Mercurial > hg > truffle
annotate src/cpu/x86/vm/stubGenerator_x86_32.cpp @ 7227:8fbb97b0016a
fixes after merge
author | Christian Haeubl <haeubl@ssw.jku.at> |
---|---|
date | Fri, 14 Dec 2012 13:10:55 +0100 |
parents | e522a00b91aa |
children | 291ffc492eb6 |
rev | line source |
---|---|
0 | 1 /* |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
2 * Copyright (c) 1999, 2012, 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:
1506
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1506
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:
1506
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1972 | 25 #include "precompiled.hpp" |
26 #include "asm/assembler.hpp" | |
27 #include "assembler_x86.inline.hpp" | |
28 #include "interpreter/interpreter.hpp" | |
29 #include "nativeInst_x86.hpp" | |
30 #include "oops/instanceOop.hpp" | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
31 #include "oops/method.hpp" |
1972 | 32 #include "oops/objArrayKlass.hpp" |
33 #include "oops/oop.inline.hpp" | |
34 #include "prims/methodHandles.hpp" | |
35 #include "runtime/frame.inline.hpp" | |
36 #include "runtime/handles.inline.hpp" | |
37 #include "runtime/sharedRuntime.hpp" | |
38 #include "runtime/stubCodeGenerator.hpp" | |
39 #include "runtime/stubRoutines.hpp" | |
40 #include "utilities/top.hpp" | |
41 #ifdef TARGET_OS_FAMILY_linux | |
42 # include "thread_linux.inline.hpp" | |
43 #endif | |
44 #ifdef TARGET_OS_FAMILY_solaris | |
45 # include "thread_solaris.inline.hpp" | |
46 #endif | |
47 #ifdef TARGET_OS_FAMILY_windows | |
48 # include "thread_windows.inline.hpp" | |
49 #endif | |
3960 | 50 #ifdef TARGET_OS_FAMILY_bsd |
51 # include "thread_bsd.inline.hpp" | |
52 #endif | |
1972 | 53 #ifdef COMPILER2 |
54 #include "opto/runtime.hpp" | |
55 #endif | |
0 | 56 |
57 // Declaration and definition of StubGenerator (no .hpp file). | |
58 // For a more detailed description of the stub routine structure | |
59 // see the comment in stubRoutines.hpp | |
60 | |
61 #define __ _masm-> | |
304 | 62 #define a__ ((Assembler*)_masm)-> |
0 | 63 |
64 #ifdef PRODUCT | |
65 #define BLOCK_COMMENT(str) /* nothing */ | |
66 #else | |
67 #define BLOCK_COMMENT(str) __ block_comment(str) | |
68 #endif | |
69 | |
70 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":") | |
71 | |
72 const int MXCSR_MASK = 0xFFC0; // Mask out any pending exceptions | |
73 const int FPU_CNTRL_WRD_MASK = 0xFFFF; | |
74 | |
75 // ------------------------------------------------------------------------------------------------------------------------- | |
76 // Stub Code definitions | |
77 | |
78 static address handle_unsafe_access() { | |
79 JavaThread* thread = JavaThread::current(); | |
80 address pc = thread->saved_exception_pc(); | |
81 // pc is the instruction which we must emulate | |
82 // doing a no-op is fine: return garbage from the load | |
83 // therefore, compute npc | |
84 address npc = Assembler::locate_next_instruction(pc); | |
85 | |
86 // request an async exception | |
87 thread->set_pending_unsafe_access_error(); | |
88 | |
89 // return address of next instruction to execute | |
90 return npc; | |
91 } | |
92 | |
93 class StubGenerator: public StubCodeGenerator { | |
94 private: | |
95 | |
96 #ifdef PRODUCT | |
97 #define inc_counter_np(counter) (0) | |
98 #else | |
99 void inc_counter_np_(int& counter) { | |
304 | 100 __ incrementl(ExternalAddress((address)&counter)); |
0 | 101 } |
102 #define inc_counter_np(counter) \ | |
103 BLOCK_COMMENT("inc_counter " #counter); \ | |
104 inc_counter_np_(counter); | |
105 #endif //PRODUCT | |
106 | |
107 void inc_copy_counter_np(BasicType t) { | |
108 #ifndef PRODUCT | |
109 switch (t) { | |
110 case T_BYTE: inc_counter_np(SharedRuntime::_jbyte_array_copy_ctr); return; | |
111 case T_SHORT: inc_counter_np(SharedRuntime::_jshort_array_copy_ctr); return; | |
112 case T_INT: inc_counter_np(SharedRuntime::_jint_array_copy_ctr); return; | |
113 case T_LONG: inc_counter_np(SharedRuntime::_jlong_array_copy_ctr); return; | |
114 case T_OBJECT: inc_counter_np(SharedRuntime::_oop_array_copy_ctr); return; | |
115 } | |
116 ShouldNotReachHere(); | |
117 #endif //PRODUCT | |
118 } | |
119 | |
120 //------------------------------------------------------------------------------------------------------------------------ | |
121 // Call stubs are used to call Java from C | |
122 // | |
123 // [ return_from_Java ] <--- rsp | |
124 // [ argument word n ] | |
125 // ... | |
126 // -N [ argument word 1 ] | |
127 // -7 [ Possible padding for stack alignment ] | |
128 // -6 [ Possible padding for stack alignment ] | |
129 // -5 [ Possible padding for stack alignment ] | |
130 // -4 [ mxcsr save ] <--- rsp_after_call | |
131 // -3 [ saved rbx, ] | |
132 // -2 [ saved rsi ] | |
133 // -1 [ saved rdi ] | |
134 // 0 [ saved rbp, ] <--- rbp, | |
135 // 1 [ return address ] | |
136 // 2 [ ptr. to call wrapper ] | |
137 // 3 [ result ] | |
138 // 4 [ result_type ] | |
139 // 5 [ method ] | |
140 // 6 [ entry_point ] | |
141 // 7 [ parameters ] | |
142 // 8 [ parameter_size ] | |
143 // 9 [ thread ] | |
144 | |
145 | |
146 address generate_call_stub(address& return_address) { | |
147 StubCodeMark mark(this, "StubRoutines", "call_stub"); | |
148 address start = __ pc(); | |
149 | |
150 // stub code parameters / addresses | |
151 assert(frame::entry_frame_call_wrapper_offset == 2, "adjust this code"); | |
152 bool sse_save = false; | |
153 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_catch_exception()! | |
154 const int locals_count_in_bytes (4*wordSize); | |
155 const Address mxcsr_save (rbp, -4 * wordSize); | |
156 const Address saved_rbx (rbp, -3 * wordSize); | |
157 const Address saved_rsi (rbp, -2 * wordSize); | |
158 const Address saved_rdi (rbp, -1 * wordSize); | |
159 const Address result (rbp, 3 * wordSize); | |
160 const Address result_type (rbp, 4 * wordSize); | |
161 const Address method (rbp, 5 * wordSize); | |
162 const Address entry_point (rbp, 6 * wordSize); | |
163 const Address parameters (rbp, 7 * wordSize); | |
164 const Address parameter_size(rbp, 8 * wordSize); | |
165 const Address thread (rbp, 9 * wordSize); // same as in generate_catch_exception()! | |
166 sse_save = UseSSE > 0; | |
167 | |
168 // stub code | |
169 __ enter(); | |
304 | 170 __ movptr(rcx, parameter_size); // parameter counter |
1506 | 171 __ shlptr(rcx, Interpreter::logStackElementSize); // convert parameter count to bytes |
304 | 172 __ addptr(rcx, locals_count_in_bytes); // reserve space for register saves |
173 __ subptr(rsp, rcx); | |
174 __ andptr(rsp, -(StackAlignmentInBytes)); // Align stack | |
0 | 175 |
176 // save rdi, rsi, & rbx, according to C calling conventions | |
304 | 177 __ movptr(saved_rdi, rdi); |
178 __ movptr(saved_rsi, rsi); | |
179 __ movptr(saved_rbx, rbx); | |
0 | 180 // save and initialize %mxcsr |
181 if (sse_save) { | |
182 Label skip_ldmx; | |
183 __ stmxcsr(mxcsr_save); | |
184 __ movl(rax, mxcsr_save); | |
185 __ andl(rax, MXCSR_MASK); // Only check control and mask bits | |
186 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); | |
187 __ cmp32(rax, mxcsr_std); | |
188 __ jcc(Assembler::equal, skip_ldmx); | |
189 __ ldmxcsr(mxcsr_std); | |
190 __ bind(skip_ldmx); | |
191 } | |
192 | |
193 // make sure the control word is correct. | |
194 __ fldcw(ExternalAddress(StubRoutines::addr_fpu_cntrl_wrd_std())); | |
195 | |
196 #ifdef ASSERT | |
197 // make sure we have no pending exceptions | |
198 { Label L; | |
304 | 199 __ movptr(rcx, thread); |
200 __ cmpptr(Address(rcx, Thread::pending_exception_offset()), (int32_t)NULL_WORD); | |
0 | 201 __ jcc(Assembler::equal, L); |
202 __ stop("StubRoutines::call_stub: entered with pending exception"); | |
203 __ bind(L); | |
204 } | |
205 #endif | |
206 | |
207 // pass parameters if any | |
208 BLOCK_COMMENT("pass parameters if any"); | |
209 Label parameters_done; | |
210 __ movl(rcx, parameter_size); // parameter counter | |
211 __ testl(rcx, rcx); | |
212 __ jcc(Assembler::zero, parameters_done); | |
213 | |
214 // parameter passing loop | |
215 | |
216 Label loop; | |
217 // Copy Java parameters in reverse order (receiver last) | |
218 // Note that the argument order is inverted in the process | |
219 // source is rdx[rcx: N-1..0] | |
220 // dest is rsp[rbx: 0..N-1] | |
221 | |
304 | 222 __ movptr(rdx, parameters); // parameter pointer |
223 __ xorptr(rbx, rbx); | |
0 | 224 |
225 __ BIND(loop); | |
226 | |
227 // get parameter | |
304 | 228 __ movptr(rax, Address(rdx, rcx, Interpreter::stackElementScale(), -wordSize)); |
229 __ movptr(Address(rsp, rbx, Interpreter::stackElementScale(), | |
0 | 230 Interpreter::expr_offset_in_bytes(0)), rax); // store parameter |
231 __ increment(rbx); | |
232 __ decrement(rcx); | |
233 __ jcc(Assembler::notZero, loop); | |
234 | |
235 // call Java function | |
236 __ BIND(parameters_done); | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
237 __ movptr(rbx, method); // get Method* |
304 | 238 __ movptr(rax, entry_point); // get entry_point |
239 __ mov(rsi, rsp); // set sender sp | |
0 | 240 BLOCK_COMMENT("call Java function"); |
241 __ call(rax); | |
242 | |
243 BLOCK_COMMENT("call_stub_return_address:"); | |
244 return_address = __ pc(); | |
245 | |
2245
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
246 #ifdef COMPILER2 |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
247 { |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
248 Label L_skip; |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
249 if (UseSSE >= 2) { |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
250 __ verify_FPU(0, "call_stub_return"); |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
251 } else { |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
252 for (int i = 1; i < 8; i++) { |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
253 __ ffree(i); |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
254 } |
0 | 255 |
2245
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
256 // UseSSE <= 1 so double result should be left on TOS |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
257 __ movl(rsi, result_type); |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
258 __ cmpl(rsi, T_DOUBLE); |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
259 __ jcc(Assembler::equal, L_skip); |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
260 if (UseSSE == 0) { |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
261 // UseSSE == 0 so float result should be left on TOS |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
262 __ cmpl(rsi, T_FLOAT); |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
263 __ jcc(Assembler::equal, L_skip); |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
264 } |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
265 __ ffree(0); |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
266 } |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
267 __ BIND(L_skip); |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
268 } |
638119ce7cfd
7009309: JSR 292: compiler/6991596/Test6991596.java crashes on fastdebug JDK7/b122
twisti
parents:
1972
diff
changeset
|
269 #endif // COMPILER2 |
0 | 270 |
271 // store result depending on type | |
272 // (everything that is not T_LONG, T_FLOAT or T_DOUBLE is treated as T_INT) | |
304 | 273 __ movptr(rdi, result); |
0 | 274 Label is_long, is_float, is_double, exit; |
275 __ movl(rsi, result_type); | |
276 __ cmpl(rsi, T_LONG); | |
277 __ jcc(Assembler::equal, is_long); | |
278 __ cmpl(rsi, T_FLOAT); | |
279 __ jcc(Assembler::equal, is_float); | |
280 __ cmpl(rsi, T_DOUBLE); | |
281 __ jcc(Assembler::equal, is_double); | |
282 | |
283 // handle T_INT case | |
284 __ movl(Address(rdi, 0), rax); | |
285 __ BIND(exit); | |
286 | |
287 // check that FPU stack is empty | |
288 __ verify_FPU(0, "generate_call_stub"); | |
289 | |
290 // pop parameters | |
304 | 291 __ lea(rsp, rsp_after_call); |
0 | 292 |
293 // restore %mxcsr | |
294 if (sse_save) { | |
295 __ ldmxcsr(mxcsr_save); | |
296 } | |
297 | |
298 // restore rdi, rsi and rbx, | |
304 | 299 __ movptr(rbx, saved_rbx); |
300 __ movptr(rsi, saved_rsi); | |
301 __ movptr(rdi, saved_rdi); | |
302 __ addptr(rsp, 4*wordSize); | |
0 | 303 |
304 // return | |
304 | 305 __ pop(rbp); |
0 | 306 __ ret(0); |
307 | |
308 // handle return types different from T_INT | |
309 __ BIND(is_long); | |
310 __ movl(Address(rdi, 0 * wordSize), rax); | |
311 __ movl(Address(rdi, 1 * wordSize), rdx); | |
312 __ jmp(exit); | |
313 | |
314 __ BIND(is_float); | |
315 // interpreter uses xmm0 for return values | |
316 if (UseSSE >= 1) { | |
317 __ movflt(Address(rdi, 0), xmm0); | |
318 } else { | |
319 __ fstp_s(Address(rdi, 0)); | |
320 } | |
321 __ jmp(exit); | |
322 | |
323 __ BIND(is_double); | |
324 // interpreter uses xmm0 for return values | |
325 if (UseSSE >= 2) { | |
326 __ movdbl(Address(rdi, 0), xmm0); | |
327 } else { | |
328 __ fstp_d(Address(rdi, 0)); | |
329 } | |
330 __ jmp(exit); | |
331 | |
332 return start; | |
333 } | |
334 | |
335 | |
336 //------------------------------------------------------------------------------------------------------------------------ | |
337 // Return point for a Java call if there's an exception thrown in Java code. | |
338 // The exception is caught and transformed into a pending exception stored in | |
339 // JavaThread that can be tested from within the VM. | |
340 // | |
341 // Note: Usually the parameters are removed by the callee. In case of an exception | |
342 // crossing an activation frame boundary, that is not the case if the callee | |
343 // is compiled code => need to setup the rsp. | |
344 // | |
345 // rax,: exception oop | |
346 | |
347 address generate_catch_exception() { | |
348 StubCodeMark mark(this, "StubRoutines", "catch_exception"); | |
349 const Address rsp_after_call(rbp, -4 * wordSize); // same as in generate_call_stub()! | |
350 const Address thread (rbp, 9 * wordSize); // same as in generate_call_stub()! | |
351 address start = __ pc(); | |
352 | |
353 // get thread directly | |
304 | 354 __ movptr(rcx, thread); |
0 | 355 #ifdef ASSERT |
356 // verify that threads correspond | |
357 { Label L; | |
358 __ get_thread(rbx); | |
304 | 359 __ cmpptr(rbx, rcx); |
0 | 360 __ jcc(Assembler::equal, L); |
361 __ stop("StubRoutines::catch_exception: threads must correspond"); | |
362 __ bind(L); | |
363 } | |
364 #endif | |
365 // set pending exception | |
366 __ verify_oop(rax); | |
304 | 367 __ movptr(Address(rcx, Thread::pending_exception_offset()), rax ); |
0 | 368 __ lea(Address(rcx, Thread::exception_file_offset ()), |
369 ExternalAddress((address)__FILE__)); | |
370 __ movl(Address(rcx, Thread::exception_line_offset ()), __LINE__ ); | |
371 // complete return to VM | |
372 assert(StubRoutines::_call_stub_return_address != NULL, "_call_stub_return_address must have been generated before"); | |
373 __ jump(RuntimeAddress(StubRoutines::_call_stub_return_address)); | |
374 | |
375 return start; | |
376 } | |
377 | |
378 | |
379 //------------------------------------------------------------------------------------------------------------------------ | |
380 // Continuation point for runtime calls returning with a pending exception. | |
381 // The pending exception check happened in the runtime or native call stub. | |
382 // The pending exception in Thread is converted into a Java-level exception. | |
383 // | |
384 // Contract with Java-level exception handlers: | |
1295 | 385 // rax: exception |
0 | 386 // rdx: throwing pc |
387 // | |
388 // NOTE: At entry of this stub, exception-pc must be on stack !! | |
389 | |
390 address generate_forward_exception() { | |
391 StubCodeMark mark(this, "StubRoutines", "forward exception"); | |
392 address start = __ pc(); | |
1295 | 393 const Register thread = rcx; |
394 | |
395 // other registers used in this stub | |
396 const Register exception_oop = rax; | |
397 const Register handler_addr = rbx; | |
398 const Register exception_pc = rdx; | |
0 | 399 |
400 // Upon entry, the sp points to the return address returning into Java | |
