |
4879 | |
4880 | |
4881 void MacroAssembler::lshr(Register hi, Register lo, bool sign_extension) { | |
4882 // Java shift right long support (semantics as described in JVM spec., p.306 & p.310) | |
4883 // (basic idea for shift counts s >= n: x >> s == (x >> n) >> (s - n)) | |
4884 assert(hi != rcx, "must not use rcx"); | |
4885 assert(lo != rcx, "must not use rcx"); | |
4886 const Register s = rcx; // shift count | |
4887 const int n = BitsPerWord; | |
4888 Label L; | |
4889 andl(s, 0x3f); // s := s & 0x3f (s < 0x40) | |
4890 cmpl(s, n); // if (s < n) | |
4891 jcc(Assembler::less, L); // else (s >= n) | |
4892 movl(lo, hi); // x := x >> n | |
4893 if (sign_extension) sarl(hi, 31); | |
4894 else xorl(hi, hi); | |
4895 // Note: subl(s, n) is not needed since the Intel shift instructions work rcx mod n! | |
4896 bind(L); // s (mod n) < n | |
4897 shrdl(lo, hi); // x := x >> s | |
4898 if (sign_extension) sarl(hi); | |
4899 else shrl(hi); | |
4900 } | |
4901 | |
304 | 4902 void MacroAssembler::movoop(Register dst, jobject obj) { |
4903 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate()); | |
4904 } | |
4905 | |
4906 void MacroAssembler::movoop(Address dst, jobject obj) { | |
4907 mov_literal32(dst, (int32_t)obj, oop_Relocation::spec_for_immediate()); | |
4908 } | |
4909 | |
4910 void MacroAssembler::movptr(Register dst, AddressLiteral src) { | |
4911 if (src.is_lval()) { | |
4912 mov_literal32(dst, (intptr_t)src.target(), src.rspec()); | |
4913 } else { | |
4914 movl(dst, as_Address(src)); | |
4915 } | |
4916 } | |
4917 | |
4918 void MacroAssembler::movptr(ArrayAddress dst, Register src) { | |
4919 movl(as_Address(dst), src); | |
4920 } | |
4921 | |
4922 void MacroAssembler::movptr(Register dst, ArrayAddress src) { | |
4923 movl(dst, as_Address(src)); | |
4924 } | |
4925 | |
4926 // src should NEVER be a real pointer. Use AddressLiteral for true pointers | |
4927 void MacroAssembler::movptr(Address dst, intptr_t src) { | |
4928 movl(dst, src); | |
4929 } | |
4930 | |
4931 | |
4932 void MacroAssembler::movsd(XMMRegister dst, AddressLiteral src) { | |
4933 movsd(dst, as_Address(src)); | |
4934 } | |
4935 | |
4936 void MacroAssembler::pop_callee_saved_registers() { | |
4937 pop(rcx); | |
4938 pop(rdx); | |
4939 pop(rdi); | |
4940 pop(rsi); | |
4941 } | |
4942 | |
4943 void MacroAssembler::pop_fTOS() { | |
4944 fld_d(Address(rsp, 0)); | |
4945 addl(rsp, 2 * wordSize); | |
4946 } | |
4947 | |
4948 void MacroAssembler::push_callee_saved_registers() { | |
4949 push(rsi); | |
4950 push(rdi); | |
4951 push(rdx); | |
4952 push(rcx); | |
4953 } | |
4954 | |
4955 void MacroAssembler::push_fTOS() { | |
4956 subl(rsp, 2 * wordSize); | |
4957 fstp_d(Address(rsp, 0)); | |
4958 } | |
4959 | |
4960 | |
4961 void MacroAssembler::pushoop(jobject obj) { | |
4962 push_literal32((int32_t)obj, oop_Relocation::spec_for_immediate()); | |
4963 } | |
4964 | |
4965 | |
4966 void MacroAssembler::pushptr(AddressLiteral src) { | |
4967 if (src.is_lval()) { | |
4968 push_literal32((int32_t)src.target(), src.rspec()); | |
4969 } else { | |
4970 pushl(as_Address(src)); | |
4971 } | |
4972 } | |
4973 | |
4974 void MacroAssembler::set_word_if_not_zero(Register dst) { | |
4975 xorl(dst, dst); | |
4976 set_byte_if_not_zero(dst); | |
4977 } | |
4978 | |
4979 static void pass_arg0(MacroAssembler* masm, Register arg) { | |
4980 masm->push(arg); | |
4981 } | |
4982 | |
4983 static void pass_arg1(MacroAssembler* masm, Register arg) { | |
4984 masm->push(arg); | |
4985 } | |
4986 | |
4987 static void pass_arg2(MacroAssembler* masm, Register arg) { | |
4988 masm->push(arg); | |
4989 } | |
4990 | |
4991 static void pass_arg3(MacroAssembler* masm, Register arg) { | |
4992 masm->push(arg); | |
4993 } | |
4994 | |
4995 #ifndef PRODUCT | |
4996 extern "C" void findpc(intptr_t x); | |
4997 #endif | |
4998 | |
4999 void MacroAssembler::debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg) { | |
5000 // In order to get locks to work, we need to fake a in_VM state | |
5001 JavaThread* thread = JavaThread::current(); | |
5002 JavaThreadState saved_state = thread->thread_state(); | |
5003 thread->set_thread_state(_thread_in_vm); | |
5004 if (ShowMessageBoxOnError) { | |
5005 JavaThread* thread = JavaThread::current(); | |
5006 JavaThreadState saved_state = thread->thread_state(); | |
5007 thread->set_thread_state(_thread_in_vm); | |
5008 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) { | |
5009 ttyLocker ttyl; | |
5010 BytecodeCounter::print(); | |
5011 } | |
5012 // To see where a verify_oop failed, get $ebx+40/X for this frame. | |
5013 // This is the value of eip which points to where verify_oop will return. | |
5014 if (os::message_box(msg, "Execution stopped, print registers?")) { | |
5015 ttyLocker ttyl; | |
5016 tty->print_cr("eip = 0x%08x", eip); | |
5017 #ifndef PRODUCT | |
1793
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
5018 if ((WizardMode || Verbose) && PrintMiscellaneous) { |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
5019 tty->cr(); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
5020 findpc(eip); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
5021 tty->cr(); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
5022 } |
304 | 5023 #endif |
1793
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
5024 tty->print_cr("rax = 0x%08x", rax); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
5025 tty->print_cr("rbx = 0x%08x", rbx); |
304 | 5026 tty->print_cr("rcx = 0x%08x", rcx); |
5027 tty->print_cr("rdx = 0x%08x", rdx); | |
5028 tty->print_cr("rdi = 0x%08x", rdi); | |
5029 tty->print_cr("rsi = 0x%08x", rsi); | |
1793
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
5030 tty->print_cr("rbp = 0x%08x", rbp); |
304 | 5031 tty->print_cr("rsp = 0x%08x", rsp); |
5032 BREAKPOINT; | |
1793
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
5033 assert(false, "start up GDB"); |
304 | 5034 } |
5035 } else { | |
5036 ttyLocker ttyl; | |
5037 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg); | |
5038 assert(false, "DEBUG MESSAGE"); | |
5039 } | |
5040 ThreadStateTransition::transition(thread, _thread_in_vm, saved_state); | |
5041 } | |
5042 | |
5043 void MacroAssembler::stop(const char* msg) { | |
5044 ExternalAddress message((address)msg); | |
5045 // push address of message | |
5046 pushptr(message.addr()); | |
5047 { Label L; call(L, relocInfo::none); bind(L); } // push eip | |
5048 pusha(); // push registers | |
5049 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32))); | |
5050 hlt(); | |
5051 } | |
5052 | |
5053 void MacroAssembler::warn(const char* msg) { | |
5054 push_CPU_state(); | |
5055 | |
5056 ExternalAddress message((address) msg); | |
5057 // push address of message | |
5058 pushptr(message.addr()); | |
5059 | |
5060 call(RuntimeAddress(CAST_FROM_FN_PTR(address, warning))); | |
5061 addl(rsp, wordSize); // discard argument | |
5062 pop_CPU_state(); | |
5063 } | |
5064 | |
5065 #else // _LP64 | |
5066 | |
5067 // 64 bit versions | |
5068 | |
5069 Address MacroAssembler::as_Address(AddressLiteral adr) { | |
5070 // amd64 always does this as a pc-rel | |
5071 // we can be absolute or disp based on the instruction type | |
5072 // jmp/call are displacements others are absolute | |
5073 assert(!adr.is_lval(), "must be rval"); | |
5074 assert(reachable(adr), "must be"); | |
5075 return Address((int32_t)(intptr_t)(adr.target() - pc()), adr.target(), adr.reloc()); | |
5076 | |
5077 } | |
5078 | |
5079 Address MacroAssembler::as_Address(ArrayAddress adr) { | |
5080 AddressLiteral base = adr.base(); | |
5081 lea(rscratch1, base); | |
5082 Address index = adr.index(); | |
5083 assert(index._disp == 0, "must not have disp"); // maybe it can? | |
5084 Address array(rscratch1, index._index, index._scale, index._disp); | |
5085 return array; | |
5086 } | |
5087 | |
5088 int MacroAssembler::biased_locking_enter(Register lock_reg, | |
5089 Register obj_reg, | |
5090 Register swap_reg, | |
5091 Register tmp_reg, | |
5092 bool swap_reg_contains_mark, | |
5093 Label& done, | |
5094 Label* slow_case, | |
5095 BiasedLockingCounters* counters) { | |
5096 assert(UseBiasedLocking, "why call this otherwise?"); | |
5097 assert(swap_reg == rax, "swap_reg must be rax for cmpxchgq"); | |
5098 assert(tmp_reg != noreg, "tmp_reg must be supplied"); | |
5099 assert_different_registers(lock_reg, obj_reg, swap_reg, tmp_reg); | |
5100 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout"); | |
5101 Address mark_addr (obj_reg, oopDesc::mark_offset_in_bytes()); | |
5102 Address saved_mark_addr(lock_reg, 0); | |
5103 | |
5104 if (PrintBiasedLockingStatistics && counters == NULL) | |
5105 counters = BiasedLocking::counters(); | |
5106 | |
5107 // Biased locking | |
5108 // See whether the lock is currently biased toward our thread and | |
5109 // whether the epoch is still valid | |
5110 // Note that the runtime guarantees sufficient alignment of JavaThread | |
5111 // pointers to allow age to be placed into low bits | |
5112 // First check to see whether biasing is even enabled for this object | |
5113 Label cas_label; | |
5114 int null_check_offset = -1; | |
5115 if (!swap_reg_contains_mark) { | |
5116 null_check_offset = offset(); | |
5117 movq(swap_reg, mark_addr); | |
5118 } | |
5119 movq(tmp_reg, swap_reg); | |
5120 andq(tmp_reg, markOopDesc::biased_lock_mask_in_place); | |
5121 cmpq(tmp_reg, markOopDesc::biased_lock_pattern); | |
5122 jcc(Assembler::notEqual, cas_label); | |
5123 // The bias pattern is present in the object's header. Need to check | |
5124 // whether the bias owner and the epoch are both still current. | |
5125 load_prototype_header(tmp_reg, obj_reg); | |
5126 orq(tmp_reg, r15_thread); | |
5127 xorq(tmp_reg, swap_reg); | |
5128 andq(tmp_reg, ~((int) markOopDesc::age_mask_in_place)); | |
5129 if (counters != NULL) { | |
5130 cond_inc32(Assembler::zero, | |
5131 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr())); | |
5132 } | |
0 | 5133 jcc(Assembler::equal, done); |
5134 | |
304 | 5135 Label try_revoke_bias; |
5136 Label try_rebias; | |
5137 | |
5138 // At this point we know that the header has the bias pattern and | |
5139 // that we are not the bias owner in the current epoch. We need to | |
5140 // figure out more details about the state of the header in order to | |
5141 // know what operations can be legally performed on the object's | |
5142 // header. | |
5143 | |
5144 // If the low three bits in the xor result aren't clear, that means | |
5145 // the prototype header is no longer biased and we have to revoke | |
5146 // the bias on this object. | |
5147 testq(tmp_reg, markOopDesc::biased_lock_mask_in_place); | |
5148 jcc(Assembler::notZero, try_revoke_bias); | |
5149 | |
5150 // Biasing is still enabled for this data type. See whether the | |
5151 // epoch of the current bias is still valid, meaning that the epoch | |
5152 // bits of the mark word are equal to the epoch bits of the | |
5153 // prototype header. (Note that the prototype header's epoch bits | |
5154 // only change at a safepoint.) If not, attempt to rebias the object | |
5155 // toward the current thread. Note that we must be absolutely sure | |
5156 // that the current epoch is invalid in order to do this because | |
5157 // otherwise the manipulations it performs on the mark word are | |
5158 // illegal. | |
5159 testq(tmp_reg, markOopDesc::epoch_mask_in_place); | |
5160 jcc(Assembler::notZero, try_rebias); | |
5161 | |
5162 // The epoch of the current bias is still valid but we know nothing | |
5163 // about the owner; it might be set or it might be clear. Try to | |
5164 // acquire the bias of the object using an atomic operation. If this | |
5165 // fails we will go in to the runtime to revoke the object's bias. | |
5166 // Note that we first construct the presumed unbiased header so we | |
5167 // don't accidentally blow away another thread's valid bias. | |
5168 andq(swap_reg, | |
5169 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place); | |
5170 movq(tmp_reg, swap_reg); | |
5171 orq(tmp_reg, r15_thread); | |
5172 if (os::is_MP()) { | |
5173 lock(); | |
5174 } | |
5175 cmpxchgq(tmp_reg, Address(obj_reg, 0)); | |
5176 // If the biasing toward our thread failed, this means that | |
5177 // another thread succeeded in biasing it toward itself and we | |
5178 // need to revoke that bias. The revocation will occur in the | |
5179 // interpreter runtime in the slow case. | |
5180 if (counters != NULL) { | |
5181 cond_inc32(Assembler::zero, | |
5182 ExternalAddress((address) counters->anonymously_biased_lock_entry_count_addr())); | |
5183 } | |
5184 if (slow_case != NULL) { | |
5185 jcc(Assembler::notZero, *slow_case); | |
5186 } | |
0 | 5187 jmp(done); |
5188 | |
304 | 5189 bind(try_rebias); |
5190 // At this point we know the epoch has expired, meaning that the | |
5191 // current "bias owner", if any, is actually invalid. Under these | |
5192 // circumstances _only_, we are allowed to use the current header's | |
5193 // value as the comparison value when doing the cas to acquire the | |
5194 // bias in the current epoch. In other words, we allow transfer of | |
5195 // the bias from one thread to another directly in this situation. | |
5196 // | |
5197 // FIXME: due to a lack of registers we currently blow away the age | |
5198 // bits in this situation. Should attempt to preserve them. | |
5199 load_prototype_header(tmp_reg, obj_reg); | |
5200 orq(tmp_reg, r15_thread); | |
5201 if (os::is_MP()) { | |
5202 lock(); | |
5203 } | |
5204 cmpxchgq(tmp_reg, Address(obj_reg, 0)); | |
5205 // If the biasing toward our thread failed, then another thread | |
5206 // succeeded in biasing it toward itself and we need to revoke that | |
5207 // bias. The revocation will occur in the runtime in the slow case. | |
5208 if (counters != NULL) { | |
5209 cond_inc32(Assembler::zero, | |
5210 ExternalAddress((address) counters->rebiased_lock_entry_count_addr())); | |
5211 } | |
5212 if (slow_case != NULL) { | |
5213 jcc(Assembler::notZero, *slow_case); | |
0 | 5214 } |
5215 jmp(done); | |
5216 | |
304 | 5217 bind(try_revoke_bias); |
5218 // The prototype mark in the klass doesn't have the bias bit set any | |
5219 // more, indicating that objects of this data type are not supposed | |
5220 // to be biased any more. We are going to try to reset the mark of | |
5221 // this object to the prototype value and fall through to the | |
5222 // CAS-based locking scheme. Note that if our CAS fails, it means | |
5223 // that another thread raced us for the privilege of revoking the | |
5224 // bias of this particular object, so it's okay to continue in the | |
5225 // normal locking code. | |
5226 // | |
5227 // FIXME: due to a lack of registers we currently blow away the age | |
5228 // bits in this situation. Should attempt to preserve them. | |
5229 load_prototype_header(tmp_reg, obj_reg); | |
5230 if (os::is_MP()) { | |
5231 lock(); | |
5232 } | |
5233 cmpxchgq(tmp_reg, Address(obj_reg, 0)); | |
5234 // Fall through to the normal CAS-based lock, because no matter what | |
5235 // the result of the above CAS, some thread must have succeeded in | |
5236 // removing the bias bit from the object's header. | |
5237 if (counters != NULL) { | |
5238 cond_inc32(Assembler::zero, | |
5239 ExternalAddress((address) counters->revoked_lock_entry_count_addr())); | |
5240 } | |
5241 | |
5242 bind(cas_label); | |
5243 | |
5244 return null_check_offset; | |
5245 } | |
5246 | |
5247 void MacroAssembler::call_VM_leaf_base(address entry_point, int num_args) { | |
5248 Label L, E; | |
5249 | |
5250 #ifdef _WIN64 | |
5251 // Windows always allocates space for it's register args | |
5252 assert(num_args <= 4, "only register arguments supported"); | |
5253 subq(rsp, frame::arg_reg_save_area_bytes); | |
5254 #endif | |
5255 | |
5256 // Align stack if necessary | |
5257 testl(rsp, 15); | |
5258 jcc(Assembler::zero, L); | |
5259 | |
5260 subq(rsp, 8); | |
5261 { | |
5262 call(RuntimeAddress(entry_point)); | |
5263 } | |
5264 addq(rsp, 8); | |
5265 jmp(E); | |
5266 | |
5267 bind(L); | |
5268 { | |
5269 call(RuntimeAddress(entry_point)); | |
5270 } | |
5271 | |
5272 bind(E); | |
5273 | |
5274 #ifdef _WIN64 | |
5275 // restore stack pointer | |
5276 addq(rsp, frame::arg_reg_save_area_bytes); | |
5277 #endif | |
5278 | |
5279 } | |
5280 | |
5281 void MacroAssembler::cmp64(Register src1, AddressLiteral src2) { | |
5282 assert(!src2.is_lval(), "should use cmpptr"); | |
5283 | |
5284 if (reachable(src2)) { | |
5285 cmpq(src1, as_Address(src2)); | |
5286 } else { | |
5287 lea(rscratch1, src2); | |
5288 Assembler::cmpq(src1, Address(rscratch1, 0)); | |
5289 } | |
5290 } | |
5291 | |
5292 int MacroAssembler::corrected_idivq(Register reg) { | |
5293 // Full implementation of Java ldiv and lrem; checks for special | |
5294 // case as described in JVM spec., p.243 & p.271. The function | |
5295 // returns the (pc) offset of the idivl instruction - may be needed | |
5296 // for implicit exceptions. | |
5297 // | |
5298 // normal case special case | |
5299 // | |
5300 // input : rax: dividend min_long | |
5301 // reg: divisor (may not be eax/edx) -1 | |
5302 // | |
5303 // output: rax: quotient (= rax idiv reg) min_long | |
5304 // rdx: remainder (= rax irem reg) 0 | |
5305 assert(reg != rax && reg != rdx, "reg cannot be rax or rdx register"); | |
5306 static const int64_t min_long = 0x8000000000000000; | |
5307 Label normal_case, special_case; | |
5308 | |
5309 // check for special case | |
5310 cmp64(rax, ExternalAddress((address) &min_long)); | |
5311 jcc(Assembler::notEqual, normal_case); | |
5312 xorl(rdx, rdx); // prepare rdx for possible special case (where | |
5313 // remainder = 0) | |
5314 cmpq(reg, -1); | |
5315 jcc(Assembler::equal, special_case); | |
5316 | |
5317 // handle normal case | |
5318 bind(normal_case); | |
5319 cdqq(); | |
5320 int idivq_offset = offset(); | |
5321 idivq(reg); | |
5322 | |
5323 // normal and special case exit | |
5324 bind(special_case); | |
5325 | |
5326 return idivq_offset; | |
5327 } | |
5328 | |
5329 void MacroAssembler::decrementq(Register reg, int value) { | |
5330 if (value == min_jint) { subq(reg, value); return; } | |
5331 if (value < 0) { incrementq(reg, -value); return; } | |
5332 if (value == 0) { ; return; } | |
5333 if (value == 1 && UseIncDec) { decq(reg) ; return; } | |
5334 /* else */ { subq(reg, value) ; return; } | |
5335 } | |
5336 | |
5337 void MacroAssembler::decrementq(Address dst, int value) { | |
5338 if (value == min_jint) { subq(dst, value); return; } | |
5339 if (value < 0) { incrementq(dst, -value); return; } | |
5340 if (value == 0) { ; return; } | |
5341 if (value == 1 && UseIncDec) { decq(dst) ; return; } | |
5342 /* else */ { subq(dst, value) ; return; } | |
5343 } | |
5344 | |
5345 void MacroAssembler::fat_nop() { | |
5346 // A 5 byte nop that is safe for patching (see patch_verified_entry) | |
5347 // Recommened sequence from 'Software Optimization Guide for the AMD | |
5348 // Hammer Processor' | |
5349 emit_byte(0x66); | |
5350 emit_byte(0x66); | |
5351 emit_byte(0x90); | |
5352 emit_byte(0x66); | |
5353 emit_byte(0x90); | |
5354 } | |
5355 | |
5356 void MacroAssembler::incrementq(Register reg, int value) { | |
5357 if (value == min_jint) { addq(reg, value); return; } | |
5358 if (value < 0) { decrementq(reg, -value); return; } | |
5359 if (value == 0) { ; return; } | |
5360 if (value == 1 && UseIncDec) { incq(reg) ; return; } | |
5361 /* else */ { addq(reg, value) ; return; } | |
5362 } | |
5363 | |
5364 void MacroAssembler::incrementq(Address dst, int value) { | |
5365 if (value == min_jint) { addq(dst, value); return; } | |
5366 if (value < 0) { decrementq(dst, -value); return; } | |
5367 if (value == 0) { ; return; } | |
5368 if (value == 1 && UseIncDec) { incq(dst) ; return; } | |
5369 /* else */ { addq(dst, value) ; return; } | |
5370 } | |
5371 | |
5372 // 32bit can do a case table jump in one instruction but we no longer allow the base | |
5373 // to be installed in the Address class | |
5374 void MacroAssembler::jump(ArrayAddress entry) { | |
5375 lea(rscratch1, entry.base()); | |
5376 Address dispatch = entry.index(); | |
5377 assert(dispatch._base == noreg, "must be"); | |
5378 dispatch._base = rscratch1; | |
5379 jmp(dispatch); | |
5380 } | |
5381 | |
5382 void MacroAssembler::lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo) { | |
5383 ShouldNotReachHere(); // 64bit doesn't use two regs | |
5384 cmpq(x_lo, y_lo); | |
5385 } | |
5386 | |
5387 void MacroAssembler::lea(Register dst, AddressLiteral src) { | |
5388 mov_literal64(dst, (intptr_t)src.target(), src.rspec()); | |
5389 } | |
5390 | |
5391 void MacroAssembler::lea(Address dst, AddressLiteral adr) { | |
5392 mov_literal64(rscratch1, (intptr_t)adr.target(), adr.rspec()); | |
5393 movptr(dst, rscratch1); | |
5394 } | |
5395 | |
5396 void MacroAssembler::leave() { | |
5397 // %%% is this really better? Why not on 32bit too? | |
5398 emit_byte(0xC9); // LEAVE | |
5399 } | |
5400 | |
5401 void MacroAssembler::lneg(Register hi, Register lo) { | |
5402 ShouldNotReachHere(); // 64bit doesn't use two regs | |
5403 negq(lo); | |
5404 } | |
5405 | |
5406 void MacroAssembler::movoop(Register dst, jobject obj) { | |
5407 mov_literal64(dst, (intptr_t)obj, oop_Relocation::spec_for_immediate()); | |
5408 } | |
5409 | |
5410 void MacroAssembler::movoop(Address dst, jobject obj) { | |
5411 mov_literal64(rscratch1, (intptr_t)obj, oop_Relocation::spec_for_immediate()); | |
5412 movq(dst, rscratch1); | |
5413 } | |
5414 | |
5415 void MacroAssembler::movptr(Register dst, AddressLiteral src) { | |
5416 if (src.is_lval()) { | |
5417 mov_literal64(dst, (intptr_t)src.target(), src.rspec()); | |
5418 } else { | |
5419 if (reachable(src)) { | |
5420 movq(dst, as_Address(src)); | |
5421 } else { | |
5422 lea(rscratch1, src); | |
5423 movq(dst, Address(rscratch1,0)); | |
0 | 5424 } |
304 | 5425 } |
5426 } | |
5427 | |
5428 void MacroAssembler::movptr(ArrayAddress dst, Register src) { | |
5429 movq(as_Address(dst), src); | |
5430 } | |
5431 | |
5432 void MacroAssembler::movptr(Register dst, ArrayAddress src) { | |
5433 movq(dst, as_Address(src)); | |
5434 } | |
5435 | |
5436 // src should NEVER be a real pointer. Use AddressLiteral for true pointers | |
5437 void MacroAssembler::movptr(Address dst, intptr_t src) { | |
5438 mov64(rscratch1, src); | |
5439 movq(dst, rscratch1); | |
5440 } | |
5441 | |
5442 // These are mostly for initializing NULL | |
5443 void MacroAssembler::movptr(Address dst, int32_t src) { | |
5444 movslq(dst, src); | |
5445 } | |
5446 | |
5447 void MacroAssembler::movptr(Register dst, int32_t src) { | |
5448 mov64(dst, (intptr_t)src); | |
5449 } | |
5450 | |
5451 void MacroAssembler::pushoop(jobject obj) { | |
5452 movoop(rscratch1, obj); | |
5453 push(rscratch1); | |
5454 } | |
5455 | |
5456 void MacroAssembler::pushptr(AddressLiteral src) { | |
5457 lea(rscratch1, src); | |
5458 if (src.is_lval()) { | |
5459 push(rscratch1); | |
5460 } else { | |
5461 pushq(Address(rscratch1, 0)); | |
5462 } | |
5463 } | |
5464 | |
5465 void MacroAssembler::reset_last_Java_frame(bool clear_fp, | |
5466 bool clear_pc) { | |
5467 // we must set sp to zero to clear frame | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
420
diff
changeset
|
5468 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), NULL_WORD); |
304 | 5469 // must clear fp, so that compiled frames are not confused; it is |
5470 // possible that we need it only for debugging | |
5471 if (clear_fp) { | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
420
diff
changeset
|
5472 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), NULL_WORD); |
304 | 5473 } |
5474 | |
5475 if (clear_pc) { | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
420
diff
changeset
|
5476 movptr(Address(r15_thread, JavaThread::last_Java_pc_offset()), NULL_WORD); |
304 | 5477 } |
5478 } | |
5479 | |
5480 void MacroAssembler::set_last_Java_frame(Register last_java_sp, | |
5481 Register last_java_fp, | |
5482 address last_java_pc) { | |
5483 // determine last_java_sp register | |
5484 if (!last_java_sp->is_valid()) { | |
5485 last_java_sp = rsp; | |
5486 } | |
5487 | |
5488 // last_java_fp is optional | |
5489 if (last_java_fp->is_valid()) { | |
5490 movptr(Address(r15_thread, JavaThread::last_Java_fp_offset()), | |
5491 last_java_fp); | |
5492 } | |
