|
1 /* |
|
2 * Copyright (c) 2014, 2017, Oracle and/or its affiliates. All rights reserved. |
|
3 * Copyright (c) 2015, 2017, SAP SE. All rights reserved. |
|
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
|
5 * |
|
6 * This code is free software; you can redistribute it and/or modify it |
|
7 * under the terms of the GNU General Public License version 2 only, as |
|
8 * published by the Free Software Foundation. |
|
9 * |
|
10 * This code is distributed in the hope that it will be useful, but WITHOUT |
|
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
13 * version 2 for more details (a copy is included in the LICENSE file that |
|
14 * accompanied this code). |
|
15 * |
|
16 * You should have received a copy of the GNU General Public License version |
|
17 * 2 along with this work; if not, write to the Free Software Foundation, |
|
18 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
19 * |
|
20 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
21 * or visit www.oracle.com if you need additional information or have any |
|
22 * questions. |
|
23 * |
|
24 */ |
|
25 |
|
26 #include "precompiled.hpp" |
|
27 #include "asm/macroAssembler.inline.hpp" |
|
28 #include "interpreter/bytecodeHistogram.hpp" |
|
29 #include "interpreter/interpreter.hpp" |
|
30 #include "interpreter/interpreterRuntime.hpp" |
|
31 #include "interpreter/interp_masm.hpp" |
|
32 #include "interpreter/templateInterpreterGenerator.hpp" |
|
33 #include "interpreter/templateTable.hpp" |
|
34 #include "oops/arrayOop.hpp" |
|
35 #include "oops/methodData.hpp" |
|
36 #include "oops/method.hpp" |
|
37 #include "oops/oop.inline.hpp" |
|
38 #include "prims/jvmtiExport.hpp" |
|
39 #include "prims/jvmtiThreadState.hpp" |
|
40 #include "runtime/arguments.hpp" |
|
41 #include "runtime/deoptimization.hpp" |
|
42 #include "runtime/frame.inline.hpp" |
|
43 #include "runtime/sharedRuntime.hpp" |
|
44 #include "runtime/stubRoutines.hpp" |
|
45 #include "runtime/synchronizer.hpp" |
|
46 #include "runtime/timer.hpp" |
|
47 #include "runtime/vframeArray.hpp" |
|
48 #include "utilities/debug.hpp" |
|
49 #include "utilities/macros.hpp" |
|
50 |
|
51 #undef __ |
|
52 #define __ _masm-> |
|
53 |
|
54 // Size of interpreter code. Increase if too small. Interpreter will |
|
55 // fail with a guarantee ("not enough space for interpreter generation"); |
|
56 // if too small. |
|
57 // Run with +PrintInterpreter to get the VM to print out the size. |
|
58 // Max size with JVMTI |
|
59 int TemplateInterpreter::InterpreterCodeSize = 256*K; |
|
60 |
|
61 #ifdef PRODUCT |
|
62 #define BLOCK_COMMENT(str) /* nothing */ |
|
63 #else |
|
64 #define BLOCK_COMMENT(str) __ block_comment(str) |
|
65 #endif |
|
66 |
|
67 #define BIND(label) __ bind(label); BLOCK_COMMENT(#label ":") |
|
68 |
|
69 //----------------------------------------------------------------------------- |
|
70 |
|
71 address TemplateInterpreterGenerator::generate_slow_signature_handler() { |
|
72 // Slow_signature handler that respects the PPC C calling conventions. |
|
73 // |
|
74 // We get called by the native entry code with our output register |
|
75 // area == 8. First we call InterpreterRuntime::get_result_handler |
|
76 // to copy the pointer to the signature string temporarily to the |
|
77 // first C-argument and to return the result_handler in |
|
78 // R3_RET. Since native_entry will copy the jni-pointer to the |
|
79 // first C-argument slot later on, it is OK to occupy this slot |
|
80 // temporarilly. Then we copy the argument list on the java |
|
81 // expression stack into native varargs format on the native stack |
|
82 // and load arguments into argument registers. Integer arguments in |
|
83 // the varargs vector will be sign-extended to 8 bytes. |
|
84 // |
|
85 // On entry: |
|
86 // R3_ARG1 - intptr_t* Address of java argument list in memory. |
|
87 // R15_prev_state - BytecodeInterpreter* Address of interpreter state for |
|
88 // this method |
|
89 // R19_method |
|
90 // |
|
91 // On exit (just before return instruction): |
|
92 // R3_RET - contains the address of the result_handler. |
|
93 // R4_ARG2 - is not updated for static methods and contains "this" otherwise. |
|
94 // R5_ARG3-R10_ARG8: - When the (i-2)th Java argument is not of type float or double, |
|
95 // ARGi contains this argument. Otherwise, ARGi is not updated. |
|
96 // F1_ARG1-F13_ARG13 - contain the first 13 arguments of type float or double. |
|
97 |
|
98 const int LogSizeOfTwoInstructions = 3; |
|
99 |
|
100 // FIXME: use Argument:: GL: Argument names different numbers! |
|
101 const int max_fp_register_arguments = 13; |
|
102 const int max_int_register_arguments = 6; // first 2 are reserved |
|
103 |
|
104 const Register arg_java = R21_tmp1; |
|
105 const Register arg_c = R22_tmp2; |
|
106 const Register signature = R23_tmp3; // is string |
|
107 const Register sig_byte = R24_tmp4; |
|
108 const Register fpcnt = R25_tmp5; |
|
109 const Register argcnt = R26_tmp6; |
|
110 const Register intSlot = R27_tmp7; |
|
111 const Register target_sp = R28_tmp8; |
|
112 const FloatRegister floatSlot = F0; |
|
113 |
|
114 address entry = __ function_entry(); |
|
115 |
|
116 __ save_LR_CR(R0); |
|
117 __ save_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14)); |
|
118 // We use target_sp for storing arguments in the C frame. |
|
119 __ mr(target_sp, R1_SP); |
|
120 __ push_frame_reg_args_nonvolatiles(0, R11_scratch1); |
|
121 |
|
122 __ mr(arg_java, R3_ARG1); |
|
123 |
|
124 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_signature), R16_thread, R19_method); |
|
125 |
|
126 // Signature is in R3_RET. Signature is callee saved. |
|
127 __ mr(signature, R3_RET); |
|
128 |
|
129 // Get the result handler. |
|
130 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::get_result_handler), R16_thread, R19_method); |
|
131 |
|
132 { |
|
133 Label L; |
|
134 // test if static |
|
135 // _access_flags._flags must be at offset 0. |
|
136 // TODO PPC port: requires change in shared code. |
|
137 //assert(in_bytes(AccessFlags::flags_offset()) == 0, |
|
138 // "MethodDesc._access_flags == MethodDesc._access_flags._flags"); |
|
139 // _access_flags must be a 32 bit value. |
|
140 assert(sizeof(AccessFlags) == 4, "wrong size"); |
|
141 __ lwa(R11_scratch1/*access_flags*/, method_(access_flags)); |
|
142 // testbit with condition register. |
|
143 __ testbitdi(CCR0, R0, R11_scratch1/*access_flags*/, JVM_ACC_STATIC_BIT); |
|
144 __ btrue(CCR0, L); |
|
145 // For non-static functions, pass "this" in R4_ARG2 and copy it |
|
146 // to 2nd C-arg slot. |
|
147 // We need to box the Java object here, so we use arg_java |
|
148 // (address of current Java stack slot) as argument and don't |
|
149 // dereference it as in case of ints, floats, etc. |
|
150 __ mr(R4_ARG2, arg_java); |
|
151 __ addi(arg_java, arg_java, -BytesPerWord); |
|
152 __ std(R4_ARG2, _abi(carg_2), target_sp); |
|
153 __ bind(L); |
|
154 } |
|
155 |
|
156 // Will be incremented directly after loop_start. argcnt=0 |
|
157 // corresponds to 3rd C argument. |
|
158 __ li(argcnt, -1); |
|
159 // arg_c points to 3rd C argument |
|
160 __ addi(arg_c, target_sp, _abi(carg_3)); |
|
161 // no floating-point args parsed so far |
|
162 __ li(fpcnt, 0); |
|
163 |
|
164 Label move_intSlot_to_ARG, move_floatSlot_to_FARG; |
|
165 Label loop_start, loop_end; |
|
166 Label do_int, do_long, do_float, do_double, do_dontreachhere, do_object, do_array, do_boxed; |
|
167 |
|
168 // signature points to '(' at entry |
|
169 #ifdef ASSERT |
|
170 __ lbz(sig_byte, 0, signature); |
|
171 __ cmplwi(CCR0, sig_byte, '('); |
|
172 __ bne(CCR0, do_dontreachhere); |
|
173 #endif |
|
174 |
|
175 __ bind(loop_start); |
|
176 |
|
177 __ addi(argcnt, argcnt, 1); |
|
178 __ lbzu(sig_byte, 1, signature); |
|
179 |
|
180 __ cmplwi(CCR0, sig_byte, ')'); // end of signature |
|
181 __ beq(CCR0, loop_end); |
|
182 |
|
183 __ cmplwi(CCR0, sig_byte, 'B'); // byte |
|
184 __ beq(CCR0, do_int); |
|
185 |
|
186 __ cmplwi(CCR0, sig_byte, 'C'); // char |
|
187 __ beq(CCR0, do_int); |
|
188 |
|
189 __ cmplwi(CCR0, sig_byte, 'D'); // double |
|
190 __ beq(CCR0, do_double); |
|
191 |
|
192 __ cmplwi(CCR0, sig_byte, 'F'); // float |
|
193 __ beq(CCR0, do_float); |
|
194 |
|
195 __ cmplwi(CCR0, sig_byte, 'I'); // int |
|
196 __ beq(CCR0, do_int); |
|
197 |
|
198 __ cmplwi(CCR0, sig_byte, 'J'); // long |
|
199 __ beq(CCR0, do_long); |
|
200 |
|
201 __ cmplwi(CCR0, sig_byte, 'S'); // short |
|
202 __ beq(CCR0, do_int); |
|
203 |
|
204 __ cmplwi(CCR0, sig_byte, 'Z'); // boolean |
|
205 __ beq(CCR0, do_int); |
|
206 |
|
207 __ cmplwi(CCR0, sig_byte, 'L'); // object |
|
208 __ beq(CCR0, do_object); |
|
209 |
|
210 __ cmplwi(CCR0, sig_byte, '['); // array |
|
211 __ beq(CCR0, do_array); |
|
212 |
|
213 // __ cmplwi(CCR0, sig_byte, 'V'); // void cannot appear since we do not parse the return type |
|
214 // __ beq(CCR0, do_void); |
|
215 |
|
216 __ bind(do_dontreachhere); |
|
217 |
|
218 __ unimplemented("ShouldNotReachHere in slow_signature_handler", 120); |
|
219 |
|
220 __ bind(do_array); |
|
221 |
|
222 { |
|
223 Label start_skip, end_skip; |
|
224 |
|
225 __ bind(start_skip); |
|
226 __ lbzu(sig_byte, 1, signature); |
|
227 __ cmplwi(CCR0, sig_byte, '['); |
|
228 __ beq(CCR0, start_skip); // skip further brackets |
|
229 __ cmplwi(CCR0, sig_byte, '9'); |
|
230 __ bgt(CCR0, end_skip); // no optional size |
|
231 __ cmplwi(CCR0, sig_byte, '0'); |
|
232 __ bge(CCR0, start_skip); // skip optional size |
|
233 __ bind(end_skip); |
|
234 |
|
235 __ cmplwi(CCR0, sig_byte, 'L'); |
|
236 __ beq(CCR0, do_object); // for arrays of objects, the name of the object must be skipped |
|
237 __ b(do_boxed); // otherwise, go directly to do_boxed |
|
238 } |
|
239 |
|
240 __ bind(do_object); |
|
241 { |
|
242 Label L; |
|
243 __ bind(L); |
|
244 __ lbzu(sig_byte, 1, signature); |
|
245 __ cmplwi(CCR0, sig_byte, ';'); |
|
246 __ bne(CCR0, L); |
|
247 } |
|
248 // Need to box the Java object here, so we use arg_java (address of |
|
249 // current Java stack slot) as argument and don't dereference it as |
|
250 // in case of ints, floats, etc. |
|
251 Label do_null; |
|
252 __ bind(do_boxed); |
|
253 __ ld(R0,0, arg_java); |
|
254 __ cmpdi(CCR0, R0, 0); |
|
255 __ li(intSlot,0); |
|
256 __ beq(CCR0, do_null); |
|
257 __ mr(intSlot, arg_java); |
|
258 __ bind(do_null); |
|
259 __ std(intSlot, 0, arg_c); |
|
260 __ addi(arg_java, arg_java, -BytesPerWord); |
|
261 __ addi(arg_c, arg_c, BytesPerWord); |
|
262 __ cmplwi(CCR0, argcnt, max_int_register_arguments); |
|
263 __ blt(CCR0, move_intSlot_to_ARG); |
|
264 __ b(loop_start); |
|
265 |
|
266 __ bind(do_int); |
|
267 __ lwa(intSlot, 0, arg_java); |
|
268 __ std(intSlot, 0, arg_c); |
|
269 __ addi(arg_java, arg_java, -BytesPerWord); |
|
270 __ addi(arg_c, arg_c, BytesPerWord); |
|
271 __ cmplwi(CCR0, argcnt, max_int_register_arguments); |
|
272 __ blt(CCR0, move_intSlot_to_ARG); |
|
273 __ b(loop_start); |
|
274 |
|
275 __ bind(do_long); |
|
276 __ ld(intSlot, -BytesPerWord, arg_java); |
|
277 __ std(intSlot, 0, arg_c); |