401 // (interpreted or compiled) code; i.e., the return address becomes the | |
402 // throwing pc. | |
403 // | |
404 // Arguments pushed before the runtime call are still on the stack but | |
405 // the exception handler will reset the stack pointer -> ignore them. | |
406 // A potential result in registers can be ignored as well. | |
407 | |
408 #ifdef ASSERT | |
409 // make sure this code is only executed if there is a pending exception | |
410 { Label L; | |
1295 | 411 __ get_thread(thread); |
412 __ cmpptr(Address(thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); | |
0 | 413 __ jcc(Assembler::notEqual, L); |
414 __ stop("StubRoutines::forward exception: no pending exception (1)"); | |
415 __ bind(L); | |
416 } | |
417 #endif | |
418 | |
419 // compute exception handler into rbx, | |
1295 | 420 __ get_thread(thread); |
421 __ movptr(exception_pc, Address(rsp, 0)); | |
0 | 422 BLOCK_COMMENT("call exception_handler_for_return_address"); |
1295 | 423 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), thread, exception_pc); |
424 __ mov(handler_addr, rax); | |
0 | 425 |
1295 | 426 // setup rax & rdx, remove return address & clear pending exception |
427 __ get_thread(thread); | |
428 __ pop(exception_pc); | |
429 __ movptr(exception_oop, Address(thread, Thread::pending_exception_offset())); | |
430 __ movptr(Address(thread, Thread::pending_exception_offset()), NULL_WORD); | |
0 | 431 |
432 #ifdef ASSERT | |
433 // make sure exception is set | |
434 { Label L; | |
1295 | 435 __ testptr(exception_oop, exception_oop); |
0 | 436 __ jcc(Assembler::notEqual, L); |
437 __ stop("StubRoutines::forward exception: no pending exception (2)"); | |
438 __ bind(L); | |
439 } | |
440 #endif | |
441 | |
1295 | 442 // Verify that there is really a valid exception in RAX. |
443 __ verify_oop(exception_oop); | |
444 | |
0 | 445 // continue at exception handler (return address removed) |
1295 | 446 // rax: exception |
447 // rbx: exception handler | |
0 | 448 // rdx: throwing pc |
1295 | 449 __ jmp(handler_addr); |
0 | 450 |
451 return start; | |
452 } | |
453 | |
454 | |
455 //---------------------------------------------------------------------------------------------------- | |
456 // Support for jint Atomic::xchg(jint exchange_value, volatile jint* dest) | |
457 // | |
458 // xchg exists as far back as 8086, lock needed for MP only | |
459 // Stack layout immediately after call: | |
460 // | |
461 // 0 [ret addr ] <--- rsp | |
462 // 1 [ ex ] | |
463 // 2 [ dest ] | |
464 // | |
465 // Result: *dest <- ex, return (old *dest) | |
466 // | |
467 // Note: win32 does not currently use this code | |
468 | |
469 address generate_atomic_xchg() { | |
470 StubCodeMark mark(this, "StubRoutines", "atomic_xchg"); | |
471 address start = __ pc(); | |
472 | |
304 | 473 __ push(rdx); |
0 | 474 Address exchange(rsp, 2 * wordSize); |
475 Address dest_addr(rsp, 3 * wordSize); | |
476 __ movl(rax, exchange); | |
304 | 477 __ movptr(rdx, dest_addr); |
478 __ xchgl(rax, Address(rdx, 0)); | |
479 __ pop(rdx); | |
0 | 480 __ ret(0); |
481 | |
482 return start; | |
483 } | |
484 | |
485 //---------------------------------------------------------------------------------------------------- | |
486 // Support for void verify_mxcsr() | |
487 // | |
488 // This routine is used with -Xcheck:jni to verify that native | |
489 // JNI code does not return to Java code without restoring the | |
490 // MXCSR register to our expected state. | |
491 | |
492 | |
493 address generate_verify_mxcsr() { | |
494 StubCodeMark mark(this, "StubRoutines", "verify_mxcsr"); | |
495 address start = __ pc(); | |
496 | |
497 const Address mxcsr_save(rsp, 0); | |
498 | |
499 if (CheckJNICalls && UseSSE > 0 ) { | |
500 Label ok_ret; | |
501 ExternalAddress mxcsr_std(StubRoutines::addr_mxcsr_std()); | |
304 | 502 __ push(rax); |
503 __ subptr(rsp, wordSize); // allocate a temp location | |
0 | 504 __ stmxcsr(mxcsr_save); |
505 __ movl(rax, mxcsr_save); | |
506 __ andl(rax, MXCSR_MASK); | |
507 __ cmp32(rax, mxcsr_std); | |
508 __ jcc(Assembler::equal, ok_ret); | |
509 | |
510 __ warn("MXCSR changed by native JNI code."); | |
511 | |
512 __ ldmxcsr(mxcsr_std); | |
513 | |
514 __ bind(ok_ret); | |
304 | 515 __ addptr(rsp, wordSize); |
516 __ pop(rax); | |
0 | 517 } |
518 | |
519 __ ret(0); | |
520 | |
521 return start; | |
522 } | |
523 | |
524 | |
525 //--------------------------------------------------------------------------- | |
526 // Support for void verify_fpu_cntrl_wrd() | |
527 // | |
528 // This routine is used with -Xcheck:jni to verify that native | |
529 // JNI code does not return to Java code without restoring the | |
530 // FP control word to our expected state. | |
531 | |
532 address generate_verify_fpu_cntrl_wrd() { | |
533 StubCodeMark mark(this, "StubRoutines", "verify_spcw"); | |
534 address start = __ pc(); | |
535 | |
536 const Address fpu_cntrl_wrd_save(rsp, 0); | |
537 | |
538 if (CheckJNICalls) { | |
539 Label ok_ret; | |
304 | 540 __ push(rax); |
541 __ subptr(rsp, wordSize); // allocate a temp location | |
0 | 542 __ fnstcw(fpu_cntrl_wrd_save); |
543 __ movl(rax, fpu_cntrl_wrd_save); | |
544 __ andl(rax, FPU_CNTRL_WRD_MASK); | |
545 ExternalAddress fpu_std(StubRoutines::addr_fpu_cntrl_wrd_std()); | |
546 __ cmp32(rax, fpu_std); | |
547 __ jcc(Assembler::equal, ok_ret); | |
548 | |
549 __ warn("Floating point control word changed by native JNI code."); | |
550 | |
551 __ fldcw(fpu_std); | |
552 | |
553 __ bind(ok_ret); | |
304 | 554 __ addptr(rsp, wordSize); |
555 __ pop(rax); | |
0 | 556 } |
557 | |
558 __ ret(0); | |
559 | |
560 return start; | |
561 } | |
562 | |
563 //--------------------------------------------------------------------------- | |
564 // Wrapper for slow-case handling of double-to-integer conversion | |
565 // d2i or f2i fast case failed either because it is nan or because | |
566 // of under/overflow. | |
567 // Input: FPU TOS: float value | |
568 // Output: rax, (rdx): integer (long) result | |
569 | |
570 address generate_d2i_wrapper(BasicType t, address fcn) { | |
571 StubCodeMark mark(this, "StubRoutines", "d2i_wrapper"); | |
572 address start = __ pc(); | |
573 | |
574 // Capture info about frame layout | |
575 enum layout { FPUState_off = 0, | |
576 rbp_off = FPUStateSizeInWords, | |
577 rdi_off, | |
578 rsi_off, | |
579 rcx_off, | |
580 rbx_off, | |
581 saved_argument_off, | |
582 saved_argument_off2, // 2nd half of double | |
583 framesize | |
584 }; | |
585 | |
586 assert(FPUStateSizeInWords == 27, "update stack layout"); | |
587 | |
588 // Save outgoing argument to stack across push_FPU_state() | |
304 | 589 __ subptr(rsp, wordSize * 2); |
0 | 590 __ fstp_d(Address(rsp, 0)); |
591 | |
592 // Save CPU & FPU state | |
304 | 593 __ push(rbx); |
594 __ push(rcx); | |
595 __ push(rsi); | |
596 __ push(rdi); | |
597 __ push(rbp); | |
0 | 598 __ push_FPU_state(); |
599 | |
600 // push_FPU_state() resets the FP top of stack | |
601 // Load original double into FP top of stack | |
602 __ fld_d(Address(rsp, saved_argument_off * wordSize)); | |
603 // Store double into stack as outgoing argument | |
304 | 604 __ subptr(rsp, wordSize*2); |
0 | 605 __ fst_d(Address(rsp, 0)); |
606 | |
607 // Prepare FPU for doing math in C-land | |
608 __ empty_FPU_stack(); | |
609 // Call the C code to massage the double. Result in EAX | |
610 if (t == T_INT) | |
611 { BLOCK_COMMENT("SharedRuntime::d2i"); } | |
612 else if (t == T_LONG) | |
613 { BLOCK_COMMENT("SharedRuntime::d2l"); } | |
614 __ call_VM_leaf( fcn, 2 ); | |
615 | |
616 // Restore CPU & FPU state | |
617 __ pop_FPU_state(); | |
304 | 618 __ pop(rbp); |
619 __ pop(rdi); | |
620 __ pop(rsi); | |
621 __ pop(rcx); | |
622 __ pop(rbx); | |
623 __ addptr(rsp, wordSize * 2); | |
0 | 624 |
625 __ ret(0); | |
626 | |
627 return start; | |
628 } | |
629 | |
630 | |
631 //--------------------------------------------------------------------------- | |
632 // The following routine generates a subroutine to throw an asynchronous | |
633 // UnknownError when an unsafe access gets a fault that could not be | |
634 // reasonably prevented by the programmer. (Example: SIGBUS/OBJERR.) | |
635 address generate_handler_for_unsafe_access() { | |
636 StubCodeMark mark(this, "StubRoutines", "handler_for_unsafe_access"); | |
637 address start = __ pc(); | |
638 | |
304 | 639 __ push(0); // hole for return address-to-be |
640 __ pusha(); // push registers | |
0 | 641 Address next_pc(rsp, RegisterImpl::number_of_registers * BytesPerWord); |
642 BLOCK_COMMENT("call handle_unsafe_access"); | |
643 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, handle_unsafe_access))); | |
304 | 644 __ movptr(next_pc, rax); // stuff next address |
645 __ popa(); | |
0 | 646 __ ret(0); // jump to next address |
647 | |
648 return start; | |
649 } | |
650 | |
651 | |
652 //---------------------------------------------------------------------------------------------------- | |
653 // Non-destructive plausibility checks for oops | |
654 | |
655 address generate_verify_oop() { | |
656 StubCodeMark mark(this, "StubRoutines", "verify_oop"); | |
657 address start = __ pc(); | |
658 | |
659 // Incoming arguments on stack after saving rax,: | |
660 // | |
661 // [tos ]: saved rdx | |
662 // [tos + 1]: saved EFLAGS | |
663 // [tos + 2]: return address | |
664 // [tos + 3]: char* error message | |
665 // [tos + 4]: oop object to verify | |
666 // [tos + 5]: saved rax, - saved by caller and bashed | |
667 | |
668 Label exit, error; | |
304 | 669 __ pushf(); |
670 __ incrementl(ExternalAddress((address) StubRoutines::verify_oop_count_addr())); | |
671 __ push(rdx); // save rdx | |
0 | 672 // make sure object is 'reasonable' |
304 | 673 __ movptr(rax, Address(rsp, 4 * wordSize)); // get object |
674 __ testptr(rax, rax); | |
0 | 675 __ jcc(Assembler::zero, exit); // if obj is NULL it is ok |
676 | |
677 // Check if the oop is in the right area of memory | |
678 const int oop_mask = Universe::verify_oop_mask(); | |
679 const int oop_bits = Universe::verify_oop_bits(); | |
304 | 680 __ mov(rdx, rax); |
681 __ andptr(rdx, oop_mask); | |
682 __ cmpptr(rdx, oop_bits); | |
0 | 683 __ jcc(Assembler::notZero, error); |
684 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
685 // make sure klass is 'reasonable', which is not zero. |
304 | 686 __ movptr(rax, Address(rax, oopDesc::klass_offset_in_bytes())); // get klass |
687 __ testptr(rax, rax); | |
0 | 688 __ jcc(Assembler::zero, error); // if klass is NULL it is broken |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6266
diff
changeset
|
689 // TODO: Future assert that klass is lower 4g memory for UseCompressedKlassPointers |
0 | 690 |
691 // return if everything seems ok | |
692 __ bind(exit); | |
304 | 693 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back |
694 __ pop(rdx); // restore rdx | |
695 __ popf(); // restore EFLAGS | |
0 | 696 __ ret(3 * wordSize); // pop arguments |
697 | |
698 // handle errors | |
699 __ bind(error); | |
304 | 700 __ movptr(rax, Address(rsp, 5 * wordSize)); // get saved rax, back |
701 __ pop(rdx); // get saved rdx back | |
702 __ popf(); // get saved EFLAGS off stack -- will be ignored | |
703 __ pusha(); // push registers (eip = return address & msg are already pushed) | |
0 | 704 BLOCK_COMMENT("call MacroAssembler::debug"); |
304 | 705 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, MacroAssembler::debug32))); |
706 __ popa(); | |
0 | 707 __ ret(3 * wordSize); // pop arguments |
708 return start; | |
709 } | |
710 | |
711 // | |
712 // Generate pre-barrier for array stores | |
713 // | |
714 // Input: | |
715 // start - starting address | |
845
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
710
diff
changeset
|
716 // count - element count |
2324 | 717 void gen_write_ref_array_pre_barrier(Register start, Register count, bool uninitialized_target) { |
0 | 718 assert_different_registers(start, count); |
719 BarrierSet* bs = Universe::heap()->barrier_set(); | |
720 switch (bs->kind()) { | |
721 case BarrierSet::G1SATBCT: | |
722 case BarrierSet::G1SATBCTLogging: | |
2324 | 723 // With G1, don't generate the call if we statically know that the target in uninitialized |
724 if (!uninitialized_target) { | |
725 __ pusha(); // push registers | |
726 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_pre), | |
727 start, count); | |
728 __ popa(); | |
729 } | |
0 | 730 break; |
731 case BarrierSet::CardTableModRef: | |
732 case BarrierSet::CardTableExtension: | |
733 case BarrierSet::ModRef: | |
734 break; | |
735 default : | |
736 ShouldNotReachHere(); | |
737 | |
738 } | |
739 } | |
740 | |
741 | |
742 // | |
743 // Generate a post-barrier for an array store | |
744 // | |
745 // start - starting address | |
746 // count - element count | |
747 // | |
748 // The two input registers are overwritten. | |
749 // | |
750 void gen_write_ref_array_post_barrier(Register start, Register count) { | |
751 BarrierSet* bs = Universe::heap()->barrier_set(); | |
752 assert_different_registers(start, count); | |
753 switch (bs->kind()) { | |
754 case BarrierSet::G1SATBCT: | |
755 case BarrierSet::G1SATBCTLogging: | |
756 { | |
304 | 757 __ pusha(); // push registers |
1192
776fb94f33cc
6918006: G1: spill space must be reserved on the stack for barrier calls on Windows x64
apetrusenko
parents:
1174
diff
changeset
|
758 __ call_VM_leaf(CAST_FROM_FN_PTR(address, BarrierSet::static_write_ref_array_post), |
776fb94f33cc
6918006: G1: spill space must be reserved on the stack for barrier calls on Windows x64
apetrusenko
parents:
1174
diff
changeset
|
759 start, count); |
304 | 760 __ popa(); |
0 | 761 } |
762 break; | |
763 | |
764 case BarrierSet::CardTableModRef: | |
765 case BarrierSet::CardTableExtension: | |
766 { | |
767 CardTableModRefBS* ct = (CardTableModRefBS*)bs; | |
768 assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code"); | |
769 | |
770 Label L_loop; | |
771 const Register end = count; // elements count; end == start+count-1 | |
772 assert_different_registers(start, end); | |
773 | |
304 | 774 __ lea(end, Address(start, count, Address::times_ptr, -wordSize)); |
775 __ shrptr(start, CardTableModRefBS::card_shift); | |
776 __ shrptr(end, CardTableModRefBS::card_shift); | |
777 __ subptr(end, start); // end --> count | |
0 | 778 __ BIND(L_loop); |
249
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
19
diff
changeset
|
779 intptr_t disp = (intptr_t) ct->byte_map_base; |
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
19
diff
changeset
|
780 Address cardtable(start, count, Address::times_1, disp); |
910a4cb98e9e
6717457: Internal Error (src/share/vm/code/relocInfo.hpp:1089)
never
parents:
19
diff
changeset
|
781 __ movb(cardtable, 0); |
0 | 782 __ decrement(count); |
783 __ jcc(Assembler::greaterEqual, L_loop); | |
784 } | |
785 break; | |
786 case BarrierSet::ModRef: | |
787 break; | |
788 default : | |
789 ShouldNotReachHere(); | |
790 | |
791 } | |
792 } | |
793 | |
405 | 794 |
795 // Copy 64 bytes chunks | |
796 // | |
797 // Inputs: | |
798 // from - source array address | |
799 // to_from - destination array address - from | |
800 // qword_count - 8-bytes element count, negative | |
801 // | |
802 void xmm_copy_forward(Register from, Register to_from, Register qword_count) { | |
803 assert( UseSSE >= 2, "supported cpu only" ); | |
804 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; | |
805 // Copy 64-byte chunks | |
806 __ jmpb(L_copy_64_bytes); | |
1365 | 807 __ align(OptoLoopAlignment); |
405 | 808 __ BIND(L_copy_64_bytes_loop); |
809 | |
810 if(UseUnalignedLoadStores) { | |
811 __ movdqu(xmm0, Address(from, 0)); | |
812 __ movdqu(Address(from, to_from, Address::times_1, 0), xmm0); | |
813 __ movdqu(xmm1, Address(from, 16)); | |
814 __ movdqu(Address(from, to_from, Address::times_1, 16), xmm1); | |
815 __ movdqu(xmm2, Address(from, 32)); | |
816 __ movdqu(Address(from, to_from, Address::times_1, 32), xmm2); | |
817 __ movdqu(xmm3, Address(from, 48)); | |
818 __ movdqu(Address(from, to_from, Address::times_1, 48), xmm3); | |