5493 | |
5494 // last_java_pc is optional | |
5495 if (last_java_pc != NULL) { | |
5496 Address java_pc(r15_thread, | |
5497 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()); | |
5498 lea(rscratch1, InternalAddress(last_java_pc)); | |
5499 movptr(java_pc, rscratch1); | |
5500 } | |
5501 | |
5502 movptr(Address(r15_thread, JavaThread::last_Java_sp_offset()), last_java_sp); | |
5503 } | |
5504 | |
5505 static void pass_arg0(MacroAssembler* masm, Register arg) { | |
5506 if (c_rarg0 != arg ) { | |
5507 masm->mov(c_rarg0, arg); | |
5508 } | |
5509 } | |
5510 | |
5511 static void pass_arg1(MacroAssembler* masm, Register arg) { | |
5512 if (c_rarg1 != arg ) { | |
5513 masm->mov(c_rarg1, arg); | |
5514 } | |
5515 } | |
5516 | |
5517 static void pass_arg2(MacroAssembler* masm, Register arg) { | |
5518 if (c_rarg2 != arg ) { | |
5519 masm->mov(c_rarg2, arg); | |
5520 } | |
5521 } | |
5522 | |
5523 static void pass_arg3(MacroAssembler* masm, Register arg) { | |
5524 if (c_rarg3 != arg ) { | |
5525 masm->mov(c_rarg3, arg); | |
5526 } | |
5527 } | |
5528 | |
5529 void MacroAssembler::stop(const char* msg) { | |
5530 address rip = pc(); | |
5531 pusha(); // get regs on stack | |
5532 lea(c_rarg0, ExternalAddress((address) msg)); | |
5533 lea(c_rarg1, InternalAddress(rip)); | |
5534 movq(c_rarg2, rsp); // pass pointer to regs array | |
5535 andq(rsp, -16); // align stack as required by ABI | |
5536 call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug64))); | |
5537 hlt(); | |
5538 } | |
5539 | |
5540 void MacroAssembler::warn(const char* msg) { | |
5541 push(r12); | |
5542 movq(r12, rsp); | |
5543 andq(rsp, -16); // align stack as required by push_CPU_state and call | |
5544 | |
5545 push_CPU_state(); // keeps alignment at 16 bytes | |
5546 lea(c_rarg0, ExternalAddress((address) msg)); | |
5547 call_VM_leaf(CAST_FROM_FN_PTR(address, warning), c_rarg0); | |
5548 pop_CPU_state(); | |
5549 | |
5550 movq(rsp, r12); | |
5551 pop(r12); | |
5552 } | |
5553 | |
5554 #ifndef PRODUCT | |
5555 extern "C" void findpc(intptr_t x); | |
5556 #endif | |
5557 | |
5558 void MacroAssembler::debug64(char* msg, int64_t pc, int64_t regs[]) { | |
5559 // In order to get locks to work, we need to fake a in_VM state | |
5560 if (ShowMessageBoxOnError ) { | |
5561 JavaThread* thread = JavaThread::current(); | |
5562 JavaThreadState saved_state = thread->thread_state(); | |
5563 thread->set_thread_state(_thread_in_vm); | |
5564 #ifndef PRODUCT | |
5565 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) { | |
5566 ttyLocker ttyl; | |
5567 BytecodeCounter::print(); | |
0 | 5568 } |
304 | 5569 #endif |
5570 // To see where a verify_oop failed, get $ebx+40/X for this frame. | |
5571 // XXX correct this offset for amd64 | |
5572 // This is the value of eip which points to where verify_oop will return. | |
5573 if (os::message_box(msg, "Execution stopped, print registers?")) { | |
5574 ttyLocker ttyl; | |
5575 tty->print_cr("rip = 0x%016lx", pc); | |
5576 #ifndef PRODUCT | |
5577 tty->cr(); | |
5578 findpc(pc); | |
5579 tty->cr(); | |
5580 #endif | |
5581 tty->print_cr("rax = 0x%016lx", regs[15]); | |
5582 tty->print_cr("rbx = 0x%016lx", regs[12]); | |
5583 tty->print_cr("rcx = 0x%016lx", regs[14]); | |
5584 tty->print_cr("rdx = 0x%016lx", regs[13]); | |
5585 tty->print_cr("rdi = 0x%016lx", regs[8]); | |
5586 tty->print_cr("rsi = 0x%016lx", regs[9]); | |
5587 tty->print_cr("rbp = 0x%016lx", regs[10]); | |
5588 tty->print_cr("rsp = 0x%016lx", regs[11]); | |
5589 tty->print_cr("r8 = 0x%016lx", regs[7]); | |
5590 tty->print_cr("r9 = 0x%016lx", regs[6]); | |
5591 tty->print_cr("r10 = 0x%016lx", regs[5]); | |
5592 tty->print_cr("r11 = 0x%016lx", regs[4]); | |
5593 tty->print_cr("r12 = 0x%016lx", regs[3]); | |
5594 tty->print_cr("r13 = 0x%016lx", regs[2]); | |
5595 tty->print_cr("r14 = 0x%016lx", regs[1]); | |
5596 tty->print_cr("r15 = 0x%016lx", regs[0]); | |
5597 BREAKPOINT; | |
0 | 5598 } |
304 | 5599 ThreadStateTransition::transition(thread, _thread_in_vm, saved_state); |
5600 } else { | |
5601 ttyLocker ttyl; | |
5602 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", | |
5603 msg); | |
5604 } | |
5605 } | |
5606 | |
5607 #endif // _LP64 | |
5608 | |
5609 // Now versions that are common to 32/64 bit | |
5610 | |
5611 void MacroAssembler::addptr(Register dst, int32_t imm32) { | |
5612 LP64_ONLY(addq(dst, imm32)) NOT_LP64(addl(dst, imm32)); | |
5613 } | |
5614 | |
5615 void MacroAssembler::addptr(Register dst, Register src) { | |
5616 LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); | |
5617 } | |
5618 | |
5619 void MacroAssembler::addptr(Address dst, Register src) { | |
5620 LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); | |
5621 } | |
5622 | |
5623 void MacroAssembler::align(int modulus) { | |
5624 if (offset() % modulus != 0) { | |
5625 nop(modulus - (offset() % modulus)); | |
5626 } | |
5627 } | |
5628 | |
5629 void MacroAssembler::andpd(XMMRegister dst, AddressLiteral src) { | |
1060 | 5630 if (reachable(src)) { |
5631 andpd(dst, as_Address(src)); | |
5632 } else { | |
5633 lea(rscratch1, src); | |
5634 andpd(dst, Address(rscratch1, 0)); | |
5635 } | |
304 | 5636 } |
5637 | |
5638 void MacroAssembler::andptr(Register dst, int32_t imm32) { | |
5639 LP64_ONLY(andq(dst, imm32)) NOT_LP64(andl(dst, imm32)); | |
5640 } | |
5641 | |
5642 void MacroAssembler::atomic_incl(AddressLiteral counter_addr) { | |
5643 pushf(); | |
5644 if (os::is_MP()) | |
5645 lock(); | |
5646 incrementl(counter_addr); | |
5647 popf(); | |
5648 } | |
5649 | |
5650 // Writes to stack successive pages until offset reached to check for | |
5651 // stack overflow + shadow pages. This clobbers tmp. | |
5652 void MacroAssembler::bang_stack_size(Register size, Register tmp) { | |
5653 movptr(tmp, rsp); | |
5654 // Bang stack for total size given plus shadow page size. | |
5655 // Bang one page at a time because large size can bang beyond yellow and | |
5656 // red zones. | |
5657 Label loop; | |
5658 bind(loop); | |
5659 movl(Address(tmp, (-os::vm_page_size())), size ); | |
5660 subptr(tmp, os::vm_page_size()); | |
5661 subl(size, os::vm_page_size()); | |
5662 jcc(Assembler::greater, loop); | |
5663 | |
5664 // Bang down shadow pages too. | |
5665 // The -1 because we already subtracted 1 page. | |
5666 for (int i = 0; i< StackShadowPages-1; i++) { | |
5667 // this could be any sized move but this is can be a debugging crumb | |
5668 // so the bigger the better. | |
5669 movptr(Address(tmp, (-i*os::vm_page_size())), size ); | |
5670 } | |
5671 } | |
5672 | |
5673 void MacroAssembler::biased_locking_exit(Register obj_reg, Register temp_reg, Label& done) { | |
5674 assert(UseBiasedLocking, "why call this otherwise?"); | |
5675 | |
5676 // Check for biased locking unlock case, which is a no-op | |
5677 // Note: we do not have to check the thread ID for two reasons. | |
5678 // First, the interpreter checks for IllegalMonitorStateException at | |
5679 // a higher level. Second, if the bias was revoked while we held the | |
5680 // lock, the object could not be rebiased toward another thread, so | |
5681 // the bias bit would be clear. | |
5682 movptr(temp_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); | |
5683 andptr(temp_reg, markOopDesc::biased_lock_mask_in_place); | |
5684 cmpptr(temp_reg, markOopDesc::biased_lock_pattern); | |
5685 jcc(Assembler::equal, done); | |
5686 } | |
5687 | |
5688 void MacroAssembler::c2bool(Register x) { | |
5689 // implements x == 0 ? 0 : 1 | |
5690 // note: must only look at least-significant byte of x | |
5691 // since C-style booleans are stored in one byte | |
5692 // only! (was bug) | |
5693 andl(x, 0xFF); | |
5694 setb(Assembler::notZero, x); | |
5695 } | |
5696 | |
5697 // Wouldn't need if AddressLiteral version had new name | |
5698 void MacroAssembler::call(Label& L, relocInfo::relocType rtype) { | |
5699 Assembler::call(L, rtype); | |
5700 } | |
5701 | |
5702 void MacroAssembler::call(Register entry) { | |
5703 Assembler::call(entry); | |
5704 } | |
5705 | |
5706 void MacroAssembler::call(AddressLiteral entry) { | |
5707 if (reachable(entry)) { | |
5708 Assembler::call_literal(entry.target(), entry.rspec()); | |
5709 } else { | |
5710 lea(rscratch1, entry); | |
5711 Assembler::call(rscratch1); | |
5712 } | |
5713 } | |
5714 | |
5715 // Implementation of call_VM versions | |
5716 | |
5717 void MacroAssembler::call_VM(Register oop_result, | |
5718 address entry_point, | |
5719 bool check_exceptions) { | |
5720 Label C, E; | |
5721 call(C, relocInfo::none); | |
5722 jmp(E); | |
5723 | |
5724 bind(C); | |
5725 call_VM_helper(oop_result, entry_point, 0, check_exceptions); | |
5726 ret(0); | |
5727 | |
5728 bind(E); | |
5729 } | |
5730 | |
5731 void MacroAssembler::call_VM(Register oop_result, | |
5732 address entry_point, | |
5733 Register arg_1, | |
5734 bool check_exceptions) { | |
5735 Label C, E; | |
5736 call(C, relocInfo::none); | |
5737 jmp(E); | |
5738 | |
5739 bind(C); | |
5740 pass_arg1(this, arg_1); | |
5741 call_VM_helper(oop_result, entry_point, 1, check_exceptions); | |
5742 ret(0); | |
5743 | |
5744 bind(E); | |
5745 } | |
5746 | |
5747 void MacroAssembler::call_VM(Register oop_result, | |
5748 address entry_point, | |
5749 Register arg_1, | |
5750 Register arg_2, | |
5751 bool check_exceptions) { | |
5752 Label C, E; | |
5753 call(C, relocInfo::none); | |
5754 jmp(E); | |
5755 | |
5756 bind(C); | |
5757 | |
5758 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); | |
5759 | |
5760 pass_arg2(this, arg_2); | |
5761 pass_arg1(this, arg_1); | |
5762 call_VM_helper(oop_result, entry_point, 2, check_exceptions); | |
5763 ret(0); | |
5764 | |
5765 bind(E); | |
5766 } | |
5767 | |
5768 void MacroAssembler::call_VM(Register oop_result, | |
5769 address entry_point, | |
5770 Register arg_1, | |
5771 Register arg_2, | |
5772 Register arg_3, | |
5773 bool check_exceptions) { | |
5774 Label C, E; | |
5775 call(C, relocInfo::none); | |
5776 jmp(E); | |
5777 | |
5778 bind(C); | |
5779 | |
5780 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg")); | |
5781 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg")); | |
5782 pass_arg3(this, arg_3); | |
5783 | |
5784 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); | |
5785 pass_arg2(this, arg_2); | |
5786 | |
5787 pass_arg1(this, arg_1); | |
5788 call_VM_helper(oop_result, entry_point, 3, check_exceptions); | |
5789 ret(0); | |
5790 | |
5791 bind(E); | |
5792 } | |
5793 | |
5794 void MacroAssembler::call_VM(Register oop_result, | |
5795 Register last_java_sp, | |
5796 address entry_point, | |
5797 int number_of_arguments, | |
5798 bool check_exceptions) { | |
5799 Register thread = LP64_ONLY(r15_thread) NOT_LP64(noreg); | |
5800 call_VM_base(oop_result, thread, last_java_sp, entry_point, number_of_arguments, check_exceptions); | |
5801 } | |
5802 | |
5803 void MacroAssembler::call_VM(Register oop_result, | |
5804 Register last_java_sp, | |
5805 address entry_point, | |
5806 Register arg_1, | |
5807 bool check_exceptions) { | |
5808 pass_arg1(this, arg_1); | |
5809 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions); | |
5810 } | |
5811 | |
5812 void MacroAssembler::call_VM(Register oop_result, | |
5813 Register last_java_sp, | |
5814 address entry_point, | |
5815 Register arg_1, | |
5816 Register arg_2, | |
5817 bool check_exceptions) { | |
5818 | |
5819 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); | |
5820 pass_arg2(this, arg_2); | |
5821 pass_arg1(this, arg_1); | |
5822 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions); | |
5823 } | |
5824 | |
5825 void MacroAssembler::call_VM(Register oop_result, | |
5826 Register last_java_sp, | |
5827 address entry_point, | |
5828 Register arg_1, | |
5829 Register arg_2, | |
5830 Register arg_3, | |
5831 bool check_exceptions) { | |
5832 LP64_ONLY(assert(arg_1 != c_rarg3, "smashed arg")); | |
5833 LP64_ONLY(assert(arg_2 != c_rarg3, "smashed arg")); | |
5834 pass_arg3(this, arg_3); | |
5835 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); | |
5836 pass_arg2(this, arg_2); | |
5837 pass_arg1(this, arg_1); | |
5838 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions); | |
5839 } | |
5840 | |
5841 void MacroAssembler::call_VM_base(Register oop_result, | |
5842 Register java_thread, | |
5843 Register last_java_sp, | |
5844 address entry_point, | |
5845 int number_of_arguments, | |
5846 bool check_exceptions) { | |
5847 // determine java_thread register | |
5848 if (!java_thread->is_valid()) { | |
5849 #ifdef _LP64 | |
5850 java_thread = r15_thread; | |
5851 #else | |
5852 java_thread = rdi; | |
5853 get_thread(java_thread); | |
5854 #endif // LP64 | |
5855 } | |
5856 // determine last_java_sp register | |
5857 if (!last_java_sp->is_valid()) { | |
5858 last_java_sp = rsp; | |
5859 } | |
5860 // debugging support | |
5861 assert(number_of_arguments >= 0 , "cannot have negative number of arguments"); | |
5862 LP64_ONLY(assert(java_thread == r15_thread, "unexpected register")); | |
5863 assert(java_thread != oop_result , "cannot use the same register for java_thread & oop_result"); | |
5864 assert(java_thread != last_java_sp, "cannot use the same register for java_thread & last_java_sp"); | |
5865 | |
5866 // push java thread (becomes first argument of C function) | |
5867 | |
5868 NOT_LP64(push(java_thread); number_of_arguments++); | |
5869 LP64_ONLY(mov(c_rarg0, r15_thread)); | |
5870 | |
5871 // set last Java frame before call | |
5872 assert(last_java_sp != rbp, "can't use ebp/rbp"); | |
5873 | |
5874 // Only interpreter should have to set fp | |
5875 set_last_Java_frame(java_thread, last_java_sp, rbp, NULL); | |
5876 | |
5877 // do the call, remove parameters | |
5878 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); | |
5879 | |
5880 // restore the thread (cannot use the pushed argument since arguments | |
5881 // may be overwritten by C code generated by an optimizing compiler); | |
5882 // however can use the register value directly if it is callee saved. | |
5883 if (LP64_ONLY(true ||) java_thread == rdi || java_thread == rsi) { | |
5884 // rdi & rsi (also r15) are callee saved -> nothing to do | |
5885 #ifdef ASSERT | |
5886 guarantee(java_thread != rax, "change this code"); | |
5887 push(rax); | |
5888 { Label L; | |
5889 get_thread(rax); | |
5890 cmpptr(java_thread, rax); | |
5891 jcc(Assembler::equal, L); | |
5892 stop("MacroAssembler::call_VM_base: rdi not callee saved?"); | |
5893 bind(L); | |
0 | 5894 } |
304 | 5895 pop(rax); |
5896 #endif | |
5897 } else { | |
5898 get_thread(java_thread); | |
5899 } | |
5900 // reset last Java frame | |
5901 // Only interpreter should have to clear fp | |
5902 reset_last_Java_frame(java_thread, true, false); | |
5903 | |
5904 #ifndef CC_INTERP | |
5905 // C++ interp handles this in the interpreter | |
5906 check_and_handle_popframe(java_thread); | |
5907 check_and_handle_earlyret(java_thread); | |
5908 #endif /* CC_INTERP */ | |
5909 | |
5910 if (check_exceptions) { | |
5911 // check for pending exceptions (java_thread is set upon return) | |
5912 cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t) NULL_WORD); | |
5913 #ifndef _LP64 | |
5914 jump_cc(Assembler::notEqual, | |
5915 RuntimeAddress(StubRoutines::forward_exception_entry())); | |
5916 #else | |
5917 // This used to conditionally jump to forward_exception however it is | |
5918 // possible if we relocate that the branch will not reach. So we must jump | |
5919 // around so we can always reach | |
5920 | |
5921 Label ok; | |
5922 jcc(Assembler::equal, ok); | |
5923 jump(RuntimeAddress(StubRoutines::forward_exception_entry())); | |
5924 bind(ok); | |
5925 #endif // LP64 | |
5926 } | |
5927 | |
5928 // get oop result if there is one and reset the value in the thread | |
5929 if (oop_result->is_valid()) { | |
5930 movptr(oop_result, Address(java_thread, JavaThread::vm_result_offset())); | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
420
diff
changeset
|
5931 movptr(Address(java_thread, JavaThread::vm_result_offset()), NULL_WORD); |
304 | 5932 verify_oop(oop_result, "broken oop in call_VM_base"); |
5933 } | |
5934 } | |
5935 | |
5936 void MacroAssembler::call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) { | |
5937 | |
5938 // Calculate the value for last_Java_sp | |
5939 // somewhat subtle. call_VM does an intermediate call | |
5940 // which places a return address on the stack just under the | |
5941 // stack pointer as the user finsihed with it. This allows | |
5942 // use to retrieve last_Java_pc from last_Java_sp[-1]. | |
5943 // On 32bit we then have to push additional args on the stack to accomplish | |
5944 // the actual requested call. On 64bit call_VM only can use register args | |
5945 // so the only extra space is the return address that call_VM created. | |
5946 // This hopefully explains the calculations here. | |
5947 | |
5948 #ifdef _LP64 | |
5949 // We've pushed one address, correct last_Java_sp | |
5950 lea(rax, Address(rsp, wordSize)); | |
5951 #else | |
5952 lea(rax, Address(rsp, (1 + number_of_arguments) * wordSize)); | |
5953 #endif // LP64 | |
5954 | |
5955 call_VM_base(oop_result, noreg, rax, entry_point, number_of_arguments, check_exceptions); | |
5956 | |
5957 } | |
5958 | |
5959 void MacroAssembler::call_VM_leaf(address entry_point, int number_of_arguments) { | |
5960 call_VM_leaf_base(entry_point, number_of_arguments); | |
5961 } | |
5962 | |
5963 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0) { | |
5964 pass_arg0(this, arg_0); | |
5965 call_VM_leaf(entry_point, 1); | |
5966 } | |
5967 | |
5968 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1) { | |
5969 | |
5970 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg")); | |
5971 pass_arg1(this, arg_1); | |
5972 pass_arg0(this, arg_0); | |
5973 call_VM_leaf(entry_point, 2); | |
5974 } | |
5975 | |
5976 void MacroAssembler::call_VM_leaf(address entry_point, Register arg_0, Register arg_1, Register arg_2) { | |
5977 LP64_ONLY(assert(arg_0 != c_rarg2, "smashed arg")); | |
5978 LP64_ONLY(assert(arg_1 != c_rarg2, "smashed arg")); | |
5979 pass_arg2(this, arg_2); | |
5980 LP64_ONLY(assert(arg_0 != c_rarg1, "smashed arg")); | |
5981 pass_arg1(this, arg_1); | |
5982 pass_arg0(this, arg_0); | |
5983 call_VM_leaf(entry_point, 3); | |
5984 } | |
5985 | |
5986 void MacroAssembler::check_and_handle_earlyret(Register java_thread) { | |
5987 } | |
5988 | |
5989 void MacroAssembler::check_and_handle_popframe(Register java_thread) { | |
5990 } | |
5991 | |
5992 void MacroAssembler::cmp32(AddressLiteral src1, int32_t imm) { | |
5993 if (reachable(src1)) { | |
5994 cmpl(as_Address(src1), imm); | |
5995 } else { | |
5996 lea(rscratch1, src1); | |
5997 cmpl(Address(rscratch1, 0), imm); | |
5998 } | |
5999 } | |
6000 | |
6001 void MacroAssembler::cmp32(Register src1, AddressLiteral src2) { | |
6002 assert(!src2.is_lval(), "use cmpptr"); | |
6003 if (reachable(src2)) { | |
6004 cmpl(src1, as_Address(src2)); | |
6005 } else { | |
6006 lea(rscratch1, src2); | |
6007 cmpl(src1, Address(rscratch1, 0)); | |
6008 } | |
6009 } | |
6010 | |
6011 void MacroAssembler::cmp32(Register src1, int32_t imm) { | |
6012 Assembler::cmpl(src1, imm); | |
6013 } | |
6014 | |
6015 void MacroAssembler::cmp32(Register src1, Address src2) { | |
6016 Assembler::cmpl(src1, src2); | |
6017 } | |
6018 | |
6019 void MacroAssembler::cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) { | |
6020 ucomisd(opr1, opr2); | |
6021 | |
6022 Label L; | |
6023 if (unordered_is_less) { | |
6024 movl(dst, -1); | |
6025 jcc(Assembler::parity, L); | |
6026 jcc(Assembler::below , L); | |
6027 movl(dst, 0); | |
6028 jcc(Assembler::equal , L); | |
6029 increment(dst); | |
6030 } else { // unordered is greater | |
6031 movl(dst, 1); | |
6032 jcc(Assembler::parity, L); | |
6033 jcc(Assembler::above , L); | |
6034 movl(dst, 0); | |
6035 jcc(Assembler::equal , L); | |
6036 decrementl(dst); | |
6037 } | |
6038 bind(L); | |
6039 } | |
6040 | |
6041 void MacroAssembler::cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less) { | |
6042 ucomiss(opr1, opr2); | |
6043 | |
6044 Label L; | |
6045 if (unordered_is_less) { | |
6046 movl(dst, -1); | |
6047 jcc(Assembler::parity, L); | |
6048 jcc(Assembler::below , L); | |
6049 movl(dst, 0); | |
6050 jcc(Assembler::equal , L); | |
6051 increment(dst); | |
6052 } else { // unordered is greater | |
6053 movl(dst, 1); | |
6054 jcc(Assembler::parity, L); | |
6055 jcc(Assembler::above , L); | |
6056 movl(dst, 0); | |
6057 jcc(Assembler::equal , L); | |
6058 decrementl(dst); | |
6059 } | |
6060 bind(L); | |
6061 } | |
6062 | |
6063 | |
6064 void MacroAssembler::cmp8(AddressLiteral src1, int imm) { | |
6065 if (reachable(src1)) { | |
6066 cmpb(as_Address(src1), imm); | |
6067 } else { | |
6068 lea(rscratch1, src1); | |
6069 cmpb(Address(rscratch1, 0), imm); | |
6070 } | |
6071 } | |
6072 | |
6073 void MacroAssembler::cmpptr(Register src1, AddressLiteral src2) { | |
6074 #ifdef _LP64 | |
6075 if (src2.is_lval()) { | |
6076 movptr(rscratch1, src2); | |
6077 Assembler::cmpq(src1, rscratch1); | |
6078 } else if (reachable(src2)) { | |
6079 cmpq(src1, as_Address(src2)); | |
6080 } else { | |
6081 lea(rscratch1, src2); | |
6082 Assembler::cmpq(src1, Address(rscratch1, 0)); | |
6083 } | |
6084 #else | |
6085 if (src2.is_lval()) { | |
6086 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec()); | |
6087 } else { | |
6088 cmpl(src1, as_Address(src2)); | |
6089 } | |
6090 #endif // _LP64 | |
6091 } | |
6092 | |
6093 void MacroAssembler::cmpptr(Address src1, AddressLiteral src2) { | |
6094 assert(src2.is_lval(), "not a mem-mem compare"); | |
6095 #ifdef _LP64 | |
6096 // moves src2's literal address | |
6097 movptr(rscratch1, src2); | |
6098 Assembler::cmpq(src1, rscratch1); | |
6099 #else | |
6100 cmp_literal32(src1, (int32_t) src2.target(), src2.rspec()); | |
6101 #endif // _LP64 | |
6102 } | |
6103 | |
6104 void MacroAssembler::locked_cmpxchgptr(Register reg, AddressLiteral adr) { | |
6105 if (reachable(adr)) { | |
6106 if (os::is_MP()) | |
6107 lock(); | |
6108 cmpxchgptr(reg, as_Address(adr)); | |
6109 } else { | |
6110 lea(rscratch1, adr); | |
6111 if (os::is_MP()) | |
6112 lock(); | |
6113 cmpxchgptr(reg, Address(rscratch1, 0)); | |
6114 } | |
6115 } | |
6116 | |
6117 void MacroAssembler::cmpxchgptr(Register reg, Address adr) { | |
6118 LP64_ONLY(cmpxchgq(reg, adr)) NOT_LP64(cmpxchgl(reg, adr)); | |
6119 } | |
6120 | |
6121 void MacroAssembler::comisd(XMMRegister dst, AddressLiteral src) { | |
1060 | 6122 if (reachable(src)) { |
6123 comisd(dst, as_Address(src)); | |
6124 } else { | |
6125 lea(rscratch1, src); | |
6126 comisd(dst, Address(rscratch1, 0)); | |
6127 } | |
304 | 6128 } |
6129 | |
6130 void MacroAssembler::comiss(XMMRegister dst, AddressLiteral src) { | |
1060 | 6131 if (reachable(src)) { |
6132 comiss(dst, as_Address(src)); | |
6133 } else { | |
6134 lea(rscratch1, src); | |
6135 comiss(dst, Address(rscratch1, 0)); | |
6136 } | |
304 | 6137 } |
6138 | |
6139 | |
6140 void MacroAssembler::cond_inc32(Condition cond, AddressLiteral counter_addr) { | |
6141 Condition negated_cond = negate_condition(cond); | |
6142 Label L; | |
6143 jcc(negated_cond, L); | |
6144 atomic_incl(counter_addr); | |
6145 bind(L); | |
6146 } | |
6147 | |
6148 int MacroAssembler::corrected_idivl(Register reg) { | |
6149 // Full implementation of Java idiv and irem; checks for | |
6150 // special case as described in JVM spec., p.243 & p.271. | |
6151 // The function returns the (pc) offset of the idivl | |
6152 // instruction - may be needed for implicit exceptions. | |
6153 // | |
6154 // normal case special case | |
6155 // | |
6156 // input : rax,: dividend min_int | |
6157 // reg: divisor (may not be rax,/rdx) -1 | |
6158 // | |
6159 // output: rax,: quotient (= rax, idiv reg) min_int | |
6160 // rdx: remainder (= rax, irem reg) 0 | |
6161 assert(reg != rax && reg != rdx, "reg cannot be rax, or rdx register"); | |
6162 const int min_int = 0x80000000; | |
6163 Label normal_case, special_case; | |
6164 | |
6165 // check for special case | |
6166 cmpl(rax, min_int); | |
6167 jcc(Assembler::notEqual, normal_case); | |
6168 xorl(rdx, rdx); // prepare rdx for possible special case (where remainder = 0) | |
6169 cmpl(reg, -1); | |
6170 jcc(Assembler::equal, special_case); | |
6171 | |
6172 // handle normal case | |
6173 bind(normal_case); | |
6174 cdql(); | |