|
278 __ addi(arg_java, arg_java, - 2 * BytesPerWord); |
|
279 __ addi(arg_c, arg_c, BytesPerWord); |
|
280 __ cmplwi(CCR0, argcnt, max_int_register_arguments); |
|
281 __ blt(CCR0, move_intSlot_to_ARG); |
|
282 __ b(loop_start); |
|
283 |
|
284 __ bind(do_float); |
|
285 __ lfs(floatSlot, 0, arg_java); |
|
286 #if defined(LINUX) |
|
287 // Linux uses ELF ABI. Both original ELF and ELFv2 ABIs have float |
|
288 // in the least significant word of an argument slot. |
|
289 #if defined(VM_LITTLE_ENDIAN) |
|
290 __ stfs(floatSlot, 0, arg_c); |
|
291 #else |
|
292 __ stfs(floatSlot, 4, arg_c); |
|
293 #endif |
|
294 #elif defined(AIX) |
|
295 // Although AIX runs on big endian CPU, float is in most significant |
|
296 // word of an argument slot. |
|
297 __ stfs(floatSlot, 0, arg_c); |
|
298 #else |
|
299 #error "unknown OS" |
|
300 #endif |
|
301 __ addi(arg_java, arg_java, -BytesPerWord); |
|
302 __ addi(arg_c, arg_c, BytesPerWord); |
|
303 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments); |
|
304 __ blt(CCR0, move_floatSlot_to_FARG); |
|
305 __ b(loop_start); |
|
306 |
|
307 __ bind(do_double); |
|
308 __ lfd(floatSlot, - BytesPerWord, arg_java); |
|
309 __ stfd(floatSlot, 0, arg_c); |
|
310 __ addi(arg_java, arg_java, - 2 * BytesPerWord); |
|
311 __ addi(arg_c, arg_c, BytesPerWord); |
|
312 __ cmplwi(CCR0, fpcnt, max_fp_register_arguments); |
|
313 __ blt(CCR0, move_floatSlot_to_FARG); |
|
314 __ b(loop_start); |
|
315 |
|
316 __ bind(loop_end); |
|
317 |
|
318 __ pop_frame(); |
|
319 __ restore_nonvolatile_gprs(R1_SP, _spill_nonvolatiles_neg(r14)); |
|
320 __ restore_LR_CR(R0); |
|
321 |
|
322 __ blr(); |
|
323 |
|
324 Label move_int_arg, move_float_arg; |
|
325 __ bind(move_int_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions) |
|
326 __ mr(R5_ARG3, intSlot); __ b(loop_start); |
|
327 __ mr(R6_ARG4, intSlot); __ b(loop_start); |
|
328 __ mr(R7_ARG5, intSlot); __ b(loop_start); |
|
329 __ mr(R8_ARG6, intSlot); __ b(loop_start); |
|
330 __ mr(R9_ARG7, intSlot); __ b(loop_start); |
|
331 __ mr(R10_ARG8, intSlot); __ b(loop_start); |
|
332 |
|
333 __ bind(move_float_arg); // each case must consist of 2 instructions (otherwise adapt LogSizeOfTwoInstructions) |
|
334 __ fmr(F1_ARG1, floatSlot); __ b(loop_start); |
|
335 __ fmr(F2_ARG2, floatSlot); __ b(loop_start); |
|
336 __ fmr(F3_ARG3, floatSlot); __ b(loop_start); |
|
337 __ fmr(F4_ARG4, floatSlot); __ b(loop_start); |
|
338 __ fmr(F5_ARG5, floatSlot); __ b(loop_start); |
|
339 __ fmr(F6_ARG6, floatSlot); __ b(loop_start); |
|
340 __ fmr(F7_ARG7, floatSlot); __ b(loop_start); |
|
341 __ fmr(F8_ARG8, floatSlot); __ b(loop_start); |
|
342 __ fmr(F9_ARG9, floatSlot); __ b(loop_start); |
|
343 __ fmr(F10_ARG10, floatSlot); __ b(loop_start); |
|
344 __ fmr(F11_ARG11, floatSlot); __ b(loop_start); |
|
345 __ fmr(F12_ARG12, floatSlot); __ b(loop_start); |
|
346 __ fmr(F13_ARG13, floatSlot); __ b(loop_start); |
|
347 |
|
348 __ bind(move_intSlot_to_ARG); |
|
349 __ sldi(R0, argcnt, LogSizeOfTwoInstructions); |
|
350 __ load_const(R11_scratch1, move_int_arg); // Label must be bound here. |
|
351 __ add(R11_scratch1, R0, R11_scratch1); |
|
352 __ mtctr(R11_scratch1/*branch_target*/); |
|
353 __ bctr(); |
|
354 __ bind(move_floatSlot_to_FARG); |
|
355 __ sldi(R0, fpcnt, LogSizeOfTwoInstructions); |
|
356 __ addi(fpcnt, fpcnt, 1); |
|
357 __ load_const(R11_scratch1, move_float_arg); // Label must be bound here. |
|
358 __ add(R11_scratch1, R0, R11_scratch1); |
|
359 __ mtctr(R11_scratch1/*branch_target*/); |
|
360 __ bctr(); |
|
361 |
|
362 return entry; |
|
363 } |
|
364 |
|
365 address TemplateInterpreterGenerator::generate_result_handler_for(BasicType type) { |
|
366 // |
|
367 // Registers alive |
|
368 // R3_RET |
|
369 // LR |
|
370 // |
|
371 // Registers updated |
|
372 // R3_RET |
|
373 // |
|
374 |
|
375 Label done; |
|
376 address entry = __ pc(); |
|
377 |
|
378 switch (type) { |
|
379 case T_BOOLEAN: |
|
380 // convert !=0 to 1 |
|
381 __ neg(R0, R3_RET); |
|
382 __ orr(R0, R3_RET, R0); |
|
383 __ srwi(R3_RET, R0, 31); |
|
384 break; |
|
385 case T_BYTE: |
|
386 // sign extend 8 bits |
|
387 __ extsb(R3_RET, R3_RET); |
|
388 break; |
|
389 case T_CHAR: |
|
390 // zero extend 16 bits |
|
391 __ clrldi(R3_RET, R3_RET, 48); |
|
392 break; |
|
393 case T_SHORT: |
|
394 // sign extend 16 bits |
|
395 __ extsh(R3_RET, R3_RET); |
|
396 break; |
|
397 case T_INT: |
|
398 // sign extend 32 bits |
|
399 __ extsw(R3_RET, R3_RET); |
|
400 break; |
|
401 case T_LONG: |
|
402 break; |
|
403 case T_OBJECT: |
|
404 // JNIHandles::resolve result. |
|
405 __ resolve_jobject(R3_RET, R11_scratch1, R12_scratch2, /* needs_frame */ true); // kills R31 |
|
406 break; |
|
407 case T_FLOAT: |
|
408 break; |
|
409 case T_DOUBLE: |
|
410 break; |
|
411 case T_VOID: |
|
412 break; |
|
413 default: ShouldNotReachHere(); |
|
414 } |
|
415 |
|
416 BIND(done); |
|
417 __ blr(); |
|
418 |
|
419 return entry; |
|
420 } |
|
421 |
|
422 // Abstract method entry. |
|
423 // |
|
424 address TemplateInterpreterGenerator::generate_abstract_entry(void) { |
|
425 address entry = __ pc(); |
|
426 |
|
427 // |
|
428 // Registers alive |
|
429 // R16_thread - JavaThread* |
|
430 // R19_method - callee's method (method to be invoked) |
|
431 // R1_SP - SP prepared such that caller's outgoing args are near top |
|
432 // LR - return address to caller |
|
433 // |
|
434 // Stack layout at this point: |
|
435 // |
|
436 // 0 [TOP_IJAVA_FRAME_ABI] <-- R1_SP |
|
437 // alignment (optional) |
|
438 // [outgoing Java arguments] |
|
439 // ... |
|
440 // PARENT [PARENT_IJAVA_FRAME_ABI] |
|
441 // ... |
|
442 // |
|
443 |
|
444 // Can't use call_VM here because we have not set up a new |
|
445 // interpreter state. Make the call to the vm and make it look like |
|
446 // our caller set up the JavaFrameAnchor. |
|
447 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); |
|
448 |
|
449 // Push a new C frame and save LR. |
|
450 __ save_LR_CR(R0); |
|
451 __ push_frame_reg_args(0, R11_scratch1); |
|
452 |
|
453 // This is not a leaf but we have a JavaFrameAnchor now and we will |
|
454 // check (create) exceptions afterward so this is ok. |
|
455 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError), |
|
456 R16_thread); |
|
457 |
|
458 // Pop the C frame and restore LR. |
|
459 __ pop_frame(); |
|
460 __ restore_LR_CR(R0); |
|
461 |
|
462 // Reset JavaFrameAnchor from call_VM_leaf above. |
|
463 __ reset_last_Java_frame(); |
|
464 |
|
465 // We don't know our caller, so jump to the general forward exception stub, |
|
466 // which will also pop our full frame off. Satisfy the interface of |
|
467 // SharedRuntime::generate_forward_exception() |
|
468 __ load_const_optimized(R11_scratch1, StubRoutines::forward_exception_entry(), R0); |
|
469 __ mtctr(R11_scratch1); |
|
470 __ bctr(); |
|
471 |
|
472 return entry; |
|
473 } |
|
474 |
|
475 // Interpreter intrinsic for WeakReference.get(). |
|
476 // 1. Don't push a full blown frame and go on dispatching, but fetch the value |
|
477 // into R8 and return quickly |
|
478 // 2. If G1 is active we *must* execute this intrinsic for corrrectness: |
|
479 // It contains a GC barrier which puts the reference into the satb buffer |
|
480 // to indicate that someone holds a strong reference to the object the |
|
481 // weak ref points to! |
|
482 address TemplateInterpreterGenerator::generate_Reference_get_entry(void) { |
|
483 // Code: _aload_0, _getfield, _areturn |
|
484 // parameter size = 1 |
|
485 // |
|
486 // The code that gets generated by this routine is split into 2 parts: |
|
487 // 1. the "intrinsified" code for G1 (or any SATB based GC), |
|
488 // 2. the slow path - which is an expansion of the regular method entry. |
|
489 // |
|
490 // Notes: |
|
491 // * In the G1 code we do not check whether we need to block for |
|
492 // a safepoint. If G1 is enabled then we must execute the specialized |
|
493 // code for Reference.get (except when the Reference object is null) |
|
494 // so that we can log the value in the referent field with an SATB |
|
495 // update buffer. |
|
496 // If the code for the getfield template is modified so that the |
|
497 // G1 pre-barrier code is executed when the current method is |
|
498 // Reference.get() then going through the normal method entry |
|
499 // will be fine. |
|
500 // * The G1 code can, however, check the receiver object (the instance |
|
501 // of java.lang.Reference) and jump to the slow path if null. If the |
|
502 // Reference object is null then we obviously cannot fetch the referent |
|
503 // and so we don't need to call the G1 pre-barrier. Thus we can use the |
|
504 // regular method entry code to generate the NPE. |
|
505 // |
|
506 |
|
507 if (UseG1GC) { |
|
508 address entry = __ pc(); |
|
509 |
|
510 const int referent_offset = java_lang_ref_Reference::referent_offset; |
|
511 guarantee(referent_offset > 0, "referent offset not initialized"); |
|
512 |
|
513 Label slow_path; |
|
514 |
|
515 // Debugging not possible, so can't use __ skip_if_jvmti_mode(slow_path, GR31_SCRATCH); |
|
516 |
|
517 // In the G1 code we don't check if we need to reach a safepoint. We |
|
518 // continue and the thread will safepoint at the next bytecode dispatch. |
|
519 |
|
520 // If the receiver is null then it is OK to jump to the slow path. |
|
521 __ ld(R3_RET, Interpreter::stackElementSize, R15_esp); // get receiver |
|
522 |
|
523 // Check if receiver == NULL and go the slow path. |
|
524 __ cmpdi(CCR0, R3_RET, 0); |
|
525 __ beq(CCR0, slow_path); |
|
526 |
|
527 // Load the value of the referent field. |
|
528 __ load_heap_oop(R3_RET, referent_offset, R3_RET); |
|
529 |
|
530 // Generate the G1 pre-barrier code to log the value of |
|
531 // the referent field in an SATB buffer. Note with |
|
532 // these parameters the pre-barrier does not generate |
|
533 // the load of the previous value. |
|
534 |
|
535 // Restore caller sp for c2i case. |
|
536 #ifdef ASSERT |
|
537 __ ld(R9_ARG7, 0, R1_SP); |
|
538 __ ld(R10_ARG8, 0, R21_sender_SP); |
|
539 __ cmpd(CCR0, R9_ARG7, R10_ARG8); |
|
540 __ asm_assert_eq("backlink", 0x544); |
|
541 #endif // ASSERT |
|
542 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. |
|
543 |
|
544 __ g1_write_barrier_pre(noreg, // obj |
|
545 noreg, // offset |
|
546 R3_RET, // pre_val |
|
547 R11_scratch1, // tmp |
|
548 R12_scratch2, // tmp |
|
549 true); // needs_frame |
|
550 |
|
551 __ blr(); |
|
552 |
|
553 // Generate regular method entry. |
|
554 __ bind(slow_path); |
|
555 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::zerolocals), R11_scratch1); |
|
556 return entry; |
|
557 } |
|
558 |
|
559 return NULL; |
|
560 } |
|
561 |
|
562 address TemplateInterpreterGenerator::generate_StackOverflowError_handler() { |
|
563 address entry = __ pc(); |
|
564 |
|
565 // Expression stack must be empty before entering the VM if an |
|
566 // exception happened. |
|
567 __ empty_expression_stack(); |
|
568 // Throw exception. |
|
569 __ call_VM(noreg, |
|
570 CAST_FROM_FN_PTR(address, |
|