819 | |
820 } else { | |
821 __ movq(xmm0, Address(from, 0)); | |
822 __ movq(Address(from, to_from, Address::times_1, 0), xmm0); | |
823 __ movq(xmm1, Address(from, 8)); | |
824 __ movq(Address(from, to_from, Address::times_1, 8), xmm1); | |
825 __ movq(xmm2, Address(from, 16)); | |
826 __ movq(Address(from, to_from, Address::times_1, 16), xmm2); | |
827 __ movq(xmm3, Address(from, 24)); | |
828 __ movq(Address(from, to_from, Address::times_1, 24), xmm3); | |
829 __ movq(xmm4, Address(from, 32)); | |
830 __ movq(Address(from, to_from, Address::times_1, 32), xmm4); | |
831 __ movq(xmm5, Address(from, 40)); | |
832 __ movq(Address(from, to_from, Address::times_1, 40), xmm5); | |
833 __ movq(xmm6, Address(from, 48)); | |
834 __ movq(Address(from, to_from, Address::times_1, 48), xmm6); | |
835 __ movq(xmm7, Address(from, 56)); | |
836 __ movq(Address(from, to_from, Address::times_1, 56), xmm7); | |
837 } | |
838 | |
839 __ addl(from, 64); | |
840 __ BIND(L_copy_64_bytes); | |
841 __ subl(qword_count, 8); | |
842 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); | |
843 __ addl(qword_count, 8); | |
844 __ jccb(Assembler::zero, L_exit); | |
845 // | |
846 // length is too short, just copy qwords | |
847 // | |
848 __ BIND(L_copy_8_bytes); | |
849 __ movq(xmm0, Address(from, 0)); | |
850 __ movq(Address(from, to_from, Address::times_1), xmm0); | |
851 __ addl(from, 8); | |
852 __ decrement(qword_count); | |
853 __ jcc(Assembler::greater, L_copy_8_bytes); | |
854 __ BIND(L_exit); | |
855 } | |
856 | |
0 | 857 // Copy 64 bytes chunks |
858 // | |
859 // Inputs: | |
860 // from - source array address | |
861 // to_from - destination array address - from | |
862 // qword_count - 8-bytes element count, negative | |
863 // | |
864 void mmx_copy_forward(Register from, Register to_from, Register qword_count) { | |
405 | 865 assert( VM_Version::supports_mmx(), "supported cpu only" ); |
0 | 866 Label L_copy_64_bytes_loop, L_copy_64_bytes, L_copy_8_bytes, L_exit; |
867 // Copy 64-byte chunks | |
868 __ jmpb(L_copy_64_bytes); | |
1365 | 869 __ align(OptoLoopAlignment); |
0 | 870 __ BIND(L_copy_64_bytes_loop); |
871 __ movq(mmx0, Address(from, 0)); | |
872 __ movq(mmx1, Address(from, 8)); | |
873 __ movq(mmx2, Address(from, 16)); | |
874 __ movq(Address(from, to_from, Address::times_1, 0), mmx0); | |
875 __ movq(mmx3, Address(from, 24)); | |
876 __ movq(Address(from, to_from, Address::times_1, 8), mmx1); | |
877 __ movq(mmx4, Address(from, 32)); | |
878 __ movq(Address(from, to_from, Address::times_1, 16), mmx2); | |
879 __ movq(mmx5, Address(from, 40)); | |
880 __ movq(Address(from, to_from, Address::times_1, 24), mmx3); | |
881 __ movq(mmx6, Address(from, 48)); | |
882 __ movq(Address(from, to_from, Address::times_1, 32), mmx4); | |
883 __ movq(mmx7, Address(from, 56)); | |
884 __ movq(Address(from, to_from, Address::times_1, 40), mmx5); | |
885 __ movq(Address(from, to_from, Address::times_1, 48), mmx6); | |
886 __ movq(Address(from, to_from, Address::times_1, 56), mmx7); | |
304 | 887 __ addptr(from, 64); |
0 | 888 __ BIND(L_copy_64_bytes); |
889 __ subl(qword_count, 8); | |
890 __ jcc(Assembler::greaterEqual, L_copy_64_bytes_loop); | |
891 __ addl(qword_count, 8); | |
892 __ jccb(Assembler::zero, L_exit); | |
893 // | |
894 // length is too short, just copy qwords | |
895 // | |
896 __ BIND(L_copy_8_bytes); | |
897 __ movq(mmx0, Address(from, 0)); | |
898 __ movq(Address(from, to_from, Address::times_1), mmx0); | |
304 | 899 __ addptr(from, 8); |
0 | 900 __ decrement(qword_count); |
901 __ jcc(Assembler::greater, L_copy_8_bytes); | |
902 __ BIND(L_exit); | |
903 __ emms(); | |
904 } | |
905 | |
906 address generate_disjoint_copy(BasicType t, bool aligned, | |
907 Address::ScaleFactor sf, | |
2324 | 908 address* entry, const char *name, |
909 bool dest_uninitialized = false) { | |
0 | 910 __ align(CodeEntryAlignment); |
911 StubCodeMark mark(this, "StubRoutines", name); | |
912 address start = __ pc(); | |
913 | |
914 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; | |
915 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_64_bytes; | |
916 | |
304 | 917 int shift = Address::times_ptr - sf; |
0 | 918 |
919 const Register from = rsi; // source array address | |
920 const Register to = rdi; // destination array address | |
921 const Register count = rcx; // elements count | |
922 const Register to_from = to; // (to - from) | |
923 const Register saved_to = rdx; // saved destination array address | |
924 | |
925 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
304 | 926 __ push(rsi); |
927 __ push(rdi); | |
928 __ movptr(from , Address(rsp, 12+ 4)); | |
929 __ movptr(to , Address(rsp, 12+ 8)); | |
0 | 930 __ movl(count, Address(rsp, 12+ 12)); |
2313
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
931 |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
932 if (entry != NULL) { |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
933 *entry = __ pc(); // Entry point from conjoint arraycopy stub. |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
934 BLOCK_COMMENT("Entry:"); |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
935 } |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
936 |
0 | 937 if (t == T_OBJECT) { |
938 __ testl(count, count); | |
939 __ jcc(Assembler::zero, L_0_count); | |
2324 | 940 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
304 | 941 __ mov(saved_to, to); // save 'to' |
0 | 942 } |
943 | |
304 | 944 __ subptr(to, from); // to --> to_from |
0 | 945 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element |
946 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp | |
405 | 947 if (!UseUnalignedLoadStores && !aligned && (t == T_BYTE || t == T_SHORT)) { |
0 | 948 // align source address at 4 bytes address boundary |
949 if (t == T_BYTE) { | |
950 // One byte misalignment happens only for byte arrays | |
951 __ testl(from, 1); | |
952 __ jccb(Assembler::zero, L_skip_align1); | |
953 __ movb(rax, Address(from, 0)); | |
954 __ movb(Address(from, to_from, Address::times_1, 0), rax); | |
955 __ increment(from); | |
956 __ decrement(count); | |
957 __ BIND(L_skip_align1); | |
958 } | |
959 // Two bytes misalignment happens only for byte and short (char) arrays | |
960 __ testl(from, 2); | |
961 __ jccb(Assembler::zero, L_skip_align2); | |
962 __ movw(rax, Address(from, 0)); | |
963 __ movw(Address(from, to_from, Address::times_1, 0), rax); | |
304 | 964 __ addptr(from, 2); |
0 | 965 __ subl(count, 1<<(shift-1)); |
966 __ BIND(L_skip_align2); | |
967 } | |
968 if (!VM_Version::supports_mmx()) { | |
304 | 969 __ mov(rax, count); // save 'count' |
970 __ shrl(count, shift); // bytes count | |
971 __ addptr(to_from, from);// restore 'to' | |
972 __ rep_mov(); | |
973 __ subptr(to_from, from);// restore 'to_from' | |
974 __ mov(count, rax); // restore 'count' | |
0 | 975 __ jmpb(L_copy_2_bytes); // all dwords were copied |
976 } else { | |
405 | 977 if (!UseUnalignedLoadStores) { |
978 // align to 8 bytes, we know we are 4 byte aligned to start | |
979 __ testptr(from, 4); | |
980 __ jccb(Assembler::zero, L_copy_64_bytes); | |
981 __ movl(rax, Address(from, 0)); | |
982 __ movl(Address(from, to_from, Address::times_1, 0), rax); | |
983 __ addptr(from, 4); | |
984 __ subl(count, 1<<shift); | |
985 } | |
0 | 986 __ BIND(L_copy_64_bytes); |
304 | 987 __ mov(rax, count); |
0 | 988 __ shrl(rax, shift+1); // 8 bytes chunk count |
989 // | |
990 // Copy 8-byte chunks through MMX registers, 8 per iteration of the loop | |
991 // | |
405 | 992 if (UseXMMForArrayCopy) { |
993 xmm_copy_forward(from, to_from, rax); | |
994 } else { | |
995 mmx_copy_forward(from, to_from, rax); | |
996 } | |
0 | 997 } |
998 // copy tailing dword | |
999 __ BIND(L_copy_4_bytes); | |
1000 __ testl(count, 1<<shift); | |
1001 __ jccb(Assembler::zero, L_copy_2_bytes); | |
1002 __ movl(rax, Address(from, 0)); | |
1003 __ movl(Address(from, to_from, Address::times_1, 0), rax); | |
1004 if (t == T_BYTE || t == T_SHORT) { | |
304 | 1005 __ addptr(from, 4); |
0 | 1006 __ BIND(L_copy_2_bytes); |
1007 // copy tailing word | |
1008 __ testl(count, 1<<(shift-1)); | |
1009 __ jccb(Assembler::zero, L_copy_byte); | |
1010 __ movw(rax, Address(from, 0)); | |
1011 __ movw(Address(from, to_from, Address::times_1, 0), rax); | |
1012 if (t == T_BYTE) { | |
304 | 1013 __ addptr(from, 2); |
0 | 1014 __ BIND(L_copy_byte); |
1015 // copy tailing byte | |
1016 __ testl(count, 1); | |
1017 __ jccb(Assembler::zero, L_exit); | |
1018 __ movb(rax, Address(from, 0)); | |
1019 __ movb(Address(from, to_from, Address::times_1, 0), rax); | |
1020 __ BIND(L_exit); | |
1021 } else { | |
1022 __ BIND(L_copy_byte); | |
1023 } | |
1024 } else { | |
1025 __ BIND(L_copy_2_bytes); | |
1026 } | |
1027 | |
1028 if (t == T_OBJECT) { | |
1029 __ movl(count, Address(rsp, 12+12)); // reread 'count' | |
304 | 1030 __ mov(to, saved_to); // restore 'to' |
0 | 1031 gen_write_ref_array_post_barrier(to, count); |
1032 __ BIND(L_0_count); | |
1033 } | |
1034 inc_copy_counter_np(t); | |
304 | 1035 __ pop(rdi); |
1036 __ pop(rsi); | |
0 | 1037 __ leave(); // required for proper stackwalking of RuntimeStub frame |
304 | 1038 __ xorptr(rax, rax); // return 0 |
0 | 1039 __ ret(0); |
1040 return start; | |
1041 } | |
1042 | |
1043 | |
1763 | 1044 address generate_fill(BasicType t, bool aligned, const char *name) { |
1045 __ align(CodeEntryAlignment); | |
1046 StubCodeMark mark(this, "StubRoutines", name); | |
1047 address start = __ pc(); | |
1048 | |
1049 BLOCK_COMMENT("Entry:"); | |
1050 | |
1051 const Register to = rdi; // source array address | |
1052 const Register value = rdx; // value | |
1053 const Register count = rsi; // elements count | |
1054 | |
1055 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
1056 __ push(rsi); | |
1057 __ push(rdi); | |
1058 __ movptr(to , Address(rsp, 12+ 4)); | |
1059 __ movl(value, Address(rsp, 12+ 8)); | |
1060 __ movl(count, Address(rsp, 12+ 12)); | |
1061 | |
1062 __ generate_fill(t, aligned, to, value, count, rax, xmm0); | |
1063 | |
1064 __ pop(rdi); | |
1065 __ pop(rsi); | |
1066 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
1067 __ ret(0); | |
1068 return start; | |
1069 } | |
1070 | |
0 | 1071 address generate_conjoint_copy(BasicType t, bool aligned, |
1072 Address::ScaleFactor sf, | |
1073 address nooverlap_target, | |
2324 | 1074 address* entry, const char *name, |
1075 bool dest_uninitialized = false) { | |
0 | 1076 __ align(CodeEntryAlignment); |
1077 StubCodeMark mark(this, "StubRoutines", name); | |
1078 address start = __ pc(); | |
1079 | |
1080 Label L_0_count, L_exit, L_skip_align1, L_skip_align2, L_copy_byte; | |
1081 Label L_copy_2_bytes, L_copy_4_bytes, L_copy_8_bytes, L_copy_8_bytes_loop; | |
1082 | |
304 | 1083 int shift = Address::times_ptr - sf; |
0 | 1084 |
1085 const Register src = rax; // source array address | |
1086 const Register dst = rdx; // destination array address | |
1087 const Register from = rsi; // source array address | |
1088 const Register to = rdi; // destination array address | |
1089 const Register count = rcx; // elements count | |
1090 const Register end = rax; // array end address | |
1091 | |
1092 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
304 | 1093 __ push(rsi); |
1094 __ push(rdi); | |
1095 __ movptr(src , Address(rsp, 12+ 4)); // from | |
1096 __ movptr(dst , Address(rsp, 12+ 8)); // to | |
1097 __ movl2ptr(count, Address(rsp, 12+12)); // count | |
0 | 1098 |
1099 if (entry != NULL) { | |
1100 *entry = __ pc(); // Entry point from generic arraycopy stub. | |
1101 BLOCK_COMMENT("Entry:"); | |
1102 } | |
1103 | |
2313
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1104 // nooverlap_target expects arguments in rsi and rdi. |
304 | 1105 __ mov(from, src); |
1106 __ mov(to , dst); | |
0 | 1107 |
2313
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1108 // arrays overlap test: dispatch to disjoint stub if necessary. |
0 | 1109 RuntimeAddress nooverlap(nooverlap_target); |
304 | 1110 __ cmpptr(dst, src); |
1111 __ lea(end, Address(src, count, sf, 0)); // src + count * elem_size | |
0 | 1112 __ jump_cc(Assembler::belowEqual, nooverlap); |
304 | 1113 __ cmpptr(dst, end); |
0 | 1114 __ jump_cc(Assembler::aboveEqual, nooverlap); |
1115 | |
2313
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1116 if (t == T_OBJECT) { |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1117 __ testl(count, count); |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1118 __ jcc(Assembler::zero, L_0_count); |
2324 | 1119 gen_write_ref_array_pre_barrier(dst, count, dest_uninitialized); |
2313
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1120 } |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1121 |
0 | 1122 // copy from high to low |
1123 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element | |
1124 __ jcc(Assembler::below, L_copy_4_bytes); // use unsigned cmp | |
1125 if (t == T_BYTE || t == T_SHORT) { | |
1126 // Align the end of destination array at 4 bytes address boundary | |
304 | 1127 __ lea(end, Address(dst, count, sf, 0)); |
0 | 1128 if (t == T_BYTE) { |
1129 // One byte misalignment happens only for byte arrays | |
1130 __ testl(end, 1); | |
1131 __ jccb(Assembler::zero, L_skip_align1); | |
1132 __ decrement(count); | |
1133 __ movb(rdx, Address(from, count, sf, 0)); | |
1134 __ movb(Address(to, count, sf, 0), rdx); | |
1135 __ BIND(L_skip_align1); | |
1136 } | |
1137 // Two bytes misalignment happens only for byte and short (char) arrays | |
1138 __ testl(end, 2); | |
1139 __ jccb(Assembler::zero, L_skip_align2); | |
304 | 1140 __ subptr(count, 1<<(shift-1)); |
0 | 1141 __ movw(rdx, Address(from, count, sf, 0)); |
1142 __ movw(Address(to, count, sf, 0), rdx); | |
1143 __ BIND(L_skip_align2); | |
1144 __ cmpl(count, 2<<shift); // Short arrays (< 8 bytes) copy by element | |
1145 __ jcc(Assembler::below, L_copy_4_bytes); | |
1146 } | |
1147 | |
1148 if (!VM_Version::supports_mmx()) { | |
1149 __ std(); | |
304 | 1150 __ mov(rax, count); // Save 'count' |
1151 __ mov(rdx, to); // Save 'to' | |
1152 __ lea(rsi, Address(from, count, sf, -4)); | |
1153 __ lea(rdi, Address(to , count, sf, -4)); | |
1154 __ shrptr(count, shift); // bytes count | |
1155 __ rep_mov(); | |
0 | 1156 __ cld(); |
304 | 1157 __ mov(count, rax); // restore 'count' |
0 | 1158 __ andl(count, (1<<shift)-1); // mask the number of rest elements |
304 | 1159 __ movptr(from, Address(rsp, 12+4)); // reread 'from' |
1160 __ mov(to, rdx); // restore 'to' | |
0 | 1161 __ jmpb(L_copy_2_bytes); // all dword were copied |
1162 } else { | |
1163 // Align to 8 bytes the end of array. It is aligned to 4 bytes already. | |
304 | 1164 __ testptr(end, 4); |
0 | 1165 __ jccb(Assembler::zero, L_copy_8_bytes); |
1166 __ subl(count, 1<<shift); | |
1167 __ movl(rdx, Address(from, count, sf, 0)); | |
1168 __ movl(Address(to, count, sf, 0), rdx); | |
1169 __ jmpb(L_copy_8_bytes); | |
1170 | |
1365 | 1171 __ align(OptoLoopAlignment); |
0 | 1172 // Move 8 bytes |
1173 __ BIND(L_copy_8_bytes_loop); | |
405 | 1174 if (UseXMMForArrayCopy) { |
1175 __ movq(xmm0, Address(from, count, sf, 0)); | |
1176 __ movq(Address(to, count, sf, 0), xmm0); | |
1177 } else { | |
1178 __ movq(mmx0, Address(from, count, sf, 0)); | |
1179 __ movq(Address(to, count, sf, 0), mmx0); | |
1180 } | |
0 | 1181 __ BIND(L_copy_8_bytes); |
1182 __ subl(count, 2<<shift); | |
1183 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); | |
1184 __ addl(count, 2<<shift); | |
405 | 1185 if (!UseXMMForArrayCopy) { |
1186 __ emms(); | |
1187 } | |
0 | 1188 } |
1189 __ BIND(L_copy_4_bytes); | |
1190 // copy prefix qword | |
1191 __ testl(count, 1<<shift); | |
1192 __ jccb(Assembler::zero, L_copy_2_bytes); | |
1193 __ movl(rdx, Address(from, count, sf, -4)); | |
1194 __ movl(Address(to, count, sf, -4), rdx); | |
1195 | |
1196 if (t == T_BYTE || t == T_SHORT) { | |
1197 __ subl(count, (1<<shift)); | |
1198 __ BIND(L_copy_2_bytes); | |
1199 // copy prefix dword | |
1200 __ testl(count, 1<<(shift-1)); | |