6175 int idivl_offset = offset(); | |
6176 idivl(reg); | |
6177 | |
6178 // normal and special case exit | |
6179 bind(special_case); | |
6180 | |
6181 return idivl_offset; | |
6182 } | |
6183 | |
6184 | |
6185 | |
6186 void MacroAssembler::decrementl(Register reg, int value) { | |
6187 if (value == min_jint) {subl(reg, value) ; return; } | |
6188 if (value < 0) { incrementl(reg, -value); return; } | |
6189 if (value == 0) { ; return; } | |
6190 if (value == 1 && UseIncDec) { decl(reg) ; return; } | |
6191 /* else */ { subl(reg, value) ; return; } | |
6192 } | |
6193 | |
6194 void MacroAssembler::decrementl(Address dst, int value) { | |
6195 if (value == min_jint) {subl(dst, value) ; return; } | |
6196 if (value < 0) { incrementl(dst, -value); return; } | |
6197 if (value == 0) { ; return; } | |
6198 if (value == 1 && UseIncDec) { decl(dst) ; return; } | |
6199 /* else */ { subl(dst, value) ; return; } | |
6200 } | |
6201 | |
6202 void MacroAssembler::division_with_shift (Register reg, int shift_value) { | |
6203 assert (shift_value > 0, "illegal shift value"); | |
6204 Label _is_positive; | |
6205 testl (reg, reg); | |
6206 jcc (Assembler::positive, _is_positive); | |
6207 int offset = (1 << shift_value) - 1 ; | |
6208 | |
6209 if (offset == 1) { | |
6210 incrementl(reg); | |
6211 } else { | |
6212 addl(reg, offset); | |
6213 } | |
6214 | |
6215 bind (_is_positive); | |
6216 sarl(reg, shift_value); | |
6217 } | |
6218 | |
6219 // !defined(COMPILER2) is because of stupid core builds | |
6220 #if !defined(_LP64) || defined(COMPILER1) || !defined(COMPILER2) | |
6221 void MacroAssembler::empty_FPU_stack() { | |
6222 if (VM_Version::supports_mmx()) { | |
6223 emms(); | |
6224 } else { | |
6225 for (int i = 8; i-- > 0; ) ffree(i); | |
6226 } | |
6227 } | |
6228 #endif // !LP64 || C1 || !C2 | |
6229 | |
6230 | |
6231 // Defines obj, preserves var_size_in_bytes | |
6232 void MacroAssembler::eden_allocate(Register obj, | |
6233 Register var_size_in_bytes, | |
6234 int con_size_in_bytes, | |
6235 Register t1, | |
6236 Label& slow_case) { | |
6237 assert(obj == rax, "obj must be in rax, for cmpxchg"); | |
6238 assert_different_registers(obj, var_size_in_bytes, t1); | |
362 | 6239 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) { |
6240 jmp(slow_case); | |
304 | 6241 } else { |
362 | 6242 Register end = t1; |
6243 Label retry; | |
6244 bind(retry); | |
6245 ExternalAddress heap_top((address) Universe::heap()->top_addr()); | |
6246 movptr(obj, heap_top); | |
6247 if (var_size_in_bytes == noreg) { | |
6248 lea(end, Address(obj, con_size_in_bytes)); | |
6249 } else { | |
6250 lea(end, Address(obj, var_size_in_bytes, Address::times_1)); | |
6251 } | |
6252 // if end < obj then we wrapped around => object too long => slow case | |
6253 cmpptr(end, obj); | |
6254 jcc(Assembler::below, slow_case); | |
6255 cmpptr(end, ExternalAddress((address) Universe::heap()->end_addr())); | |
6256 jcc(Assembler::above, slow_case); | |
6257 // Compare obj with the top addr, and if still equal, store the new top addr in | |
6258 // end at the address of the top addr pointer. Sets ZF if was equal, and clears | |
6259 // it otherwise. Use lock prefix for atomicity on MPs. | |
6260 locked_cmpxchgptr(end, heap_top); | |
6261 jcc(Assembler::notEqual, retry); | |
6262 } | |
304 | 6263 } |
6264 | |
6265 void MacroAssembler::enter() { | |
6266 push(rbp); | |
6267 mov(rbp, rsp); | |
6268 } | |
0 | 6269 |
6270 void MacroAssembler::fcmp(Register tmp) { | |
6271 fcmp(tmp, 1, true, true); | |
6272 } | |
6273 | |
6274 void MacroAssembler::fcmp(Register tmp, int index, bool pop_left, bool pop_right) { | |
6275 assert(!pop_right || pop_left, "usage error"); | |
6276 if (VM_Version::supports_cmov()) { | |
6277 assert(tmp == noreg, "unneeded temp"); | |
6278 if (pop_left) { | |
6279 fucomip(index); | |
6280 } else { | |
6281 fucomi(index); | |
6282 } | |
6283 if (pop_right) { | |
6284 fpop(); | |
6285 } | |
6286 } else { | |
6287 assert(tmp != noreg, "need temp"); | |
6288 if (pop_left) { | |
6289 if (pop_right) { | |
6290 fcompp(); | |
6291 } else { | |
6292 fcomp(index); | |
6293 } | |
6294 } else { | |
6295 fcom(index); | |
6296 } | |
6297 // convert FPU condition into eflags condition via rax, | |
6298 save_rax(tmp); | |
6299 fwait(); fnstsw_ax(); | |
6300 sahf(); | |
6301 restore_rax(tmp); | |
6302 } | |
6303 // condition codes set as follows: | |
6304 // | |
6305 // CF (corresponds to C0) if x < y | |
6306 // PF (corresponds to C2) if unordered | |
6307 // ZF (corresponds to C3) if x = y | |
6308 } | |
6309 | |
6310 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less) { | |
6311 fcmp2int(dst, unordered_is_less, 1, true, true); | |
6312 } | |
6313 | |
6314 void MacroAssembler::fcmp2int(Register dst, bool unordered_is_less, int index, bool pop_left, bool pop_right) { | |
6315 fcmp(VM_Version::supports_cmov() ? noreg : dst, index, pop_left, pop_right); | |
6316 Label L; | |
6317 if (unordered_is_less) { | |
6318 movl(dst, -1); | |
6319 jcc(Assembler::parity, L); | |
6320 jcc(Assembler::below , L); | |
6321 movl(dst, 0); | |
6322 jcc(Assembler::equal , L); | |
6323 increment(dst); | |
6324 } else { // unordered is greater | |
6325 movl(dst, 1); | |
6326 jcc(Assembler::parity, L); | |
6327 jcc(Assembler::above , L); | |
6328 movl(dst, 0); | |
6329 jcc(Assembler::equal , L); | |
304 | 6330 decrementl(dst); |
0 | 6331 } |
6332 bind(L); | |
6333 } | |
6334 | |
304 | 6335 void MacroAssembler::fld_d(AddressLiteral src) { |
6336 fld_d(as_Address(src)); | |
6337 } | |
6338 | |
6339 void MacroAssembler::fld_s(AddressLiteral src) { | |
6340 fld_s(as_Address(src)); | |
6341 } | |
6342 | |
6343 void MacroAssembler::fld_x(AddressLiteral src) { | |
6344 Assembler::fld_x(as_Address(src)); | |
6345 } | |
6346 | |
6347 void MacroAssembler::fldcw(AddressLiteral src) { | |
6348 Assembler::fldcw(as_Address(src)); | |
6349 } | |
0 | 6350 |
6351 void MacroAssembler::fpop() { | |
6352 ffree(); | |
6353 fincstp(); | |
6354 } | |
6355 | |
304 | 6356 void MacroAssembler::fremr(Register tmp) { |
6357 save_rax(tmp); | |
6358 { Label L; | |
6359 bind(L); | |
6360 fprem(); | |
6361 fwait(); fnstsw_ax(); | |
6362 #ifdef _LP64 | |
6363 testl(rax, 0x400); | |
6364 jcc(Assembler::notEqual, L); | |
6365 #else | |
6366 sahf(); | |
6367 jcc(Assembler::parity, L); | |
6368 #endif // _LP64 | |
6369 } | |
6370 restore_rax(tmp); | |
6371 // Result is in ST0. | |
6372 // Note: fxch & fpop to get rid of ST1 | |
6373 // (otherwise FPU stack could overflow eventually) | |
6374 fxch(1); | |
6375 fpop(); | |
6376 } | |
6377 | |
6378 | |
6379 void MacroAssembler::incrementl(AddressLiteral dst) { | |
6380 if (reachable(dst)) { | |
6381 incrementl(as_Address(dst)); | |
0 | 6382 } else { |
304 | 6383 lea(rscratch1, dst); |
6384 incrementl(Address(rscratch1, 0)); | |
6385 } | |
6386 } | |
6387 | |
6388 void MacroAssembler::incrementl(ArrayAddress dst) { | |
6389 incrementl(as_Address(dst)); | |
6390 } | |
6391 | |
6392 void MacroAssembler::incrementl(Register reg, int value) { | |
6393 if (value == min_jint) {addl(reg, value) ; return; } | |
6394 if (value < 0) { decrementl(reg, -value); return; } | |
6395 if (value == 0) { ; return; } | |
6396 if (value == 1 && UseIncDec) { incl(reg) ; return; } | |
6397 /* else */ { addl(reg, value) ; return; } | |
6398 } | |
6399 | |
6400 void MacroAssembler::incrementl(Address dst, int value) { | |
6401 if (value == min_jint) {addl(dst, value) ; return; } | |
6402 if (value < 0) { decrementl(dst, -value); return; } | |
6403 if (value == 0) { ; return; } | |
6404 if (value == 1 && UseIncDec) { incl(dst) ; return; } | |
6405 /* else */ { addl(dst, value) ; return; } | |
6406 } | |
6407 | |
6408 void MacroAssembler::jump(AddressLiteral dst) { | |
6409 if (reachable(dst)) { | |
6410 jmp_literal(dst.target(), dst.rspec()); | |
6411 } else { | |
6412 lea(rscratch1, dst); | |
6413 jmp(rscratch1); | |
6414 } | |
6415 } | |
6416 | |
6417 void MacroAssembler::jump_cc(Condition cc, AddressLiteral dst) { | |
6418 if (reachable(dst)) { | |
6419 InstructionMark im(this); | |
6420 relocate(dst.reloc()); | |
6421 const int short_size = 2; | |
6422 const int long_size = 6; | |
6423 int offs = (intptr_t)dst.target() - ((intptr_t)_code_pos); | |
6424 if (dst.reloc() == relocInfo::none && is8bit(offs - short_size)) { | |
6425 // 0111 tttn #8-bit disp | |
6426 emit_byte(0x70 | cc); | |
6427 emit_byte((offs - short_size) & 0xFF); | |
6428 } else { | |
6429 // 0000 1111 1000 tttn #32-bit disp | |
6430 emit_byte(0x0F); | |
6431 emit_byte(0x80 | cc); | |
6432 emit_long(offs - long_size); | |
6433 } | |
0 | 6434 } else { |
304 | 6435 #ifdef ASSERT |
6436 warning("reversing conditional branch"); | |
6437 #endif /* ASSERT */ | |
6438 Label skip; | |
6439 jccb(reverse[cc], skip); | |
6440 lea(rscratch1, dst); | |
6441 Assembler::jmp(rscratch1); | |
6442 bind(skip); | |
6443 } | |
6444 } | |
6445 | |
6446 void MacroAssembler::ldmxcsr(AddressLiteral src) { | |
6447 if (reachable(src)) { | |
6448 Assembler::ldmxcsr(as_Address(src)); | |
6449 } else { | |
6450 lea(rscratch1, src); | |
6451 Assembler::ldmxcsr(Address(rscratch1, 0)); | |
6452 } | |
6453 } | |
6454 | |
6455 int MacroAssembler::load_signed_byte(Register dst, Address src) { | |
6456 int off; | |
6457 if (LP64_ONLY(true ||) VM_Version::is_P6()) { | |
6458 off = offset(); | |
6459 movsbl(dst, src); // movsxb | |
6460 } else { | |
6461 off = load_unsigned_byte(dst, src); | |
6462 shll(dst, 24); | |
6463 sarl(dst, 24); | |
6464 } | |
6465 return off; | |
6466 } | |
6467 | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6468 // Note: load_signed_short used to be called load_signed_word. |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6469 // Although the 'w' in x86 opcodes refers to the term "word" in the assembler |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6470 // manual, which means 16 bits, that usage is found nowhere in HotSpot code. |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6471 // The term "word" in HotSpot means a 32- or 64-bit machine word. |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6472 int MacroAssembler::load_signed_short(Register dst, Address src) { |
304 | 6473 int off; |
6474 if (LP64_ONLY(true ||) VM_Version::is_P6()) { | |
6475 // This is dubious to me since it seems safe to do a signed 16 => 64 bit | |
6476 // version but this is what 64bit has always done. This seems to imply | |
6477 // that users are only using 32bits worth. | |
6478 off = offset(); | |
6479 movswl(dst, src); // movsxw | |
6480 } else { | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6481 off = load_unsigned_short(dst, src); |
304 | 6482 shll(dst, 16); |
6483 sarl(dst, 16); | |
6484 } | |
6485 return off; | |
6486 } | |
6487 | |
6488 int MacroAssembler::load_unsigned_byte(Register dst, Address src) { | |
6489 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16, | |
6490 // and "3.9 Partial Register Penalties", p. 22). | |
6491 int off; | |
6492 if (LP64_ONLY(true || ) VM_Version::is_P6() || src.uses(dst)) { | |
6493 off = offset(); | |
6494 movzbl(dst, src); // movzxb | |
6495 } else { | |
6496 xorl(dst, dst); | |
6497 off = offset(); | |
6498 movb(dst, src); | |
6499 } | |
6500 return off; | |
6501 } | |
6502 | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6503 // Note: load_unsigned_short used to be called load_unsigned_word. |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6504 int MacroAssembler::load_unsigned_short(Register dst, Address src) { |
304 | 6505 // According to Intel Doc. AP-526, "Zero-Extension of Short", p.16, |
6506 // and "3.9 Partial Register Penalties", p. 22). | |
6507 int off; | |
6508 if (LP64_ONLY(true ||) VM_Version::is_P6() || src.uses(dst)) { | |
6509 off = offset(); | |
6510 movzwl(dst, src); // movzxw | |
6511 } else { | |
6512 xorl(dst, dst); | |
6513 off = offset(); | |
6514 movw(dst, src); | |
6515 } | |
6516 return off; | |
6517 } | |
6518 | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6519 void MacroAssembler::load_sized_value(Register dst, Address src, |
1503 | 6520 size_t size_in_bytes, bool is_signed) { |
6521 switch (size_in_bytes) { | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6522 #ifndef _LP64 |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6523 // For case 8, caller is responsible for manually loading |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6524 // the second word into another register. |
1503 | 6525 case 8: movl(dst, src); break; |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6526 #else |
1503 | 6527 case 8: movq(dst, src); break; |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6528 #endif |
1503 | 6529 case 4: movl(dst, src); break; |
6530 case 2: is_signed ? load_signed_short(dst, src) : load_unsigned_short(dst, src); break; | |
6531 case 1: is_signed ? load_signed_byte( dst, src) : load_unsigned_byte( dst, src); break; | |
6532 default: ShouldNotReachHere(); | |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6533 } |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6534 } |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
6535 |
304 | 6536 void MacroAssembler::mov32(AddressLiteral dst, Register src) { |
6537 if (reachable(dst)) { | |
6538 movl(as_Address(dst), src); | |
6539 } else { | |
6540 lea(rscratch1, dst); | |
6541 movl(Address(rscratch1, 0), src); | |
6542 } | |
6543 } | |
6544 | |
6545 void MacroAssembler::mov32(Register dst, AddressLiteral src) { | |
6546 if (reachable(src)) { | |
6547 movl(dst, as_Address(src)); | |
6548 } else { | |
6549 lea(rscratch1, src); | |
6550 movl(dst, Address(rscratch1, 0)); | |
6551 } | |
0 | 6552 } |
6553 | |
6554 // C++ bool manipulation | |
6555 | |
6556 void MacroAssembler::movbool(Register dst, Address src) { | |
6557 if(sizeof(bool) == 1) | |
6558 movb(dst, src); | |
6559 else if(sizeof(bool) == 2) | |
6560 movw(dst, src); | |
6561 else if(sizeof(bool) == 4) | |
6562 movl(dst, src); | |
6563 else | |
6564 // unsupported | |
6565 ShouldNotReachHere(); | |
6566 } | |
6567 | |
6568 void MacroAssembler::movbool(Address dst, bool boolconst) { | |
6569 if(sizeof(bool) == 1) | |
6570 movb(dst, (int) boolconst); | |
6571 else if(sizeof(bool) == 2) | |
6572 movw(dst, (int) boolconst); | |
6573 else if(sizeof(bool) == 4) | |
6574 movl(dst, (int) boolconst); | |
6575 else | |
6576 // unsupported | |
6577 ShouldNotReachHere(); | |
6578 } | |
6579 | |
6580 void MacroAssembler::movbool(Address dst, Register src) { | |
6581 if(sizeof(bool) == 1) | |
6582 movb(dst, src); | |
6583 else if(sizeof(bool) == 2) | |
6584 movw(dst, src); | |
6585 else if(sizeof(bool) == 4) | |
6586 movl(dst, src); | |
6587 else | |
6588 // unsupported | |
6589 ShouldNotReachHere(); | |
6590 } | |
6591 | |
304 | 6592 void MacroAssembler::movbyte(ArrayAddress dst, int src) { |
6593 movb(as_Address(dst), src); | |
6594 } | |
6595 | |
6596 void MacroAssembler::movdbl(XMMRegister dst, AddressLiteral src) { | |
6597 if (reachable(src)) { | |
6598 if (UseXmmLoadAndClearUpper) { | |
6599 movsd (dst, as_Address(src)); | |
6600 } else { | |
6601 movlpd(dst, as_Address(src)); | |
6602 } | |
6603 } else { | |
6604 lea(rscratch1, src); | |
6605 if (UseXmmLoadAndClearUpper) { | |
6606 movsd (dst, Address(rscratch1, 0)); | |
6607 } else { | |
6608 movlpd(dst, Address(rscratch1, 0)); | |
6609 } | |
6610 } | |
6611 } | |
6612 | |
6613 void MacroAssembler::movflt(XMMRegister dst, AddressLiteral src) { | |
6614 if (reachable(src)) { | |
6615 movss(dst, as_Address(src)); | |
6616 } else { | |
6617 lea(rscratch1, src); | |
6618 movss(dst, Address(rscratch1, 0)); | |
6619 } | |
6620 } | |
6621 | |
6622 void MacroAssembler::movptr(Register dst, Register src) { | |
6623 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src)); | |
6624 } | |
6625 | |
6626 void MacroAssembler::movptr(Register dst, Address src) { | |
6627 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src)); | |
6628 } | |
6629 | |
6630 // src should NEVER be a real pointer. Use AddressLiteral for true pointers | |
6631 void MacroAssembler::movptr(Register dst, intptr_t src) { | |
6632 LP64_ONLY(mov64(dst, src)) NOT_LP64(movl(dst, src)); | |
6633 } | |
6634 | |
6635 void MacroAssembler::movptr(Address dst, Register src) { | |
6636 LP64_ONLY(movq(dst, src)) NOT_LP64(movl(dst, src)); | |
6637 } | |
6638 | |
6639 void MacroAssembler::movss(XMMRegister dst, AddressLiteral src) { | |
6640 if (reachable(src)) { | |
6641 movss(dst, as_Address(src)); | |
6642 } else { | |
6643 lea(rscratch1, src); | |
6644 movss(dst, Address(rscratch1, 0)); | |
6645 } | |
6646 } | |
6647 | |
6648 void MacroAssembler::null_check(Register reg, int offset) { | |
6649 if (needs_explicit_null_check(offset)) { | |
6650 // provoke OS NULL exception if reg = NULL by | |
6651 // accessing M[reg] w/o changing any (non-CC) registers | |
6652 // NOTE: cmpl is plenty here to provoke a segv | |
6653 cmpptr(rax, Address(reg, 0)); | |
6654 // Note: should probably use testl(rax, Address(reg, 0)); | |
6655 // may be shorter code (however, this version of | |
6656 // testl needs to be implemented first) | |
6657 } else { | |
6658 // nothing to do, (later) access of M[reg + offset] | |
6659 // will provoke OS NULL exception if reg = NULL | |
6660 } | |
6661 } | |
6662 | |
6663 void MacroAssembler::os_breakpoint() { | |
6664 // instead of directly emitting a breakpoint, call os:breakpoint for better debugability | |
6665 // (e.g., MSVC can't call ps() otherwise) | |
6666 call(RuntimeAddress(CAST_FROM_FN_PTR(address, os::breakpoint))); | |
6667 } | |
6668 | |
6669 void MacroAssembler::pop_CPU_state() { | |
6670 pop_FPU_state(); | |
6671 pop_IU_state(); | |
6672 } | |
6673 | |
6674 void MacroAssembler::pop_FPU_state() { | |
6675 NOT_LP64(frstor(Address(rsp, 0));) | |
6676 LP64_ONLY(fxrstor(Address(rsp, 0));) | |
6677 addptr(rsp, FPUStateSizeInWords * wordSize); | |
6678 } | |
6679 | |
6680 void MacroAssembler::pop_IU_state() { | |
6681 popa(); | |
6682 LP64_ONLY(addq(rsp, 8)); | |
6683 popf(); | |
6684 } | |
6685 | |
6686 // Save Integer and Float state | |
6687 // Warning: Stack must be 16 byte aligned (64bit) | |
6688 void MacroAssembler::push_CPU_state() { | |
6689 push_IU_state(); | |
6690 push_FPU_state(); | |
6691 } | |
6692 | |
6693 void MacroAssembler::push_FPU_state() { | |
6694 subptr(rsp, FPUStateSizeInWords * wordSize); | |
6695 #ifndef _LP64 | |
6696 fnsave(Address(rsp, 0)); | |
6697 fwait(); | |
6698 #else | |
6699 fxsave(Address(rsp, 0)); | |
6700 #endif // LP64 | |
6701 } | |
6702 | |
6703 void MacroAssembler::push_IU_state() { | |
6704 // Push flags first because pusha kills them | |
6705 pushf(); | |
6706 // Make sure rsp stays 16-byte aligned | |
6707 LP64_ONLY(subq(rsp, 8)); | |
6708 pusha(); | |
6709 } | |
6710 | |
6711 void MacroAssembler::reset_last_Java_frame(Register java_thread, bool clear_fp, bool clear_pc) { | |
6712 // determine java_thread register | |
6713 if (!java_thread->is_valid()) { | |
6714 java_thread = rdi; | |
6715 get_thread(java_thread); | |
6716 } | |
6717 // we must set sp to zero to clear frame | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
420
diff
changeset
|
6718 movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), NULL_WORD); |
304 | 6719 if (clear_fp) { |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
420
diff
changeset
|
6720 movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), NULL_WORD); |
304 | 6721 } |
6722 | |
6723 if (clear_pc) | |
512
db4caa99ef11
6787106: Hotspot 32 bit build fails on platforms having different definitions for intptr_t & int32_t
xlu
parents:
420
diff
changeset
|
6724 movptr(Address(java_thread, JavaThread::last_Java_pc_offset()), NULL_WORD); |
304 | 6725 |
6726 } | |
6727 | |
6728 void MacroAssembler::restore_rax(Register tmp) { | |
6729 if (tmp == noreg) pop(rax); | |
6730 else if (tmp != rax) mov(rax, tmp); | |
6731 } | |
6732 | |
6733 void MacroAssembler::round_to(Register reg, int modulus) { | |
6734 addptr(reg, modulus - 1); | |
6735 andptr(reg, -modulus); | |
6736 } | |
6737 | |
6738 void MacroAssembler::save_rax(Register tmp) { | |
6739 if (tmp == noreg) push(rax); | |
6740 else if (tmp != rax) mov(tmp, rax); | |
6741 } | |
6742 | |
6743 // Write serialization page so VM thread can do a pseudo remote membar. | |
6744 // We use the current thread pointer to calculate a thread specific | |
6745 // offset to write to within the page. This minimizes bus traffic | |
6746 // due to cache line collision. | |
6747 void MacroAssembler::serialize_memory(Register thread, Register tmp) { | |
6748 movl(tmp, thread); | |
6749 shrl(tmp, os::get_serialize_page_shift_count()); | |
6750 andl(tmp, (os::vm_page_size() - sizeof(int))); | |
6751 | |
6752 Address index(noreg, tmp, Address::times_1); | |
6753 ExternalAddress page(os::get_memory_serialize_page()); | |
6754 | |
606
19962e74284f
6811384: MacroAssembler::serialize_memory may touch next page on amd64
never
parents:
520
diff
changeset
|
6755 // Size of store must match masking code above |
19962e74284f
6811384: MacroAssembler::serialize_memory may touch next page on amd64
never
parents:
520
diff
changeset
|
6756 movl(as_Address(ArrayAddress(page, index)), tmp); |
304 | 6757 } |
6758 | |
6759 // Calls to C land | |
6760 // | |
6761 // When entering C land, the rbp, & rsp of the last Java frame have to be recorded | |
6762 // in the (thread-local) JavaThread object. When leaving C land, the last Java fp | |
6763 // has to be reset to 0. This is required to allow proper stack traversal. | |
6764 void MacroAssembler::set_last_Java_frame(Register java_thread, | |
6765 Register last_java_sp, | |
6766 Register last_java_fp, | |
6767 address last_java_pc) { | |
6768 // determine java_thread register | |
6769 if (!java_thread->is_valid()) { | |
6770 java_thread = rdi; | |
6771 get_thread(java_thread); | |
6772 } | |
6773 // determine last_java_sp register | |
6774 if (!last_java_sp->is_valid()) { | |
6775 last_java_sp = rsp; | |
6776 } | |
6777 | |
6778 // last_java_fp is optional | |
6779 | |
6780 if (last_java_fp->is_valid()) { | |
6781 movptr(Address(java_thread, JavaThread::last_Java_fp_offset()), last_java_fp); | |
6782 } | |
6783 | |
6784 // last_java_pc is optional | |
6785 | |
6786 if (last_java_pc != NULL) { | |
6787 lea(Address(java_thread, | |
6788 JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()), | |
6789 InternalAddress(last_java_pc)); | |
6790 | |
6791 } | |
6792 movptr(Address(java_thread, JavaThread::last_Java_sp_offset()), last_java_sp); | |
6793 } | |
6794 | |
6795 void MacroAssembler::shlptr(Register dst, int imm8) { | |
6796 LP64_ONLY(shlq(dst, imm8)) NOT_LP64(shll(dst, imm8)); | |
6797 } | |
6798 | |
6799 void MacroAssembler::shrptr(Register dst, int imm8) { | |
6800 LP64_ONLY(shrq(dst, imm8)) NOT_LP64(shrl(dst, imm8)); | |
6801 } | |
6802 | |
6803 void MacroAssembler::sign_extend_byte(Register reg) { | |
6804 if (LP64_ONLY(true ||) (VM_Version::is_P6() && reg->has_byte_register())) { | |
6805 movsbl(reg, reg); // movsxb | |
6806 } else { | |
6807 shll(reg, 24); | |
6808 sarl(reg, 24); | |
6809 } | |
6810 } | |
6811 | |
6812 void MacroAssembler::sign_extend_short(Register reg) { | |
6813 if (LP64_ONLY(true ||) VM_Version::is_P6()) { | |
6814 movswl(reg, reg); // movsxw | |
6815 } else { | |
6816 shll(reg, 16); | |
6817 sarl(reg, 16); | |
6818 } | |
6819 } | |
6820 | |
362 | 6821 ////////////////////////////////////////////////////////////////////////////////// |
6822 #ifndef SERIALGC | |
6823 | |
6824 void MacroAssembler::g1_write_barrier_pre(Register obj, | |
6825 #ifndef _LP64 | |
6826 Register thread, | |
6827 #endif | |
6828 Register tmp, | |
6829 Register tmp2, | |
6830 bool tosca_live) { | |
6831 LP64_ONLY(Register thread = r15_thread;) | |
6832 Address in_progress(thread, in_bytes(JavaThread::satb_mark_queue_offset() + | |
6833 PtrQueue::byte_offset_of_active())); | |
6834 | |
6835 Address index(thread, in_bytes(JavaThread::satb_mark_queue_offset() + | |
6836 PtrQueue::byte_offset_of_index())); | |
6837 Address buffer(thread, in_bytes(JavaThread::satb_mark_queue_offset() + | |
6838 PtrQueue::byte_offset_of_buf())); | |
6839 | |
6840 | |
6841 Label done; | |
6842 Label runtime; | |
6843 | |
6844 // if (!marking_in_progress) goto done; | |
6845 if (in_bytes(PtrQueue::byte_width_of_active()) == 4) { | |
6846 cmpl(in_progress, 0); | |
6847 } else { | |