571 InterpreterRuntime::throw_StackOverflowError)); |
|
572 return entry; |
|
573 } |
|
574 |
|
575 address TemplateInterpreterGenerator::generate_ArrayIndexOutOfBounds_handler(const char* name) { |
|
576 address entry = __ pc(); |
|
577 __ empty_expression_stack(); |
|
578 __ load_const_optimized(R4_ARG2, (address) name); |
|
579 // Index is in R17_tos. |
|
580 __ mr(R5_ARG3, R17_tos); |
|
581 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ArrayIndexOutOfBoundsException)); |
|
582 return entry; |
|
583 } |
|
584 |
|
585 #if 0 |
|
586 // Call special ClassCastException constructor taking object to cast |
|
587 // and target class as arguments. |
|
588 address TemplateInterpreterGenerator::generate_ClassCastException_verbose_handler() { |
|
589 address entry = __ pc(); |
|
590 |
|
591 // Expression stack must be empty before entering the VM if an |
|
592 // exception happened. |
|
593 __ empty_expression_stack(); |
|
594 |
|
595 // Thread will be loaded to R3_ARG1. |
|
596 // Target class oop is in register R5_ARG3 by convention! |
|
597 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException_verbose), R17_tos, R5_ARG3); |
|
598 // Above call must not return here since exception pending. |
|
599 DEBUG_ONLY(__ should_not_reach_here();) |
|
600 return entry; |
|
601 } |
|
602 #endif |
|
603 |
|
604 address TemplateInterpreterGenerator::generate_ClassCastException_handler() { |
|
605 address entry = __ pc(); |
|
606 // Expression stack must be empty before entering the VM if an |
|
607 // exception happened. |
|
608 __ empty_expression_stack(); |
|
609 |
|
610 // Load exception object. |
|
611 // Thread will be loaded to R3_ARG1. |
|
612 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_ClassCastException), R17_tos); |
|
613 #ifdef ASSERT |
|
614 // Above call must not return here since exception pending. |
|
615 __ should_not_reach_here(); |
|
616 #endif |
|
617 return entry; |
|
618 } |
|
619 |
|
620 address TemplateInterpreterGenerator::generate_exception_handler_common(const char* name, const char* message, bool pass_oop) { |
|
621 address entry = __ pc(); |
|
622 //__ untested("generate_exception_handler_common"); |
|
623 Register Rexception = R17_tos; |
|
624 |
|
625 // Expression stack must be empty before entering the VM if an exception happened. |
|
626 __ empty_expression_stack(); |
|
627 |
|
628 __ load_const_optimized(R4_ARG2, (address) name, R11_scratch1); |
|
629 if (pass_oop) { |
|
630 __ mr(R5_ARG3, Rexception); |
|
631 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_klass_exception), false); |
|
632 } else { |
|
633 __ load_const_optimized(R5_ARG3, (address) message, R11_scratch1); |
|
634 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::create_exception), false); |
|
635 } |
|
636 |
|
637 // Throw exception. |
|
638 __ mr(R3_ARG1, Rexception); |
|
639 __ load_const_optimized(R11_scratch1, Interpreter::throw_exception_entry(), R12_scratch2); |
|
640 __ mtctr(R11_scratch1); |
|
641 __ bctr(); |
|
642 |
|
643 return entry; |
|
644 } |
|
645 |
|
646 // This entry is returned to when a call returns to the interpreter. |
|
647 // When we arrive here, we expect that the callee stack frame is already popped. |
|
648 address TemplateInterpreterGenerator::generate_return_entry_for(TosState state, int step, size_t index_size) { |
|
649 address entry = __ pc(); |
|
650 |
|
651 // Move the value out of the return register back to the TOS cache of current frame. |
|
652 switch (state) { |
|
653 case ltos: |
|
654 case btos: |
|
655 case ztos: |
|
656 case ctos: |
|
657 case stos: |
|
658 case atos: |
|
659 case itos: __ mr(R17_tos, R3_RET); break; // RET -> TOS cache |
|
660 case ftos: |
|
661 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET |
|
662 case vtos: break; // Nothing to do, this was a void return. |
|
663 default : ShouldNotReachHere(); |
|
664 } |
|
665 |
|
666 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. |
|
667 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); |
|
668 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); |
|
669 |
|
670 // Compiled code destroys templateTableBase, reload. |
|
671 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R12_scratch2); |
|
672 |
|
673 if (state == atos) { |
|
674 __ profile_return_type(R3_RET, R11_scratch1, R12_scratch2); |
|
675 } |
|
676 |
|
677 const Register cache = R11_scratch1; |
|
678 const Register size = R12_scratch2; |
|
679 __ get_cache_and_index_at_bcp(cache, 1, index_size); |
|
680 |
|
681 // Get least significant byte of 64 bit value: |
|
682 #if defined(VM_LITTLE_ENDIAN) |
|
683 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()), cache); |
|
684 #else |
|
685 __ lbz(size, in_bytes(ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::flags_offset()) + 7, cache); |
|
686 #endif |
|
687 __ sldi(size, size, Interpreter::logStackElementSize); |
|
688 __ add(R15_esp, R15_esp, size); |
|
689 |
|
690 __ check_and_handle_popframe(R11_scratch1); |
|
691 __ check_and_handle_earlyret(R11_scratch1); |
|
692 |
|
693 __ dispatch_next(state, step); |
|
694 return entry; |
|
695 } |
|
696 |
|
697 address TemplateInterpreterGenerator::generate_deopt_entry_for(TosState state, int step) { |
|
698 address entry = __ pc(); |
|
699 // If state != vtos, we're returning from a native method, which put it's result |
|
700 // into the result register. So move the value out of the return register back |
|
701 // to the TOS cache of current frame. |
|
702 |
|
703 switch (state) { |
|
704 case ltos: |
|
705 case btos: |
|
706 case ztos: |
|
707 case ctos: |
|
708 case stos: |
|
709 case atos: |
|
710 case itos: __ mr(R17_tos, R3_RET); break; // GR_RET -> TOS cache |
|
711 case ftos: |
|
712 case dtos: __ fmr(F15_ftos, F1_RET); break; // TOS cache -> GR_FRET |
|
713 case vtos: break; // Nothing to do, this was a void return. |
|
714 default : ShouldNotReachHere(); |
|
715 } |
|
716 |
|
717 // Load LcpoolCache @@@ should be already set! |
|
718 __ get_constant_pool_cache(R27_constPoolCache); |
|
719 |
|
720 // Handle a pending exception, fall through if none. |
|
721 __ check_and_forward_exception(R11_scratch1, R12_scratch2); |
|
722 |
|
723 // Start executing bytecodes. |
|
724 __ dispatch_next(state, step); |
|
725 |
|
726 return entry; |
|
727 } |
|
728 |
|
729 address TemplateInterpreterGenerator::generate_safept_entry_for(TosState state, address runtime_entry) { |
|
730 address entry = __ pc(); |
|
731 |
|
732 __ push(state); |
|
733 __ call_VM(noreg, runtime_entry); |
|
734 __ dispatch_via(vtos, Interpreter::_normal_table.table_for(vtos)); |
|
735 |
|
736 return entry; |
|
737 } |
|
738 |
|
739 // Helpers for commoning out cases in the various type of method entries. |
|
740 |
|
741 // Increment invocation count & check for overflow. |
|
742 // |
|
743 // Note: checking for negative value instead of overflow |
|
744 // so we have a 'sticky' overflow test. |
|
745 // |
|
746 void TemplateInterpreterGenerator::generate_counter_incr(Label* overflow, Label* profile_method, Label* profile_method_continue) { |
|
747 // Note: In tiered we increment either counters in method or in MDO depending if we're profiling or not. |
|
748 Register Rscratch1 = R11_scratch1; |
|
749 Register Rscratch2 = R12_scratch2; |
|
750 Register R3_counters = R3_ARG1; |
|
751 Label done; |
|
752 |
|
753 if (TieredCompilation) { |
|
754 const int increment = InvocationCounter::count_increment; |
|
755 Label no_mdo; |
|
756 if (ProfileInterpreter) { |
|
757 const Register Rmdo = R3_counters; |
|
758 // If no method data exists, go to profile_continue. |
|
759 __ ld(Rmdo, in_bytes(Method::method_data_offset()), R19_method); |
|
760 __ cmpdi(CCR0, Rmdo, 0); |
|
761 __ beq(CCR0, no_mdo); |
|
762 |
|
763 // Increment invocation counter in the MDO. |
|
764 const int mdo_ic_offs = in_bytes(MethodData::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); |
|
765 __ lwz(Rscratch2, mdo_ic_offs, Rmdo); |
|
766 __ lwz(Rscratch1, in_bytes(MethodData::invoke_mask_offset()), Rmdo); |
|
767 __ addi(Rscratch2, Rscratch2, increment); |
|
768 __ stw(Rscratch2, mdo_ic_offs, Rmdo); |
|
769 __ and_(Rscratch1, Rscratch2, Rscratch1); |
|
770 __ bne(CCR0, done); |
|
771 __ b(*overflow); |
|
772 } |
|
773 |
|
774 // Increment counter in MethodCounters*. |
|
775 const int mo_ic_offs = in_bytes(MethodCounters::invocation_counter_offset()) + in_bytes(InvocationCounter::counter_offset()); |
|
776 __ bind(no_mdo); |
|
777 __ get_method_counters(R19_method, R3_counters, done); |
|
778 __ lwz(Rscratch2, mo_ic_offs, R3_counters); |
|
779 __ lwz(Rscratch1, in_bytes(MethodCounters::invoke_mask_offset()), R3_counters); |
|
780 __ addi(Rscratch2, Rscratch2, increment); |
|
781 __ stw(Rscratch2, mo_ic_offs, R3_counters); |
|
782 __ and_(Rscratch1, Rscratch2, Rscratch1); |
|
783 __ beq(CCR0, *overflow); |
|
784 |
|
785 __ bind(done); |
|
786 |
|
787 } else { |
|
788 |
|
789 // Update standard invocation counters. |
|
790 Register Rsum_ivc_bec = R4_ARG2; |
|
791 __ get_method_counters(R19_method, R3_counters, done); |
|
792 __ increment_invocation_counter(R3_counters, Rsum_ivc_bec, R12_scratch2); |
|
793 // Increment interpreter invocation counter. |
|
794 if (ProfileInterpreter) { // %%% Merge this into methodDataOop. |
|
795 __ lwz(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); |
|
796 __ addi(R12_scratch2, R12_scratch2, 1); |
|
797 __ stw(R12_scratch2, in_bytes(MethodCounters::interpreter_invocation_counter_offset()), R3_counters); |
|
798 } |
|
799 // Check if we must create a method data obj. |
|
800 if (ProfileInterpreter && profile_method != NULL) { |
|
801 const Register profile_limit = Rscratch1; |
|
802 __ lwz(profile_limit, in_bytes(MethodCounters::interpreter_profile_limit_offset()), R3_counters); |
|
803 // Test to see if we should create a method data oop. |
|
804 __ cmpw(CCR0, Rsum_ivc_bec, profile_limit); |
|
805 __ blt(CCR0, *profile_method_continue); |
|
806 // If no method data exists, go to profile_method. |
|
807 __ test_method_data_pointer(*profile_method); |
|
808 } |
|
809 // Finally check for counter overflow. |
|
810 if (overflow) { |
|
811 const Register invocation_limit = Rscratch1; |
|
812 __ lwz(invocation_limit, in_bytes(MethodCounters::interpreter_invocation_limit_offset()), R3_counters); |
|
813 __ cmpw(CCR0, Rsum_ivc_bec, invocation_limit); |
|
814 __ bge(CCR0, *overflow); |
|
815 } |
|
816 |
|
817 __ bind(done); |
|
818 } |
|
819 } |
|
820 |
|
821 // Generate code to initiate compilation on invocation counter overflow. |
|
822 void TemplateInterpreterGenerator::generate_counter_overflow(Label& continue_entry) { |
|
823 // Generate code to initiate compilation on the counter overflow. |
|
824 |
|
825 // InterpreterRuntime::frequency_counter_overflow takes one arguments, |
|
826 // which indicates if the counter overflow occurs at a backwards branch (NULL bcp) |
|
827 // We pass zero in. |
|
828 // The call returns the address of the verified entry point for the method or NULL |
|
829 // if the compilation did not complete (either went background or bailed out). |
|
830 // |
|