1201 __ jccb(Assembler::zero, L_copy_byte); | |
1202 __ movw(rdx, Address(from, count, sf, -2)); | |
1203 __ movw(Address(to, count, sf, -2), rdx); | |
1204 if (t == T_BYTE) { | |
1205 __ subl(count, 1<<(shift-1)); | |
1206 __ BIND(L_copy_byte); | |
1207 // copy prefix byte | |
1208 __ testl(count, 1); | |
1209 __ jccb(Assembler::zero, L_exit); | |
1210 __ movb(rdx, Address(from, 0)); | |
1211 __ movb(Address(to, 0), rdx); | |
1212 __ BIND(L_exit); | |
1213 } else { | |
1214 __ BIND(L_copy_byte); | |
1215 } | |
1216 } else { | |
1217 __ BIND(L_copy_2_bytes); | |
1218 } | |
1219 if (t == T_OBJECT) { | |
304 | 1220 __ movl2ptr(count, Address(rsp, 12+12)); // reread count |
0 | 1221 gen_write_ref_array_post_barrier(to, count); |
1222 __ BIND(L_0_count); | |
1223 } | |
1224 inc_copy_counter_np(t); | |
304 | 1225 __ pop(rdi); |
1226 __ pop(rsi); | |
0 | 1227 __ leave(); // required for proper stackwalking of RuntimeStub frame |
304 | 1228 __ xorptr(rax, rax); // return 0 |
0 | 1229 __ ret(0); |
1230 return start; | |
1231 } | |
1232 | |
1233 | |
1234 address generate_disjoint_long_copy(address* entry, const char *name) { | |
1235 __ align(CodeEntryAlignment); | |
1236 StubCodeMark mark(this, "StubRoutines", name); | |
1237 address start = __ pc(); | |
1238 | |
1239 Label L_copy_8_bytes, L_copy_8_bytes_loop; | |
1240 const Register from = rax; // source array address | |
1241 const Register to = rdx; // destination array address | |
1242 const Register count = rcx; // elements count | |
1243 const Register to_from = rdx; // (to - from) | |
1244 | |
1245 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
304 | 1246 __ movptr(from , Address(rsp, 8+0)); // from |
1247 __ movptr(to , Address(rsp, 8+4)); // to | |
1248 __ movl2ptr(count, Address(rsp, 8+8)); // count | |
0 | 1249 |
1250 *entry = __ pc(); // Entry point from conjoint arraycopy stub. | |
1251 BLOCK_COMMENT("Entry:"); | |
1252 | |
304 | 1253 __ subptr(to, from); // to --> to_from |
0 | 1254 if (VM_Version::supports_mmx()) { |
405 | 1255 if (UseXMMForArrayCopy) { |
1256 xmm_copy_forward(from, to_from, count); | |
1257 } else { | |
1258 mmx_copy_forward(from, to_from, count); | |
1259 } | |
0 | 1260 } else { |
1261 __ jmpb(L_copy_8_bytes); | |
1365 | 1262 __ align(OptoLoopAlignment); |
0 | 1263 __ BIND(L_copy_8_bytes_loop); |
1264 __ fild_d(Address(from, 0)); | |
1265 __ fistp_d(Address(from, to_from, Address::times_1)); | |
304 | 1266 __ addptr(from, 8); |
0 | 1267 __ BIND(L_copy_8_bytes); |
1268 __ decrement(count); | |
1269 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); | |
1270 } | |
1271 inc_copy_counter_np(T_LONG); | |
1272 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
304 | 1273 __ xorptr(rax, rax); // return 0 |
0 | 1274 __ ret(0); |
1275 return start; | |
1276 } | |
1277 | |
1278 address generate_conjoint_long_copy(address nooverlap_target, | |
1279 address* entry, const char *name) { | |
1280 __ align(CodeEntryAlignment); | |
1281 StubCodeMark mark(this, "StubRoutines", name); | |
1282 address start = __ pc(); | |
1283 | |
1284 Label L_copy_8_bytes, L_copy_8_bytes_loop; | |
1285 const Register from = rax; // source array address | |
1286 const Register to = rdx; // destination array address | |
1287 const Register count = rcx; // elements count | |
1288 const Register end_from = rax; // source array end address | |
1289 | |
1290 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
304 | 1291 __ movptr(from , Address(rsp, 8+0)); // from |
1292 __ movptr(to , Address(rsp, 8+4)); // to | |
1293 __ movl2ptr(count, Address(rsp, 8+8)); // count | |
0 | 1294 |
1295 *entry = __ pc(); // Entry point from generic arraycopy stub. | |
1296 BLOCK_COMMENT("Entry:"); | |
1297 | |
1298 // arrays overlap test | |
304 | 1299 __ cmpptr(to, from); |
0 | 1300 RuntimeAddress nooverlap(nooverlap_target); |
1301 __ jump_cc(Assembler::belowEqual, nooverlap); | |
304 | 1302 __ lea(end_from, Address(from, count, Address::times_8, 0)); |
1303 __ cmpptr(to, end_from); | |
1304 __ movptr(from, Address(rsp, 8)); // from | |
0 | 1305 __ jump_cc(Assembler::aboveEqual, nooverlap); |
1306 | |
1307 __ jmpb(L_copy_8_bytes); | |
1308 | |
1365 | 1309 __ align(OptoLoopAlignment); |
0 | 1310 __ BIND(L_copy_8_bytes_loop); |
1311 if (VM_Version::supports_mmx()) { | |
405 | 1312 if (UseXMMForArrayCopy) { |
1313 __ movq(xmm0, Address(from, count, Address::times_8)); | |
1314 __ movq(Address(to, count, Address::times_8), xmm0); | |
1315 } else { | |
1316 __ movq(mmx0, Address(from, count, Address::times_8)); | |
1317 __ movq(Address(to, count, Address::times_8), mmx0); | |
1318 } | |
0 | 1319 } else { |
1320 __ fild_d(Address(from, count, Address::times_8)); | |
1321 __ fistp_d(Address(to, count, Address::times_8)); | |
1322 } | |
1323 __ BIND(L_copy_8_bytes); | |
1324 __ decrement(count); | |
1325 __ jcc(Assembler::greaterEqual, L_copy_8_bytes_loop); | |
1326 | |
405 | 1327 if (VM_Version::supports_mmx() && !UseXMMForArrayCopy) { |
0 | 1328 __ emms(); |
1329 } | |
1330 inc_copy_counter_np(T_LONG); | |
1331 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
304 | 1332 __ xorptr(rax, rax); // return 0 |
0 | 1333 __ ret(0); |
1334 return start; | |
1335 } | |
1336 | |
1337 | |
1338 // Helper for generating a dynamic type check. | |
1339 // The sub_klass must be one of {rbx, rdx, rsi}. | |
1340 // The temp is killed. | |
1341 void generate_type_check(Register sub_klass, | |
1342 Address& super_check_offset_addr, | |
1343 Address& super_klass_addr, | |
1344 Register temp, | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1345 Label* L_success, Label* L_failure) { |
0 | 1346 BLOCK_COMMENT("type_check:"); |
1347 | |
1348 Label L_fallthrough; | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1349 #define LOCAL_JCC(assembler_con, label_ptr) \ |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1350 if (label_ptr != NULL) __ jcc(assembler_con, *(label_ptr)); \ |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1351 else __ jcc(assembler_con, L_fallthrough) /*omit semi*/ |
0 | 1352 |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1353 // The following is a strange variation of the fast path which requires |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1354 // one less register, because needed values are on the argument stack. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1355 // __ check_klass_subtype_fast_path(sub_klass, *super_klass*, temp, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1356 // L_success, L_failure, NULL); |
0 | 1357 assert_different_registers(sub_klass, temp); |
1358 | |
4762
069ab3f976d3
7118863: Move sizeof(klassOopDesc) into the *Klass::*_offset_in_bytes() functions
stefank
parents:
3960
diff
changeset
|
1359 int sc_offset = in_bytes(Klass::secondary_super_cache_offset()); |
0 | 1360 |
1361 // if the pointers are equal, we are done (e.g., String[] elements) | |
304 | 1362 __ cmpptr(sub_klass, super_klass_addr); |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1363 LOCAL_JCC(Assembler::equal, L_success); |
0 | 1364 |
1365 // check the supertype display: | |
304 | 1366 __ movl2ptr(temp, super_check_offset_addr); |
0 | 1367 Address super_check_addr(sub_klass, temp, Address::times_1, 0); |
304 | 1368 __ movptr(temp, super_check_addr); // load displayed supertype |
1369 __ cmpptr(temp, super_klass_addr); // test the super type | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1370 LOCAL_JCC(Assembler::equal, L_success); |
0 | 1371 |
1372 // if it was a primary super, we can just fail immediately | |
1373 __ cmpl(super_check_offset_addr, sc_offset); | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1374 LOCAL_JCC(Assembler::notEqual, L_failure); |
0 | 1375 |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1376 // The repne_scan instruction uses fixed registers, which will get spilled. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1377 // We happen to know this works best when super_klass is in rax. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1378 Register super_klass = temp; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1379 __ movptr(super_klass, super_klass_addr); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1380 __ check_klass_subtype_slow_path(sub_klass, super_klass, noreg, noreg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1381 L_success, L_failure); |
0 | 1382 |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1383 __ bind(L_fallthrough); |
0 | 1384 |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1385 if (L_success == NULL) { BLOCK_COMMENT("L_success:"); } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1386 if (L_failure == NULL) { BLOCK_COMMENT("L_failure:"); } |
0 | 1387 |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
533
diff
changeset
|
1388 #undef LOCAL_JCC |
0 | 1389 } |
1390 | |
1391 // | |
1392 // Generate checkcasting array copy stub | |
1393 // | |
1394 // Input: | |
1395 // 4(rsp) - source array address | |
1396 // 8(rsp) - destination array address | |
1397 // 12(rsp) - element count, can be zero | |
1398 // 16(rsp) - size_t ckoff (super_check_offset) | |
1399 // 20(rsp) - oop ckval (super_klass) | |
1400 // | |
1401 // Output: | |
1402 // rax, == 0 - success | |
1403 // rax, == -1^K - failure, where K is partial transfer count | |
1404 // | |
2324 | 1405 address generate_checkcast_copy(const char *name, address* entry, bool dest_uninitialized = false) { |
0 | 1406 __ align(CodeEntryAlignment); |
1407 StubCodeMark mark(this, "StubRoutines", name); | |
1408 address start = __ pc(); | |
1409 | |
1410 Label L_load_element, L_store_element, L_do_card_marks, L_done; | |
1411 | |
1412 // register use: | |
1413 // rax, rdx, rcx -- loop control (end_from, end_to, count) | |
1414 // rdi, rsi -- element access (oop, klass) | |
1415 // rbx, -- temp | |
1416 const Register from = rax; // source array address | |
1417 const Register to = rdx; // destination array address | |
1418 const Register length = rcx; // elements count | |
1419 const Register elem = rdi; // each oop copied | |
1420 const Register elem_klass = rsi; // each elem._klass (sub_klass) | |
1421 const Register temp = rbx; // lone remaining temp | |
1422 | |
1423 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
1424 | |
304 | 1425 __ push(rsi); |
1426 __ push(rdi); | |
1427 __ push(rbx); | |
0 | 1428 |
1429 Address from_arg(rsp, 16+ 4); // from | |
1430 Address to_arg(rsp, 16+ 8); // to | |
1431 Address length_arg(rsp, 16+12); // elements count | |
1432 Address ckoff_arg(rsp, 16+16); // super_check_offset | |
1433 Address ckval_arg(rsp, 16+20); // super_klass | |
1434 | |
1435 // Load up: | |
304 | 1436 __ movptr(from, from_arg); |
1437 __ movptr(to, to_arg); | |
1438 __ movl2ptr(length, length_arg); | |
0 | 1439 |
2313
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1440 if (entry != NULL) { |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1441 *entry = __ pc(); // Entry point from generic arraycopy stub. |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1442 BLOCK_COMMENT("Entry:"); |
d89a22843c62
7020521: arraycopy stubs place prebarriers incorrectly
iveresov
parents:
2245
diff
changeset
|
1443 } |
0 | 1444 |
1445 //--------------------------------------------------------------- | |
1446 // Assembler stub will be used for this call to arraycopy | |
1447 // if the two arrays are subtypes of Object[] but the | |
1448 // destination array type is not equal to or a supertype | |
1449 // of the source type. Each element must be separately | |
1450 // checked. | |
1451 | |
1452 // Loop-invariant addresses. They are exclusive end pointers. | |
304 | 1453 Address end_from_addr(from, length, Address::times_ptr, 0); |
1454 Address end_to_addr(to, length, Address::times_ptr, 0); | |
0 | 1455 |
1456 Register end_from = from; // re-use | |
1457 Register end_to = to; // re-use | |
1458 Register count = length; // re-use | |
1459 | |
1460 // Loop-variant addresses. They assume post-incremented count < 0. | |
304 | 1461 Address from_element_addr(end_from, count, Address::times_ptr, 0); |
1462 Address to_element_addr(end_to, count, Address::times_ptr, 0); | |
0 | 1463 Address elem_klass_addr(elem, oopDesc::klass_offset_in_bytes()); |
1464 | |
1465 // Copy from low to high addresses, indexed from the end of each array. | |
2324 | 1466 gen_write_ref_array_pre_barrier(to, count, dest_uninitialized); |
304 | 1467 __ lea(end_from, end_from_addr); |
1468 __ lea(end_to, end_to_addr); | |
0 | 1469 assert(length == count, ""); // else fix next line: |
304 | 1470 __ negptr(count); // negate and test the length |
0 | 1471 __ jccb(Assembler::notZero, L_load_element); |
1472 | |
1473 // Empty array: Nothing to do. | |
304 | 1474 __ xorptr(rax, rax); // return 0 on (trivial) success |
0 | 1475 __ jmp(L_done); |
1476 | |
1477 // ======== begin loop ======== | |
1478 // (Loop is rotated; its entry is L_load_element.) | |
1479 // Loop control: | |
1480 // for (count = -count; count != 0; count++) | |
1481 // Base pointers src, dst are biased by 8*count,to last element. | |
1365 | 1482 __ align(OptoLoopAlignment); |
0 | 1483 |
1484 __ BIND(L_store_element); | |
304 | 1485 __ movptr(to_element_addr, elem); // store the oop |
0 | 1486 __ increment(count); // increment the count toward zero |
1487 __ jccb(Assembler::zero, L_do_card_marks); | |
1488 | |
1489 // ======== loop entry is here ======== | |
1490 __ BIND(L_load_element); | |
304 | 1491 __ movptr(elem, from_element_addr); // load the oop |
1492 __ testptr(elem, elem); | |
0 | 1493 __ jccb(Assembler::zero, L_store_element); |
1494 | |
1495 // (Could do a trick here: Remember last successful non-null | |
1496 // element stored and make a quick oop equality check on it.) | |
1497 | |
304 | 1498 __ movptr(elem_klass, elem_klass_addr); // query the object klass |
0 | 1499 generate_type_check(elem_klass, ckoff_arg, ckval_arg, temp, |
1500 &L_store_element, NULL); | |
1501 // (On fall-through, we have failed the element type check.) | |
1502 // ======== end loop ======== | |
1503 | |
1504 // It was a real error; we must depend on the caller to finish the job. | |
19
a73cc31728fe
6614036: REGRESSION: Java server x86 VM intermittently crash with SIGSEGV (0xb)
rasbold
parents:
16
diff
changeset
|
1505 // Register "count" = -1 * number of *remaining* oops, length_arg = *total* oops. |
a73cc31728fe
6614036: REGRESSION: Java server x86 VM intermittently crash with SIGSEGV (0xb)
rasbold
parents:
16
diff
changeset
|
1506 // Emit GC store barriers for the oops we have copied (length_arg + count), |
0 | 1507 // and report their number to the caller. |
1508 __ addl(count, length_arg); // transfers = (length - remaining) | |
304 | 1509 __ movl2ptr(rax, count); // save the value |
1510 __ notptr(rax); // report (-1^K) to caller | |
1511 __ movptr(to, to_arg); // reload | |
0 | 1512 assert_different_registers(to, count, rax); |
1513 gen_write_ref_array_post_barrier(to, count); | |
1514 __ jmpb(L_done); | |
1515 | |
1516 // Come here on success only. | |
1517 __ BIND(L_do_card_marks); | |
304 | 1518 __ movl2ptr(count, length_arg); |
1519 __ movptr(to, to_arg); // reload | |
0 | 1520 gen_write_ref_array_post_barrier(to, count); |
304 | 1521 __ xorptr(rax, rax); // return 0 on success |
0 | 1522 |
1523 // Common exit point (success or failure). | |
1524 __ BIND(L_done); | |
304 | 1525 __ pop(rbx); |
1526 __ pop(rdi); | |
1527 __ pop(rsi); | |
0 | 1528 inc_counter_np(SharedRuntime::_checkcast_array_copy_ctr); |
1529 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
1530 __ ret(0); | |
1531 | |
1532 return start; | |
1533 } | |
1534 | |
1535 // | |
1536 // Generate 'unsafe' array copy stub | |
1537 // Though just as safe as the other stubs, it takes an unscaled | |
1538 // size_t argument instead of an element count. | |
1539 // | |
1540 // Input: | |
1541 // 4(rsp) - source array address | |
1542 // 8(rsp) - destination array address | |
1543 // 12(rsp) - byte count, can be zero | |
1544 // | |
1545 // Output: | |
1546 // rax, == 0 - success | |
1547 // rax, == -1 - need to call System.arraycopy | |