6848 assert(in_bytes(PtrQueue::byte_width_of_active()) == 1, "Assumption"); | |
6849 cmpb(in_progress, 0); | |
6850 } | |
6851 jcc(Assembler::equal, done); | |
6852 | |
6853 // if (x.f == NULL) goto done; | |
845
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
775
diff
changeset
|
6854 #ifdef _LP64 |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
775
diff
changeset
|
6855 load_heap_oop(tmp2, Address(obj, 0)); |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
775
diff
changeset
|
6856 #else |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
775
diff
changeset
|
6857 movptr(tmp2, Address(obj, 0)); |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
775
diff
changeset
|
6858 #endif |
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
775
diff
changeset
|
6859 cmpptr(tmp2, (int32_t) NULL_WORD); |
362 | 6860 jcc(Assembler::equal, done); |
6861 | |
6862 // Can we store original value in the thread's buffer? | |
6863 | |
6864 #ifdef _LP64 | |
845
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
775
diff
changeset
|
6865 movslq(tmp, index); |
362 | 6866 cmpq(tmp, 0); |
6867 #else | |
6868 cmpl(index, 0); | |
6869 #endif | |
6870 jcc(Assembler::equal, runtime); | |
6871 #ifdef _LP64 | |
6872 subq(tmp, wordSize); | |
6873 movl(index, tmp); | |
6874 addq(tmp, buffer); | |
6875 #else | |
6876 subl(index, wordSize); | |
6877 movl(tmp, buffer); | |
6878 addl(tmp, index); | |
6879 #endif | |
6880 movptr(Address(tmp, 0), tmp2); | |
6881 jmp(done); | |
6882 bind(runtime); | |
6883 // save the live input values | |
6884 if(tosca_live) push(rax); | |
6885 push(obj); | |
6886 #ifdef _LP64 | |
845
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
775
diff
changeset
|
6887 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, r15_thread); |
362 | 6888 #else |
6889 push(thread); | |
6890 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_pre), tmp2, thread); | |
6891 pop(thread); | |
6892 #endif | |
6893 pop(obj); | |
6894 if(tosca_live) pop(rax); | |
6895 bind(done); | |
6896 | |
6897 } | |
6898 | |
6899 void MacroAssembler::g1_write_barrier_post(Register store_addr, | |
6900 Register new_val, | |
6901 #ifndef _LP64 | |
6902 Register thread, | |
6903 #endif | |
6904 Register tmp, | |
6905 Register tmp2) { | |
6906 | |
6907 LP64_ONLY(Register thread = r15_thread;) | |
6908 Address queue_index(thread, in_bytes(JavaThread::dirty_card_queue_offset() + | |
6909 PtrQueue::byte_offset_of_index())); | |
6910 Address buffer(thread, in_bytes(JavaThread::dirty_card_queue_offset() + | |
6911 PtrQueue::byte_offset_of_buf())); | |
6912 BarrierSet* bs = Universe::heap()->barrier_set(); | |
6913 CardTableModRefBS* ct = (CardTableModRefBS*)bs; | |
6914 Label done; | |
6915 Label runtime; | |
6916 | |
6917 // Does store cross heap regions? | |
6918 | |
6919 movptr(tmp, store_addr); | |
6920 xorptr(tmp, new_val); | |
6921 shrptr(tmp, HeapRegion::LogOfHRGrainBytes); | |
6922 jcc(Assembler::equal, done); | |
6923 | |
6924 // crosses regions, storing NULL? | |
6925 | |
6926 cmpptr(new_val, (int32_t) NULL_WORD); | |
6927 jcc(Assembler::equal, done); | |
6928 | |
6929 // storing region crossing non-NULL, is card already dirty? | |
6930 | |
6931 ExternalAddress cardtable((address) ct->byte_map_base); | |
6932 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); | |
6933 #ifdef _LP64 | |
6934 const Register card_addr = tmp; | |
6935 | |
6936 movq(card_addr, store_addr); | |
6937 shrq(card_addr, CardTableModRefBS::card_shift); | |
6938 | |
6939 lea(tmp2, cardtable); | |
6940 | |
6941 // get the address of the card | |
6942 addq(card_addr, tmp2); | |
6943 #else | |
6944 const Register card_index = tmp; | |
6945 | |
6946 movl(card_index, store_addr); | |
6947 shrl(card_index, CardTableModRefBS::card_shift); | |
6948 | |
6949 Address index(noreg, card_index, Address::times_1); | |
6950 const Register card_addr = tmp; | |
6951 lea(card_addr, as_Address(ArrayAddress(cardtable, index))); | |
6952 #endif | |
6953 cmpb(Address(card_addr, 0), 0); | |
6954 jcc(Assembler::equal, done); | |
6955 | |
6956 // storing a region crossing, non-NULL oop, card is clean. | |
6957 // dirty card and log. | |
6958 | |
6959 movb(Address(card_addr, 0), 0); | |
6960 | |
6961 cmpl(queue_index, 0); | |
6962 jcc(Assembler::equal, runtime); | |
6963 subl(queue_index, wordSize); | |
6964 movptr(tmp2, buffer); | |
6965 #ifdef _LP64 | |
6966 movslq(rscratch1, queue_index); | |
6967 addq(tmp2, rscratch1); | |
6968 movq(Address(tmp2, 0), card_addr); | |
6969 #else | |
6970 addl(tmp2, queue_index); | |
6971 movl(Address(tmp2, 0), card_index); | |
6972 #endif | |
6973 jmp(done); | |
6974 | |
6975 bind(runtime); | |
6976 // save the live input values | |
6977 push(store_addr); | |
6978 push(new_val); | |
6979 #ifdef _LP64 | |
6980 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, r15_thread); | |
6981 #else | |
6982 push(thread); | |
6983 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::g1_wb_post), card_addr, thread); | |
6984 pop(thread); | |
6985 #endif | |
6986 pop(new_val); | |
6987 pop(store_addr); | |
6988 | |
6989 bind(done); | |
6990 | |
6991 } | |
6992 | |
6993 #endif // SERIALGC | |
6994 ////////////////////////////////////////////////////////////////////////////////// | |
6995 | |
6996 | |
304 | 6997 void MacroAssembler::store_check(Register obj) { |
6998 // Does a store check for the oop in register obj. The content of | |
6999 // register obj is destroyed afterwards. | |
7000 store_check_part_1(obj); | |
7001 store_check_part_2(obj); | |
7002 } | |
7003 | |
7004 void MacroAssembler::store_check(Register obj, Address dst) { | |
7005 store_check(obj); | |
7006 } | |
7007 | |
7008 | |
7009 // split the store check operation so that other instructions can be scheduled inbetween | |
7010 void MacroAssembler::store_check_part_1(Register obj) { | |
7011 BarrierSet* bs = Universe::heap()->barrier_set(); | |
7012 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind"); | |
7013 shrptr(obj, CardTableModRefBS::card_shift); | |
7014 } | |
7015 | |
7016 void MacroAssembler::store_check_part_2(Register obj) { | |
7017 BarrierSet* bs = Universe::heap()->barrier_set(); | |
7018 assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind"); | |
7019 CardTableModRefBS* ct = (CardTableModRefBS*)bs; | |
7020 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); | |
7021 | |
7022 // The calculation for byte_map_base is as follows: | |
7023 // byte_map_base = _byte_map - (uintptr_t(low_bound) >> card_shift); | |
7024 // So this essentially converts an address to a displacement and | |
7025 // it will never need to be relocated. On 64bit however the value may be too | |
7026 // large for a 32bit displacement | |
7027 | |
7028 intptr_t disp = (intptr_t) ct->byte_map_base; | |
7029 if (is_simm32(disp)) { | |
7030 Address cardtable(noreg, obj, Address::times_1, disp); | |
7031 movb(cardtable, 0); | |
7032 } else { | |
7033 // By doing it as an ExternalAddress disp could be converted to a rip-relative | |
7034 // displacement and done in a single instruction given favorable mapping and | |
7035 // a smarter version of as_Address. Worst case it is two instructions which | |
7036 // is no worse off then loading disp into a register and doing as a simple | |
7037 // Address() as above. | |
7038 // We can't do as ExternalAddress as the only style since if disp == 0 we'll | |
7039 // assert since NULL isn't acceptable in a reloci (see 6644928). In any case | |
7040 // in some cases we'll get a single instruction version. | |
7041 | |
7042 ExternalAddress cardtable((address)disp); | |
7043 Address index(noreg, obj, Address::times_1); | |
7044 movb(as_Address(ArrayAddress(cardtable, index)), 0); | |
7045 } | |
7046 } | |
7047 | |
7048 void MacroAssembler::subptr(Register dst, int32_t imm32) { | |
7049 LP64_ONLY(subq(dst, imm32)) NOT_LP64(subl(dst, imm32)); | |
7050 } | |
7051 | |
7052 void MacroAssembler::subptr(Register dst, Register src) { | |
7053 LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); | |
7054 } | |
7055 | |
7056 void MacroAssembler::test32(Register src1, AddressLiteral src2) { | |
7057 // src2 must be rval | |
7058 | |
7059 if (reachable(src2)) { | |
7060 testl(src1, as_Address(src2)); | |
7061 } else { | |
7062 lea(rscratch1, src2); | |
7063 testl(src1, Address(rscratch1, 0)); | |
7064 } | |
7065 } | |
7066 | |
7067 // C++ bool manipulation | |
0 | 7068 void MacroAssembler::testbool(Register dst) { |
7069 if(sizeof(bool) == 1) | |
304 | 7070 testb(dst, 0xff); |
0 | 7071 else if(sizeof(bool) == 2) { |
7072 // testw implementation needed for two byte bools | |
7073 ShouldNotReachHere(); | |
7074 } else if(sizeof(bool) == 4) | |
7075 testl(dst, dst); | |
7076 else | |
7077 // unsupported | |
7078 ShouldNotReachHere(); | |
7079 } | |
7080 | |
304 | 7081 void MacroAssembler::testptr(Register dst, Register src) { |
7082 LP64_ONLY(testq(dst, src)) NOT_LP64(testl(dst, src)); | |
7083 } | |
7084 | |
7085 // Defines obj, preserves var_size_in_bytes, okay for t2 == var_size_in_bytes. | |
7086 void MacroAssembler::tlab_allocate(Register obj, | |
7087 Register var_size_in_bytes, | |
7088 int con_size_in_bytes, | |
7089 Register t1, | |
7090 Register t2, | |
7091 Label& slow_case) { | |
7092 assert_different_registers(obj, t1, t2); | |
7093 assert_different_registers(obj, var_size_in_bytes, t1); | |
7094 Register end = t2; | |
7095 Register thread = NOT_LP64(t1) LP64_ONLY(r15_thread); | |
7096 | |
7097 verify_tlab(); | |
7098 | |
7099 NOT_LP64(get_thread(thread)); | |
7100 | |
7101 movptr(obj, Address(thread, JavaThread::tlab_top_offset())); | |
7102 if (var_size_in_bytes == noreg) { | |
7103 lea(end, Address(obj, con_size_in_bytes)); | |
7104 } else { | |
7105 lea(end, Address(obj, var_size_in_bytes, Address::times_1)); | |
7106 } | |
7107 cmpptr(end, Address(thread, JavaThread::tlab_end_offset())); | |
7108 jcc(Assembler::above, slow_case); | |
7109 | |
7110 // update the tlab top pointer | |
7111 movptr(Address(thread, JavaThread::tlab_top_offset()), end); | |
7112 | |
7113 // recover var_size_in_bytes if necessary | |
7114 if (var_size_in_bytes == end) { | |
7115 subptr(var_size_in_bytes, obj); | |
7116 } | |
7117 verify_tlab(); | |
7118 } | |
7119 | |
7120 // Preserves rbx, and rdx. | |
7121 void MacroAssembler::tlab_refill(Label& retry, | |
7122 Label& try_eden, | |
7123 Label& slow_case) { | |
7124 Register top = rax; | |
7125 Register t1 = rcx; | |
7126 Register t2 = rsi; | |
7127 Register thread_reg = NOT_LP64(rdi) LP64_ONLY(r15_thread); | |
7128 assert_different_registers(top, thread_reg, t1, t2, /* preserve: */ rbx, rdx); | |
7129 Label do_refill, discard_tlab; | |
7130 | |
7131 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) { | |
7132 // No allocation in the shared eden. | |
7133 jmp(slow_case); | |
7134 } | |
7135 | |
7136 NOT_LP64(get_thread(thread_reg)); | |
7137 | |
7138 movptr(top, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); | |
7139 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset()))); | |
7140 | |
7141 // calculate amount of free space | |
7142 subptr(t1, top); | |
7143 shrptr(t1, LogHeapWordSize); | |
7144 | |
7145 // Retain tlab and allocate object in shared space if | |
7146 // the amount free in the tlab is too large to discard. | |
7147 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset()))); | |
7148 jcc(Assembler::lessEqual, discard_tlab); | |
7149 | |
7150 // Retain | |
7151 // %%% yuck as movptr... | |
7152 movptr(t2, (int32_t) ThreadLocalAllocBuffer::refill_waste_limit_increment()); | |
7153 addptr(Address(thread_reg, in_bytes(JavaThread::tlab_refill_waste_limit_offset())), t2); | |
7154 if (TLABStats) { | |
7155 // increment number of slow_allocations | |
7156 addl(Address(thread_reg, in_bytes(JavaThread::tlab_slow_allocations_offset())), 1); | |
7157 } | |
7158 jmp(try_eden); | |
7159 | |
7160 bind(discard_tlab); | |
7161 if (TLABStats) { | |
7162 // increment number of refills | |
7163 addl(Address(thread_reg, in_bytes(JavaThread::tlab_number_of_refills_offset())), 1); | |
7164 // accumulate wastage -- t1 is amount free in tlab | |
7165 addl(Address(thread_reg, in_bytes(JavaThread::tlab_fast_refill_waste_offset())), t1); | |
7166 } | |
7167 | |
7168 // if tlab is currently allocated (top or end != null) then | |
7169 // fill [top, end + alignment_reserve) with array object | |
7170 testptr (top, top); | |
7171 jcc(Assembler::zero, do_refill); | |
7172 | |
7173 // set up the mark word | |
7174 movptr(Address(top, oopDesc::mark_offset_in_bytes()), (intptr_t)markOopDesc::prototype()->copy_set_hash(0x2)); | |
7175 // set the length to the remaining space | |
7176 subptr(t1, typeArrayOopDesc::header_size(T_INT)); | |
7177 addptr(t1, (int32_t)ThreadLocalAllocBuffer::alignment_reserve()); | |
7178 shlptr(t1, log2_intptr(HeapWordSize/sizeof(jint))); | |
1690 | 7179 movl(Address(top, arrayOopDesc::length_offset_in_bytes()), t1); |
304 | 7180 // set klass to intArrayKlass |
7181 // dubious reloc why not an oop reloc? | |
7182 movptr(t1, ExternalAddress((address) Universe::intArrayKlassObj_addr())); | |
7183 // store klass last. concurrent gcs assumes klass length is valid if | |
7184 // klass field is not null. | |
7185 store_klass(top, t1); | |
7186 | |
7187 // refill the tlab with an eden allocation | |
7188 bind(do_refill); | |
7189 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset()))); | |
7190 shlptr(t1, LogHeapWordSize); | |
7191 // add object_size ?? | |
7192 eden_allocate(top, t1, 0, t2, slow_case); | |
7193 | |
7194 // Check that t1 was preserved in eden_allocate. | |
7195 #ifdef ASSERT | |
7196 if (UseTLAB) { | |
7197 Label ok; | |
7198 Register tsize = rsi; | |
7199 assert_different_registers(tsize, thread_reg, t1); | |
7200 push(tsize); | |
7201 movptr(tsize, Address(thread_reg, in_bytes(JavaThread::tlab_size_offset()))); | |
7202 shlptr(tsize, LogHeapWordSize); | |
7203 cmpptr(t1, tsize); | |
7204 jcc(Assembler::equal, ok); | |
7205 stop("assert(t1 != tlab size)"); | |
7206 should_not_reach_here(); | |
7207 | |
7208 bind(ok); | |
7209 pop(tsize); | |
7210 } | |
7211 #endif | |
7212 movptr(Address(thread_reg, in_bytes(JavaThread::tlab_start_offset())), top); | |
7213 movptr(Address(thread_reg, in_bytes(JavaThread::tlab_top_offset())), top); | |
7214 addptr(top, t1); | |
7215 subptr(top, (int32_t)ThreadLocalAllocBuffer::alignment_reserve_in_bytes()); | |
7216 movptr(Address(thread_reg, in_bytes(JavaThread::tlab_end_offset())), top); | |
7217 verify_tlab(); | |
7218 jmp(retry); | |
7219 } | |
7220 | |
7221 static const double pi_4 = 0.7853981633974483; | |
7222 | |
7223 void MacroAssembler::trigfunc(char trig, int num_fpu_regs_in_use) { | |
7224 // A hand-coded argument reduction for values in fabs(pi/4, pi/2) | |
7225 // was attempted in this code; unfortunately it appears that the | |
7226 // switch to 80-bit precision and back causes this to be | |
7227 // unprofitable compared with simply performing a runtime call if | |
7228 // the argument is out of the (-pi/4, pi/4) range. | |
7229 | |
7230 Register tmp = noreg; | |
7231 if (!VM_Version::supports_cmov()) { | |
7232 // fcmp needs a temporary so preserve rbx, | |
7233 tmp = rbx; | |
7234 push(tmp); | |
7235 } | |
7236 | |
7237 Label slow_case, done; | |
7238 | |
520
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7239 ExternalAddress pi4_adr = (address)&pi_4; |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7240 if (reachable(pi4_adr)) { |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7241 // x ?<= pi/4 |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7242 fld_d(pi4_adr); |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7243 fld_s(1); // Stack: X PI/4 X |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7244 fabs(); // Stack: |X| PI/4 X |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7245 fcmp(tmp); |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7246 jcc(Assembler::above, slow_case); |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7247 |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7248 // fastest case: -pi/4 <= x <= pi/4 |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7249 switch(trig) { |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7250 case 's': |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7251 fsin(); |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7252 break; |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7253 case 'c': |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7254 fcos(); |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7255 break; |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7256 case 't': |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7257 ftan(); |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7258 break; |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7259 default: |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7260 assert(false, "bad intrinsic"); |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7261 break; |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7262 } |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7263 jmp(done); |
52a431267315
6791168: Fix invalid code in bytecodeInterpreter that can cause gcc ICE
coleenp
parents:
512
diff
changeset
|
7264 } |
304 | 7265 |
7266 // slow case: runtime call | |
7267 bind(slow_case); | |
7268 // Preserve registers across runtime call | |
7269 pusha(); | |
7270 int incoming_argument_and_return_value_offset = -1; | |
7271 if (num_fpu_regs_in_use > 1) { | |
7272 // Must preserve all other FPU regs (could alternatively convert | |
7273 // SharedRuntime::dsin and dcos into assembly routines known not to trash | |
7274 // FPU state, but can not trust C compiler) | |
7275 NEEDS_CLEANUP; | |
7276 // NOTE that in this case we also push the incoming argument to | |
7277 // the stack and restore it later; we also use this stack slot to | |
7278 // hold the return value from dsin or dcos. | |
7279 for (int i = 0; i < num_fpu_regs_in_use; i++) { | |
7280 subptr(rsp, sizeof(jdouble)); | |
7281 fstp_d(Address(rsp, 0)); | |
7282 } | |
7283 incoming_argument_and_return_value_offset = sizeof(jdouble)*(num_fpu_regs_in_use-1); | |
7284 fld_d(Address(rsp, incoming_argument_and_return_value_offset)); | |
7285 } | |
7286 subptr(rsp, sizeof(jdouble)); | |
7287 fstp_d(Address(rsp, 0)); | |
7288 #ifdef _LP64 | |
7289 movdbl(xmm0, Address(rsp, 0)); | |
7290 #endif // _LP64 | |
7291 | |
7292 // NOTE: we must not use call_VM_leaf here because that requires a | |
7293 // complete interpreter frame in debug mode -- same bug as 4387334 | |
7294 // MacroAssembler::call_VM_leaf_base is perfectly safe and will | |
7295 // do proper 64bit abi | |
7296 | |
7297 NEEDS_CLEANUP; | |
7298 // Need to add stack banging before this runtime call if it needs to | |
7299 // be taken; however, there is no generic stack banging routine at | |
7300 // the MacroAssembler level | |
7301 switch(trig) { | |
7302 case 's': | |
7303 { | |
7304 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dsin), 0); | |
7305 } | |
7306 break; | |
7307 case 'c': | |
7308 { | |
7309 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dcos), 0); | |
7310 } | |
7311 break; | |
7312 case 't': | |
7313 { | |
7314 MacroAssembler::call_VM_leaf_base(CAST_FROM_FN_PTR(address, SharedRuntime::dtan), 0); | |
7315 } | |
7316 break; | |
7317 default: | |
7318 assert(false, "bad intrinsic"); | |
7319 break; | |
7320 } | |
7321 #ifdef _LP64 | |
7322 movsd(Address(rsp, 0), xmm0); | |
7323 fld_d(Address(rsp, 0)); | |
7324 #endif // _LP64 | |
7325 addptr(rsp, sizeof(jdouble)); | |
7326 if (num_fpu_regs_in_use > 1) { | |
7327 // Must save return value to stack and then restore entire FPU stack | |
7328 fstp_d(Address(rsp, incoming_argument_and_return_value_offset)); | |
7329 for (int i = 0; i < num_fpu_regs_in_use; i++) { | |
7330 fld_d(Address(rsp, 0)); | |
7331 addptr(rsp, sizeof(jdouble)); | |
7332 } | |
7333 } | |
7334 popa(); | |
7335 | |
7336 // Come here with result in F-TOS | |
7337 bind(done); | |
7338 | |
7339 if (tmp != noreg) { | |
7340 pop(tmp); | |
7341 } | |
7342 } | |
7343 | |
7344 | |
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7345 // Look up the method for a megamorphic invokeinterface call. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7346 // The target method is determined by <intf_klass, itable_index>. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7347 // The receiver klass is in recv_klass. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7348 // On success, the result will be in method_result, and execution falls through. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7349 // On failure, execution transfers to the given label. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7350 void MacroAssembler::lookup_interface_method(Register recv_klass, |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7351 Register intf_klass, |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
647
diff
changeset
|
7352 RegisterOrConstant itable_index, |
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7353 Register method_result, |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7354 Register scan_temp, |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7355 Label& L_no_such_interface) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7356 assert_different_registers(recv_klass, intf_klass, method_result, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7357 assert(itable_index.is_constant() || itable_index.as_register() == method_result, |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7358 "caller must use same register for non-constant itable index as for method"); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7359 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7360 // Compute start of first itableOffsetEntry (which is at the end of the vtable) |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7361 int vtable_base = instanceKlass::vtable_start_offset() * wordSize; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7362 int itentry_off = itableMethodEntry::method_offset_in_bytes(); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7363 int scan_step = itableOffsetEntry::size() * wordSize; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7364 int vte_size = vtableEntry::size() * wordSize; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7365 Address::ScaleFactor times_vte_scale = Address::times_ptr; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7366 assert(vte_size == wordSize, "else adjust times_vte_scale"); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7367 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7368 movl(scan_temp, Address(recv_klass, instanceKlass::vtable_length_offset() * wordSize)); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7369 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7370 // %%% Could store the aligned, prescaled offset in the klassoop. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7371 lea(scan_temp, Address(recv_klass, scan_temp, times_vte_scale, vtable_base)); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7372 if (HeapWordsPerLong > 1) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7373 // Round up to align_object_offset boundary |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7374 // see code for instanceKlass::start_of_itable! |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7375 round_to(scan_temp, BytesPerLong); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7376 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7377 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7378 // Adjust recv_klass by scaled itable_index, so we can free itable_index. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7379 assert(itableMethodEntry::size() * wordSize == wordSize, "adjust the scaling in the code below"); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7380 lea(recv_klass, Address(recv_klass, itable_index, Address::times_ptr, itentry_off)); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7381 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7382 // for (scan = klass->itable(); scan->interface() != NULL; scan += scan_step) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7383 // if (scan->interface() == intf) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7384 // result = (klass + scan->offset() + itable_index); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7385 // } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7386 // } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7387 Label search, found_method; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7388 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7389 for (int peel = 1; peel >= 0; peel--) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7390 movptr(method_result, Address(scan_temp, itableOffsetEntry::interface_offset_in_bytes())); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7391 cmpptr(intf_klass, method_result); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7392 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7393 if (peel) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7394 jccb(Assembler::equal, found_method); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7395 } else { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7396 jccb(Assembler::notEqual, search); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7397 // (invert the test to fall through to found_method...) |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7398 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7399 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7400 if (!peel) break; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7401 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7402 bind(search); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7403 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7404 // Check that the previous entry is non-null. A null entry means that |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7405 // the receiver class doesn't implement the interface, and wasn't the |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7406 // same as when the caller was compiled. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7407 testptr(method_result, method_result); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7408 jcc(Assembler::zero, L_no_such_interface); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7409 addptr(scan_temp, scan_step); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7410 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7411 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7412 bind(found_method); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7413 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7414 // Got a hit. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7415 movl(scan_temp, Address(scan_temp, itableOffsetEntry::offset_offset_in_bytes())); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7416 movptr(method_result, Address(recv_klass, scan_temp, Address::times_1)); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7417 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7418 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