831 // Unlike the C++ interpreter above: Check exceptions! |
|
832 // Assumption: Caller must set the flag "do_not_unlock_if_sychronized" if the monitor of a sync'ed |
|
833 // method has not yet been created. Thus, no unlocking of a non-existing monitor can occur. |
|
834 |
|
835 __ li(R4_ARG2, 0); |
|
836 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), R4_ARG2, true); |
|
837 |
|
838 // Returns verified_entry_point or NULL. |
|
839 // We ignore it in any case. |
|
840 __ b(continue_entry); |
|
841 } |
|
842 |
|
843 // See if we've got enough room on the stack for locals plus overhead below |
|
844 // JavaThread::stack_overflow_limit(). If not, throw a StackOverflowError |
|
845 // without going through the signal handler, i.e., reserved and yellow zones |
|
846 // will not be made usable. The shadow zone must suffice to handle the |
|
847 // overflow. |
|
848 // |
|
849 // Kills Rmem_frame_size, Rscratch1. |
|
850 void TemplateInterpreterGenerator::generate_stack_overflow_check(Register Rmem_frame_size, Register Rscratch1) { |
|
851 Label done; |
|
852 assert_different_registers(Rmem_frame_size, Rscratch1); |
|
853 |
|
854 BLOCK_COMMENT("stack_overflow_check_with_compare {"); |
|
855 __ sub(Rmem_frame_size, R1_SP, Rmem_frame_size); |
|
856 __ ld(Rscratch1, thread_(stack_overflow_limit)); |
|
857 __ cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1); |
|
858 __ bgt(CCR0/*is_stack_overflow*/, done); |
|
859 |
|
860 // The stack overflows. Load target address of the runtime stub and call it. |
|
861 assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order"); |
|
862 __ load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0); |
|
863 __ mtctr(Rscratch1); |
|
864 // Restore caller_sp. |
|
865 #ifdef ASSERT |
|
866 __ ld(Rscratch1, 0, R1_SP); |
|
867 __ ld(R0, 0, R21_sender_SP); |
|
868 __ cmpd(CCR0, R0, Rscratch1); |
|
869 __ asm_assert_eq("backlink", 0x547); |
|
870 #endif // ASSERT |
|
871 __ mr(R1_SP, R21_sender_SP); |
|
872 __ bctr(); |
|
873 |
|
874 __ align(32, 12); |
|
875 __ bind(done); |
|
876 BLOCK_COMMENT("} stack_overflow_check_with_compare"); |
|
877 } |
|
878 |
|
879 // Lock the current method, interpreter register window must be set up! |
|
880 void TemplateInterpreterGenerator::lock_method(Register Rflags, Register Rscratch1, Register Rscratch2, bool flags_preloaded) { |
|
881 const Register Robj_to_lock = Rscratch2; |
|
882 |
|
883 { |
|
884 if (!flags_preloaded) { |
|
885 __ lwz(Rflags, method_(access_flags)); |
|
886 } |
|
887 |
|
888 #ifdef ASSERT |
|
889 // Check if methods needs synchronization. |
|
890 { |
|
891 Label Lok; |
|
892 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_SYNCHRONIZED_BIT); |
|
893 __ btrue(CCR0,Lok); |
|
894 __ stop("method doesn't need synchronization"); |
|
895 __ bind(Lok); |
|
896 } |
|
897 #endif // ASSERT |
|
898 } |
|
899 |
|
900 // Get synchronization object to Rscratch2. |
|
901 { |
|
902 Label Lstatic; |
|
903 Label Ldone; |
|
904 |
|
905 __ testbitdi(CCR0, R0, Rflags, JVM_ACC_STATIC_BIT); |
|
906 __ btrue(CCR0, Lstatic); |
|
907 |
|
908 // Non-static case: load receiver obj from stack and we're done. |
|
909 __ ld(Robj_to_lock, R18_locals); |
|
910 __ b(Ldone); |
|
911 |
|
912 __ bind(Lstatic); // Static case: Lock the java mirror |
|
913 // Load mirror from interpreter frame. |
|
914 __ ld(Robj_to_lock, _abi(callers_sp), R1_SP); |
|
915 __ ld(Robj_to_lock, _ijava_state_neg(mirror), Robj_to_lock); |
|
916 |
|
917 __ bind(Ldone); |
|
918 __ verify_oop(Robj_to_lock); |
|
919 } |
|
920 |
|
921 // Got the oop to lock => execute! |
|
922 __ add_monitor_to_stack(true, Rscratch1, R0); |
|
923 |
|
924 __ std(Robj_to_lock, BasicObjectLock::obj_offset_in_bytes(), R26_monitor); |
|
925 __ lock_object(R26_monitor, Robj_to_lock); |
|
926 } |
|
927 |
|
928 // Generate a fixed interpreter frame for pure interpreter |
|
929 // and I2N native transition frames. |
|
930 // |
|
931 // Before (stack grows downwards): |
|
932 // |
|
933 // | ... | |
|
934 // |------------- | |
|
935 // | java arg0 | |
|
936 // | ... | |
|
937 // | java argn | |
|
938 // | | <- R15_esp |
|
939 // | | |
|
940 // |--------------| |
|
941 // | abi_112 | |
|
942 // | | <- R1_SP |
|
943 // |==============| |
|
944 // |
|
945 // |
|
946 // After: |
|
947 // |
|
948 // | ... | |
|
949 // | java arg0 |<- R18_locals |
|
950 // | ... | |
|
951 // | java argn | |
|
952 // |--------------| |
|
953 // | | |
|
954 // | java locals | |
|
955 // | | |
|
956 // |--------------| |
|
957 // | abi_48 | |
|
958 // |==============| |
|
959 // | | |
|
960 // | istate | |
|
961 // | | |
|
962 // |--------------| |
|
963 // | monitor |<- R26_monitor |
|
964 // |--------------| |
|
965 // | |<- R15_esp |
|
966 // | expression | |
|
967 // | stack | |
|
968 // | | |
|
969 // |--------------| |
|
970 // | | |
|
971 // | abi_112 |<- R1_SP |
|
972 // |==============| |
|
973 // |
|
974 // The top most frame needs an abi space of 112 bytes. This space is needed, |
|
975 // since we call to c. The c function may spill their arguments to the caller |
|
976 // frame. When we call to java, we don't need these spill slots. In order to save |
|
977 // space on the stack, we resize the caller. However, java locals reside in |
|
978 // the caller frame and the frame has to be increased. The frame_size for the |
|
979 // current frame was calculated based on max_stack as size for the expression |
|
980 // stack. At the call, just a part of the expression stack might be used. |
|
981 // We don't want to waste this space and cut the frame back accordingly. |
|
982 // The resulting amount for resizing is calculated as follows: |
|
983 // resize = (number_of_locals - number_of_arguments) * slot_size |
|
984 // + (R1_SP - R15_esp) + 48 |
|
985 // |
|
986 // The size for the callee frame is calculated: |
|
987 // framesize = 112 + max_stack + monitor + state_size |
|
988 // |
|
989 // maxstack: Max number of slots on the expression stack, loaded from the method. |
|
990 // monitor: We statically reserve room for one monitor object. |
|
991 // state_size: We save the current state of the interpreter to this area. |
|
992 // |
|
993 void TemplateInterpreterGenerator::generate_fixed_frame(bool native_call, Register Rsize_of_parameters, Register Rsize_of_locals) { |
|
994 Register parent_frame_resize = R6_ARG4, // Frame will grow by this number of bytes. |
|
995 top_frame_size = R7_ARG5, |
|
996 Rconst_method = R8_ARG6; |
|
997 |
|
998 assert_different_registers(Rsize_of_parameters, Rsize_of_locals, parent_frame_resize, top_frame_size); |
|
999 |
|
1000 __ ld(Rconst_method, method_(const)); |
|
1001 __ lhz(Rsize_of_parameters /* number of params */, |
|
1002 in_bytes(ConstMethod::size_of_parameters_offset()), Rconst_method); |
|
1003 if (native_call) { |
|
1004 // If we're calling a native method, we reserve space for the worst-case signature |
|
1005 // handler varargs vector, which is max(Argument::n_register_parameters, parameter_count+2). |
|
1006 // We add two slots to the parameter_count, one for the jni |
|
1007 // environment and one for a possible native mirror. |
|
1008 Label skip_native_calculate_max_stack; |
|
1009 __ addi(top_frame_size, Rsize_of_parameters, 2); |
|
1010 __ cmpwi(CCR0, top_frame_size, Argument::n_register_parameters); |
|
1011 __ bge(CCR0, skip_native_calculate_max_stack); |
|
1012 __ li(top_frame_size, Argument::n_register_parameters); |
|
1013 __ bind(skip_native_calculate_max_stack); |
|
1014 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); |
|
1015 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); |
|
1016 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! |
|
1017 assert(Rsize_of_locals == noreg, "Rsize_of_locals not initialized"); // Only relevant value is Rsize_of_parameters. |
|
1018 } else { |
|
1019 __ lhz(Rsize_of_locals /* number of params */, in_bytes(ConstMethod::size_of_locals_offset()), Rconst_method); |
|
1020 __ sldi(Rsize_of_parameters, Rsize_of_parameters, Interpreter::logStackElementSize); |
|
1021 __ sldi(Rsize_of_locals, Rsize_of_locals, Interpreter::logStackElementSize); |
|
1022 __ lhz(top_frame_size, in_bytes(ConstMethod::max_stack_offset()), Rconst_method); |
|
1023 __ sub(R11_scratch1, Rsize_of_locals, Rsize_of_parameters); // >=0 |
|
1024 __ sub(parent_frame_resize, R1_SP, R15_esp); // <0, off by Interpreter::stackElementSize! |
|
1025 __ sldi(top_frame_size, top_frame_size, Interpreter::logStackElementSize); |
|
1026 __ add(parent_frame_resize, parent_frame_resize, R11_scratch1); |
|
1027 } |
|
1028 |
|
1029 // Compute top frame size. |
|
1030 __ addi(top_frame_size, top_frame_size, frame::abi_reg_args_size + frame::ijava_state_size); |
|
1031 |
|
1032 // Cut back area between esp and max_stack. |
|
1033 __ addi(parent_frame_resize, parent_frame_resize, frame::abi_minframe_size - Interpreter::stackElementSize); |
|
1034 |
|
1035 __ round_to(top_frame_size, frame::alignment_in_bytes); |
|
1036 __ round_to(parent_frame_resize, frame::alignment_in_bytes); |
|
1037 // parent_frame_resize = (locals-parameters) - (ESP-SP-ABI48) Rounded to frame alignment size. |
|
1038 // Enlarge by locals-parameters (not in case of native_call), shrink by ESP-SP-ABI48. |
|
1039 |
|
1040 if (!native_call) { |
|
1041 // Stack overflow check. |
|
1042 // Native calls don't need the stack size check since they have no |
|
1043 // expression stack and the arguments are already on the stack and |
|
1044 // we only add a handful of words to the stack. |
|
1045 __ add(R11_scratch1, parent_frame_resize, top_frame_size); |
|
1046 generate_stack_overflow_check(R11_scratch1, R12_scratch2); |
|
1047 } |
|
1048 |
|
1049 // Set up interpreter state registers. |
|
1050 |
|
1051 __ add(R18_locals, R15_esp, Rsize_of_parameters); |
|
1052 __ ld(R27_constPoolCache, in_bytes(ConstMethod::constants_offset()), Rconst_method); |
|
1053 __ ld(R27_constPoolCache, ConstantPool::cache_offset_in_bytes(), R27_constPoolCache); |
|
1054 |
|
1055 // Set method data pointer. |
|
1056 if (ProfileInterpreter) { |
|
1057 Label zero_continue; |
|
1058 __ ld(R28_mdx, method_(method_data)); |
|
1059 __ cmpdi(CCR0, R28_mdx, 0); |
|
1060 __ beq(CCR0, zero_continue); |
|
1061 __ addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); |
|
1062 __ bind(zero_continue); |
|
1063 } |
|
1064 |
|
1065 if (native_call) { |
|
1066 __ li(R14_bcp, 0); // Must initialize. |
|
1067 } else { |
|
1068 __ add(R14_bcp, in_bytes(ConstMethod::codes_offset()), Rconst_method); |
|
1069 } |
|
1070 |
|
1071 // Resize parent frame. |
|
1072 __ mflr(R12_scratch2); |
|
1073 __ neg(parent_frame_resize, parent_frame_resize); |
|
1074 __ resize_frame(parent_frame_resize, R11_scratch1); |
|
1075 __ std(R12_scratch2, _abi(lr), R1_SP); |
|
1076 |
|
1077 // Get mirror and store it in the frame as GC root for this Method*. |
|
1078 __ load_mirror_from_const_method(R12_scratch2, Rconst_method); |
|
1079 |
|
1080 __ addi(R26_monitor, R1_SP, - frame::ijava_state_size); |
|
1081 __ addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); |
|
1082 |
|
1083 // Store values. |
|