1548 // | |
1549 // Examines the alignment of the operands and dispatches | |
1550 // to a long, int, short, or byte copy loop. | |
1551 // | |
1552 address generate_unsafe_copy(const char *name, | |
1553 address byte_copy_entry, | |
1554 address short_copy_entry, | |
1555 address int_copy_entry, | |
1556 address long_copy_entry) { | |
1557 | |
1558 Label L_long_aligned, L_int_aligned, L_short_aligned; | |
1559 | |
1560 __ align(CodeEntryAlignment); | |
1561 StubCodeMark mark(this, "StubRoutines", name); | |
1562 address start = __ pc(); | |
1563 | |
1564 const Register from = rax; // source array address | |
1565 const Register to = rdx; // destination array address | |
1566 const Register count = rcx; // elements count | |
1567 | |
1568 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
304 | 1569 __ push(rsi); |
1570 __ push(rdi); | |
0 | 1571 Address from_arg(rsp, 12+ 4); // from |
1572 Address to_arg(rsp, 12+ 8); // to | |
1573 Address count_arg(rsp, 12+12); // byte count | |
1574 | |
1575 // Load up: | |
304 | 1576 __ movptr(from , from_arg); |
1577 __ movptr(to , to_arg); | |
1578 __ movl2ptr(count, count_arg); | |
0 | 1579 |
1580 // bump this on entry, not on exit: | |
1581 inc_counter_np(SharedRuntime::_unsafe_array_copy_ctr); | |
1582 | |
1583 const Register bits = rsi; | |
304 | 1584 __ mov(bits, from); |
1585 __ orptr(bits, to); | |
1586 __ orptr(bits, count); | |
0 | 1587 |
1588 __ testl(bits, BytesPerLong-1); | |
1589 __ jccb(Assembler::zero, L_long_aligned); | |
1590 | |
1591 __ testl(bits, BytesPerInt-1); | |
1592 __ jccb(Assembler::zero, L_int_aligned); | |
1593 | |
1594 __ testl(bits, BytesPerShort-1); | |
1595 __ jump_cc(Assembler::notZero, RuntimeAddress(byte_copy_entry)); | |
1596 | |
1597 __ BIND(L_short_aligned); | |
304 | 1598 __ shrptr(count, LogBytesPerShort); // size => short_count |
0 | 1599 __ movl(count_arg, count); // update 'count' |
1600 __ jump(RuntimeAddress(short_copy_entry)); | |
1601 | |
1602 __ BIND(L_int_aligned); | |
304 | 1603 __ shrptr(count, LogBytesPerInt); // size => int_count |
0 | 1604 __ movl(count_arg, count); // update 'count' |
1605 __ jump(RuntimeAddress(int_copy_entry)); | |
1606 | |
1607 __ BIND(L_long_aligned); | |
304 | 1608 __ shrptr(count, LogBytesPerLong); // size => qword_count |
0 | 1609 __ movl(count_arg, count); // update 'count' |
304 | 1610 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. |
1611 __ pop(rsi); | |
0 | 1612 __ jump(RuntimeAddress(long_copy_entry)); |
1613 | |
1614 return start; | |
1615 } | |
1616 | |
1617 | |
1618 // Perform range checks on the proposed arraycopy. | |
1619 // Smashes src_pos and dst_pos. (Uses them up for temps.) | |
1620 void arraycopy_range_checks(Register src, | |
1621 Register src_pos, | |
1622 Register dst, | |
1623 Register dst_pos, | |
1624 Address& length, | |
1625 Label& L_failed) { | |
1626 BLOCK_COMMENT("arraycopy_range_checks:"); | |
1627 const Register src_end = src_pos; // source array end position | |
1628 const Register dst_end = dst_pos; // destination array end position | |
1629 __ addl(src_end, length); // src_pos + length | |
1630 __ addl(dst_end, length); // dst_pos + length | |
1631 | |
1632 // if (src_pos + length > arrayOop(src)->length() ) FAIL; | |
1633 __ cmpl(src_end, Address(src, arrayOopDesc::length_offset_in_bytes())); | |
1634 __ jcc(Assembler::above, L_failed); | |
1635 | |
1636 // if (dst_pos + length > arrayOop(dst)->length() ) FAIL; | |
1637 __ cmpl(dst_end, Address(dst, arrayOopDesc::length_offset_in_bytes())); | |
1638 __ jcc(Assembler::above, L_failed); | |
1639 | |
1640 BLOCK_COMMENT("arraycopy_range_checks done"); | |
1641 } | |
1642 | |
1643 | |
1644 // | |
1645 // Generate generic array copy stubs | |
1646 // | |
1647 // Input: | |
1648 // 4(rsp) - src oop | |
1649 // 8(rsp) - src_pos | |
1650 // 12(rsp) - dst oop | |
1651 // 16(rsp) - dst_pos | |
1652 // 20(rsp) - element count | |
1653 // | |
1654 // Output: | |
1655 // rax, == 0 - success | |
1656 // rax, == -1^K - failure, where K is partial transfer count | |
1657 // | |
1658 address generate_generic_copy(const char *name, | |
1659 address entry_jbyte_arraycopy, | |
1660 address entry_jshort_arraycopy, | |
1661 address entry_jint_arraycopy, | |
1662 address entry_oop_arraycopy, | |
1663 address entry_jlong_arraycopy, | |
1664 address entry_checkcast_arraycopy) { | |
1665 Label L_failed, L_failed_0, L_objArray; | |
1666 | |
1667 { int modulus = CodeEntryAlignment; | |
1668 int target = modulus - 5; // 5 = sizeof jmp(L_failed) | |
1669 int advance = target - (__ offset() % modulus); | |
1670 if (advance < 0) advance += modulus; | |
1671 if (advance > 0) __ nop(advance); | |
1672 } | |
1673 StubCodeMark mark(this, "StubRoutines", name); | |
1674 | |
1675 // Short-hop target to L_failed. Makes for denser prologue code. | |
1676 __ BIND(L_failed_0); | |
1677 __ jmp(L_failed); | |
1678 assert(__ offset() % CodeEntryAlignment == 0, "no further alignment needed"); | |
1679 | |
1680 __ align(CodeEntryAlignment); | |
1681 address start = __ pc(); | |
1682 | |
1683 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
304 | 1684 __ push(rsi); |
1685 __ push(rdi); | |
0 | 1686 |
1687 // bump this on entry, not on exit: | |
1688 inc_counter_np(SharedRuntime::_generic_array_copy_ctr); | |
1689 | |
1690 // Input values | |
1691 Address SRC (rsp, 12+ 4); | |
1692 Address SRC_POS (rsp, 12+ 8); | |
1693 Address DST (rsp, 12+12); | |
1694 Address DST_POS (rsp, 12+16); | |
1695 Address LENGTH (rsp, 12+20); | |
1696 | |
1697 //----------------------------------------------------------------------- | |
1698 // Assembler stub will be used for this call to arraycopy | |
1699 // if the following conditions are met: | |
1700 // | |
1701 // (1) src and dst must not be null. | |
1702 // (2) src_pos must not be negative. | |
1703 // (3) dst_pos must not be negative. | |
1704 // (4) length must not be negative. | |
1705 // (5) src klass and dst klass should be the same and not NULL. | |
1706 // (6) src and dst should be arrays. | |
1707 // (7) src_pos + length must not exceed length of src. | |
1708 // (8) dst_pos + length must not exceed length of dst. | |
1709 // | |
1710 | |
1711 const Register src = rax; // source array oop | |
1712 const Register src_pos = rsi; | |
1713 const Register dst = rdx; // destination array oop | |
1714 const Register dst_pos = rdi; | |
1715 const Register length = rcx; // transfer count | |
1716 | |
1717 // if (src == NULL) return -1; | |
304 | 1718 __ movptr(src, SRC); // src oop |
1719 __ testptr(src, src); | |
0 | 1720 __ jccb(Assembler::zero, L_failed_0); |
1721 | |
1722 // if (src_pos < 0) return -1; | |
304 | 1723 __ movl2ptr(src_pos, SRC_POS); // src_pos |
0 | 1724 __ testl(src_pos, src_pos); |
1725 __ jccb(Assembler::negative, L_failed_0); | |
1726 | |
1727 // if (dst == NULL) return -1; | |
304 | 1728 __ movptr(dst, DST); // dst oop |
1729 __ testptr(dst, dst); | |
0 | 1730 __ jccb(Assembler::zero, L_failed_0); |
1731 | |
1732 // if (dst_pos < 0) return -1; | |
304 | 1733 __ movl2ptr(dst_pos, DST_POS); // dst_pos |
0 | 1734 __ testl(dst_pos, dst_pos); |
1735 __ jccb(Assembler::negative, L_failed_0); | |
1736 | |
1737 // if (length < 0) return -1; | |
304 | 1738 __ movl2ptr(length, LENGTH); // length |
0 | 1739 __ testl(length, length); |
1740 __ jccb(Assembler::negative, L_failed_0); | |
1741 | |
1742 // if (src->klass() == NULL) return -1; | |
1743 Address src_klass_addr(src, oopDesc::klass_offset_in_bytes()); | |
1744 Address dst_klass_addr(dst, oopDesc::klass_offset_in_bytes()); | |
1745 const Register rcx_src_klass = rcx; // array klass | |
304 | 1746 __ movptr(rcx_src_klass, Address(src, oopDesc::klass_offset_in_bytes())); |
0 | 1747 |
1748 #ifdef ASSERT | |
1749 // assert(src->klass() != NULL); | |
1750 BLOCK_COMMENT("assert klasses not null"); | |
1751 { Label L1, L2; | |
304 | 1752 __ testptr(rcx_src_klass, rcx_src_klass); |
0 | 1753 __ jccb(Assembler::notZero, L2); // it is broken if klass is NULL |
1754 __ bind(L1); | |
1755 __ stop("broken null klass"); | |
1756 __ bind(L2); | |
304 | 1757 __ cmpptr(dst_klass_addr, (int32_t)NULL_WORD); |
0 | 1758 __ jccb(Assembler::equal, L1); // this would be broken also |
1759 BLOCK_COMMENT("assert done"); | |
1760 } | |
1761 #endif //ASSERT | |
1762 | |
1763 // Load layout helper (32-bits) | |
1764 // | |
1765 // |array_tag| | header_size | element_type | |log2_element_size| | |
1766 // 32 30 24 16 8 2 0 | |
1767 // | |
1768 // array_tag: typeArray = 0x3, objArray = 0x2, non-array = 0x0 | |
1769 // | |
1770 | |
4762
069ab3f976d3
7118863: Move sizeof(klassOopDesc) into the *Klass::*_offset_in_bytes() functions
stefank
parents:
3960
diff
changeset
|
1771 int lh_offset = in_bytes(Klass::layout_helper_offset()); |
0 | 1772 Address src_klass_lh_addr(rcx_src_klass, lh_offset); |
1773 | |
1774 // Handle objArrays completely differently... | |
1775 jint objArray_lh = Klass::array_layout_helper(T_OBJECT); | |
1776 __ cmpl(src_klass_lh_addr, objArray_lh); | |
1777 __ jcc(Assembler::equal, L_objArray); | |
1778 | |
1779 // if (src->klass() != dst->klass()) return -1; | |
304 | 1780 __ cmpptr(rcx_src_klass, dst_klass_addr); |
0 | 1781 __ jccb(Assembler::notEqual, L_failed_0); |
1782 | |
1783 const Register rcx_lh = rcx; // layout helper | |
1784 assert(rcx_lh == rcx_src_klass, "known alias"); | |
1785 __ movl(rcx_lh, src_klass_lh_addr); | |
1786 | |
1787 // if (!src->is_Array()) return -1; | |
1788 __ cmpl(rcx_lh, Klass::_lh_neutral_value); | |
1789 __ jcc(Assembler::greaterEqual, L_failed_0); // signed cmp | |
1790 | |
1791 // At this point, it is known to be a typeArray (array_tag 0x3). | |
1792 #ifdef ASSERT | |
1793 { Label L; | |
1794 __ cmpl(rcx_lh, (Klass::_lh_array_tag_type_value << Klass::_lh_array_tag_shift)); | |
1795 __ jcc(Assembler::greaterEqual, L); // signed cmp | |
1796 __ stop("must be a primitive array"); | |
1797 __ bind(L); | |
1798 } | |
1799 #endif | |
1800 | |
1801 assert_different_registers(src, src_pos, dst, dst_pos, rcx_lh); | |
1802 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); | |
1803 | |
6831
d8ce2825b193
8000213: NPG: Should have renamed arrayKlass and typeArrayKlass
coleenp
parents:
6725
diff
changeset
|
1804 // TypeArrayKlass |
0 | 1805 // |
1806 // src_addr = (src + array_header_in_bytes()) + (src_pos << log2elemsize); | |
1807 // dst_addr = (dst + array_header_in_bytes()) + (dst_pos << log2elemsize); | |
1808 // | |
1809 const Register rsi_offset = rsi; // array offset | |
1810 const Register src_array = src; // src array offset | |
1811 const Register dst_array = dst; // dst array offset | |
1812 const Register rdi_elsize = rdi; // log2 element size | |
1813 | |
304 | 1814 __ mov(rsi_offset, rcx_lh); |
1815 __ shrptr(rsi_offset, Klass::_lh_header_size_shift); | |
1816 __ andptr(rsi_offset, Klass::_lh_header_size_mask); // array_offset | |
1817 __ addptr(src_array, rsi_offset); // src array offset | |
1818 __ addptr(dst_array, rsi_offset); // dst array offset | |
1819 __ andptr(rcx_lh, Klass::_lh_log2_element_size_mask); // log2 elsize | |
0 | 1820 |
1821 // next registers should be set before the jump to corresponding stub | |
1822 const Register from = src; // source array address | |
1823 const Register to = dst; // destination array address | |
1824 const Register count = rcx; // elements count | |
1825 // some of them should be duplicated on stack | |
1826 #define FROM Address(rsp, 12+ 4) | |
1827 #define TO Address(rsp, 12+ 8) // Not used now | |
1828 #define COUNT Address(rsp, 12+12) // Only for oop arraycopy | |
1829 | |
1830 BLOCK_COMMENT("scale indexes to element size"); | |
304 | 1831 __ movl2ptr(rsi, SRC_POS); // src_pos |
1832 __ shlptr(rsi); // src_pos << rcx (log2 elsize) | |
0 | 1833 assert(src_array == from, ""); |
304 | 1834 __ addptr(from, rsi); // from = src_array + SRC_POS << log2 elsize |
1835 __ movl2ptr(rdi, DST_POS); // dst_pos | |
1836 __ shlptr(rdi); // dst_pos << rcx (log2 elsize) | |
0 | 1837 assert(dst_array == to, ""); |
304 | 1838 __ addptr(to, rdi); // to = dst_array + DST_POS << log2 elsize |
1839 __ movptr(FROM, from); // src_addr | |
1840 __ mov(rdi_elsize, rcx_lh); // log2 elsize | |
1841 __ movl2ptr(count, LENGTH); // elements count | |
0 | 1842 |
1843 BLOCK_COMMENT("choose copy loop based on element size"); | |
1844 __ cmpl(rdi_elsize, 0); | |
1845 | |
1846 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jbyte_arraycopy)); | |
1847 __ cmpl(rdi_elsize, LogBytesPerShort); | |
1848 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jshort_arraycopy)); | |
1849 __ cmpl(rdi_elsize, LogBytesPerInt); | |
1850 __ jump_cc(Assembler::equal, RuntimeAddress(entry_jint_arraycopy)); | |
1851 #ifdef ASSERT | |
1852 __ cmpl(rdi_elsize, LogBytesPerLong); | |
1853 __ jccb(Assembler::notEqual, L_failed); | |
1854 #endif | |
304 | 1855 __ pop(rdi); // Do pops here since jlong_arraycopy stub does not do it. |
1856 __ pop(rsi); | |
0 | 1857 __ jump(RuntimeAddress(entry_jlong_arraycopy)); |
1858 | |
1859 __ BIND(L_failed); | |
304 | 1860 __ xorptr(rax, rax); |
1861 __ notptr(rax); // return -1 | |
1862 __ pop(rdi); | |
1863 __ pop(rsi); | |
0 | 1864 __ leave(); // required for proper stackwalking of RuntimeStub frame |
1865 __ ret(0); | |
1866 | |
6831
d8ce2825b193
8000213: NPG: Should have renamed arrayKlass and typeArrayKlass
coleenp
parents:
6725
diff
changeset
|
1867 // ObjArrayKlass |
0 | 1868 __ BIND(L_objArray); |
1869 // live at this point: rcx_src_klass, src[_pos], dst[_pos] | |
1870 | |
1871 Label L_plain_copy, L_checkcast_copy; | |
1872 // test array classes for subtyping | |
304 | 1873 __ cmpptr(rcx_src_klass, dst_klass_addr); // usual case is exact equality |
0 | 1874 __ jccb(Assembler::notEqual, L_checkcast_copy); |
1875 | |
1876 // Identically typed arrays can be copied without element-wise checks. | |
1877 assert_different_registers(src, src_pos, dst, dst_pos, rcx_src_klass); | |
1878 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); | |
1879 | |
1880 __ BIND(L_plain_copy); | |
304 | 1881 __ movl2ptr(count, LENGTH); // elements count |
1882 __ movl2ptr(src_pos, SRC_POS); // reload src_pos | |
1883 __ lea(from, Address(src, src_pos, Address::times_ptr, | |
1884 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // src_addr | |
1885 __ movl2ptr(dst_pos, DST_POS); // reload dst_pos | |
1886 __ lea(to, Address(dst, dst_pos, Address::times_ptr, | |
1887 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); // dst_addr | |
1888 __ movptr(FROM, from); // src_addr | |
1889 __ movptr(TO, to); // dst_addr | |
0 | 1890 __ movl(COUNT, count); // count |
1891 __ jump(RuntimeAddress(entry_oop_arraycopy)); | |
1892 | |
1893 __ BIND(L_checkcast_copy); | |
1894 // live at this point: rcx_src_klass, dst[_pos], src[_pos] | |
1895 { | |
1896 // Handy offsets: | |
6831
d8ce2825b193
8000213: NPG: Should have renamed arrayKlass and typeArrayKlass
coleenp
parents:
6725
diff
changeset
|
1897 int ek_offset = in_bytes(ObjArrayKlass::element_klass_offset()); |
4762
069ab3f976d3
7118863: Move sizeof(klassOopDesc) into the *Klass::*_offset_in_bytes() functions
stefank
parents:
3960
diff
changeset
|
1898 int sco_offset = in_bytes(Klass::super_check_offset_offset()); |
0 | 1899 |
1900 Register rsi_dst_klass = rsi; | |
1901 Register rdi_temp = rdi; | |
1902 assert(rsi_dst_klass == src_pos, "expected alias w/ src_pos"); | |
1903 assert(rdi_temp == dst_pos, "expected alias w/ dst_pos"); | |
1904 Address dst_klass_lh_addr(rsi_dst_klass, lh_offset); | |
1905 | |
1906 // Before looking at dst.length, make sure dst is also an objArray. | |
304 | 1907 __ movptr(rsi_dst_klass, dst_klass_addr); |
0 | 1908 __ cmpl(dst_klass_lh_addr, objArray_lh); |
1909 __ jccb(Assembler::notEqual, L_failed); | |
1910 | |
1911 // It is safe to examine both src.length and dst.length. | |
304 | 1912 __ movl2ptr(src_pos, SRC_POS); // reload rsi |
0 | 1913 arraycopy_range_checks(src, src_pos, dst, dst_pos, LENGTH, L_failed); |
1914 // (Now src_pos and dst_pos are killed, but not src and dst.) | |
1915 | |
1916 // We'll need this temp (don't forget to pop it after the type check). | |