7419 |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7420 void MacroAssembler::check_klass_subtype(Register sub_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7421 Register super_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7422 Register temp_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7423 Label& L_success) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7424 Label L_failure; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7425 check_klass_subtype_fast_path(sub_klass, super_klass, temp_reg, &L_success, &L_failure, NULL); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7426 check_klass_subtype_slow_path(sub_klass, super_klass, temp_reg, noreg, &L_success, NULL); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7427 bind(L_failure); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7428 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7429 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7430 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7431 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7432 Register super_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7433 Register temp_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7434 Label* L_success, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7435 Label* L_failure, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7436 Label* L_slow_path, |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
647
diff
changeset
|
7437 RegisterOrConstant super_check_offset) { |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7438 assert_different_registers(sub_klass, super_klass, temp_reg); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7439 bool must_load_sco = (super_check_offset.constant_or_zero() == -1); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7440 if (super_check_offset.is_register()) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7441 assert_different_registers(sub_klass, super_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7442 super_check_offset.as_register()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7443 } else if (must_load_sco) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7444 assert(temp_reg != noreg, "supply either a temp or a register offset"); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7445 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7446 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7447 Label L_fallthrough; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7448 int label_nulls = 0; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7449 if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7450 if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7451 if (L_slow_path == NULL) { L_slow_path = &L_fallthrough; label_nulls++; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7452 assert(label_nulls <= 1, "at most one NULL in the batch"); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7453 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7454 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7455 Klass::secondary_super_cache_offset_in_bytes()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7456 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7457 Klass::super_check_offset_offset_in_bytes()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7458 Address super_check_offset_addr(super_klass, sco_offset); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7459 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7460 // Hacked jcc, which "knows" that L_fallthrough, at least, is in |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7461 // range of a jccb. If this routine grows larger, reconsider at |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7462 // least some of these. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7463 #define local_jcc(assembler_cond, label) \ |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7464 if (&(label) == &L_fallthrough) jccb(assembler_cond, label); \ |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7465 else jcc( assembler_cond, label) /*omit semi*/ |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7466 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7467 // Hacked jmp, which may only be used just before L_fallthrough. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7468 #define final_jmp(label) \ |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7469 if (&(label) == &L_fallthrough) { /*do nothing*/ } \ |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7470 else jmp(label) /*omit semi*/ |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7471 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7472 // If the pointers are equal, we are done (e.g., String[] elements). |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7473 // This self-check enables sharing of secondary supertype arrays among |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7474 // non-primary types such as array-of-interface. Otherwise, each such |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7475 // type would need its own customized SSA. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7476 // We move this check to the front of the fast path because many |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7477 // type checks are in fact trivially successful in this manner, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7478 // so we get a nicely predicted branch right at the start of the check. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7479 cmpptr(sub_klass, super_klass); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7480 local_jcc(Assembler::equal, *L_success); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7481 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7482 // Check the supertype display: |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7483 if (must_load_sco) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7484 // Positive movl does right thing on LP64. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7485 movl(temp_reg, super_check_offset_addr); |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
647
diff
changeset
|
7486 super_check_offset = RegisterOrConstant(temp_reg); |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7487 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7488 Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7489 cmpptr(super_klass, super_check_addr); // load displayed supertype |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7490 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7491 // This check has worked decisively for primary supers. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7492 // Secondary supers are sought in the super_cache ('super_cache_addr'). |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7493 // (Secondary supers are interfaces and very deeply nested subtypes.) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7494 // This works in the same check above because of a tricky aliasing |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7495 // between the super_cache and the primary super display elements. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7496 // (The 'super_check_addr' can address either, as the case requires.) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7497 // Note that the cache is updated below if it does not help us find |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7498 // what we need immediately. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7499 // So if it was a primary super, we can just fail immediately. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7500 // Otherwise, it's the slow path for us (no success at this point). |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7501 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7502 if (super_check_offset.is_register()) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7503 local_jcc(Assembler::equal, *L_success); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7504 cmpl(super_check_offset.as_register(), sc_offset); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7505 if (L_failure == &L_fallthrough) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7506 local_jcc(Assembler::equal, *L_slow_path); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7507 } else { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7508 local_jcc(Assembler::notEqual, *L_failure); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7509 final_jmp(*L_slow_path); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7510 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7511 } else if (super_check_offset.as_constant() == sc_offset) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7512 // Need a slow path; fast failure is impossible. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7513 if (L_slow_path == &L_fallthrough) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7514 local_jcc(Assembler::equal, *L_success); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7515 } else { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7516 local_jcc(Assembler::notEqual, *L_slow_path); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7517 final_jmp(*L_success); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7518 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7519 } else { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7520 // No slow path; it's a fast decision. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7521 if (L_failure == &L_fallthrough) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7522 local_jcc(Assembler::equal, *L_success); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7523 } else { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7524 local_jcc(Assembler::notEqual, *L_failure); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7525 final_jmp(*L_success); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7526 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7527 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7528 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7529 bind(L_fallthrough); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7530 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7531 #undef local_jcc |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7532 #undef final_jmp |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7533 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7534 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7535 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7536 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7537 Register super_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7538 Register temp_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7539 Register temp2_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7540 Label* L_success, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7541 Label* L_failure, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7542 bool set_cond_codes) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7543 assert_different_registers(sub_klass, super_klass, temp_reg); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7544 if (temp2_reg != noreg) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7545 assert_different_registers(sub_klass, super_klass, temp_reg, temp2_reg); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7546 #define IS_A_TEMP(reg) ((reg) == temp_reg || (reg) == temp2_reg) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7547 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7548 Label L_fallthrough; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7549 int label_nulls = 0; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7550 if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7551 if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7552 assert(label_nulls <= 1, "at most one NULL in the batch"); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7553 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7554 // a couple of useful fields in sub_klass: |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7555 int ss_offset = (klassOopDesc::header_size() * HeapWordSize + |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7556 Klass::secondary_supers_offset_in_bytes()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7557 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7558 Klass::secondary_super_cache_offset_in_bytes()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7559 Address secondary_supers_addr(sub_klass, ss_offset); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7560 Address super_cache_addr( sub_klass, sc_offset); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7561 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7562 // Do a linear scan of the secondary super-klass chain. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7563 // This code is rarely used, so simplicity is a virtue here. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7564 // The repne_scan instruction uses fixed registers, which we must spill. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7565 // Don't worry too much about pre-existing connections with the input regs. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7566 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7567 assert(sub_klass != rax, "killed reg"); // killed by mov(rax, super) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7568 assert(sub_klass != rcx, "killed reg"); // killed by lea(rcx, &pst_counter) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7569 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7570 // Get super_klass value into rax (even if it was in rdi or rcx). |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7571 bool pushed_rax = false, pushed_rcx = false, pushed_rdi = false; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7572 if (super_klass != rax || UseCompressedOops) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7573 if (!IS_A_TEMP(rax)) { push(rax); pushed_rax = true; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7574 mov(rax, super_klass); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7575 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7576 if (!IS_A_TEMP(rcx)) { push(rcx); pushed_rcx = true; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7577 if (!IS_A_TEMP(rdi)) { push(rdi); pushed_rdi = true; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7578 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7579 #ifndef PRODUCT |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7580 int* pst_counter = &SharedRuntime::_partial_subtype_ctr; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7581 ExternalAddress pst_counter_addr((address) pst_counter); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7582 NOT_LP64( incrementl(pst_counter_addr) ); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7583 LP64_ONLY( lea(rcx, pst_counter_addr) ); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7584 LP64_ONLY( incrementl(Address(rcx, 0)) ); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7585 #endif //PRODUCT |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7586 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7587 // We will consult the secondary-super array. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7588 movptr(rdi, secondary_supers_addr); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7589 // Load the array length. (Positive movl does right thing on LP64.) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7590 movl(rcx, Address(rdi, arrayOopDesc::length_offset_in_bytes())); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7591 // Skip to start of data. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7592 addptr(rdi, arrayOopDesc::base_offset_in_bytes(T_OBJECT)); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7593 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7594 // Scan RCX words at [RDI] for an occurrence of RAX. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7595 // Set NZ/Z based on last compare. |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
7596 // Z flag value will not be set by 'repne' if RCX == 0 since 'repne' does |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
7597 // not change flags (only scas instruction which is repeated sets flags). |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
7598 // Set Z = 0 (not equal) before 'repne' to indicate that class was not found. |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7599 #ifdef _LP64 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7600 // This part is tricky, as values in supers array could be 32 or 64 bit wide |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7601 // and we store values in objArrays always encoded, thus we need to encode |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7602 // the value of rax before repne. Note that rax is dead after the repne. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7603 if (UseCompressedOops) { |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
7604 encode_heap_oop_not_null(rax); // Changes flags. |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7605 // The superclass is never null; it would be a basic system error if a null |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7606 // pointer were to sneak in here. Note that we have already loaded the |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7607 // Klass::super_check_offset from the super_klass in the fast path, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7608 // so if there is a null in that register, we are already in the afterlife. |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
7609 testl(rax,rax); // Set Z = 0 |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7610 repne_scanl(); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7611 } else |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7612 #endif // _LP64 |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
7613 { |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
7614 testptr(rax,rax); // Set Z = 0 |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7615 repne_scan(); |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
7616 } |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7617 // Unspill the temp. registers: |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7618 if (pushed_rdi) pop(rdi); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7619 if (pushed_rcx) pop(rcx); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7620 if (pushed_rax) pop(rax); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7621 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7622 if (set_cond_codes) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7623 // Special hack for the AD files: rdi is guaranteed non-zero. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7624 assert(!pushed_rdi, "rdi must be left non-NULL"); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7625 // Also, the condition codes are properly set Z/NZ on succeed/failure. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7626 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7627 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7628 if (L_failure == &L_fallthrough) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7629 jccb(Assembler::notEqual, *L_failure); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7630 else jcc(Assembler::notEqual, *L_failure); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7631 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7632 // Success. Cache the super we found and proceed in triumph. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7633 movptr(super_cache_addr, super_klass); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7634 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7635 if (L_success != &L_fallthrough) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7636 jmp(*L_success); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7637 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7638 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7639 #undef IS_A_TEMP |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7640 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7641 bind(L_fallthrough); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7642 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7643 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
643
diff
changeset
|
7644 |
304 | 7645 void MacroAssembler::ucomisd(XMMRegister dst, AddressLiteral src) { |
7646 ucomisd(dst, as_Address(src)); | |
7647 } | |
7648 | |
7649 void MacroAssembler::ucomiss(XMMRegister dst, AddressLiteral src) { | |
7650 ucomiss(dst, as_Address(src)); | |
7651 } | |
7652 | |
7653 void MacroAssembler::xorpd(XMMRegister dst, AddressLiteral src) { | |
7654 if (reachable(src)) { | |
7655 xorpd(dst, as_Address(src)); | |
7656 } else { | |
7657 lea(rscratch1, src); | |
7658 xorpd(dst, Address(rscratch1, 0)); | |
7659 } | |
7660 } | |
7661 | |
7662 void MacroAssembler::xorps(XMMRegister dst, AddressLiteral src) { | |
7663 if (reachable(src)) { | |
7664 xorps(dst, as_Address(src)); | |
7665 } else { | |
7666 lea(rscratch1, src); | |
7667 xorps(dst, Address(rscratch1, 0)); | |
7668 } | |
7669 } | |
7670 | |