1084 // R15_esp, R14_bcp, R26_monitor, R28_mdx are saved at java calls |
|
1085 // in InterpreterMacroAssembler::call_from_interpreter. |
|
1086 __ std(R19_method, _ijava_state_neg(method), R1_SP); |
|
1087 __ std(R12_scratch2, _ijava_state_neg(mirror), R1_SP); |
|
1088 __ std(R21_sender_SP, _ijava_state_neg(sender_sp), R1_SP); |
|
1089 __ std(R27_constPoolCache, _ijava_state_neg(cpoolCache), R1_SP); |
|
1090 __ std(R18_locals, _ijava_state_neg(locals), R1_SP); |
|
1091 |
|
1092 // Note: esp, bcp, monitor, mdx live in registers. Hence, the correct version can only |
|
1093 // be found in the frame after save_interpreter_state is done. This is always true |
|
1094 // for non-top frames. But when a signal occurs, dumping the top frame can go wrong, |
|
1095 // because e.g. frame::interpreter_frame_bcp() will not access the correct value |
|
1096 // (Enhanced Stack Trace). |
|
1097 // The signal handler does not save the interpreter state into the frame. |
|
1098 __ li(R0, 0); |
|
1099 #ifdef ASSERT |
|
1100 // Fill remaining slots with constants. |
|
1101 __ load_const_optimized(R11_scratch1, 0x5afe); |
|
1102 __ load_const_optimized(R12_scratch2, 0xdead); |
|
1103 #endif |
|
1104 // We have to initialize some frame slots for native calls (accessed by GC). |
|
1105 if (native_call) { |
|
1106 __ std(R26_monitor, _ijava_state_neg(monitors), R1_SP); |
|
1107 __ std(R14_bcp, _ijava_state_neg(bcp), R1_SP); |
|
1108 if (ProfileInterpreter) { __ std(R28_mdx, _ijava_state_neg(mdx), R1_SP); } |
|
1109 } |
|
1110 #ifdef ASSERT |
|
1111 else { |
|
1112 __ std(R12_scratch2, _ijava_state_neg(monitors), R1_SP); |
|
1113 __ std(R12_scratch2, _ijava_state_neg(bcp), R1_SP); |
|
1114 __ std(R12_scratch2, _ijava_state_neg(mdx), R1_SP); |
|
1115 } |
|
1116 __ std(R11_scratch1, _ijava_state_neg(ijava_reserved), R1_SP); |
|
1117 __ std(R12_scratch2, _ijava_state_neg(esp), R1_SP); |
|
1118 __ std(R12_scratch2, _ijava_state_neg(lresult), R1_SP); |
|
1119 __ std(R12_scratch2, _ijava_state_neg(fresult), R1_SP); |
|
1120 #endif |
|
1121 __ subf(R12_scratch2, top_frame_size, R1_SP); |
|
1122 __ std(R0, _ijava_state_neg(oop_tmp), R1_SP); |
|
1123 __ std(R12_scratch2, _ijava_state_neg(top_frame_sp), R1_SP); |
|
1124 |
|
1125 // Push top frame. |
|
1126 __ push_frame(top_frame_size, R11_scratch1); |
|
1127 } |
|
1128 |
|
1129 // End of helpers |
|
1130 |
|
1131 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { |
|
1132 |
|
1133 // Decide what to do: Use same platform specific instructions and runtime calls as compilers. |
|
1134 bool use_instruction = false; |
|
1135 address runtime_entry = NULL; |
|
1136 int num_args = 1; |
|
1137 bool double_precision = true; |
|
1138 |
|
1139 // PPC64 specific: |
|
1140 switch (kind) { |
|
1141 case Interpreter::java_lang_math_sqrt: use_instruction = VM_Version::has_fsqrt(); break; |
|
1142 case Interpreter::java_lang_math_abs: use_instruction = true; break; |
|
1143 case Interpreter::java_lang_math_fmaF: |
|
1144 case Interpreter::java_lang_math_fmaD: use_instruction = UseFMA; break; |
|
1145 default: break; // Fall back to runtime call. |
|
1146 } |
|
1147 |
|
1148 switch (kind) { |
|
1149 case Interpreter::java_lang_math_sin : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsin); break; |
|
1150 case Interpreter::java_lang_math_cos : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dcos); break; |
|
1151 case Interpreter::java_lang_math_tan : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dtan); break; |
|
1152 case Interpreter::java_lang_math_abs : /* run interpreted */ break; |
|
1153 case Interpreter::java_lang_math_sqrt : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dsqrt); break; |
|
1154 case Interpreter::java_lang_math_log : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog); break; |
|
1155 case Interpreter::java_lang_math_log10: runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dlog10); break; |
|
1156 case Interpreter::java_lang_math_pow : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dpow); num_args = 2; break; |
|
1157 case Interpreter::java_lang_math_exp : runtime_entry = CAST_FROM_FN_PTR(address, SharedRuntime::dexp); break; |
|
1158 case Interpreter::java_lang_math_fmaF : /* run interpreted */ num_args = 3; double_precision = false; break; |
|
1159 case Interpreter::java_lang_math_fmaD : /* run interpreted */ num_args = 3; break; |
|
1160 default: ShouldNotReachHere(); |
|
1161 } |
|
1162 |
|
1163 // Use normal entry if neither instruction nor runtime call is used. |
|
1164 if (!use_instruction && runtime_entry == NULL) return NULL; |
|
1165 |
|
1166 address entry = __ pc(); |
|
1167 |
|
1168 // Load arguments |
|
1169 assert(num_args <= 13, "passed in registers"); |
|
1170 if (double_precision) { |
|
1171 int offset = (2 * num_args - 1) * Interpreter::stackElementSize; |
|
1172 for (int i = 0; i < num_args; ++i) { |
|
1173 __ lfd(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp); |
|
1174 offset -= 2 * Interpreter::stackElementSize; |
|
1175 } |
|
1176 } else { |
|
1177 int offset = num_args * Interpreter::stackElementSize; |
|
1178 for (int i = 0; i < num_args; ++i) { |
|
1179 __ lfs(as_FloatRegister(F1_ARG1->encoding() + i), offset, R15_esp); |
|
1180 offset -= Interpreter::stackElementSize; |
|
1181 } |
|
1182 } |
|
1183 |
|
1184 // Pop c2i arguments (if any) off when we return. |
|
1185 #ifdef ASSERT |
|
1186 __ ld(R9_ARG7, 0, R1_SP); |
|
1187 __ ld(R10_ARG8, 0, R21_sender_SP); |
|
1188 __ cmpd(CCR0, R9_ARG7, R10_ARG8); |
|
1189 __ asm_assert_eq("backlink", 0x545); |
|
1190 #endif // ASSERT |
|
1191 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. |
|
1192 |
|
1193 if (use_instruction) { |
|
1194 switch (kind) { |
|
1195 case Interpreter::java_lang_math_sqrt: __ fsqrt(F1_RET, F1); break; |
|
1196 case Interpreter::java_lang_math_abs: __ fabs(F1_RET, F1); break; |
|
1197 case Interpreter::java_lang_math_fmaF: __ fmadds(F1_RET, F1, F2, F3); break; |
|
1198 case Interpreter::java_lang_math_fmaD: __ fmadd(F1_RET, F1, F2, F3); break; |
|
1199 default: ShouldNotReachHere(); |
|
1200 } |
|
1201 } else { |
|
1202 // Comment: Can use tail call if the unextended frame is always C ABI compliant: |
|
1203 //__ load_const_optimized(R12_scratch2, runtime_entry, R0); |
|
1204 //__ call_c_and_return_to_caller(R12_scratch2); |
|
1205 |
|
1206 // Push a new C frame and save LR. |
|
1207 __ save_LR_CR(R0); |
|
1208 __ push_frame_reg_args(0, R11_scratch1); |
|
1209 |
|
1210 __ call_VM_leaf(runtime_entry); |
|
1211 |
|
1212 // Pop the C frame and restore LR. |
|
1213 __ pop_frame(); |
|
1214 __ restore_LR_CR(R0); |
|
1215 } |
|
1216 |
|
1217 __ blr(); |
|
1218 |
|
1219 __ flush(); |
|
1220 |
|
1221 return entry; |
|
1222 } |
|
1223 |
|
1224 void TemplateInterpreterGenerator::bang_stack_shadow_pages(bool native_call) { |
|
1225 // Quick & dirty stack overflow checking: bang the stack & handle trap. |
|
1226 // Note that we do the banging after the frame is setup, since the exception |
|
1227 // handling code expects to find a valid interpreter frame on the stack. |
|
1228 // Doing the banging earlier fails if the caller frame is not an interpreter |
|
1229 // frame. |
|
1230 // (Also, the exception throwing code expects to unlock any synchronized |
|
1231 // method receiever, so do the banging after locking the receiver.) |
|
1232 |
|
1233 // Bang each page in the shadow zone. We can't assume it's been done for |
|
1234 // an interpreter frame with greater than a page of locals, so each page |
|
1235 // needs to be checked. Only true for non-native. |
|
1236 if (UseStackBanging) { |
|
1237 const int page_size = os::vm_page_size(); |
|
1238 const int n_shadow_pages = ((int)JavaThread::stack_shadow_zone_size()) / page_size; |
|
1239 const int start_page = native_call ? n_shadow_pages : 1; |
|
1240 BLOCK_COMMENT("bang_stack_shadow_pages:"); |
|
1241 for (int pages = start_page; pages <= n_shadow_pages; pages++) { |
|
1242 __ bang_stack_with_offset(pages*page_size); |
|
1243 } |
|
1244 } |
|
1245 } |
|
1246 |
|
1247 // Interpreter stub for calling a native method. (asm interpreter) |
|
1248 // This sets up a somewhat different looking stack for calling the |
|
1249 // native method than the typical interpreter frame setup. |
|
1250 // |
|
1251 // On entry: |
|
1252 // R19_method - method |
|
1253 // R16_thread - JavaThread* |
|
1254 // R15_esp - intptr_t* sender tos |
|
1255 // |
|
1256 // abstract stack (grows up) |
|
1257 // [ IJava (caller of JNI callee) ] <-- ASP |
|
1258 // ... |
|
1259 address TemplateInterpreterGenerator::generate_native_entry(bool synchronized) { |
|
1260 |
|
1261 address entry = __ pc(); |
|
1262 |
|
1263 const bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; |
|
1264 |
|
1265 // ----------------------------------------------------------------------------- |
|
1266 // Allocate a new frame that represents the native callee (i2n frame). |
|
1267 // This is not a full-blown interpreter frame, but in particular, the |
|
1268 // following registers are valid after this: |
|
1269 // - R19_method |
|
1270 // - R18_local (points to start of arguments to native function) |
|
1271 // |
|
1272 // abstract stack (grows up) |
|
1273 // [ IJava (caller of JNI callee) ] <-- ASP |
|
1274 // ... |
|
1275 |
|
1276 const Register signature_handler_fd = R11_scratch1; |
|
1277 const Register pending_exception = R0; |
|
1278 const Register result_handler_addr = R31; |
|
1279 const Register native_method_fd = R11_scratch1; |
|
1280 const Register access_flags = R22_tmp2; |
|
1281 const Register active_handles = R11_scratch1; // R26_monitor saved to state. |
|
1282 const Register sync_state = R12_scratch2; |
|
1283 const Register sync_state_addr = sync_state; // Address is dead after use. |
|
1284 const Register suspend_flags = R11_scratch1; |
|
1285 |
|
1286 //============================================================================= |
|
1287 // Allocate new frame and initialize interpreter state. |
|
1288 |
|
1289 Label exception_return; |
|
1290 Label exception_return_sync_check; |
|
1291 Label stack_overflow_return; |
|
1292 |
|
1293 // Generate new interpreter state and jump to stack_overflow_return in case of |
|
1294 // a stack overflow. |
|
1295 //generate_compute_interpreter_state(stack_overflow_return); |
|
1296 |
|
1297 Register size_of_parameters = R22_tmp2; |
|
1298 |
|
1299 generate_fixed_frame(true, size_of_parameters, noreg /* unused */); |
|
1300 |
|
1301 //============================================================================= |
|
1302 // Increment invocation counter. On overflow, entry to JNI method |
|
1303 // will be compiled. |
|
1304 Label invocation_counter_overflow, continue_after_compile; |
|
1305 if (inc_counter) { |
|
1306 if (synchronized) { |
|
1307 // Since at this point in the method invocation the exception handler |
|
1308 // would try to exit the monitor of synchronized methods which hasn't |
|
1309 // been entered yet, we set the thread local variable |
|
1310 // _do_not_unlock_if_synchronized to true. If any exception was thrown by |
|