304 | 1917 __ push(rbx); |
0 | 1918 Register rbx_src_klass = rbx; |
1919 | |
304 | 1920 __ mov(rbx_src_klass, rcx_src_klass); // spill away from rcx |
1921 __ movptr(rsi_dst_klass, dst_klass_addr); | |
0 | 1922 Address super_check_offset_addr(rsi_dst_klass, sco_offset); |
1923 Label L_fail_array_check; | |
1924 generate_type_check(rbx_src_klass, | |
1925 super_check_offset_addr, dst_klass_addr, | |
1926 rdi_temp, NULL, &L_fail_array_check); | |
1927 // (On fall-through, we have passed the array type check.) | |
304 | 1928 __ pop(rbx); |
0 | 1929 __ jmp(L_plain_copy); |
1930 | |
1931 __ BIND(L_fail_array_check); | |
1932 // Reshuffle arguments so we can call checkcast_arraycopy: | |
1933 | |
1934 // match initial saves for checkcast_arraycopy | |
304 | 1935 // push(rsi); // already done; see above |
1936 // push(rdi); // already done; see above | |
1937 // push(rbx); // already done; see above | |
0 | 1938 |
1939 // Marshal outgoing arguments now, freeing registers. | |
1940 Address from_arg(rsp, 16+ 4); // from | |
1941 Address to_arg(rsp, 16+ 8); // to | |
1942 Address length_arg(rsp, 16+12); // elements count | |
1943 Address ckoff_arg(rsp, 16+16); // super_check_offset | |
1944 Address ckval_arg(rsp, 16+20); // super_klass | |
1945 | |
1946 Address SRC_POS_arg(rsp, 16+ 8); | |
1947 Address DST_POS_arg(rsp, 16+16); | |
1948 Address LENGTH_arg(rsp, 16+20); | |
1949 // push rbx, changed the incoming offsets (why not just use rbp,??) | |
1950 // assert(SRC_POS_arg.disp() == SRC_POS.disp() + 4, ""); | |
1951 | |
304 | 1952 __ movptr(rbx, Address(rsi_dst_klass, ek_offset)); |
1953 __ movl2ptr(length, LENGTH_arg); // reload elements count | |
1954 __ movl2ptr(src_pos, SRC_POS_arg); // reload src_pos | |
1955 __ movl2ptr(dst_pos, DST_POS_arg); // reload dst_pos | |
0 | 1956 |
304 | 1957 __ movptr(ckval_arg, rbx); // destination element type |
0 | 1958 __ movl(rbx, Address(rbx, sco_offset)); |
1959 __ movl(ckoff_arg, rbx); // corresponding class check offset | |
1960 | |
1961 __ movl(length_arg, length); // outgoing length argument | |
1962 | |
304 | 1963 __ lea(from, Address(src, src_pos, Address::times_ptr, |
0 | 1964 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
304 | 1965 __ movptr(from_arg, from); |
0 | 1966 |
304 | 1967 __ lea(to, Address(dst, dst_pos, Address::times_ptr, |
0 | 1968 arrayOopDesc::base_offset_in_bytes(T_OBJECT))); |
304 | 1969 __ movptr(to_arg, to); |
0 | 1970 __ jump(RuntimeAddress(entry_checkcast_arraycopy)); |
1971 } | |
1972 | |
1973 return start; | |
1974 } | |
1975 | |
1976 void generate_arraycopy_stubs() { | |
1977 address entry; | |
1978 address entry_jbyte_arraycopy; | |
1979 address entry_jshort_arraycopy; | |
1980 address entry_jint_arraycopy; | |
1981 address entry_oop_arraycopy; | |
1982 address entry_jlong_arraycopy; | |
1983 address entry_checkcast_arraycopy; | |
1984 | |
1985 StubRoutines::_arrayof_jbyte_disjoint_arraycopy = | |
1986 generate_disjoint_copy(T_BYTE, true, Address::times_1, &entry, | |
1987 "arrayof_jbyte_disjoint_arraycopy"); | |
1988 StubRoutines::_arrayof_jbyte_arraycopy = | |
1989 generate_conjoint_copy(T_BYTE, true, Address::times_1, entry, | |
1990 NULL, "arrayof_jbyte_arraycopy"); | |
1991 StubRoutines::_jbyte_disjoint_arraycopy = | |
1992 generate_disjoint_copy(T_BYTE, false, Address::times_1, &entry, | |
1993 "jbyte_disjoint_arraycopy"); | |
1994 StubRoutines::_jbyte_arraycopy = | |
1995 generate_conjoint_copy(T_BYTE, false, Address::times_1, entry, | |
1996 &entry_jbyte_arraycopy, "jbyte_arraycopy"); | |
1997 | |
1998 StubRoutines::_arrayof_jshort_disjoint_arraycopy = | |
1999 generate_disjoint_copy(T_SHORT, true, Address::times_2, &entry, | |
2000 "arrayof_jshort_disjoint_arraycopy"); | |
2001 StubRoutines::_arrayof_jshort_arraycopy = | |
2002 generate_conjoint_copy(T_SHORT, true, Address::times_2, entry, | |
2003 NULL, "arrayof_jshort_arraycopy"); | |
2004 StubRoutines::_jshort_disjoint_arraycopy = | |
2005 generate_disjoint_copy(T_SHORT, false, Address::times_2, &entry, | |
2006 "jshort_disjoint_arraycopy"); | |
2007 StubRoutines::_jshort_arraycopy = | |
2008 generate_conjoint_copy(T_SHORT, false, Address::times_2, entry, | |
2009 &entry_jshort_arraycopy, "jshort_arraycopy"); | |
2010 | |
2011 // Next arrays are always aligned on 4 bytes at least. | |
2012 StubRoutines::_jint_disjoint_arraycopy = | |
2013 generate_disjoint_copy(T_INT, true, Address::times_4, &entry, | |
2014 "jint_disjoint_arraycopy"); | |
2015 StubRoutines::_jint_arraycopy = | |
2016 generate_conjoint_copy(T_INT, true, Address::times_4, entry, | |
2017 &entry_jint_arraycopy, "jint_arraycopy"); | |
2018 | |
2019 StubRoutines::_oop_disjoint_arraycopy = | |
304 | 2020 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry, |
0 | 2021 "oop_disjoint_arraycopy"); |
2022 StubRoutines::_oop_arraycopy = | |
304 | 2023 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry, |
0 | 2024 &entry_oop_arraycopy, "oop_arraycopy"); |
2025 | |
2324 | 2026 StubRoutines::_oop_disjoint_arraycopy_uninit = |
2027 generate_disjoint_copy(T_OBJECT, true, Address::times_ptr, &entry, | |
2028 "oop_disjoint_arraycopy_uninit", | |
2029 /*dest_uninitialized*/true); | |
2030 StubRoutines::_oop_arraycopy_uninit = | |
2031 generate_conjoint_copy(T_OBJECT, true, Address::times_ptr, entry, | |
2032 NULL, "oop_arraycopy_uninit", | |
2033 /*dest_uninitialized*/true); | |
2034 | |
0 | 2035 StubRoutines::_jlong_disjoint_arraycopy = |
2036 generate_disjoint_long_copy(&entry, "jlong_disjoint_arraycopy"); | |
2037 StubRoutines::_jlong_arraycopy = | |
2038 generate_conjoint_long_copy(entry, &entry_jlong_arraycopy, | |
2039 "jlong_arraycopy"); | |
2040 | |
1763 | 2041 StubRoutines::_jbyte_fill = generate_fill(T_BYTE, false, "jbyte_fill"); |
2042 StubRoutines::_jshort_fill = generate_fill(T_SHORT, false, "jshort_fill"); | |
2043 StubRoutines::_jint_fill = generate_fill(T_INT, false, "jint_fill"); | |
2044 StubRoutines::_arrayof_jbyte_fill = generate_fill(T_BYTE, true, "arrayof_jbyte_fill"); | |
2045 StubRoutines::_arrayof_jshort_fill = generate_fill(T_SHORT, true, "arrayof_jshort_fill"); | |
2046 StubRoutines::_arrayof_jint_fill = generate_fill(T_INT, true, "arrayof_jint_fill"); | |
2047 | |
2324 | 2048 StubRoutines::_arrayof_jint_disjoint_arraycopy = StubRoutines::_jint_disjoint_arraycopy; |
2049 StubRoutines::_arrayof_oop_disjoint_arraycopy = StubRoutines::_oop_disjoint_arraycopy; | |
2050 StubRoutines::_arrayof_oop_disjoint_arraycopy_uninit = StubRoutines::_oop_disjoint_arraycopy_uninit; | |
2051 StubRoutines::_arrayof_jlong_disjoint_arraycopy = StubRoutines::_jlong_disjoint_arraycopy; | |
0 | 2052 |
2324 | 2053 StubRoutines::_arrayof_jint_arraycopy = StubRoutines::_jint_arraycopy; |
2054 StubRoutines::_arrayof_oop_arraycopy = StubRoutines::_oop_arraycopy; | |
2055 StubRoutines::_arrayof_oop_arraycopy_uninit = StubRoutines::_oop_arraycopy_uninit; | |
2056 StubRoutines::_arrayof_jlong_arraycopy = StubRoutines::_jlong_arraycopy; | |
0 | 2057 |
2058 StubRoutines::_checkcast_arraycopy = | |
2324 | 2059 generate_checkcast_copy("checkcast_arraycopy", &entry_checkcast_arraycopy); |
2060 StubRoutines::_checkcast_arraycopy_uninit = | |
2061 generate_checkcast_copy("checkcast_arraycopy_uninit", NULL, /*dest_uninitialized*/true); | |
0 | 2062 |
2063 StubRoutines::_unsafe_arraycopy = | |
2064 generate_unsafe_copy("unsafe_arraycopy", | |
2065 entry_jbyte_arraycopy, | |
2066 entry_jshort_arraycopy, | |
2067 entry_jint_arraycopy, | |
2068 entry_jlong_arraycopy); | |
2069 | |
2070 StubRoutines::_generic_arraycopy = | |
2071 generate_generic_copy("generic_arraycopy", | |
2072 entry_jbyte_arraycopy, | |
2073 entry_jshort_arraycopy, | |
2074 entry_jint_arraycopy, | |
2075 entry_oop_arraycopy, | |
2076 entry_jlong_arraycopy, | |
2077 entry_checkcast_arraycopy); | |
2078 } | |
2079 | |
1174
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2080 void generate_math_stubs() { |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2081 { |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2082 StubCodeMark mark(this, "StubRoutines", "log"); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2083 StubRoutines::_intrinsic_log = (double (*)(double)) __ pc(); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2084 |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2085 __ fld_d(Address(rsp, 4)); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2086 __ flog(); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2087 __ ret(0); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2088 } |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2089 { |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2090 StubCodeMark mark(this, "StubRoutines", "log10"); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2091 StubRoutines::_intrinsic_log10 = (double (*)(double)) __ pc(); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2092 |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2093 __ fld_d(Address(rsp, 4)); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2094 __ flog10(); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2095 __ ret(0); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2096 } |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2097 { |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2098 StubCodeMark mark(this, "StubRoutines", "sin"); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2099 StubRoutines::_intrinsic_sin = (double (*)(double)) __ pc(); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2100 |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2101 __ fld_d(Address(rsp, 4)); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2102 __ trigfunc('s'); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2103 __ ret(0); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2104 } |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2105 { |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2106 StubCodeMark mark(this, "StubRoutines", "cos"); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2107 StubRoutines::_intrinsic_cos = (double (*)(double)) __ pc(); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2108 |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2109 __ fld_d(Address(rsp, 4)); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2110 __ trigfunc('c'); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2111 __ ret(0); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2112 } |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2113 { |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2114 StubCodeMark mark(this, "StubRoutines", "tan"); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2115 StubRoutines::_intrinsic_tan = (double (*)(double)) __ pc(); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2116 |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2117 __ fld_d(Address(rsp, 4)); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2118 __ trigfunc('t'); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2119 __ ret(0); |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2120 } |
6084
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2121 { |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2122 StubCodeMark mark(this, "StubRoutines", "exp"); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2123 StubRoutines::_intrinsic_exp = (double (*)(double)) __ pc(); |
1174
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2124 |
6084
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2125 __ fld_d(Address(rsp, 4)); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2126 __ exp_with_fallback(0); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2127 __ ret(0); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2128 } |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2129 { |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2130 StubCodeMark mark(this, "StubRoutines", "pow"); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2131 StubRoutines::_intrinsic_pow = (double (*)(double,double)) __ pc(); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2132 |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2133 __ fld_d(Address(rsp, 12)); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2134 __ fld_d(Address(rsp, 4)); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2135 __ pow_with_fallback(0); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2136 __ ret(0); |
6759698e3140
7133857: exp() and pow() should use the x87 ISA on x86
roland
parents:
4771
diff
changeset
|
2137 } |
1174
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2138 } |
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2139 |
6894 | 2140 // AES intrinsic stubs |
2141 enum {AESBlockSize = 16}; | |
2142 | |
2143 address generate_key_shuffle_mask() { | |
2144 __ align(16); | |
2145 StubCodeMark mark(this, "StubRoutines", "key_shuffle_mask"); | |
2146 address start = __ pc(); | |
2147 __ emit_data(0x00010203, relocInfo::none, 0 ); | |
2148 __ emit_data(0x04050607, relocInfo::none, 0 ); | |
2149 __ emit_data(0x08090a0b, relocInfo::none, 0 ); | |
2150 __ emit_data(0x0c0d0e0f, relocInfo::none, 0 ); | |
2151 return start; | |
2152 } | |
2153 | |
2154 // Utility routine for loading a 128-bit key word in little endian format | |
2155 // can optionally specify that the shuffle mask is already in an xmmregister | |
2156 void load_key(XMMRegister xmmdst, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { | |
2157 __ movdqu(xmmdst, Address(key, offset)); | |
2158 if (xmm_shuf_mask != NULL) { | |
2159 __ pshufb(xmmdst, xmm_shuf_mask); | |
2160 } else { | |
2161 __ pshufb(xmmdst, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); | |
2162 } | |
2163 } | |
2164 | |
2165 // aesenc using specified key+offset | |
2166 // can optionally specify that the shuffle mask is already in an xmmregister | |
2167 void aes_enc_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { | |
2168 load_key(xmmtmp, key, offset, xmm_shuf_mask); | |
2169 __ aesenc(xmmdst, xmmtmp); | |
2170 } | |
2171 | |
2172 // aesdec using specified key+offset | |
2173 // can optionally specify that the shuffle mask is already in an xmmregister | |
2174 void aes_dec_key(XMMRegister xmmdst, XMMRegister xmmtmp, Register key, int offset, XMMRegister xmm_shuf_mask=NULL) { | |
2175 load_key(xmmtmp, key, offset, xmm_shuf_mask); | |
2176 __ aesdec(xmmdst, xmmtmp); | |
2177 } | |
2178 | |
2179 | |
2180 // Arguments: | |
2181 // | |
2182 // Inputs: | |
2183 // c_rarg0 - source byte array address | |
2184 // c_rarg1 - destination byte array address | |
2185 // c_rarg2 - K (key) in little endian int array | |
2186 // | |
2187 address generate_aescrypt_encryptBlock() { | |
2188 assert(UseAES && (UseAVX > 0), "need AES instructions and misaligned SSE support"); | |
2189 __ align(CodeEntryAlignment); | |
2190 StubCodeMark mark(this, "StubRoutines", "aescrypt_encryptBlock"); | |
2191 Label L_doLast; | |
2192 address start = __ pc(); | |
2193 | |
2194 const Register from = rsi; // source array address | |
2195 const Register to = rdx; // destination array address | |
2196 const Register key = rcx; // key array address | |
2197 const Register keylen = rax; | |
2198 const Address from_param(rbp, 8+0); | |
2199 const Address to_param (rbp, 8+4); | |
2200 const Address key_param (rbp, 8+8); | |
2201 | |
2202 const XMMRegister xmm_result = xmm0; | |
2203 const XMMRegister xmm_temp = xmm1; | |
2204 const XMMRegister xmm_key_shuf_mask = xmm2; | |
2205 | |
2206 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
2207 __ push(rsi); | |
2208 __ movptr(from , from_param); | |
2209 __ movptr(to , to_param); | |
2210 __ movptr(key , key_param); | |
2211 | |
2212 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); | |
2213 // keylen = # of 32-bit words, convert to 128-bit words | |
2214 __ shrl(keylen, 2); | |
2215 __ subl(keylen, 11); // every key has at least 11 128-bit words, some have more | |
2216 | |
2217 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); | |
2218 __ movdqu(xmm_result, Address(from, 0)); // get 16 bytes of input | |