0 | 7671 void MacroAssembler::verify_oop(Register reg, const char* s) { |
7672 if (!VerifyOops) return; | |
304 | 7673 |
0 | 7674 // Pass register number to verify_oop_subroutine |
7675 char* b = new char[strlen(s) + 50]; | |
7676 sprintf(b, "verify_oop: %s: %s", reg->name(), s); | |
1583 | 7677 #ifdef _LP64 |
7678 push(rscratch1); // save r10, trashed by movptr() | |
7679 #endif | |
304 | 7680 push(rax); // save rax, |
7681 push(reg); // pass register argument | |
0 | 7682 ExternalAddress buffer((address) b); |
304 | 7683 // avoid using pushptr, as it modifies scratch registers |
7684 // and our contract is not to modify anything | |
7685 movptr(rax, buffer.addr()); | |
7686 push(rax); | |
0 | 7687 // call indirectly to solve generation ordering problem |
7688 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address())); | |
7689 call(rax); | |
1583 | 7690 // Caller pops the arguments (oop, message) and restores rax, r10 |
0 | 7691 } |
7692 | |
7693 | |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
647
diff
changeset
|
7694 RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_addr, |
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
647
diff
changeset
|
7695 Register tmp, |
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
647
diff
changeset
|
7696 int offset) { |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7697 intptr_t value = *delayed_value_addr; |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7698 if (value != 0) |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
647
diff
changeset
|
7699 return RegisterOrConstant(value + offset); |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7700 |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7701 // load indirectly to solve generation ordering problem |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7702 movptr(tmp, ExternalAddress((address) delayed_value_addr)); |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7703 |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7704 #ifdef ASSERT |
1793
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7705 { Label L; |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7706 testptr(tmp, tmp); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7707 if (WizardMode) { |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7708 jcc(Assembler::notZero, L); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7709 char* buf = new char[40]; |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7710 sprintf(buf, "DelayedValue="INTPTR_FORMAT, delayed_value_addr[1]); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7711 stop(buf); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7712 } else { |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7713 jccb(Assembler::notZero, L); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7714 hlt(); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7715 } |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7716 bind(L); |
d257356e35f0
6939224: MethodHandle.invokeGeneric needs to perform the correct set of conversions
jrose
parents:
1763
diff
changeset
|
7717 } |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7718 #endif |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7719 |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7720 if (offset != 0) |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7721 addptr(tmp, offset); |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7722 |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
647
diff
changeset
|
7723 return RegisterOrConstant(tmp); |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7724 } |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7725 |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
606
diff
changeset
|
7726 |
710 | 7727 // registers on entry: |
7728 // - rax ('check' register): required MethodType | |
7729 // - rcx: method handle | |
7730 // - rdx, rsi, or ?: killable temp | |
7731 void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg, | |
7732 Register temp_reg, | |
7733 Label& wrong_method_type) { | |
1846 | 7734 Address type_addr(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg)); |
710 | 7735 // compare method type against that of the receiver |
1846 | 7736 if (UseCompressedOops) { |
7737 load_heap_oop(temp_reg, type_addr); | |
7738 cmpptr(mtype_reg, temp_reg); | |
7739 } else { | |
7740 cmpptr(mtype_reg, type_addr); | |
7741 } | |
710 | 7742 jcc(Assembler::notEqual, wrong_method_type); |
7743 } | |
7744 | |
7745 | |
7746 // A method handle has a "vmslots" field which gives the size of its | |
7747 // argument list in JVM stack slots. This field is either located directly | |
7748 // in every method handle, or else is indirectly accessed through the | |
7749 // method handle's MethodType. This macro hides the distinction. | |
7750 void MacroAssembler::load_method_handle_vmslots(Register vmslots_reg, Register mh_reg, | |
7751 Register temp_reg) { | |
1503 | 7752 assert_different_registers(vmslots_reg, mh_reg, temp_reg); |
710 | 7753 // load mh.type.form.vmslots |
7754 if (java_dyn_MethodHandle::vmslots_offset_in_bytes() != 0) { | |
7755 // hoist vmslots into every mh to avoid dependent load chain | |
7756 movl(vmslots_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmslots_offset_in_bytes, temp_reg))); | |
7757 } else { | |
7758 Register temp2_reg = vmslots_reg; | |
1846 | 7759 load_heap_oop(temp2_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::type_offset_in_bytes, temp_reg))); |
7760 load_heap_oop(temp2_reg, Address(temp2_reg, delayed_value(java_dyn_MethodType::form_offset_in_bytes, temp_reg))); | |
710 | 7761 movl(vmslots_reg, Address(temp2_reg, delayed_value(java_dyn_MethodTypeForm::vmslots_offset_in_bytes, temp_reg))); |
7762 } | |
7763 } | |
7764 | |
7765 | |
7766 // registers on entry: | |
7767 // - rcx: method handle | |
7768 // - rdx: killable temp (interpreted only) | |
7769 // - rax: killable temp (compiled only) | |
7770 void MacroAssembler::jump_to_method_handle_entry(Register mh_reg, Register temp_reg) { | |
7771 assert(mh_reg == rcx, "caller must put MH object in rcx"); | |
7772 assert_different_registers(mh_reg, temp_reg); | |
7773 | |
7774 // pick out the interpreted side of the handler | |
1846 | 7775 // NOTE: vmentry is not an oop! |
710 | 7776 movptr(temp_reg, Address(mh_reg, delayed_value(java_dyn_MethodHandle::vmentry_offset_in_bytes, temp_reg))); |
7777 | |
7778 // off we go... | |
7779 jmp(Address(temp_reg, MethodHandleEntry::from_interpreted_entry_offset_in_bytes())); | |
7780 | |
7781 // for the various stubs which take control at this point, | |
7782 // see MethodHandles::generate_method_handle_stub | |
7783 } | |
7784 | |
7785 | |
7786 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot, | |
7787 int extra_slot_offset) { | |
7788 // cf. TemplateTable::prepare_invoke(), if (load_receiver). | |
1506 | 7789 int stackElementSize = Interpreter::stackElementSize; |
710 | 7790 int offset = Interpreter::expr_offset_in_bytes(extra_slot_offset+0); |
7791 #ifdef ASSERT | |
7792 int offset1 = Interpreter::expr_offset_in_bytes(extra_slot_offset+1); | |
7793 assert(offset1 - offset == stackElementSize, "correct arithmetic"); | |
7794 #endif | |
7795 Register scale_reg = noreg; | |
7796 Address::ScaleFactor scale_factor = Address::no_scale; | |
7797 if (arg_slot.is_constant()) { | |
7798 offset += arg_slot.as_constant() * stackElementSize; | |
7799 } else { | |
7800 scale_reg = arg_slot.as_register(); | |
7801 scale_factor = Address::times(stackElementSize); | |
7802 } | |
7803 offset += wordSize; // return PC is on stack | |
7804 return Address(rsp, scale_reg, scale_factor, offset); | |
7805 } | |
7806 | |
7807 | |
0 | 7808 void MacroAssembler::verify_oop_addr(Address addr, const char* s) { |
7809 if (!VerifyOops) return; | |
304 | 7810 |
0 | 7811 // Address adjust(addr.base(), addr.index(), addr.scale(), addr.disp() + BytesPerWord); |
7812 // Pass register number to verify_oop_subroutine | |
7813 char* b = new char[strlen(s) + 50]; | |
7814 sprintf(b, "verify_oop_addr: %s", s); | |
304 | 7815 |
1583 | 7816 #ifdef _LP64 |
7817 push(rscratch1); // save r10, trashed by movptr() | |
7818 #endif | |
304 | 7819 push(rax); // save rax, |
0 | 7820 // addr may contain rsp so we will have to adjust it based on the push |
7821 // we just did | |
304 | 7822 // NOTE: 64bit seemed to have had a bug in that it did movq(addr, rax); which |
7823 // stores rax into addr which is backwards of what was intended. | |
0 | 7824 if (addr.uses(rsp)) { |
304 | 7825 lea(rax, addr); |
7826 pushptr(Address(rax, BytesPerWord)); | |
0 | 7827 } else { |
304 | 7828 pushptr(addr); |
7829 } | |
7830 | |
0 | 7831 ExternalAddress buffer((address) b); |
7832 // pass msg argument | |
304 | 7833 // avoid using pushptr, as it modifies scratch registers |
7834 // and our contract is not to modify anything | |
7835 movptr(rax, buffer.addr()); | |
7836 push(rax); | |
7837 | |
0 | 7838 // call indirectly to solve generation ordering problem |
7839 movptr(rax, ExternalAddress(StubRoutines::verify_oop_subroutine_entry_address())); | |
7840 call(rax); | |
1583 | 7841 // Caller pops the arguments (addr, message) and restores rax, r10. |
0 | 7842 } |
7843 | |
304 | 7844 void MacroAssembler::verify_tlab() { |
7845 #ifdef ASSERT | |
7846 if (UseTLAB && VerifyOops) { | |
7847 Label next, ok; | |
7848 Register t1 = rsi; | |
7849 Register thread_reg = NOT_LP64(rbx) LP64_ONLY(r15_thread); | |
7850 | |
7851 push(t1); | |
7852 NOT_LP64(push(thread_reg)); | |
7853 NOT_LP64(get_thread(thread_reg)); | |
7854 | |
7855 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); | |
7856 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_start_offset()))); | |
7857 jcc(Assembler::aboveEqual, next); | |
7858 stop("assert(top >= start)"); | |
7859 should_not_reach_here(); | |
7860 | |
7861 bind(next); | |
7862 movptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_end_offset()))); | |
7863 cmpptr(t1, Address(thread_reg, in_bytes(JavaThread::tlab_top_offset()))); | |
7864 jcc(Assembler::aboveEqual, ok); | |
7865 stop("assert(top <= end)"); | |
7866 should_not_reach_here(); | |
7867 | |
7868 bind(ok); | |
7869 NOT_LP64(pop(thread_reg)); | |
7870 pop(t1); | |
7871 } | |
7872 #endif | |
7873 } | |
0 | 7874 |
7875 class ControlWord { | |
7876 public: | |
7877 int32_t _value; | |
7878 | |
7879 int rounding_control() const { return (_value >> 10) & 3 ; } | |
7880 int precision_control() const { return (_value >> 8) & 3 ; } | |
7881 bool precision() const { return ((_value >> 5) & 1) != 0; } | |
7882 bool underflow() const { return ((_value >> 4) & 1) != 0; } | |
7883 bool overflow() const { return ((_value >> 3) & 1) != 0; } | |
7884 bool zero_divide() const { return ((_value >> 2) & 1) != 0; } | |
7885 bool denormalized() const { return ((_value >> 1) & 1) != 0; } | |
7886 bool invalid() const { return ((_value >> 0) & 1) != 0; } | |
7887 | |
7888 void print() const { | |
7889 // rounding control | |
7890 const char* rc; | |
7891 switch (rounding_control()) { | |
7892 case 0: rc = "round near"; break; | |
7893 case 1: rc = "round down"; break; | |
7894 case 2: rc = "round up "; break; | |
7895 case 3: rc = "chop "; break; | |
7896 }; | |
7897 // precision control | |
7898 const char* pc; | |
7899 switch (precision_control()) { | |
7900 case 0: pc = "24 bits "; break; | |
7901 case 1: pc = "reserved"; break; | |
7902 case 2: pc = "53 bits "; break; | |
7903 case 3: pc = "64 bits "; break; | |
7904 }; | |
7905 // flags | |
7906 char f[9]; | |
7907 f[0] = ' '; | |
7908 f[1] = ' '; | |
7909 f[2] = (precision ()) ? 'P' : 'p'; | |
7910 f[3] = (underflow ()) ? 'U' : 'u'; | |
7911 f[4] = (overflow ()) ? 'O' : 'o'; | |
7912 f[5] = (zero_divide ()) ? 'Z' : 'z'; | |
7913 f[6] = (denormalized()) ? 'D' : 'd'; | |
7914 f[7] = (invalid ()) ? 'I' : 'i'; | |
7915 f[8] = '\x0'; | |
7916 // output | |
7917 printf("%04x masks = %s, %s, %s", _value & 0xFFFF, f, rc, pc); | |
7918 } | |
7919 | |
7920 }; | |
7921 | |
7922 class StatusWord { | |
7923 public: | |
7924 int32_t _value; | |
7925 | |
7926 bool busy() const { return ((_value >> 15) & 1) != 0; } | |
7927 bool C3() const { return ((_value >> 14) & 1) != 0; } | |
7928 bool C2() const { return ((_value >> 10) & 1) != 0; } | |
7929 bool C1() const { return ((_value >> 9) & 1) != 0; } | |
7930 bool C0() const { return ((_value >> 8) & 1) != 0; } | |
7931 int top() const { return (_value >> 11) & 7 ; } | |
7932 bool error_status() const { return ((_value >> 7) & 1) != 0; } | |
7933 bool stack_fault() const { return ((_value >> 6) & 1) != 0; } | |
7934 bool precision() const { return ((_value >> 5) & 1) != 0; } | |
7935 bool underflow() const { return ((_value >> 4) & 1) != 0; } | |
7936 bool overflow() const { return ((_value >> 3) & 1) != 0; } | |
7937 bool zero_divide() const { return ((_value >> 2) & 1) != 0; } | |
7938 bool denormalized() const { return ((_value >> 1) & 1) != 0; } | |
7939 bool invalid() const { return ((_value >> 0) & 1) != 0; } | |
7940 | |
7941 void print() const { | |
7942 // condition codes | |
7943 char c[5]; | |
7944 c[0] = (C3()) ? '3' : '-'; | |
7945 c[1] = (C2()) ? '2' : '-'; | |
7946 c[2] = (C1()) ? '1' : '-'; | |
7947 c[3] = (C0()) ? '0' : '-'; | |
7948 c[4] = '\x0'; | |
7949 // flags | |
7950 char f[9]; | |
7951 f[0] = (error_status()) ? 'E' : '-'; | |
7952 f[1] = (stack_fault ()) ? 'S' : '-'; | |
7953 f[2] = (precision ()) ? 'P' : '-'; | |
7954 f[3] = (underflow ()) ? 'U' : '-'; | |
7955 f[4] = (overflow ()) ? 'O' : '-'; | |
7956 f[5] = (zero_divide ()) ? 'Z' : '-'; | |
7957 f[6] = (denormalized()) ? 'D' : '-'; | |
7958 f[7] = (invalid ()) ? 'I' : '-'; | |
7959 f[8] = '\x0'; | |
7960 // output | |
7961 printf("%04x flags = %s, cc = %s, top = %d", _value & 0xFFFF, f, c, top()); | |
7962 } | |
7963 | |
7964 }; | |
7965 | |
7966 class TagWord { | |
7967 public: | |
7968 int32_t _value; | |
7969 | |
7970 int tag_at(int i) const { return (_value >> (i*2)) & 3; } | |
7971 | |
7972 void print() const { | |
7973 printf("%04x", _value & 0xFFFF); | |
7974 } | |
7975 | |
7976 }; | |
7977 | |
7978 class FPU_Register { | |
7979 public: | |
7980 int32_t _m0; | |
7981 int32_t _m1; | |
7982 int16_t _ex; | |
7983 | |
7984 bool is_indefinite() const { | |
7985 return _ex == -1 && _m1 == (int32_t)0xC0000000 && _m0 == 0; | |
7986 } | |
7987 | |
7988 void print() const { | |
7989 char sign = (_ex < 0) ? '-' : '+'; | |
7990 const char* kind = (_ex == 0x7FFF || _ex == (int16_t)-1) ? "NaN" : " "; | |
7991 printf("%c%04hx.%08x%08x %s", sign, _ex, _m1, _m0, kind); | |
7992 }; | |
7993 | |
7994 }; | |
7995 | |
7996 class FPU_State { | |
7997 public: | |
7998 enum { | |
7999 register_size = 10, | |
8000 number_of_registers = 8, | |
8001 register_mask = 7 | |
8002 }; | |
8003 | |
8004 ControlWord _control_word; | |
8005 StatusWord _status_word; | |
8006 TagWord _tag_word; | |
8007 int32_t _error_offset; | |
8008 int32_t _error_selector; | |
8009 int32_t _data_offset; | |
8010 int32_t _data_selector; | |
8011 int8_t _register[register_size * number_of_registers]; | |
8012 | |
8013 int tag_for_st(int i) const { return _tag_word.tag_at((_status_word.top() + i) & register_mask); } | |
8014 FPU_Register* st(int i) const { return (FPU_Register*)&_register[register_size * i]; } | |
8015 | |
8016 const char* tag_as_string(int tag) const { | |
8017 switch (tag) { | |
8018 case 0: return "valid"; | |
8019 case 1: return "zero"; | |
8020 case 2: return "special"; | |
8021 case 3: return "empty"; | |
8022 } | |
1489
cff162798819
6888953: some calls to function-like macros are missing semicolons
jcoomes
parents:
1369
diff
changeset
|
8023 ShouldNotReachHere(); |
0 | 8024 return NULL; |
8025 } | |
8026 | |
8027 void print() const { | |
8028 // print computation registers | |
8029 { int t = _status_word.top(); | |
8030 for (int i = 0; i < number_of_registers; i++) { | |
8031 int j = (i - t) & register_mask; | |
8032 printf("%c r%d = ST%d = ", (j == 0 ? '*' : ' '), i, j); | |
8033 st(j)->print(); | |
8034 printf(" %s\n", tag_as_string(_tag_word.tag_at(i))); | |
8035 } | |
8036 } | |
8037 printf("\n"); | |
8038 // print control registers | |
8039 printf("ctrl = "); _control_word.print(); printf("\n"); | |
8040 printf("stat = "); _status_word .print(); printf("\n"); | |
8041 printf("tags = "); _tag_word .print(); printf("\n"); | |
8042 } | |
8043 | |
8044 }; | |
8045 | |
8046 class Flag_Register { | |
8047 public: | |
8048 int32_t _value; | |
8049 | |
8050 bool overflow() const { return ((_value >> 11) & 1) != 0; } | |
8051 bool direction() const { return ((_value >> 10) & 1) != 0; } | |
8052 bool sign() const { return ((_value >> 7) & 1) != 0; } | |
8053 bool zero() const { return ((_value >> 6) & 1) != 0; } | |
8054 bool auxiliary_carry() const { return ((_value >> 4) & 1) != 0; } | |
8055 bool parity() const { return ((_value >> 2) & 1) != 0; } | |
8056 bool carry() const { return ((_value >> 0) & 1) != 0; } | |
8057 | |
8058 void print() const { | |
8059 // flags | |
8060 char f[8]; | |
8061 f[0] = (overflow ()) ? 'O' : '-'; | |
8062 f[1] = (direction ()) ? 'D' : '-'; | |
8063 f[2] = (sign ()) ? 'S' : '-'; | |
8064 f[3] = (zero ()) ? 'Z' : '-'; | |
8065 f[4] = (auxiliary_carry()) ? 'A' : '-'; | |
8066 f[5] = (parity ()) ? 'P' : '-'; | |
8067 f[6] = (carry ()) ? 'C' : '-'; | |
8068 f[7] = '\x0'; | |
8069 // output | |
8070 printf("%08x flags = %s", _value, f); | |
8071 } | |
8072 | |
8073 }; | |
8074 | |
8075 class IU_Register { | |
8076 public: | |
8077 int32_t _value; | |
8078 | |
8079 void print() const { | |
8080 printf("%08x %11d", _value, _value); | |
8081 } | |
8082 | |
8083 }; | |
8084 | |
8085 class IU_State { | |
8086 public: | |
8087 Flag_Register _eflags; | |
8088 IU_Register _rdi; | |
8089 IU_Register _rsi; | |
8090 IU_Register _rbp; | |
8091 IU_Register _rsp; | |
8092 IU_Register _rbx; | |
8093 IU_Register _rdx; | |
8094 IU_Register _rcx; | |
8095 IU_Register _rax; | |
8096 | |
8097 void print() const { | |
8098 // computation registers | |
8099 printf("rax, = "); _rax.print(); printf("\n"); | |
8100 printf("rbx, = "); _rbx.print(); printf("\n"); | |
8101 printf("rcx = "); _rcx.print(); printf("\n"); | |
8102 printf("rdx = "); _rdx.print(); printf("\n"); | |
8103 printf("rdi = "); _rdi.print(); printf("\n"); | |
8104 printf("rsi = "); _rsi.print(); printf("\n"); | |
8105 printf("rbp, = "); _rbp.print(); printf("\n"); | |
8106 printf("rsp = "); _rsp.print(); printf("\n"); | |
8107 printf("\n"); | |
8108 // control registers | |
8109 printf("flgs = "); _eflags.print(); printf("\n"); | |
8110 } | |
8111 }; | |
8112 | |
8113 | |
8114 class CPU_State { | |
8115 public: | |
8116 FPU_State _fpu_state; | |
8117 IU_State _iu_state; | |
8118 | |
8119 void print() const { | |
8120 printf("--------------------------------------------------\n"); | |
8121 _iu_state .print(); | |
8122 printf("\n"); | |
8123 _fpu_state.print(); | |
8124 printf("--------------------------------------------------\n"); | |
8125 } | |
8126 | |
8127 }; | |
8128 | |
8129 | |
8130 static void _print_CPU_state(CPU_State* state) { | |
8131 state->print(); | |
8132 }; | |
8133 | |
8134 | |
8135 void MacroAssembler::print_CPU_state() { | |
8136 push_CPU_state(); | |
304 | 8137 push(rsp); // pass CPU state |
0 | 8138 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _print_CPU_state))); |
304 | 8139 addptr(rsp, wordSize); // discard argument |
0 | 8140 pop_CPU_state(); |
8141 } | |
8142 | |
8143 | |
8144 static bool _verify_FPU(int stack_depth, char* s, CPU_State* state) { | |
8145 static int counter = 0; | |
8146 FPU_State* fs = &state->_fpu_state; | |
8147 counter++; | |
8148 // For leaf calls, only verify that the top few elements remain empty. | |
8149 // We only need 1 empty at the top for C2 code. | |
8150 if( stack_depth < 0 ) { | |
8151 if( fs->tag_for_st(7) != 3 ) { | |
8152 printf("FPR7 not empty\n"); | |
8153 state->print(); | |
8154 assert(false, "error"); | |
8155 return false; | |
8156 } | |
8157 return true; // All other stack states do not matter | |
8158 } | |
8159 | |
8160 assert((fs->_control_word._value & 0xffff) == StubRoutines::_fpu_cntrl_wrd_std, | |
8161 "bad FPU control word"); | |
8162 | |
8163 // compute stack depth | |
8164 int i = 0; | |
8165 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) < 3) i++; | |
8166 int d = i; | |
8167 while (i < FPU_State::number_of_registers && fs->tag_for_st(i) == 3) i++; | |
8168 // verify findings | |
8169 if (i != FPU_State::number_of_registers) { | |
8170 // stack not contiguous | |
8171 printf("%s: stack not contiguous at ST%d\n", s, i); | |
8172 state->print(); | |
8173 assert(false, "error"); | |
8174 return false; | |
8175 } | |
8176 // check if computed stack depth corresponds to expected stack depth | |
8177 if (stack_depth < 0) { | |
8178 // expected stack depth is -stack_depth or less | |
8179 if (d > -stack_depth) { | |
8180 // too many elements on the stack | |
8181 printf("%s: <= %d stack elements expected but found %d\n", s, -stack_depth, d); | |
8182 state->print(); | |
8183 assert(false, "error"); | |
8184 return false; | |
8185 } | |
8186 } else { | |
8187 // expected stack depth is stack_depth | |
8188 if (d != stack_depth) { | |
8189 // wrong stack depth | |
8190 printf("%s: %d stack elements expected but found %d\n", s, stack_depth, d); | |
8191 state->print(); | |
8192 assert(false, "error"); | |
8193 return false; | |
8194 } | |
8195 } | |
8196 // everything is cool | |
8197 return true; | |
8198 } | |
8199 | |
8200 | |
8201 void MacroAssembler::verify_FPU(int stack_depth, const char* s) { | |
8202 if (!VerifyFPU) return; | |
8203 push_CPU_state(); | |
304 | 8204 push(rsp); // pass CPU state |
0 | 8205 ExternalAddress msg((address) s); |
8206 // pass message string s | |
8207 pushptr(msg.addr()); | |
304 | 8208 push(stack_depth); // pass stack depth |
0 | 8209 call(RuntimeAddress(CAST_FROM_FN_PTR(address, _verify_FPU))); |
304 | 8210 addptr(rsp, 3 * wordSize); // discard arguments |
0 | 8211 // check for error |
8212 { Label L; | |
8213 testl(rax, rax); | |
8214 jcc(Assembler::notZero, L); | |
8215 int3(); // break if error condition | |
8216 bind(L); | |
8217 } | |
8218 pop_CPU_state(); | |
8219 } | |
8220 | |
304 | 8221 void MacroAssembler::load_klass(Register dst, Register src) { |
8222 #ifdef _LP64 | |
8223 if (UseCompressedOops) { | |
8224 movl(dst, Address(src, oopDesc::klass_offset_in_bytes())); | |
8225 decode_heap_oop_not_null(dst); | |
8226 } else | |
8227 #endif | |
8228 movptr(dst, Address(src, oopDesc::klass_offset_in_bytes())); | |
8229 } | |
8230 | |
8231 void MacroAssembler::load_prototype_header(Register dst, Register src) { | |
8232 #ifdef _LP64 | |
8233 if (UseCompressedOops) { | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8234 assert (Universe::heap() != NULL, "java heap should be initialized"); |
304 | 8235 movl(dst, Address(src, oopDesc::klass_offset_in_bytes())); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8236 if (Universe::narrow_oop_shift() != 0) { |
1571
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8237 assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8238 if (LogMinObjAlignmentInBytes == Address::times_8) { |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8239 movq(dst, Address(r12_heapbase, dst, Address::times_8, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8240 } else { |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8241 // OK to use shift since we don't need to preserve flags. |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8242 shlq(dst, LogMinObjAlignmentInBytes); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8243 movq(dst, Address(r12_heapbase, dst, Address::times_1, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8244 } |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8245 } else { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8246 movq(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8247 } |
304 | 8248 } else |
8249 #endif | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8250 { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8251 movptr(dst, Address(src, oopDesc::klass_offset_in_bytes())); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8252 movptr(dst, Address(dst, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes())); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8253 } |
304 | 8254 } |
8255 | |