1311 // runtime, exception handling i.e. unlock_if_synchronized_method will |
|
1312 // check this thread local flag. |
|
1313 // This flag has two effects, one is to force an unwind in the topmost |
|
1314 // interpreter frame and not perform an unlock while doing so. |
|
1315 __ li(R0, 1); |
|
1316 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); |
|
1317 } |
|
1318 generate_counter_incr(&invocation_counter_overflow, NULL, NULL); |
|
1319 |
|
1320 BIND(continue_after_compile); |
|
1321 } |
|
1322 |
|
1323 bang_stack_shadow_pages(true); |
|
1324 |
|
1325 if (inc_counter) { |
|
1326 // Reset the _do_not_unlock_if_synchronized flag. |
|
1327 if (synchronized) { |
|
1328 __ li(R0, 0); |
|
1329 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); |
|
1330 } |
|
1331 } |
|
1332 |
|
1333 // access_flags = method->access_flags(); |
|
1334 // Load access flags. |
|
1335 assert(access_flags->is_nonvolatile(), |
|
1336 "access_flags must be in a non-volatile register"); |
|
1337 // Type check. |
|
1338 assert(4 == sizeof(AccessFlags), "unexpected field size"); |
|
1339 __ lwz(access_flags, method_(access_flags)); |
|
1340 |
|
1341 // We don't want to reload R19_method and access_flags after calls |
|
1342 // to some helper functions. |
|
1343 assert(R19_method->is_nonvolatile(), |
|
1344 "R19_method must be a non-volatile register"); |
|
1345 |
|
1346 // Check for synchronized methods. Must happen AFTER invocation counter |
|
1347 // check, so method is not locked if counter overflows. |
|
1348 |
|
1349 if (synchronized) { |
|
1350 lock_method(access_flags, R11_scratch1, R12_scratch2, true); |
|
1351 |
|
1352 // Update monitor in state. |
|
1353 __ ld(R11_scratch1, 0, R1_SP); |
|
1354 __ std(R26_monitor, _ijava_state_neg(monitors), R11_scratch1); |
|
1355 } |
|
1356 |
|
1357 // jvmti/jvmpi support |
|
1358 __ notify_method_entry(); |
|
1359 |
|
1360 //============================================================================= |
|
1361 // Get and call the signature handler. |
|
1362 |
|
1363 __ ld(signature_handler_fd, method_(signature_handler)); |
|
1364 Label call_signature_handler; |
|
1365 |
|
1366 __ cmpdi(CCR0, signature_handler_fd, 0); |
|
1367 __ bne(CCR0, call_signature_handler); |
|
1368 |
|
1369 // Method has never been called. Either generate a specialized |
|
1370 // handler or point to the slow one. |
|
1371 // |
|
1372 // Pass parameter 'false' to avoid exception check in call_VM. |
|
1373 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::prepare_native_call), R19_method, false); |
|
1374 |
|
1375 // Check for an exception while looking up the target method. If we |
|
1376 // incurred one, bail. |
|
1377 __ ld(pending_exception, thread_(pending_exception)); |
|
1378 __ cmpdi(CCR0, pending_exception, 0); |
|
1379 __ bne(CCR0, exception_return_sync_check); // Has pending exception. |
|
1380 |
|
1381 // Reload signature handler, it may have been created/assigned in the meanwhile. |
|
1382 __ ld(signature_handler_fd, method_(signature_handler)); |
|
1383 __ twi_0(signature_handler_fd); // Order wrt. load of klass mirror and entry point (isync is below). |
|
1384 |
|
1385 BIND(call_signature_handler); |
|
1386 |
|
1387 // Before we call the signature handler we push a new frame to |
|
1388 // protect the interpreter frame volatile registers when we return |
|
1389 // from jni but before we can get back to Java. |
|
1390 |
|
1391 // First set the frame anchor while the SP/FP registers are |
|
1392 // convenient and the slow signature handler can use this same frame |
|
1393 // anchor. |
|
1394 |
|
1395 // We have a TOP_IJAVA_FRAME here, which belongs to us. |
|
1396 __ set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R12_scratch2/*tmp*/); |
|
1397 |
|
1398 // Now the interpreter frame (and its call chain) have been |
|
1399 // invalidated and flushed. We are now protected against eager |
|
1400 // being enabled in native code. Even if it goes eager the |
|
1401 // registers will be reloaded as clean and we will invalidate after |
|
1402 // the call so no spurious flush should be possible. |
|
1403 |
|
1404 // Call signature handler and pass locals address. |
|
1405 // |
|
1406 // Our signature handlers copy required arguments to the C stack |
|
1407 // (outgoing C args), R3_ARG1 to R10_ARG8, and FARG1 to FARG13. |
|
1408 __ mr(R3_ARG1, R18_locals); |
|
1409 #if !defined(ABI_ELFv2) |
|
1410 __ ld(signature_handler_fd, 0, signature_handler_fd); |
|
1411 #endif |
|
1412 |
|
1413 __ call_stub(signature_handler_fd); |
|
1414 |
|
1415 // Remove the register parameter varargs slots we allocated in |
|
1416 // compute_interpreter_state. SP+16 ends up pointing to the ABI |
|
1417 // outgoing argument area. |
|
1418 // |
|
1419 // Not needed on PPC64. |
|
1420 //__ add(SP, SP, Argument::n_register_parameters*BytesPerWord); |
|
1421 |
|
1422 assert(result_handler_addr->is_nonvolatile(), "result_handler_addr must be in a non-volatile register"); |
|
1423 // Save across call to native method. |
|
1424 __ mr(result_handler_addr, R3_RET); |
|
1425 |
|
1426 __ isync(); // Acquire signature handler before trying to fetch the native entry point and klass mirror. |
|
1427 |
|
1428 // Set up fixed parameters and call the native method. |
|
1429 // If the method is static, get mirror into R4_ARG2. |
|
1430 { |
|
1431 Label method_is_not_static; |
|
1432 // Access_flags is non-volatile and still, no need to restore it. |
|
1433 |
|
1434 // Restore access flags. |
|
1435 __ testbitdi(CCR0, R0, access_flags, JVM_ACC_STATIC_BIT); |
|
1436 __ bfalse(CCR0, method_is_not_static); |
|
1437 |
|
1438 __ ld(R11_scratch1, _abi(callers_sp), R1_SP); |
|
1439 // Load mirror from interpreter frame. |
|
1440 __ ld(R12_scratch2, _ijava_state_neg(mirror), R11_scratch1); |
|
1441 // R4_ARG2 = &state->_oop_temp; |
|
1442 __ addi(R4_ARG2, R11_scratch1, _ijava_state_neg(oop_tmp)); |
|
1443 __ std(R12_scratch2/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); |
|
1444 BIND(method_is_not_static); |
|
1445 } |
|
1446 |
|
1447 // At this point, arguments have been copied off the stack into |
|
1448 // their JNI positions. Oops are boxed in-place on the stack, with |
|
1449 // handles copied to arguments. The result handler address is in a |
|
1450 // register. |
|
1451 |
|
1452 // Pass JNIEnv address as first parameter. |
|
1453 __ addir(R3_ARG1, thread_(jni_environment)); |
|
1454 |
|
1455 // Load the native_method entry before we change the thread state. |
|
1456 __ ld(native_method_fd, method_(native_function)); |
|
1457 |
|
1458 //============================================================================= |
|
1459 // Transition from _thread_in_Java to _thread_in_native. As soon as |
|
1460 // we make this change the safepoint code needs to be certain that |
|
1461 // the last Java frame we established is good. The pc in that frame |
|
1462 // just needs to be near here not an actual return address. |
|
1463 |
|
1464 // We use release_store_fence to update values like the thread state, where |
|
1465 // we don't want the current thread to continue until all our prior memory |
|
1466 // accesses (including the new thread state) are visible to other threads. |
|
1467 __ li(R0, _thread_in_native); |
|
1468 __ release(); |
|
1469 |
|
1470 // TODO PPC port assert(4 == JavaThread::sz_thread_state(), "unexpected field size"); |
|
1471 __ stw(R0, thread_(thread_state)); |
|
1472 |
|
1473 if (UseMembar) { |
|
1474 __ fence(); |
|
1475 } |
|
1476 |
|
1477 //============================================================================= |
|
1478 // Call the native method. Argument registers must not have been |
|
1479 // overwritten since "__ call_stub(signature_handler);" (except for |
|
1480 // ARG1 and ARG2 for static methods). |
|
1481 __ call_c(native_method_fd); |
|
1482 |
|
1483 __ li(R0, 0); |
|
1484 __ ld(R11_scratch1, 0, R1_SP); |
|
1485 __ std(R3_RET, _ijava_state_neg(lresult), R11_scratch1); |
|
1486 __ stfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); |
|
1487 __ std(R0/*mirror*/, _ijava_state_neg(oop_tmp), R11_scratch1); // reset |
|
1488 |
|
1489 // Note: C++ interpreter needs the following here: |
|
1490 // The frame_manager_lr field, which we use for setting the last |
|
1491 // java frame, gets overwritten by the signature handler. Restore |
|
1492 // it now. |
|
1493 //__ get_PC_trash_LR(R11_scratch1); |
|
1494 //__ std(R11_scratch1, _top_ijava_frame_abi(frame_manager_lr), R1_SP); |
|
1495 |
|
1496 // Because of GC R19_method may no longer be valid. |
|
1497 |
|
1498 // Block, if necessary, before resuming in _thread_in_Java state. |
|
1499 // In order for GC to work, don't clear the last_Java_sp until after |
|
1500 // blocking. |
|
1501 |
|
1502 //============================================================================= |
|
1503 // Switch thread to "native transition" state before reading the |
|
1504 // synchronization state. This additional state is necessary |
|
1505 // because reading and testing the synchronization state is not |
|
1506 // atomic w.r.t. GC, as this scenario demonstrates: Java thread A, |
|
1507 // in _thread_in_native state, loads _not_synchronized and is |
|
1508 // preempted. VM thread changes sync state to synchronizing and |
|
1509 // suspends threads for GC. Thread A is resumed to finish this |
|
1510 // native method, but doesn't block here since it didn't see any |
|
1511 // synchronization in progress, and escapes. |
|
1512 |
|
1513 // We use release_store_fence to update values like the thread state, where |
|
1514 // we don't want the current thread to continue until all our prior memory |
|
1515 // accesses (including the new thread state) are visible to other threads. |
|
1516 __ li(R0/*thread_state*/, _thread_in_native_trans); |
|
1517 __ release(); |
|
1518 __ stw(R0/*thread_state*/, thread_(thread_state)); |
|
1519 if (UseMembar) { |
|
1520 __ fence(); |
|
1521 } |
|
1522 // Write serialization page so that the VM thread can do a pseudo remote |
|
1523 // membar. We use the current thread pointer to calculate a thread |
|
1524 // specific offset to write to within the page. This minimizes bus |
|
1525 // traffic due to cache line collision. |
|
1526 else { |
|
1527 __ serialize_memory(R16_thread, R11_scratch1, R12_scratch2); |
|
1528 } |
|
1529 |
|
1530 // Now before we return to java we must look for a current safepoint |
|
1531 // (a new safepoint can not start since we entered native_trans). |
|
1532 // We must check here because a current safepoint could be modifying |
|
1533 // the callers registers right this moment. |
|
1534 |
|
1535 // Acquire isn't strictly necessary here because of the fence, but |
|
1536 // sync_state is declared to be volatile, so we do it anyway |
|
1537 // (cmp-br-isync on one path, release (same as acquire on PPC64) on the other path). |
|
1538 int sync_state_offs = __ load_const_optimized(sync_state_addr, SafepointSynchronize::address_of_state(), /*temp*/R0, true); |
|
1539 |
|
1540 // TODO PPC port assert(4 == SafepointSynchronize::sz_state(), "unexpected field size"); |
|