2219 | |
2220 // For encryption, the java expanded key ordering is just what we need | |
2221 | |
2222 load_key(xmm_temp, key, 0x00, xmm_key_shuf_mask); | |
2223 __ pxor(xmm_result, xmm_temp); | |
2224 for (int offset = 0x10; offset <= 0x90; offset += 0x10) { | |
2225 aes_enc_key(xmm_result, xmm_temp, key, offset, xmm_key_shuf_mask); | |
2226 } | |
2227 load_key (xmm_temp, key, 0xa0, xmm_key_shuf_mask); | |
2228 __ cmpl(keylen, 0); | |
2229 __ jcc(Assembler::equal, L_doLast); | |
2230 __ aesenc(xmm_result, xmm_temp); // only in 192 and 256 bit keys | |
2231 aes_enc_key(xmm_result, xmm_temp, key, 0xb0, xmm_key_shuf_mask); | |
2232 load_key(xmm_temp, key, 0xc0, xmm_key_shuf_mask); | |
2233 __ subl(keylen, 2); | |
2234 __ jcc(Assembler::equal, L_doLast); | |
2235 __ aesenc(xmm_result, xmm_temp); // only in 256 bit keys | |
2236 aes_enc_key(xmm_result, xmm_temp, key, 0xd0, xmm_key_shuf_mask); | |
2237 load_key(xmm_temp, key, 0xe0, xmm_key_shuf_mask); | |
2238 | |
2239 __ BIND(L_doLast); | |
2240 __ aesenclast(xmm_result, xmm_temp); | |
2241 __ movdqu(Address(to, 0), xmm_result); // store the result | |
2242 __ xorptr(rax, rax); // return 0 | |
2243 __ pop(rsi); | |
2244 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
2245 __ ret(0); | |
2246 | |
2247 return start; | |
2248 } | |
2249 | |
2250 | |
2251 // Arguments: | |
2252 // | |
2253 // Inputs: | |
2254 // c_rarg0 - source byte array address | |
2255 // c_rarg1 - destination byte array address | |
2256 // c_rarg2 - K (key) in little endian int array | |
2257 // | |
2258 address generate_aescrypt_decryptBlock() { | |
2259 assert(UseAES && (UseAVX > 0), "need AES instructions and misaligned SSE support"); | |
2260 __ align(CodeEntryAlignment); | |
2261 StubCodeMark mark(this, "StubRoutines", "aescrypt_decryptBlock"); | |
2262 Label L_doLast; | |
2263 address start = __ pc(); | |
2264 | |
2265 const Register from = rsi; // source array address | |
2266 const Register to = rdx; // destination array address | |
2267 const Register key = rcx; // key array address | |
2268 const Register keylen = rax; | |
2269 const Address from_param(rbp, 8+0); | |
2270 const Address to_param (rbp, 8+4); | |
2271 const Address key_param (rbp, 8+8); | |
2272 | |
2273 const XMMRegister xmm_result = xmm0; | |
2274 const XMMRegister xmm_temp = xmm1; | |
2275 const XMMRegister xmm_key_shuf_mask = xmm2; | |
2276 | |
2277 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
2278 __ push(rsi); | |
2279 __ movptr(from , from_param); | |
2280 __ movptr(to , to_param); | |
2281 __ movptr(key , key_param); | |
2282 | |
2283 __ movl(keylen, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); | |
2284 // keylen = # of 32-bit words, convert to 128-bit words | |
2285 __ shrl(keylen, 2); | |
2286 __ subl(keylen, 11); // every key has at least 11 128-bit words, some have more | |
2287 | |
2288 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); | |
2289 __ movdqu(xmm_result, Address(from, 0)); | |
2290 | |
2291 // for decryption java expanded key ordering is rotated one position from what we want | |
2292 // so we start from 0x10 here and hit 0x00 last | |
2293 // we don't know if the key is aligned, hence not using load-execute form | |
2294 load_key(xmm_temp, key, 0x10, xmm_key_shuf_mask); | |
2295 __ pxor (xmm_result, xmm_temp); | |
2296 for (int offset = 0x20; offset <= 0xa0; offset += 0x10) { | |
2297 aes_dec_key(xmm_result, xmm_temp, key, offset, xmm_key_shuf_mask); | |
2298 } | |
2299 __ cmpl(keylen, 0); | |
2300 __ jcc(Assembler::equal, L_doLast); | |
2301 // only in 192 and 256 bit keys | |
2302 aes_dec_key(xmm_result, xmm_temp, key, 0xb0, xmm_key_shuf_mask); | |
2303 aes_dec_key(xmm_result, xmm_temp, key, 0xc0, xmm_key_shuf_mask); | |
2304 __ subl(keylen, 2); | |
2305 __ jcc(Assembler::equal, L_doLast); | |
2306 // only in 256 bit keys | |
2307 aes_dec_key(xmm_result, xmm_temp, key, 0xd0, xmm_key_shuf_mask); | |
2308 aes_dec_key(xmm_result, xmm_temp, key, 0xe0, xmm_key_shuf_mask); | |
2309 | |
2310 __ BIND(L_doLast); | |
2311 // for decryption the aesdeclast operation is always on key+0x00 | |
2312 load_key(xmm_temp, key, 0x00, xmm_key_shuf_mask); | |
2313 __ aesdeclast(xmm_result, xmm_temp); | |
2314 | |
2315 __ movdqu(Address(to, 0), xmm_result); // store the result | |
2316 | |
2317 __ xorptr(rax, rax); // return 0 | |
2318 __ pop(rsi); | |
2319 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
2320 __ ret(0); | |
2321 | |
2322 return start; | |
2323 } | |
2324 | |
2325 void handleSOERegisters(bool saving) { | |
2326 const int saveFrameSizeInBytes = 4 * wordSize; | |
2327 const Address saved_rbx (rbp, -3 * wordSize); | |
2328 const Address saved_rsi (rbp, -2 * wordSize); | |
2329 const Address saved_rdi (rbp, -1 * wordSize); | |
2330 | |
2331 if (saving) { | |
2332 __ subptr(rsp, saveFrameSizeInBytes); | |
2333 __ movptr(saved_rsi, rsi); | |
2334 __ movptr(saved_rdi, rdi); | |
2335 __ movptr(saved_rbx, rbx); | |
2336 } else { | |
2337 // restoring | |
2338 __ movptr(rsi, saved_rsi); | |
2339 __ movptr(rdi, saved_rdi); | |
2340 __ movptr(rbx, saved_rbx); | |
2341 } | |
2342 } | |
2343 | |
2344 // Arguments: | |
2345 // | |
2346 // Inputs: | |
2347 // c_rarg0 - source byte array address | |
2348 // c_rarg1 - destination byte array address | |
2349 // c_rarg2 - K (key) in little endian int array | |
2350 // c_rarg3 - r vector byte array address | |
2351 // c_rarg4 - input length | |
2352 // | |
2353 address generate_cipherBlockChaining_encryptAESCrypt() { | |
2354 assert(UseAES && (UseAVX > 0), "need AES instructions and misaligned SSE support"); | |
2355 __ align(CodeEntryAlignment); | |
2356 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_encryptAESCrypt"); | |
2357 address start = __ pc(); | |
2358 | |
2359 Label L_exit, L_key_192_256, L_key_256, L_loopTop_128, L_loopTop_192, L_loopTop_256; | |
2360 const Register from = rsi; // source array address | |
2361 const Register to = rdx; // destination array address | |
2362 const Register key = rcx; // key array address | |
2363 const Register rvec = rdi; // r byte array initialized from initvector array address | |
2364 // and left with the results of the last encryption block | |
2365 const Register len_reg = rbx; // src len (must be multiple of blocksize 16) | |
2366 const Register pos = rax; | |
2367 | |
2368 // xmm register assignments for the loops below | |
2369 const XMMRegister xmm_result = xmm0; | |
2370 const XMMRegister xmm_temp = xmm1; | |
2371 // first 6 keys preloaded into xmm2-xmm7 | |
2372 const int XMM_REG_NUM_KEY_FIRST = 2; | |
2373 const int XMM_REG_NUM_KEY_LAST = 7; | |
2374 const XMMRegister xmm_key0 = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); | |
2375 | |
2376 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
2377 handleSOERegisters(true /*saving*/); | |
2378 | |
2379 // load registers from incoming parameters | |
2380 const Address from_param(rbp, 8+0); | |
2381 const Address to_param (rbp, 8+4); | |
2382 const Address key_param (rbp, 8+8); | |
2383 const Address rvec_param (rbp, 8+12); | |
2384 const Address len_param (rbp, 8+16); | |
2385 __ movptr(from , from_param); | |
2386 __ movptr(to , to_param); | |
2387 __ movptr(key , key_param); | |
2388 __ movptr(rvec , rvec_param); | |
2389 __ movptr(len_reg , len_param); | |
2390 | |
2391 const XMMRegister xmm_key_shuf_mask = xmm_temp; // used temporarily to swap key bytes up front | |
2392 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); | |
2393 // load up xmm regs 2 thru 7 with keys 0-5 | |
2394 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x00; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { | |
2395 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); | |
2396 offset += 0x10; | |
2397 } | |
2398 | |
2399 __ movdqu(xmm_result, Address(rvec, 0x00)); // initialize xmm_result with r vec | |
2400 | |
2401 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) | |
2402 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); | |
2403 __ cmpl(rax, 44); | |
2404 __ jcc(Assembler::notEqual, L_key_192_256); | |
2405 | |
2406 // 128 bit code follows here | |
2407 __ movptr(pos, 0); | |
2408 __ align(OptoLoopAlignment); | |
2409 __ BIND(L_loopTop_128); | |
2410 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input | |
2411 __ pxor (xmm_result, xmm_temp); // xor with the current r vector | |
2412 | |
2413 __ pxor (xmm_result, xmm_key0); // do the aes rounds | |
2414 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { | |
2415 __ aesenc(xmm_result, as_XMMRegister(rnum)); | |
2416 } | |
2417 for (int key_offset = 0x60; key_offset <= 0x90; key_offset += 0x10) { | |
2418 aes_enc_key(xmm_result, xmm_temp, key, key_offset); | |
2419 } | |
2420 load_key(xmm_temp, key, 0xa0); | |
2421 __ aesenclast(xmm_result, xmm_temp); | |
2422 | |
2423 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output | |
2424 // no need to store r to memory until we exit | |
2425 __ addptr(pos, AESBlockSize); | |
2426 __ subptr(len_reg, AESBlockSize); | |
2427 __ jcc(Assembler::notEqual, L_loopTop_128); | |
2428 | |
2429 __ BIND(L_exit); | |
2430 __ movdqu(Address(rvec, 0), xmm_result); // final value of r stored in rvec of CipherBlockChaining object | |
2431 | |
2432 handleSOERegisters(false /*restoring*/); | |
2433 __ movl(rax, 0); // return 0 (why?) | |
2434 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
2435 __ ret(0); | |
2436 | |
2437 __ BIND(L_key_192_256); | |
2438 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) | |
2439 __ cmpl(rax, 52); | |
2440 __ jcc(Assembler::notEqual, L_key_256); | |
2441 | |
2442 // 192-bit code follows here (could be changed to use more xmm registers) | |
2443 __ movptr(pos, 0); | |
2444 __ align(OptoLoopAlignment); | |
2445 __ BIND(L_loopTop_192); | |
2446 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input | |
2447 __ pxor (xmm_result, xmm_temp); // xor with the current r vector | |
2448 | |
2449 __ pxor (xmm_result, xmm_key0); // do the aes rounds | |
2450 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { | |
2451 __ aesenc(xmm_result, as_XMMRegister(rnum)); | |
2452 } | |
2453 for (int key_offset = 0x60; key_offset <= 0xb0; key_offset += 0x10) { | |
2454 aes_enc_key(xmm_result, xmm_temp, key, key_offset); | |
2455 } | |
2456 load_key(xmm_temp, key, 0xc0); | |
2457 __ aesenclast(xmm_result, xmm_temp); | |
2458 | |
2459 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output | |
2460 // no need to store r to memory until we exit | |
2461 __ addptr(pos, AESBlockSize); | |
2462 __ subptr(len_reg, AESBlockSize); | |
2463 __ jcc(Assembler::notEqual, L_loopTop_192); | |
2464 __ jmp(L_exit); | |
2465 | |
2466 __ BIND(L_key_256); | |
2467 // 256-bit code follows here (could be changed to use more xmm registers) | |
2468 __ movptr(pos, 0); | |
2469 __ align(OptoLoopAlignment); | |
2470 __ BIND(L_loopTop_256); | |
2471 __ movdqu(xmm_temp, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of input | |
2472 __ pxor (xmm_result, xmm_temp); // xor with the current r vector | |
2473 | |
2474 __ pxor (xmm_result, xmm_key0); // do the aes rounds | |
2475 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { | |
2476 __ aesenc(xmm_result, as_XMMRegister(rnum)); | |
2477 } | |
2478 for (int key_offset = 0x60; key_offset <= 0xd0; key_offset += 0x10) { | |
2479 aes_enc_key(xmm_result, xmm_temp, key, key_offset); | |
2480 } | |
2481 load_key(xmm_temp, key, 0xe0); | |
2482 __ aesenclast(xmm_result, xmm_temp); | |
2483 | |
2484 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output | |
2485 // no need to store r to memory until we exit | |
2486 __ addptr(pos, AESBlockSize); | |
2487 __ subptr(len_reg, AESBlockSize); | |
2488 __ jcc(Assembler::notEqual, L_loopTop_256); | |
2489 __ jmp(L_exit); | |
2490 | |
2491 return start; | |
2492 } | |
2493 | |
2494 | |
2495 // CBC AES Decryption. | |
2496 // In 32-bit stub, because of lack of registers we do not try to parallelize 4 blocks at a time. | |
2497 // | |
2498 // Arguments: | |
2499 // | |
2500 // Inputs: | |
2501 // c_rarg0 - source byte array address | |
2502 // c_rarg1 - destination byte array address | |
2503 // c_rarg2 - K (key) in little endian int array | |
2504 // c_rarg3 - r vector byte array address | |
2505 // c_rarg4 - input length | |
2506 // | |
2507 | |
2508 address generate_cipherBlockChaining_decryptAESCrypt() { | |
2509 assert(UseAES && (UseAVX > 0), "need AES instructions and misaligned SSE support"); | |
2510 __ align(CodeEntryAlignment); | |
2511 StubCodeMark mark(this, "StubRoutines", "cipherBlockChaining_decryptAESCrypt"); | |
2512 address start = __ pc(); | |
2513 | |
2514 Label L_exit, L_key_192_256, L_key_256; | |
2515 Label L_singleBlock_loopTop_128; | |
2516 Label L_singleBlock_loopTop_192, L_singleBlock_loopTop_256; | |
2517 const Register from = rsi; // source array address | |
2518 const Register to = rdx; // destination array address | |
2519 const Register key = rcx; // key array address | |
2520 const Register rvec = rdi; // r byte array initialized from initvector array address | |
2521 // and left with the results of the last encryption block | |
2522 const Register len_reg = rbx; // src len (must be multiple of blocksize 16) | |
2523 const Register pos = rax; | |
2524 | |
2525 // xmm register assignments for the loops below | |
2526 const XMMRegister xmm_result = xmm0; | |
2527 const XMMRegister xmm_temp = xmm1; | |
2528 // first 6 keys preloaded into xmm2-xmm7 | |
2529 const int XMM_REG_NUM_KEY_FIRST = 2; | |
2530 const int XMM_REG_NUM_KEY_LAST = 7; | |
2531 const int FIRST_NON_REG_KEY_offset = 0x70; | |
2532 const XMMRegister xmm_key_first = as_XMMRegister(XMM_REG_NUM_KEY_FIRST); | |
2533 | |
2534 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
2535 handleSOERegisters(true /*saving*/); | |
2536 | |
2537 // load registers from incoming parameters | |
2538 const Address from_param(rbp, 8+0); | |
2539 const Address to_param (rbp, 8+4); | |
2540 const Address key_param (rbp, 8+8); | |
2541 const Address rvec_param (rbp, 8+12); | |
2542 const Address len_param (rbp, 8+16); | |
2543 __ movptr(from , from_param); | |
2544 __ movptr(to , to_param); | |
2545 __ movptr(key , key_param); | |
2546 __ movptr(rvec , rvec_param); | |
2547 __ movptr(len_reg , len_param); | |
2548 | |
2549 // the java expanded key ordering is rotated one position from what we want | |
2550 // so we start from 0x10 here and hit 0x00 last | |
2551 const XMMRegister xmm_key_shuf_mask = xmm1; // used temporarily to swap key bytes up front | |
2552 __ movdqu(xmm_key_shuf_mask, ExternalAddress(StubRoutines::x86::key_shuffle_mask_addr())); | |
2553 // load up xmm regs 2 thru 6 with first 5 keys | |
2554 for (int rnum = XMM_REG_NUM_KEY_FIRST, offset = 0x10; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { | |
2555 load_key(as_XMMRegister(rnum), key, offset, xmm_key_shuf_mask); | |
2556 offset += 0x10; | |
2557 } | |
2558 | |
2559 // inside here, use the rvec register to point to previous block cipher | |
2560 // with which we xor at the end of each newly decrypted block | |
2561 const Register prev_block_cipher_ptr = rvec; | |
2562 | |
2563 // now split to different paths depending on the keylen (len in ints of AESCrypt.KLE array (52=192, or 60=256)) | |
2564 __ movl(rax, Address(key, arrayOopDesc::length_offset_in_bytes() - arrayOopDesc::base_offset_in_bytes(T_INT))); | |
2565 __ cmpl(rax, 44); | |
2566 __ jcc(Assembler::notEqual, L_key_192_256); | |
2567 | |
2568 | |
2569 // 128-bit code follows here, parallelized | |
2570 __ movptr(pos, 0); | |
2571 __ align(OptoLoopAlignment); | |
2572 __ BIND(L_singleBlock_loopTop_128); | |
2573 __ cmpptr(len_reg, 0); // any blocks left?? | |
2574 __ jcc(Assembler::equal, L_exit); | |
2575 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input | |
2576 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds | |
2577 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { | |
2578 __ aesdec(xmm_result, as_XMMRegister(rnum)); | |
2579 } | |