8256 void MacroAssembler::store_klass(Register dst, Register src) { | |
8257 #ifdef _LP64 | |
8258 if (UseCompressedOops) { | |
8259 encode_heap_oop_not_null(src); | |
8260 movl(Address(dst, oopDesc::klass_offset_in_bytes()), src); | |
8261 } else | |
8262 #endif | |
8263 movptr(Address(dst, oopDesc::klass_offset_in_bytes()), src); | |
8264 } | |
8265 | |
1846 | 8266 void MacroAssembler::load_heap_oop(Register dst, Address src) { |
8267 #ifdef _LP64 | |
8268 if (UseCompressedOops) { | |
8269 movl(dst, src); | |
8270 decode_heap_oop(dst); | |
8271 } else | |
8272 #endif | |
8273 movptr(dst, src); | |
8274 } | |
8275 | |
8276 void MacroAssembler::store_heap_oop(Address dst, Register src) { | |
8277 #ifdef _LP64 | |
8278 if (UseCompressedOops) { | |
8279 assert(!dst.uses(src), "not enough registers"); | |
8280 encode_heap_oop(src); | |
8281 movl(dst, src); | |
8282 } else | |
8283 #endif | |
8284 movptr(dst, src); | |
8285 } | |
8286 | |
8287 // Used for storing NULLs. | |
8288 void MacroAssembler::store_heap_oop_null(Address dst) { | |
8289 #ifdef _LP64 | |
8290 if (UseCompressedOops) { | |
8291 movl(dst, (int32_t)NULL_WORD); | |
8292 } else { | |
8293 movslq(dst, (int32_t)NULL_WORD); | |
8294 } | |
8295 #else | |
8296 movl(dst, (int32_t)NULL_WORD); | |
8297 #endif | |
8298 } | |
8299 | |
304 | 8300 #ifdef _LP64 |
8301 void MacroAssembler::store_klass_gap(Register dst, Register src) { | |
8302 if (UseCompressedOops) { | |
8303 // Store to klass gap in destination | |
8304 movl(Address(dst, oopDesc::klass_gap_offset_in_bytes()), src); | |
8305 } | |
8306 } | |
8307 | |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8308 #ifdef ASSERT |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8309 void MacroAssembler::verify_heapbase(const char* msg) { |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8310 assert (UseCompressedOops, "should be compressed"); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8311 assert (Universe::heap() != NULL, "java heap should be initialized"); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8312 if (CheckCompressedOops) { |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8313 Label ok; |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8314 push(rscratch1); // cmpptr trashes rscratch1 |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8315 cmpptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr())); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8316 jcc(Assembler::equal, ok); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8317 stop(msg); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8318 bind(ok); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8319 pop(rscratch1); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8320 } |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8321 } |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8322 #endif |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8323 |
304 | 8324 // Algorithm must match oop.inline.hpp encode_heap_oop. |
8325 void MacroAssembler::encode_heap_oop(Register r) { | |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8326 #ifdef ASSERT |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8327 verify_heapbase("MacroAssembler::encode_heap_oop: heap base corrupted?"); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8328 #endif |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8329 verify_oop(r, "broken oop in encode_heap_oop"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8330 if (Universe::narrow_oop_base() == NULL) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8331 if (Universe::narrow_oop_shift() != 0) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8332 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8333 shrq(r, LogMinObjAlignmentInBytes); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8334 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8335 return; |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8336 } |
304 | 8337 testq(r, r); |
8338 cmovq(Assembler::equal, r, r12_heapbase); | |
8339 subq(r, r12_heapbase); | |
8340 shrq(r, LogMinObjAlignmentInBytes); | |
8341 } | |
8342 | |
8343 void MacroAssembler::encode_heap_oop_not_null(Register r) { | |
0 | 8344 #ifdef ASSERT |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8345 verify_heapbase("MacroAssembler::encode_heap_oop_not_null: heap base corrupted?"); |
304 | 8346 if (CheckCompressedOops) { |
0 | 8347 Label ok; |
304 | 8348 testq(r, r); |
8349 jcc(Assembler::notEqual, ok); | |
8350 stop("null oop passed to encode_heap_oop_not_null"); | |
0 | 8351 bind(ok); |
304 | 8352 } |
8353 #endif | |
8354 verify_oop(r, "broken oop in encode_heap_oop_not_null"); | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8355 if (Universe::narrow_oop_base() != NULL) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8356 subq(r, r12_heapbase); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8357 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8358 if (Universe::narrow_oop_shift() != 0) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8359 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8360 shrq(r, LogMinObjAlignmentInBytes); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8361 } |
304 | 8362 } |
8363 | |
8364 void MacroAssembler::encode_heap_oop_not_null(Register dst, Register src) { | |
8365 #ifdef ASSERT | |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8366 verify_heapbase("MacroAssembler::encode_heap_oop_not_null2: heap base corrupted?"); |
304 | 8367 if (CheckCompressedOops) { |
8368 Label ok; | |
8369 testq(src, src); | |
8370 jcc(Assembler::notEqual, ok); | |
8371 stop("null oop passed to encode_heap_oop_not_null2"); | |
8372 bind(ok); | |
0 | 8373 } |
8374 #endif | |
304 | 8375 verify_oop(src, "broken oop in encode_heap_oop_not_null2"); |
8376 if (dst != src) { | |
8377 movq(dst, src); | |
8378 } | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8379 if (Universe::narrow_oop_base() != NULL) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8380 subq(dst, r12_heapbase); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8381 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8382 if (Universe::narrow_oop_shift() != 0) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8383 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8384 shrq(dst, LogMinObjAlignmentInBytes); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8385 } |
304 | 8386 } |
8387 | |
8388 void MacroAssembler::decode_heap_oop(Register r) { | |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8389 #ifdef ASSERT |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8390 verify_heapbase("MacroAssembler::decode_heap_oop: heap base corrupted?"); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8391 #endif |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8392 if (Universe::narrow_oop_base() == NULL) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8393 if (Universe::narrow_oop_shift() != 0) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8394 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8395 shlq(r, LogMinObjAlignmentInBytes); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8396 } |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8397 } else { |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8398 Label done; |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8399 shlq(r, LogMinObjAlignmentInBytes); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8400 jccb(Assembler::equal, done); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8401 addq(r, r12_heapbase); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8402 bind(done); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8403 } |
304 | 8404 verify_oop(r, "broken oop in decode_heap_oop"); |
8405 } | |
8406 | |
8407 void MacroAssembler::decode_heap_oop_not_null(Register r) { | |
1571
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8408 // Note: it will change flags |
304 | 8409 assert (UseCompressedOops, "should only be used for compressed headers"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8410 assert (Universe::heap() != NULL, "java heap should be initialized"); |
304 | 8411 // Cannot assert, unverified entry point counts instructions (see .ad file) |
8412 // vtableStubs also counts instructions in pd_code_size_limit. | |
8413 // Also do not verify_oop as this is called by verify_oop. | |
898
60fea60a6db5
6864914: SPECjvm2008 produces invalid result with zero based Compressed Oops
kvn
parents:
845
diff
changeset
|
8414 if (Universe::narrow_oop_shift() != 0) { |
1571
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8415 assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8416 shlq(r, LogMinObjAlignmentInBytes); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8417 if (Universe::narrow_oop_base() != NULL) { |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8418 addq(r, r12_heapbase); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8419 } |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8420 } else { |
898
60fea60a6db5
6864914: SPECjvm2008 produces invalid result with zero based Compressed Oops
kvn
parents:
845
diff
changeset
|
8421 assert (Universe::narrow_oop_base() == NULL, "sanity"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8422 } |
304 | 8423 } |
8424 | |
8425 void MacroAssembler::decode_heap_oop_not_null(Register dst, Register src) { | |
1571
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8426 // Note: it will change flags |
304 | 8427 assert (UseCompressedOops, "should only be used for compressed headers"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8428 assert (Universe::heap() != NULL, "java heap should be initialized"); |
304 | 8429 // Cannot assert, unverified entry point counts instructions (see .ad file) |
8430 // vtableStubs also counts instructions in pd_code_size_limit. | |
8431 // Also do not verify_oop as this is called by verify_oop. | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8432 if (Universe::narrow_oop_shift() != 0) { |
1571
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8433 assert(LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8434 if (LogMinObjAlignmentInBytes == Address::times_8) { |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8435 leaq(dst, Address(r12_heapbase, src, Address::times_8, 0)); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8436 } else { |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8437 if (dst != src) { |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8438 movq(dst, src); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8439 } |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8440 shlq(dst, LogMinObjAlignmentInBytes); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8441 if (Universe::narrow_oop_base() != NULL) { |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8442 addq(dst, r12_heapbase); |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8443 } |
2d127394260e
6916623: Align object to 16 bytes to use Compressed Oops with java heap up to 64Gb
kvn
parents:
1513
diff
changeset
|
8444 } |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8445 } else { |
898
60fea60a6db5
6864914: SPECjvm2008 produces invalid result with zero based Compressed Oops
kvn
parents:
845
diff
changeset
|
8446 assert (Universe::narrow_oop_base() == NULL, "sanity"); |
1684
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8447 if (dst != src) { |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8448 movq(dst, src); |
66c5dadb4d61
6973308: Missing zero length check before repne scas in check_klass_subtype_slow_path()
kvn
parents:
1583
diff
changeset
|
8449 } |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8450 } |
304 | 8451 } |
8452 | |
8453 void MacroAssembler::set_narrow_oop(Register dst, jobject obj) { | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8454 assert (UseCompressedOops, "should only be used for compressed headers"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8455 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8456 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8457 int oop_index = oop_recorder()->find_index(obj); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8458 RelocationHolder rspec = oop_Relocation::spec(oop_index); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8459 mov_narrow_oop(dst, oop_index, rspec); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8460 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8461 |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8462 void MacroAssembler::set_narrow_oop(Address dst, jobject obj) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8463 assert (UseCompressedOops, "should only be used for compressed headers"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8464 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8465 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); |
304 | 8466 int oop_index = oop_recorder()->find_index(obj); |
8467 RelocationHolder rspec = oop_Relocation::spec(oop_index); | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8468 mov_narrow_oop(dst, oop_index, rspec); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8469 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8470 |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8471 void MacroAssembler::cmp_narrow_oop(Register dst, jobject obj) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8472 assert (UseCompressedOops, "should only be used for compressed headers"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8473 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8474 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8475 int oop_index = oop_recorder()->find_index(obj); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8476 RelocationHolder rspec = oop_Relocation::spec(oop_index); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8477 Assembler::cmp_narrow_oop(dst, oop_index, rspec); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8478 } |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8479 |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8480 void MacroAssembler::cmp_narrow_oop(Address dst, jobject obj) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8481 assert (UseCompressedOops, "should only be used for compressed headers"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8482 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8483 assert (oop_recorder() != NULL, "this assembler needs an OopRecorder"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8484 int oop_index = oop_recorder()->find_index(obj); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8485 RelocationHolder rspec = oop_Relocation::spec(oop_index); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8486 Assembler::cmp_narrow_oop(dst, oop_index, rspec); |
304 | 8487 } |
8488 | |
8489 void MacroAssembler::reinit_heapbase() { | |
8490 if (UseCompressedOops) { | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
8491 movptr(r12_heapbase, ExternalAddress((address)Universe::narrow_oop_base_addr())); |
304 | 8492 } |
8493 } | |
8494 #endif // _LP64 | |
0 | 8495 |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8496 // IndexOf substring. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8497 void MacroAssembler::string_indexof(Register str1, Register str2, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8498 Register cnt1, Register cnt2, Register result, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8499 XMMRegister vec, Register tmp) { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8500 assert(UseSSE42Intrinsics, "SSE4.2 is required"); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8501 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8502 Label RELOAD_SUBSTR, PREP_FOR_SCAN, SCAN_TO_SUBSTR, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8503 SCAN_SUBSTR, RET_NOT_FOUND, CLEANUP; |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8504 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8505 push(str1); // string addr |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8506 push(str2); // substr addr |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8507 push(cnt2); // substr count |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8508 jmpb(PREP_FOR_SCAN); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8509 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8510 // Substr count saved at sp |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8511 // Substr saved at sp+1*wordSize |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8512 // String saved at sp+2*wordSize |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8513 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8514 // Reload substr for rescan |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8515 bind(RELOAD_SUBSTR); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8516 movl(cnt2, Address(rsp, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8517 movptr(str2, Address(rsp, wordSize)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8518 // We came here after the beginninig of the substring was |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8519 // matched but the rest of it was not so we need to search |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8520 // again. Start from the next element after the previous match. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8521 subptr(str1, result); // Restore counter |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8522 shrl(str1, 1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8523 addl(cnt1, str1); |
1302
2484f4d6a54e
6935535: String.indexOf() returns incorrect result on x86 with SSE4.2
kvn
parents:
1108
diff
changeset
|
8524 decrementl(cnt1); |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8525 lea(str1, Address(result, 2)); // Reload string |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8526 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8527 // Load substr |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8528 bind(PREP_FOR_SCAN); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8529 movdqu(vec, Address(str2, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8530 addl(cnt1, 8); // prime the loop |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8531 subptr(str1, 16); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8532 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8533 // Scan string for substr in 16-byte vectors |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8534 bind(SCAN_TO_SUBSTR); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8535 subl(cnt1, 8); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8536 addptr(str1, 16); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8537 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8538 // pcmpestri |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8539 // inputs: |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8540 // xmm - substring |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8541 // rax - substring length (elements count) |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8542 // mem - scaned string |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8543 // rdx - string length (elements count) |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8544 // 0xd - mode: 1100 (substring search) + 01 (unsigned shorts) |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8545 // outputs: |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8546 // rcx - matched index in string |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8547 assert(cnt1 == rdx && cnt2 == rax && tmp == rcx, "pcmpestri"); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8548 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8549 pcmpestri(vec, Address(str1, 0), 0x0d); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8550 jcc(Assembler::above, SCAN_TO_SUBSTR); // CF == 0 && ZF == 0 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8551 jccb(Assembler::aboveEqual, RET_NOT_FOUND); // CF == 0 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8552 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8553 // Fallthrough: found a potential substr |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8554 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8555 // Make sure string is still long enough |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8556 subl(cnt1, tmp); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8557 cmpl(cnt1, cnt2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8558 jccb(Assembler::negative, RET_NOT_FOUND); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8559 // Compute start addr of substr |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8560 lea(str1, Address(str1, tmp, Address::times_2)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8561 movptr(result, str1); // save |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8562 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8563 // Compare potential substr |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8564 addl(cnt1, 8); // prime the loop |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8565 addl(cnt2, 8); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8566 subptr(str1, 16); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8567 subptr(str2, 16); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8568 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8569 // Scan 16-byte vectors of string and substr |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8570 bind(SCAN_SUBSTR); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8571 subl(cnt1, 8); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8572 subl(cnt2, 8); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8573 addptr(str1, 16); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8574 addptr(str2, 16); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8575 movdqu(vec, Address(str2, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8576 pcmpestri(vec, Address(str1, 0), 0x0d); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8577 jcc(Assembler::noOverflow, RELOAD_SUBSTR); // OF == 0 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8578 jcc(Assembler::positive, SCAN_SUBSTR); // SF == 0 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8579 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8580 // Compute substr offset |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8581 subptr(result, Address(rsp, 2*wordSize)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8582 shrl(result, 1); // index |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8583 jmpb(CLEANUP); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8584 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8585 bind(RET_NOT_FOUND); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8586 movl(result, -1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8587 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8588 bind(CLEANUP); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8589 addptr(rsp, 3*wordSize); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8590 } |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8591 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8592 // Compare strings. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8593 void MacroAssembler::string_compare(Register str1, Register str2, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8594 Register cnt1, Register cnt2, Register result, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8595 XMMRegister vec1, XMMRegister vec2) { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8596 Label LENGTH_DIFF_LABEL, POP_LABEL, DONE_LABEL, WHILE_HEAD_LABEL; |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8597 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8598 // Compute the minimum of the string lengths and the |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8599 // difference of the string lengths (stack). |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8600 // Do the conditional move stuff |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8601 movl(result, cnt1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8602 subl(cnt1, cnt2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8603 push(cnt1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8604 if (VM_Version::supports_cmov()) { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8605 cmovl(Assembler::lessEqual, cnt2, result); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8606 } else { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8607 Label GT_LABEL; |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8608 jccb(Assembler::greater, GT_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8609 movl(cnt2, result); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8610 bind(GT_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8611 } |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8612 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8613 // Is the minimum length zero? |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8614 testl(cnt2, cnt2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8615 jcc(Assembler::zero, LENGTH_DIFF_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8616 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8617 // Load first characters |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8618 load_unsigned_short(result, Address(str1, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8619 load_unsigned_short(cnt1, Address(str2, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8620 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8621 // Compare first characters |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8622 subl(result, cnt1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8623 jcc(Assembler::notZero, POP_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8624 decrementl(cnt2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8625 jcc(Assembler::zero, LENGTH_DIFF_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8626 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8627 { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8628 // Check after comparing first character to see if strings are equivalent |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8629 Label LSkip2; |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8630 // Check if the strings start at same location |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8631 cmpptr(str1, str2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8632 jccb(Assembler::notEqual, LSkip2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8633 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8634 // Check if the length difference is zero (from stack) |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8635 cmpl(Address(rsp, 0), 0x0); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8636 jcc(Assembler::equal, LENGTH_DIFF_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8637 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8638 // Strings might not be equivalent |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8639 bind(LSkip2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8640 } |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8641 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8642 // Advance to next character |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8643 addptr(str1, 2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8644 addptr(str2, 2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8645 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8646 if (UseSSE42Intrinsics) { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8647 // With SSE4.2, use double quad vector compare |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8648 Label COMPARE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL; |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8649 // Setup to compare 16-byte vectors |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8650 movl(cnt1, cnt2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8651 andl(cnt2, 0xfffffff8); // cnt2 holds the vector count |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8652 andl(cnt1, 0x00000007); // cnt1 holds the tail count |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8653 testl(cnt2, cnt2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8654 jccb(Assembler::zero, COMPARE_TAIL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8655 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8656 lea(str2, Address(str2, cnt2, Address::times_2)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8657 lea(str1, Address(str1, cnt2, Address::times_2)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8658 negptr(cnt2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8659 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8660 bind(COMPARE_VECTORS); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8661 movdqu(vec1, Address(str1, cnt2, Address::times_2)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8662 movdqu(vec2, Address(str2, cnt2, Address::times_2)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8663 pxor(vec1, vec2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8664 ptest(vec1, vec1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8665 jccb(Assembler::notZero, VECTOR_NOT_EQUAL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8666 addptr(cnt2, 8); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8667 jcc(Assembler::notZero, COMPARE_VECTORS); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8668 jmpb(COMPARE_TAIL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8669 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8670 // Mismatched characters in the vectors |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8671 bind(VECTOR_NOT_EQUAL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8672 lea(str1, Address(str1, cnt2, Address::times_2)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8673 lea(str2, Address(str2, cnt2, Address::times_2)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8674 movl(cnt1, 8); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8675 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8676 // Compare tail (< 8 chars), or rescan last vectors to |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8677 // find 1st mismatched characters |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8678 bind(COMPARE_TAIL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8679 testl(cnt1, cnt1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8680 jccb(Assembler::zero, LENGTH_DIFF_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8681 movl(cnt2, cnt1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8682 // Fallthru to tail compare |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8683 } |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8684 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8685 // Shift str2 and str1 to the end of the arrays, negate min |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8686 lea(str1, Address(str1, cnt2, Address::times_2, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8687 lea(str2, Address(str2, cnt2, Address::times_2, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8688 negptr(cnt2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8689 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8690 // Compare the rest of the characters |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8691 bind(WHILE_HEAD_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8692 load_unsigned_short(result, Address(str1, cnt2, Address::times_2, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8693 load_unsigned_short(cnt1, Address(str2, cnt2, Address::times_2, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8694 subl(result, cnt1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8695 jccb(Assembler::notZero, POP_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8696 increment(cnt2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8697 jcc(Assembler::notZero, WHILE_HEAD_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8698 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8699 // Strings are equal up to min length. Return the length difference. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8700 bind(LENGTH_DIFF_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8701 pop(result); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8702 jmpb(DONE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8703 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8704 // Discard the stored length difference |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8705 bind(POP_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8706 addptr(rsp, wordSize); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8707 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8708 // That's it |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8709 bind(DONE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8710 } |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8711 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8712 // Compare char[] arrays aligned to 4 bytes or substrings. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8713 void MacroAssembler::char_arrays_equals(bool is_array_equ, Register ary1, Register ary2, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8714 Register limit, Register result, Register chr, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8715 XMMRegister vec1, XMMRegister vec2) { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8716 Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR; |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8717 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8718 int length_offset = arrayOopDesc::length_offset_in_bytes(); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8719 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8720 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8721 // Check the input args |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8722 cmpptr(ary1, ary2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8723 jcc(Assembler::equal, TRUE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8724 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8725 if (is_array_equ) { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8726 // Need additional checks for arrays_equals. |
1016 | 8727 testptr(ary1, ary1); |
8728 jcc(Assembler::zero, FALSE_LABEL); | |
8729 testptr(ary2, ary2); | |
8730 jcc(Assembler::zero, FALSE_LABEL); | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8731 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8732 // Check the lengths |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8733 movl(limit, Address(ary1, length_offset)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8734 cmpl(limit, Address(ary2, length_offset)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8735 jcc(Assembler::notEqual, FALSE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8736 } |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8737 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8738 // count == 0 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8739 testl(limit, limit); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8740 jcc(Assembler::zero, TRUE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8741 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8742 if (is_array_equ) { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8743 // Load array address |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8744 lea(ary1, Address(ary1, base_offset)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8745 lea(ary2, Address(ary2, base_offset)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8746 } |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8747 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8748 shll(limit, 1); // byte count != 0 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8749 movl(result, limit); // copy |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8750 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8751 if (UseSSE42Intrinsics) { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8752 // With SSE4.2, use double quad vector compare |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8753 Label COMPARE_WIDE_VECTORS, COMPARE_TAIL; |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8754 // Compare 16-byte vectors |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8755 andl(result, 0x0000000e); // tail count (in bytes) |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8756 andl(limit, 0xfffffff0); // vector count (in bytes) |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8757 jccb(Assembler::zero, COMPARE_TAIL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8758 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8759 lea(ary1, Address(ary1, limit, Address::times_1)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8760 lea(ary2, Address(ary2, limit, Address::times_1)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8761 negptr(limit); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8762 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8763 bind(COMPARE_WIDE_VECTORS); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8764 movdqu(vec1, Address(ary1, limit, Address::times_1)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8765 movdqu(vec2, Address(ary2, limit, Address::times_1)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8766 pxor(vec1, vec2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8767 ptest(vec1, vec1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8768 jccb(Assembler::notZero, FALSE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8769 addptr(limit, 16); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8770 jcc(Assembler::notZero, COMPARE_WIDE_VECTORS); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8771 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8772 bind(COMPARE_TAIL); // limit is zero |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8773 movl(limit, result); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8774 // Fallthru to tail compare |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8775 } |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8776 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8777 // Compare 4-byte vectors |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8778 andl(limit, 0xfffffffc); // vector count (in bytes) |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8779 jccb(Assembler::zero, COMPARE_CHAR); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8780 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8781 lea(ary1, Address(ary1, limit, Address::times_1)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8782 lea(ary2, Address(ary2, limit, Address::times_1)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8783 negptr(limit); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8784 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8785 bind(COMPARE_VECTORS); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8786 movl(chr, Address(ary1, limit, Address::times_1)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8787 cmpl(chr, Address(ary2, limit, Address::times_1)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8788 jccb(Assembler::notEqual, FALSE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8789 addptr(limit, 4); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8790 jcc(Assembler::notZero, COMPARE_VECTORS); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8791 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8792 // Compare trailing char (final 2 bytes), if any |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8793 bind(COMPARE_CHAR); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8794 testl(result, 0x2); // tail char |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8795 jccb(Assembler::zero, TRUE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8796 load_unsigned_short(chr, Address(ary1, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8797 load_unsigned_short(limit, Address(ary2, 0)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8798 cmpl(chr, limit); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8799 jccb(Assembler::notEqual, FALSE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8800 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8801 bind(TRUE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8802 movl(result, 1); // return true |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8803 jmpb(DONE); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8804 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8805 bind(FALSE_LABEL); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8806 xorl(result, result); // return false |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8807 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8808 // That's it |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8809 bind(DONE); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8810 } |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
898
diff
changeset
|
8811 |
1763 | 8812 #ifdef PRODUCT |
8813 #define BLOCK_COMMENT(str) /* nothing */ | |
8814 #else | |
8815 #define BLOCK_COMMENT(str) block_comment(str) | |
8816 #endif | |
8817 | |
8818 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") | |
8819 void MacroAssembler::generate_fill(BasicType t, bool aligned, | |
8820 Register to, Register value, Register count, | |
8821 Register rtmp, XMMRegister xtmp) { | |
8822 assert_different_registers(to, value, count, rtmp); | |
8823 Label L_exit, L_skip_align1, L_skip_align2, L_fill_byte; | |
8824 Label L_fill_2_bytes, L_fill_4_bytes; | |
8825 | |
8826 int shift = -1; | |
8827 switch (t) { | |
8828 case T_BYTE: | |
8829 shift = 2; | |
8830 break; | |
8831 case T_SHORT: | |
8832 shift = 1; | |
8833 break; | |
8834 case T_INT: | |
8835 shift = 0; | |
8836 break; | |
8837 default: ShouldNotReachHere(); | |
8838 } | |
8839 | |
8840 if (t == T_BYTE) { | |
8841 andl(value, 0xff); | |
8842 movl(rtmp, value); | |
8843 shll(rtmp, 8); | |
8844 orl(value, rtmp); | |
8845 } | |
8846 if (t == T_SHORT) { | |
8847 andl(value, 0xffff); | |
8848 } | |
8849 if (t == T_BYTE || t == T_SHORT) { | |
8850 movl(rtmp, value); | |
8851 shll(rtmp, 16); | |
8852 orl(value, rtmp); | |
8853 } | |
8854 | |
8855 cmpl(count, 2<<shift); // Short arrays (< 8 bytes) fill by element | |
8856 jcc(Assembler::below, L_fill_4_bytes); // use unsigned cmp | |
8857 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) { | |
8858 // align source address at 4 bytes address boundary | |
8859 if (t == T_BYTE) { | |
8860 // One byte misalignment happens only for byte arrays | |
8861 testptr(to, 1); | |
8862 jccb(Assembler::zero, L_skip_align1); | |
8863 movb(Address(to, 0), value); | |
8864 increment(to); | |
8865 decrement(count); | |
8866 BIND(L_skip_align1); | |
8867 } | |
8868 // Two bytes misalignment happens only for byte and short (char) arrays | |
8869 testptr(to, 2); | |
8870 jccb(Assembler::zero, L_skip_align2); | |
8871 movw(Address(to, 0), value); | |
8872 addptr(to, 2); | |
8873 subl(count, 1<<(shift-1)); | |
8874 BIND(L_skip_align2); | |
8875 } | |
8876 if (UseSSE < 2) { | |
8877 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes; | |
8878 // Fill 32-byte chunks | |
8879 subl(count, 8 << shift); | |
8880 jcc(Assembler::less, L_check_fill_8_bytes); | |
8881 align(16); | |
8882 | |
8883 BIND(L_fill_32_bytes_loop); | |
8884 | |
8885 for (int i = 0; i < 32; i += 4) { | |
8886 movl(Address(to, i), value); | |
8887 } | |
8888 | |
8889 addptr(to, 32); | |
8890 subl(count, 8 << shift); | |
8891 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop); | |
8892 BIND(L_check_fill_8_bytes); | |
8893 addl(count, 8 << shift); | |
8894 jccb(Assembler::zero, L_exit); | |
8895 jmpb(L_fill_8_bytes); | |
8896 | |
8897 // | |
8898 // length is too short, just fill qwords | |
8899 // | |
8900 BIND(L_fill_8_bytes_loop); | |
8901 movl(Address(to, 0), value); | |
8902 movl(Address(to, 4), value); | |
8903 addptr(to, 8); | |
8904 BIND(L_fill_8_bytes); | |
8905 subl(count, 1 << (shift + 1)); | |
8906 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop); | |
8907 // fall through to fill 4 bytes | |
8908 } else { | |
8909 Label L_fill_32_bytes; | |
8910 if (!UseUnalignedLoadStores) { | |
8911 // align to 8 bytes, we know we are 4 byte aligned to start | |
8912 testptr(to, 4); | |
8913 jccb(Assembler::zero, L_fill_32_bytes); | |
8914 movl(Address(to, 0), value); | |
8915 addptr(to, 4); | |
8916 subl(count, 1<<shift); | |
8917 } | |
8918 BIND(L_fill_32_bytes); | |
8919 { | |
8920 assert( UseSSE >= 2, "supported cpu only" ); | |
8921 Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes; | |
8922 // Fill 32-byte chunks | |
8923 movdl(xtmp, value); | |
8924 pshufd(xtmp, xtmp, 0); | |
8925 | |
8926 subl(count, 8 << shift); | |
8927 jcc(Assembler::less, L_check_fill_8_bytes); | |
8928 align(16); | |
8929 | |
8930 BIND(L_fill_32_bytes_loop); | |
8931 | |
8932 if (UseUnalignedLoadStores) { | |
8933 movdqu(Address(to, 0), xtmp); | |
8934 movdqu(Address(to, 16), xtmp); | |
8935 } else { | |
8936 movq(Address(to, 0), xtmp); | |
8937 movq(Address(to, 8), xtmp); | |
8938 movq(Address(to, 16), xtmp); | |
8939 movq(Address(to, 24), xtmp); | |
8940 } | |
8941 | |
8942 addptr(to, 32); | |
8943 subl(count, 8 << shift); | |
8944 jcc(Assembler::greaterEqual, L_fill_32_bytes_loop); | |
8945 BIND(L_check_fill_8_bytes); | |
8946 addl(count, 8 << shift); | |
8947 jccb(Assembler::zero, L_exit); | |
8948 jmpb(L_fill_8_bytes); | |
8949 | |
8950 // | |
8951 // length is too short, just fill qwords | |
8952 // | |
8953 BIND(L_fill_8_bytes_loop); | |
8954 movq(Address(to, 0), xtmp); | |
8955 addptr(to, 8); | |
8956 BIND(L_fill_8_bytes); | |
8957 subl(count, 1 << (shift + 1)); | |
8958 jcc(Assembler::greaterEqual, L_fill_8_bytes_loop); | |
8959 } | |
8960 } | |
8961 // fill trailing 4 bytes | |
8962 BIND(L_fill_4_bytes); | |
8963 testl(count, 1<<shift); | |
8964 jccb(Assembler::zero, L_fill_2_bytes); | |
8965 movl(Address(to, 0), value); | |
8966 if (t == T_BYTE || t == T_SHORT) { | |
8967 addptr(to, 4); | |
8968 BIND(L_fill_2_bytes); | |
8969 // fill trailing 2 bytes | |
8970 testl(count, 1<<(shift-1)); | |
8971 jccb(Assembler::zero, L_fill_byte); | |
8972 movw(Address(to, 0), value); | |
8973 if (t == T_BYTE) { | |
8974 addptr(to, 2); | |
8975 BIND(L_fill_byte); | |
8976 // fill trailing byte | |
8977 testl(count, 1); | |
8978 jccb(Assembler::zero, L_exit); | |
8979 movb(Address(to, 0), value); | |
8980 } else { | |
8981 BIND(L_fill_byte); | |
8982 } | |
8983 } else { | |
8984 BIND(L_fill_2_bytes); | |
8985 } | |
8986 BIND(L_exit); | |
8987 } | |
8988 #undef BIND | |
8989 #undef BLOCK_COMMENT | |
8990 | |
8991 | |
0 | 8992 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) { |
8993 switch (cond) { | |
8994 // Note some conditions are synonyms for others | |
8995 case Assembler::zero: return Assembler::notZero; | |
8996 case Assembler::notZero: return Assembler::zero; | |
8997 case Assembler::less: return Assembler::greaterEqual; | |
8998 case Assembler::lessEqual: return Assembler::greater; | |
8999 case Assembler::greater: return Assembler::lessEqual; | |
9000 case Assembler::greaterEqual: return Assembler::less; | |
9001 case Assembler::below: return Assembler::aboveEqual; | |
9002 case Assembler::belowEqual: return Assembler::above; | |
9003 case Assembler::above: return Assembler::belowEqual; | |
9004 case Assembler::aboveEqual: return Assembler::below; | |
9005 case Assembler::overflow: return Assembler::noOverflow; | |
9006 case Assembler::noOverflow: return Assembler::overflow; | |
9007 case Assembler::negative: return Assembler::positive; | |
9008 case Assembler::positive: return Assembler::negative; | |
9009 case Assembler::parity: return Assembler::noParity; | |
9010 case Assembler::noParity: return Assembler::parity; | |
9011 } | |
9012 ShouldNotReachHere(); return Assembler::overflow; | |
9013 } | |
9014 | |
9015 SkipIfEqual::SkipIfEqual( | |
9016 MacroAssembler* masm, const bool* flag_addr, bool value) { | |
9017 _masm = masm; | |
9018 _masm->cmp8(ExternalAddress((address)flag_addr), value); | |
9019 _masm->jcc(Assembler::equal, _label); | |
9020 } | |
9021 | |
9022 SkipIfEqual::~SkipIfEqual() { | |
9023 _masm->bind(_label); | |
9024 } |