1541 __ lwz(sync_state, sync_state_offs, sync_state_addr); |
|
1542 |
|
1543 // TODO PPC port assert(4 == Thread::sz_suspend_flags(), "unexpected field size"); |
|
1544 __ lwz(suspend_flags, thread_(suspend_flags)); |
|
1545 |
|
1546 Label sync_check_done; |
|
1547 Label do_safepoint; |
|
1548 // No synchronization in progress nor yet synchronized. |
|
1549 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); |
|
1550 // Not suspended. |
|
1551 __ cmpwi(CCR1, suspend_flags, 0); |
|
1552 |
|
1553 __ bne(CCR0, do_safepoint); |
|
1554 __ beq(CCR1, sync_check_done); |
|
1555 __ bind(do_safepoint); |
|
1556 __ isync(); |
|
1557 // Block. We do the call directly and leave the current |
|
1558 // last_Java_frame setup undisturbed. We must save any possible |
|
1559 // native result across the call. No oop is present. |
|
1560 |
|
1561 __ mr(R3_ARG1, R16_thread); |
|
1562 #if defined(ABI_ELFv2) |
|
1563 __ call_c(CAST_FROM_FN_PTR(address, JavaThread::check_special_condition_for_native_trans), |
|
1564 relocInfo::none); |
|
1565 #else |
|
1566 __ call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, JavaThread::check_special_condition_for_native_trans), |
|
1567 relocInfo::none); |
|
1568 #endif |
|
1569 |
|
1570 __ bind(sync_check_done); |
|
1571 |
|
1572 //============================================================================= |
|
1573 // <<<<<< Back in Interpreter Frame >>>>> |
|
1574 |
|
1575 // We are in thread_in_native_trans here and back in the normal |
|
1576 // interpreter frame. We don't have to do anything special about |
|
1577 // safepoints and we can switch to Java mode anytime we are ready. |
|
1578 |
|
1579 // Note: frame::interpreter_frame_result has a dependency on how the |
|
1580 // method result is saved across the call to post_method_exit. For |
|
1581 // native methods it assumes that the non-FPU/non-void result is |
|
1582 // saved in _native_lresult and a FPU result in _native_fresult. If |
|
1583 // this changes then the interpreter_frame_result implementation |
|
1584 // will need to be updated too. |
|
1585 |
|
1586 // On PPC64, we have stored the result directly after the native call. |
|
1587 |
|
1588 //============================================================================= |
|
1589 // Back in Java |
|
1590 |
|
1591 // We use release_store_fence to update values like the thread state, where |
|
1592 // we don't want the current thread to continue until all our prior memory |
|
1593 // accesses (including the new thread state) are visible to other threads. |
|
1594 __ li(R0/*thread_state*/, _thread_in_Java); |
|
1595 __ release(); |
|
1596 __ stw(R0/*thread_state*/, thread_(thread_state)); |
|
1597 if (UseMembar) { |
|
1598 __ fence(); |
|
1599 } |
|
1600 |
|
1601 if (CheckJNICalls) { |
|
1602 // clear_pending_jni_exception_check |
|
1603 __ load_const_optimized(R0, 0L); |
|
1604 __ st_ptr(R0, JavaThread::pending_jni_exception_check_fn_offset(), R16_thread); |
|
1605 } |
|
1606 |
|
1607 __ reset_last_Java_frame(); |
|
1608 |
|
1609 // Jvmdi/jvmpi support. Whether we've got an exception pending or |
|
1610 // not, and whether unlocking throws an exception or not, we notify |
|
1611 // on native method exit. If we do have an exception, we'll end up |
|
1612 // in the caller's context to handle it, so if we don't do the |
|
1613 // notify here, we'll drop it on the floor. |
|
1614 __ notify_method_exit(true/*native method*/, |
|
1615 ilgl /*illegal state (not used for native methods)*/, |
|
1616 InterpreterMacroAssembler::NotifyJVMTI, |
|
1617 false /*check_exceptions*/); |
|
1618 |
|
1619 //============================================================================= |
|
1620 // Handle exceptions |
|
1621 |
|
1622 if (synchronized) { |
|
1623 // Don't check for exceptions since we're still in the i2n frame. Do that |
|
1624 // manually afterwards. |
|
1625 __ unlock_object(R26_monitor, false); // Can also unlock methods. |
|
1626 } |
|
1627 |
|
1628 // Reset active handles after returning from native. |
|
1629 // thread->active_handles()->clear(); |
|
1630 __ ld(active_handles, thread_(active_handles)); |
|
1631 // TODO PPC port assert(4 == JNIHandleBlock::top_size_in_bytes(), "unexpected field size"); |
|
1632 __ li(R0, 0); |
|
1633 __ stw(R0, JNIHandleBlock::top_offset_in_bytes(), active_handles); |
|
1634 |
|
1635 Label exception_return_sync_check_already_unlocked; |
|
1636 __ ld(R0/*pending_exception*/, thread_(pending_exception)); |
|
1637 __ cmpdi(CCR0, R0/*pending_exception*/, 0); |
|
1638 __ bne(CCR0, exception_return_sync_check_already_unlocked); |
|
1639 |
|
1640 //----------------------------------------------------------------------------- |
|
1641 // No exception pending. |
|
1642 |
|
1643 // Move native method result back into proper registers and return. |
|
1644 // Invoke result handler (may unbox/promote). |
|
1645 __ ld(R11_scratch1, 0, R1_SP); |
|
1646 __ ld(R3_RET, _ijava_state_neg(lresult), R11_scratch1); |
|
1647 __ lfd(F1_RET, _ijava_state_neg(fresult), R11_scratch1); |
|
1648 __ call_stub(result_handler_addr); |
|
1649 |
|
1650 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); |
|
1651 |
|
1652 // Must use the return pc which was loaded from the caller's frame |
|
1653 // as the VM uses return-pc-patching for deoptimization. |
|
1654 __ mtlr(R0); |
|
1655 __ blr(); |
|
1656 |
|
1657 //----------------------------------------------------------------------------- |
|
1658 // An exception is pending. We call into the runtime only if the |
|
1659 // caller was not interpreted. If it was interpreted the |
|
1660 // interpreter will do the correct thing. If it isn't interpreted |
|
1661 // (call stub/compiled code) we will change our return and continue. |
|
1662 |
|
1663 BIND(exception_return_sync_check); |
|
1664 |
|
1665 if (synchronized) { |
|
1666 // Don't check for exceptions since we're still in the i2n frame. Do that |
|
1667 // manually afterwards. |
|
1668 __ unlock_object(R26_monitor, false); // Can also unlock methods. |
|
1669 } |
|
1670 BIND(exception_return_sync_check_already_unlocked); |
|
1671 |
|
1672 const Register return_pc = R31; |
|
1673 |
|
1674 __ ld(return_pc, 0, R1_SP); |
|
1675 __ ld(return_pc, _abi(lr), return_pc); |
|
1676 |
|
1677 // Get the address of the exception handler. |
|
1678 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), |
|
1679 R16_thread, |
|
1680 return_pc /* return pc */); |
|
1681 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, noreg, R11_scratch1, R12_scratch2); |
|
1682 |
|
1683 // Load the PC of the the exception handler into LR. |
|
1684 __ mtlr(R3_RET); |
|
1685 |
|
1686 // Load exception into R3_ARG1 and clear pending exception in thread. |
|
1687 __ ld(R3_ARG1/*exception*/, thread_(pending_exception)); |
|
1688 __ li(R4_ARG2, 0); |
|
1689 __ std(R4_ARG2, thread_(pending_exception)); |
|
1690 |
|
1691 // Load the original return pc into R4_ARG2. |
|
1692 __ mr(R4_ARG2/*issuing_pc*/, return_pc); |
|
1693 |
|
1694 // Return to exception handler. |
|
1695 __ blr(); |
|
1696 |
|
1697 //============================================================================= |
|
1698 // Counter overflow. |
|
1699 |
|
1700 if (inc_counter) { |
|
1701 // Handle invocation counter overflow. |
|
1702 __ bind(invocation_counter_overflow); |
|
1703 |
|
1704 generate_counter_overflow(continue_after_compile); |
|
1705 } |
|
1706 |
|
1707 return entry; |
|
1708 } |
|
1709 |
|
1710 // Generic interpreted method entry to (asm) interpreter. |
|
1711 // |
|
1712 address TemplateInterpreterGenerator::generate_normal_entry(bool synchronized) { |
|
1713 bool inc_counter = UseCompiler || CountCompiledCalls || LogTouchedMethods; |
|
1714 address entry = __ pc(); |
|
1715 // Generate the code to allocate the interpreter stack frame. |
|
1716 Register Rsize_of_parameters = R4_ARG2, // Written by generate_fixed_frame. |
|
1717 Rsize_of_locals = R5_ARG3; // Written by generate_fixed_frame. |
|
1718 |
|
1719 // Does also a stack check to assure this frame fits on the stack. |
|
1720 generate_fixed_frame(false, Rsize_of_parameters, Rsize_of_locals); |
|
1721 |
|
1722 // -------------------------------------------------------------------------- |
|
1723 // Zero out non-parameter locals. |
|
1724 // Note: *Always* zero out non-parameter locals as Sparc does. It's not |
|
1725 // worth to ask the flag, just do it. |
|
1726 Register Rslot_addr = R6_ARG4, |
|
1727 Rnum = R7_ARG5; |
|
1728 Label Lno_locals, Lzero_loop; |
|
1729 |
|
1730 // Set up the zeroing loop. |
|
1731 __ subf(Rnum, Rsize_of_parameters, Rsize_of_locals); |
|
1732 __ subf(Rslot_addr, Rsize_of_parameters, R18_locals); |
|
1733 __ srdi_(Rnum, Rnum, Interpreter::logStackElementSize); |
|
1734 __ beq(CCR0, Lno_locals); |
|
1735 __ li(R0, 0); |
|
1736 __ mtctr(Rnum); |
|
1737 |
|
1738 // The zero locals loop. |
|
1739 __ bind(Lzero_loop); |
|
1740 __ std(R0, 0, Rslot_addr); |
|
1741 __ addi(Rslot_addr, Rslot_addr, -Interpreter::stackElementSize); |
|
1742 __ bdnz(Lzero_loop); |
|
1743 |
|
1744 __ bind(Lno_locals); |
|
1745 |
|
1746 // -------------------------------------------------------------------------- |
|
1747 // Counter increment and overflow check. |
|
1748 Label invocation_counter_overflow, |
|
1749 profile_method, |
|
1750 profile_method_continue; |
|
1751 if (inc_counter || ProfileInterpreter) { |
|
1752 |
|
1753 Register Rdo_not_unlock_if_synchronized_addr = R11_scratch1; |
|
1754 if (synchronized) { |
|
1755 // Since at this point in the method invocation the exception handler |
|
1756 // would try to exit the monitor of synchronized methods which hasn't |
|
1757 // been entered yet, we set the thread local variable |
|
1758 // _do_not_unlock_if_synchronized to true. If any exception was thrown by |
|
1759 // runtime, exception handling i.e. unlock_if_synchronized_method will |
|
1760 // check this thread local flag. |
|
1761 // This flag has two effects, one is to force an unwind in the topmost |
|
1762 // interpreter frame and not perform an unlock while doing so. |
|
1763 __ li(R0, 1); |
|
1764 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); |
|
1765 } |
|
1766 |
|
1767 // Argument and return type profiling. |
|
1768 __ profile_parameters_type(R3_ARG1, R4_ARG2, R5_ARG3, R6_ARG4); |
|
1769 |
|
1770 // Increment invocation counter and check for overflow. |
|
1771 if (inc_counter) { |
|
1772 generate_counter_incr(&invocation_counter_overflow, &profile_method, &profile_method_continue); |
|
1773 } |
|
1774 |
|
1775 __ bind(profile_method_continue); |
|
1776 } |
|
1777 |
|
1778 bang_stack_shadow_pages(false); |
|
1779 |
|
1780 if (inc_counter || ProfileInterpreter) { |
|
1781 // Reset the _do_not_unlock_if_synchronized flag. |
|
1782 if (synchronized) { |
|
1783 __ li(R0, 0); |
|
1784 __ stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); |
|
1785 } |
|
1786 } |
|
1787 |
|
1788 // -------------------------------------------------------------------------- |
|
1789 // Locking of synchronized methods. Must happen AFTER invocation_counter |
|
1790 // check and stack overflow check, so method is not locked if overflows. |