2580 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xa0; key_offset += 0x10) { // 128-bit runs up to key offset a0 | |
2581 aes_dec_key(xmm_result, xmm_temp, key, key_offset); | |
2582 } | |
2583 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0 | |
2584 __ aesdeclast(xmm_result, xmm_temp); | |
2585 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); | |
2586 __ pxor (xmm_result, xmm_temp); // xor with the current r vector | |
2587 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output | |
2588 // no need to store r to memory until we exit | |
2589 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr | |
2590 __ addptr(pos, AESBlockSize); | |
2591 __ subptr(len_reg, AESBlockSize); | |
2592 __ jmp(L_singleBlock_loopTop_128); | |
2593 | |
2594 | |
2595 __ BIND(L_exit); | |
2596 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); | |
2597 __ movptr(rvec , rvec_param); // restore this since used in loop | |
2598 __ movdqu(Address(rvec, 0), xmm_temp); // final value of r stored in rvec of CipherBlockChaining object | |
2599 handleSOERegisters(false /*restoring*/); | |
2600 __ movl(rax, 0); // return 0 (why?) | |
2601 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
2602 __ ret(0); | |
2603 | |
2604 | |
2605 __ BIND(L_key_192_256); | |
2606 // here rax = len in ints of AESCrypt.KLE array (52=192, or 60=256) | |
2607 __ cmpl(rax, 52); | |
2608 __ jcc(Assembler::notEqual, L_key_256); | |
2609 | |
2610 // 192-bit code follows here (could be optimized to use parallelism) | |
2611 __ movptr(pos, 0); | |
2612 __ align(OptoLoopAlignment); | |
2613 __ BIND(L_singleBlock_loopTop_192); | |
2614 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input | |
2615 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds | |
2616 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { | |
2617 __ aesdec(xmm_result, as_XMMRegister(rnum)); | |
2618 } | |
2619 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xc0; key_offset += 0x10) { // 192-bit runs up to key offset c0 | |
2620 aes_dec_key(xmm_result, xmm_temp, key, key_offset); | |
2621 } | |
2622 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0 | |
2623 __ aesdeclast(xmm_result, xmm_temp); | |
2624 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); | |
2625 __ pxor (xmm_result, xmm_temp); // xor with the current r vector | |
2626 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output | |
2627 // no need to store r to memory until we exit | |
2628 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr | |
2629 __ addptr(pos, AESBlockSize); | |
2630 __ subptr(len_reg, AESBlockSize); | |
2631 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_192); | |
2632 __ jmp(L_exit); | |
2633 | |
2634 __ BIND(L_key_256); | |
2635 // 256-bit code follows here (could be optimized to use parallelism) | |
2636 __ movptr(pos, 0); | |
2637 __ align(OptoLoopAlignment); | |
2638 __ BIND(L_singleBlock_loopTop_256); | |
2639 __ movdqu(xmm_result, Address(from, pos, Address::times_1, 0)); // get next 16 bytes of cipher input | |
2640 __ pxor (xmm_result, xmm_key_first); // do the aes dec rounds | |
2641 for (int rnum = XMM_REG_NUM_KEY_FIRST + 1; rnum <= XMM_REG_NUM_KEY_LAST; rnum++) { | |
2642 __ aesdec(xmm_result, as_XMMRegister(rnum)); | |
2643 } | |
2644 for (int key_offset = FIRST_NON_REG_KEY_offset; key_offset <= 0xe0; key_offset += 0x10) { // 256-bit runs up to key offset e0 | |
2645 aes_dec_key(xmm_result, xmm_temp, key, key_offset); | |
2646 } | |
2647 load_key(xmm_temp, key, 0x00); // final key is stored in java expanded array at offset 0 | |
2648 __ aesdeclast(xmm_result, xmm_temp); | |
2649 __ movdqu(xmm_temp, Address(prev_block_cipher_ptr, 0x00)); | |
2650 __ pxor (xmm_result, xmm_temp); // xor with the current r vector | |
2651 __ movdqu(Address(to, pos, Address::times_1, 0), xmm_result); // store into the next 16 bytes of output | |
2652 // no need to store r to memory until we exit | |
2653 __ lea(prev_block_cipher_ptr, Address(from, pos, Address::times_1, 0)); // set up new ptr | |
2654 __ addptr(pos, AESBlockSize); | |
2655 __ subptr(len_reg, AESBlockSize); | |
2656 __ jcc(Assembler::notEqual,L_singleBlock_loopTop_256); | |
2657 __ jmp(L_exit); | |
2658 | |
2659 return start; | |
2660 } | |
2661 | |
2662 | |
0 | 2663 public: |
2664 // Information about frame layout at time of blocking runtime call. | |
2665 // Note that we only have to preserve callee-saved registers since | |
2666 // the compilers are responsible for supplying a continuation point | |
2667 // if they expect all registers to be preserved. | |
2668 enum layout { | |
2669 thread_off, // last_java_sp | |
3451
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2670 arg1_off, |
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2671 arg2_off, |
0 | 2672 rbp_off, // callee saved register |
2673 ret_pc, | |
2674 framesize | |
2675 }; | |
2676 | |
2677 private: | |
2678 | |
2679 #undef __ | |
2680 #define __ masm-> | |
2681 | |
2682 //------------------------------------------------------------------------------------------------------------------------ | |
2683 // Continuation point for throwing of implicit exceptions that are not handled in | |
2684 // the current activation. Fabricates an exception oop and initiates normal | |
2685 // exception dispatching in this frame. | |
2686 // | |
2687 // Previously the compiler (c2) allowed for callee save registers on Java calls. | |
2688 // This is no longer true after adapter frames were removed but could possibly | |
2689 // be brought back in the future if the interpreter code was reworked and it | |
2690 // was deemed worthwhile. The comment below was left to describe what must | |
2691 // happen here if callee saves were resurrected. As it stands now this stub | |
2692 // could actually be a vanilla BufferBlob and have now oopMap at all. | |
2693 // Since it doesn't make much difference we've chosen to leave it the | |
2694 // way it was in the callee save days and keep the comment. | |
2695 | |
2696 // If we need to preserve callee-saved values we need a callee-saved oop map and | |
2697 // therefore have to make these stubs into RuntimeStubs rather than BufferBlobs. | |
2698 // If the compiler needs all registers to be preserved between the fault | |
2699 // point and the exception handler then it must assume responsibility for that in | |
2700 // AbstractCompiler::continuation_for_implicit_null_exception or | |
2701 // continuation_for_implicit_division_by_zero_exception. All other implicit | |
2702 // exceptions (e.g., NullPointerException or AbstractMethodError on entry) are | |
2703 // either at call sites or otherwise assume that stack unwinding will be initiated, | |
2704 // so caller saved registers were assumed volatile in the compiler. | |
2705 address generate_throw_exception(const char* name, address runtime_entry, | |
3937 | 2706 Register arg1 = noreg, Register arg2 = noreg) { |
0 | 2707 |
2708 int insts_size = 256; | |
2709 int locs_size = 32; | |
2710 | |
2711 CodeBuffer code(name, insts_size, locs_size); | |
2712 OopMapSet* oop_maps = new OopMapSet(); | |
2713 MacroAssembler* masm = new MacroAssembler(&code); | |
2714 | |
2715 address start = __ pc(); | |
2716 | |
2717 // This is an inlined and slightly modified version of call_VM | |
2718 // which has the ability to fetch the return PC out of | |
2719 // thread-local storage and also sets up last_Java_sp slightly | |
2720 // differently than the real call_VM | |
2721 Register java_thread = rbx; | |
2722 __ get_thread(java_thread); | |
2723 | |
2724 __ enter(); // required for proper stackwalking of RuntimeStub frame | |
2725 | |
2726 // pc and rbp, already pushed | |
304 | 2727 __ subptr(rsp, (framesize-2) * wordSize); // prolog |
0 | 2728 |
2729 // Frame is now completed as far as size and linkage. | |
2730 | |
2731 int frame_complete = __ pc() - start; | |
2732 | |
2733 // push java thread (becomes first argument of C function) | |
304 | 2734 __ movptr(Address(rsp, thread_off * wordSize), java_thread); |
3451
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2735 if (arg1 != noreg) { |
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2736 __ movptr(Address(rsp, arg1_off * wordSize), arg1); |
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2737 } |
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2738 if (arg2 != noreg) { |
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2739 assert(arg1 != noreg, "missing reg arg"); |
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2740 __ movptr(Address(rsp, arg2_off * wordSize), arg2); |
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2741 } |
0 | 2742 |
2743 // Set up last_Java_sp and last_Java_fp | |
2744 __ set_last_Java_frame(java_thread, rsp, rbp, NULL); | |
2745 | |
2746 // Call runtime | |
2747 BLOCK_COMMENT("call runtime_entry"); | |
2748 __ call(RuntimeAddress(runtime_entry)); | |
2749 // Generate oop map | |
2750 OopMap* map = new OopMap(framesize, 0); | |
2751 oop_maps->add_gc_map(__ pc() - start, map); | |
2752 | |
2753 // restore the thread (cannot use the pushed argument since arguments | |
2754 // may be overwritten by C code generated by an optimizing compiler); | |
2755 // however can use the register value directly if it is callee saved. | |
2756 __ get_thread(java_thread); | |
2757 | |
2758 __ reset_last_Java_frame(java_thread, true, false); | |
2759 | |
2760 __ leave(); // required for proper stackwalking of RuntimeStub frame | |
2761 | |
2762 // check for pending exceptions | |
2763 #ifdef ASSERT | |
2764 Label L; | |
304 | 2765 __ cmpptr(Address(java_thread, Thread::pending_exception_offset()), (int32_t)NULL_WORD); |
0 | 2766 __ jcc(Assembler::notEqual, L); |
2767 __ should_not_reach_here(); | |
2768 __ bind(L); | |
2769 #endif /* ASSERT */ | |
2770 __ jump(RuntimeAddress(StubRoutines::forward_exception_entry())); | |
2771 | |
2772 | |
2773 RuntimeStub* stub = RuntimeStub::new_runtime_stub(name, &code, frame_complete, framesize, oop_maps, false); | |
2774 return stub->entry_point(); | |
2775 } | |
2776 | |
2777 | |
2778 void create_control_words() { | |
2779 // Round to nearest, 53-bit mode, exceptions masked | |
2780 StubRoutines::_fpu_cntrl_wrd_std = 0x027F; | |
2781 // Round to zero, 53-bit mode, exception mased | |
2782 StubRoutines::_fpu_cntrl_wrd_trunc = 0x0D7F; | |
2783 // Round to nearest, 24-bit mode, exceptions masked | |
2784 StubRoutines::_fpu_cntrl_wrd_24 = 0x007F; | |
2785 // Round to nearest, 64-bit mode, exceptions masked | |
2786 StubRoutines::_fpu_cntrl_wrd_64 = 0x037F; | |
2787 // Round to nearest, 64-bit mode, exceptions masked | |
2788 StubRoutines::_mxcsr_std = 0x1F80; | |
2789 // Note: the following two constants are 80-bit values | |
2790 // layout is critical for correct loading by FPU. | |
2791 // Bias for strict fp multiply/divide | |
2792 StubRoutines::_fpu_subnormal_bias1[0]= 0x00000000; // 2^(-15360) == 0x03ff 8000 0000 0000 0000 | |
2793 StubRoutines::_fpu_subnormal_bias1[1]= 0x80000000; | |
2794 StubRoutines::_fpu_subnormal_bias1[2]= 0x03ff; | |
2795 // Un-Bias for strict fp multiply/divide | |
2796 StubRoutines::_fpu_subnormal_bias2[0]= 0x00000000; // 2^(+15360) == 0x7bff 8000 0000 0000 0000 | |
2797 StubRoutines::_fpu_subnormal_bias2[1]= 0x80000000; | |
2798 StubRoutines::_fpu_subnormal_bias2[2]= 0x7bff; | |
2799 } | |
2800 | |
2801 //--------------------------------------------------------------------------- | |
2802 // Initialization | |
2803 | |
2804 void generate_initial() { | |
2805 // Generates all stubs and initializes the entry points | |
2806 | |
2807 //------------------------------------------------------------------------------------------------------------------------ | |
2808 // entry points that exist in all platforms | |
2809 // Note: This is code that could be shared among different platforms - however the benefit seems to be smaller than | |
2810 // the disadvantage of having a much more complicated generator structure. See also comment in stubRoutines.hpp. | |
2811 StubRoutines::_forward_exception_entry = generate_forward_exception(); | |
2812 | |
2813 StubRoutines::_call_stub_entry = | |
2814 generate_call_stub(StubRoutines::_call_stub_return_address); | |
2815 // is referenced by megamorphic call | |
2816 StubRoutines::_catch_exception_entry = generate_catch_exception(); | |
2817 | |
2818 // These are currently used by Solaris/Intel | |
2819 StubRoutines::_atomic_xchg_entry = generate_atomic_xchg(); | |
2820 | |
2821 StubRoutines::_handler_for_unsafe_access_entry = | |
2822 generate_handler_for_unsafe_access(); | |
2823 | |
2824 // platform dependent | |
2825 create_control_words(); | |
2826 | |
304 | 2827 StubRoutines::x86::_verify_mxcsr_entry = generate_verify_mxcsr(); |
2828 StubRoutines::x86::_verify_fpu_cntrl_wrd_entry = generate_verify_fpu_cntrl_wrd(); | |
0 | 2829 StubRoutines::_d2i_wrapper = generate_d2i_wrapper(T_INT, |
2830 CAST_FROM_FN_PTR(address, SharedRuntime::d2i)); | |
2831 StubRoutines::_d2l_wrapper = generate_d2i_wrapper(T_LONG, | |
2832 CAST_FROM_FN_PTR(address, SharedRuntime::d2l)); | |
3451
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2833 |
38fa55e5e792
7055355: JSR 292: crash while throwing WrongMethodTypeException
never
parents:
2324
diff
changeset
|
2834 // Build this early so it's available for the interpreter |
4743 | 2835 StubRoutines::_throw_StackOverflowError_entry = generate_throw_exception("StackOverflowError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_StackOverflowError)); |
0 | 2836 } |
2837 | |
2838 | |
2839 void generate_all() { | |
2840 // Generates all stubs and initializes the entry points | |
2841 | |
2842 // These entry points require SharedInfo::stack0 to be set up in non-core builds | |
2843 // and need to be relocatable, so they each fabricate a RuntimeStub internally. | |
3937 | 2844 StubRoutines::_throw_AbstractMethodError_entry = generate_throw_exception("AbstractMethodError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_AbstractMethodError)); |
2845 StubRoutines::_throw_IncompatibleClassChangeError_entry= generate_throw_exception("IncompatibleClassChangeError throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_IncompatibleClassChangeError)); | |
2846 StubRoutines::_throw_NullPointerException_at_call_entry= generate_throw_exception("NullPointerException at call throw_exception", CAST_FROM_FN_PTR(address, SharedRuntime::throw_NullPointerException_at_call)); | |
0 | 2847 |
2848 //------------------------------------------------------------------------------------------------------------------------ | |
2849 // entry points that are platform specific | |
2850 | |
2851 // support for verify_oop (must happen after universe_init) | |
2852 StubRoutines::_verify_oop_subroutine_entry = generate_verify_oop(); | |
2853 | |
2854 // arraycopy stubs used by compilers | |
2855 generate_arraycopy_stubs(); | |
710 | 2856 |
1174
ddb7834449d0
6849984: Value methods for platform dependent math functions constant fold incorrectly
never
parents:
845
diff
changeset
|
2857 generate_math_stubs(); |
6894 | 2858 |
2859 // don't bother generating these AES intrinsic stubs unless global flag is set | |
2860 if (UseAESIntrinsics) { | |
2861 StubRoutines::x86::_key_shuffle_mask_addr = generate_key_shuffle_mask(); // might be needed by the others | |
2862 | |
2863 StubRoutines::_aescrypt_encryptBlock = generate_aescrypt_encryptBlock(); | |
2864 StubRoutines::_aescrypt_decryptBlock = generate_aescrypt_decryptBlock(); | |
2865 StubRoutines::_cipherBlockChaining_encryptAESCrypt = generate_cipherBlockChaining_encryptAESCrypt(); | |
2866 StubRoutines::_cipherBlockChaining_decryptAESCrypt = generate_cipherBlockChaining_decryptAESCrypt(); | |
2867 } | |
0 | 2868 } |
2869 | |
2870 | |
2871 public: | |
2872 StubGenerator(CodeBuffer* code, bool all) : StubCodeGenerator(code) { | |
2873 if (all) { | |
2874 generate_all(); | |
2875 } else { | |
2876 generate_initial(); | |
2877 } | |
2878 } | |
2879 }; // end class declaration | |
2880 | |
2881 | |
2882 void StubGenerator_generate(CodeBuffer* code, bool all) { | |
2883 StubGenerator g(code, all); | |
2884 } |