|
1791 if (synchronized) { |
|
1792 lock_method(R3_ARG1, R4_ARG2, R5_ARG3); |
|
1793 } |
|
1794 #ifdef ASSERT |
|
1795 else { |
|
1796 Label Lok; |
|
1797 __ lwz(R0, in_bytes(Method::access_flags_offset()), R19_method); |
|
1798 __ andi_(R0, R0, JVM_ACC_SYNCHRONIZED); |
|
1799 __ asm_assert_eq("method needs synchronization", 0x8521); |
|
1800 __ bind(Lok); |
|
1801 } |
|
1802 #endif // ASSERT |
|
1803 |
|
1804 __ verify_thread(); |
|
1805 |
|
1806 // -------------------------------------------------------------------------- |
|
1807 // JVMTI support |
|
1808 __ notify_method_entry(); |
|
1809 |
|
1810 // -------------------------------------------------------------------------- |
|
1811 // Start executing instructions. |
|
1812 __ dispatch_next(vtos); |
|
1813 |
|
1814 // -------------------------------------------------------------------------- |
|
1815 // Out of line counter overflow and MDO creation code. |
|
1816 if (ProfileInterpreter) { |
|
1817 // We have decided to profile this method in the interpreter. |
|
1818 __ bind(profile_method); |
|
1819 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); |
|
1820 __ set_method_data_pointer_for_bcp(); |
|
1821 __ b(profile_method_continue); |
|
1822 } |
|
1823 |
|
1824 if (inc_counter) { |
|
1825 // Handle invocation counter overflow. |
|
1826 __ bind(invocation_counter_overflow); |
|
1827 generate_counter_overflow(profile_method_continue); |
|
1828 } |
|
1829 return entry; |
|
1830 } |
|
1831 |
|
1832 // CRC32 Intrinsics. |
|
1833 // |
|
1834 // Contract on scratch and work registers. |
|
1835 // ======================================= |
|
1836 // |
|
1837 // On ppc, the register set {R2..R12} is available in the interpreter as scratch/work registers. |
|
1838 // You should, however, keep in mind that {R3_ARG1..R10_ARG8} is the C-ABI argument register set. |
|
1839 // You can't rely on these registers across calls. |
|
1840 // |
|
1841 // The generators for CRC32_update and for CRC32_updateBytes use the |
|
1842 // scratch/work register set internally, passing the work registers |
|
1843 // as arguments to the MacroAssembler emitters as required. |
|
1844 // |
|
1845 // R3_ARG1..R6_ARG4 are preset to hold the incoming java arguments. |
|
1846 // Their contents is not constant but may change according to the requirements |
|
1847 // of the emitted code. |
|
1848 // |
|
1849 // All other registers from the scratch/work register set are used "internally" |
|
1850 // and contain garbage (i.e. unpredictable values) once blr() is reached. |
|
1851 // Basically, only R3_RET contains a defined value which is the function result. |
|
1852 // |
|
1853 /** |
|
1854 * Method entry for static native methods: |
|
1855 * int java.util.zip.CRC32.update(int crc, int b) |
|
1856 */ |
|
1857 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { |
|
1858 if (UseCRC32Intrinsics) { |
|
1859 address start = __ pc(); // Remember stub start address (is rtn value). |
|
1860 Label slow_path; |
|
1861 |
|
1862 // Safepoint check |
|
1863 const Register sync_state = R11_scratch1; |
|
1864 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true); |
|
1865 __ lwz(sync_state, sync_state_offs, sync_state); |
|
1866 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); |
|
1867 __ bne(CCR0, slow_path); |
|
1868 |
|
1869 // We don't generate local frame and don't align stack because |
|
1870 // we not even call stub code (we generate the code inline) |
|
1871 // and there is no safepoint on this path. |
|
1872 |
|
1873 // Load java parameters. |
|
1874 // R15_esp is callers operand stack pointer, i.e. it points to the parameters. |
|
1875 const Register argP = R15_esp; |
|
1876 const Register crc = R3_ARG1; // crc value |
|
1877 const Register data = R4_ARG2; // address of java byte value (kernel_crc32 needs address) |
|
1878 const Register dataLen = R5_ARG3; // source data len (1 byte). Not used because calling the single-byte emitter. |
|
1879 const Register table = R6_ARG4; // address of crc32 table |
|
1880 const Register tmp = dataLen; // Reuse unused len register to show we don't actually need a separate tmp here. |
|
1881 |
|
1882 BLOCK_COMMENT("CRC32_update {"); |
|
1883 |
|
1884 // Arguments are reversed on java expression stack |
|
1885 #ifdef VM_LITTLE_ENDIAN |
|
1886 __ addi(data, argP, 0+1*wordSize); // (stack) address of byte value. Emitter expects address, not value. |
|
1887 // Being passed as an int, the single byte is at offset +0. |
|
1888 #else |
|
1889 __ addi(data, argP, 3+1*wordSize); // (stack) address of byte value. Emitter expects address, not value. |
|
1890 // Being passed from java as an int, the single byte is at offset +3. |
|
1891 #endif |
|
1892 __ lwz(crc, 2*wordSize, argP); // Current crc state, zero extend to 64 bit to have a clean register. |
|
1893 |
|
1894 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table); |
|
1895 __ kernel_crc32_singleByte(crc, data, dataLen, table, tmp, true); |
|
1896 |
|
1897 // Restore caller sp for c2i case and return. |
|
1898 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. |
|
1899 __ blr(); |
|
1900 |
|
1901 // Generate a vanilla native entry as the slow path. |
|
1902 BLOCK_COMMENT("} CRC32_update"); |
|
1903 BIND(slow_path); |
|
1904 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); |
|
1905 return start; |
|
1906 } |
|
1907 |
|
1908 return NULL; |
|
1909 } |
|
1910 |
|
1911 |
|
1912 /** |
|
1913 * Method entry for static native methods: |
|
1914 * int java.util.zip.CRC32.updateBytes( int crc, byte[] b, int off, int len) |
|
1915 * int java.util.zip.CRC32.updateByteBuffer(int crc, long* buf, int off, int len) |
|
1916 */ |
|
1917 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { |
|
1918 if (UseCRC32Intrinsics) { |
|
1919 address start = __ pc(); // Remember stub start address (is rtn value). |
|
1920 Label slow_path; |
|
1921 |
|
1922 // Safepoint check |
|
1923 const Register sync_state = R11_scratch1; |
|
1924 int sync_state_offs = __ load_const_optimized(sync_state, SafepointSynchronize::address_of_state(), /*temp*/R0, true); |
|
1925 __ lwz(sync_state, sync_state_offs, sync_state); |
|
1926 __ cmpwi(CCR0, sync_state, SafepointSynchronize::_not_synchronized); |
|
1927 __ bne(CCR0, slow_path); |
|
1928 |
|
1929 // We don't generate local frame and don't align stack because |
|
1930 // we not even call stub code (we generate the code inline) |
|
1931 // and there is no safepoint on this path. |
|
1932 |
|
1933 // Load parameters. |
|
1934 // Z_esp is callers operand stack pointer, i.e. it points to the parameters. |
|
1935 const Register argP = R15_esp; |
|
1936 const Register crc = R3_ARG1; // crc value |
|
1937 const Register data = R4_ARG2; // address of java byte array |
|
1938 const Register dataLen = R5_ARG3; // source data len |
|
1939 const Register table = R6_ARG4; // address of crc32 table |
|
1940 |
|
1941 const Register t0 = R9; // scratch registers for crc calculation |
|
1942 const Register t1 = R10; |
|
1943 const Register t2 = R11; |
|
1944 const Register t3 = R12; |
|
1945 |
|
1946 const Register tc0 = R2; // registers to hold pre-calculated column addresses |
|
1947 const Register tc1 = R7; |
|
1948 const Register tc2 = R8; |
|
1949 const Register tc3 = table; // table address is reconstructed at the end of kernel_crc32_* emitters |
|
1950 |
|
1951 const Register tmp = t0; // Only used very locally to calculate byte buffer address. |
|
1952 |
|
1953 // Arguments are reversed on java expression stack. |
|
1954 // Calculate address of start element. |
|
1955 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { // Used for "updateByteBuffer direct". |
|
1956 BLOCK_COMMENT("CRC32_updateByteBuffer {"); |
|
1957 // crc @ (SP + 5W) (32bit) |
|
1958 // buf @ (SP + 3W) (64bit ptr to long array) |
|
1959 // off @ (SP + 2W) (32bit) |
|
1960 // dataLen @ (SP + 1W) (32bit) |
|
1961 // data = buf + off |
|
1962 __ ld( data, 3*wordSize, argP); // start of byte buffer |
|
1963 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset |
|
1964 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process |
|
1965 __ lwz( crc, 5*wordSize, argP); // current crc state |
|
1966 __ add( data, data, tmp); // Add byte buffer offset. |
|
1967 } else { // Used for "updateBytes update". |
|
1968 BLOCK_COMMENT("CRC32_updateBytes {"); |
|
1969 // crc @ (SP + 4W) (32bit) |
|
1970 // buf @ (SP + 3W) (64bit ptr to byte array) |
|
1971 // off @ (SP + 2W) (32bit) |
|
1972 // dataLen @ (SP + 1W) (32bit) |
|
1973 // data = buf + off + base_offset |
|
1974 __ ld( data, 3*wordSize, argP); // start of byte buffer |
|
1975 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset |
|
1976 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process |
|
1977 __ add( data, data, tmp); // add byte buffer offset |
|
1978 __ lwz( crc, 4*wordSize, argP); // current crc state |
|
1979 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE)); |
|
1980 } |
|
1981 |
|
1982 StubRoutines::ppc64::generate_load_crc_table_addr(_masm, table); |
|
1983 |
|
1984 // Performance measurements show the 1word and 2word variants to be almost equivalent, |
|
1985 // with very light advantages for the 1word variant. We chose the 1word variant for |
|
1986 // code compactness. |
|
1987 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3, true); |
|
1988 |
|
1989 // Restore caller sp for c2i case and return. |
|
1990 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. |
|
1991 __ blr(); |
|
1992 |
|
1993 // Generate a vanilla native entry as the slow path. |
|
1994 BLOCK_COMMENT("} CRC32_updateBytes(Buffer)"); |
|
1995 BIND(slow_path); |
|
1996 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native), R11_scratch1); |
|
1997 return start; |
|
1998 } |
|
1999 |
|
2000 return NULL; |
|
2001 } |
|
2002 |
|
2003 |
|
2004 /** |
|
2005 * Method entry for intrinsic-candidate (non-native) methods: |
|
2006 * int java.util.zip.CRC32C.updateBytes( int crc, byte[] b, int off, int end) |
|
2007 * int java.util.zip.CRC32C.updateDirectByteBuffer(int crc, long* buf, int off, int end) |
|
2008 * Unlike CRC32, CRC32C does not have any methods marked as native |
|
2009 * CRC32C also uses an "end" variable instead of the length variable CRC32 uses |
|
2010 **/ |
|
2011 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { |
|
2012 if (UseCRC32CIntrinsics) { |
|
2013 address start = __ pc(); // Remember stub start address (is rtn value). |
|
2014 |
|
2015 // We don't generate local frame and don't align stack because |
|
2016 // we not even call stub code (we generate the code inline) |
|
2017 // and there is no safepoint on this path. |
|
2018 |
|
2019 // Load parameters. |
|
2020 // Z_esp is callers operand stack pointer, i.e. it points to the parameters. |
|
2021 const Register argP = R15_esp; |
|
2022 const Register crc = R3_ARG1; // crc value |
|
2023 const Register data = R4_ARG2; // address of java byte array |
|
2024 const Register dataLen = R5_ARG3; // source data len |
|
2025 const Register table = R6_ARG4; // address of crc32c table |
|
2026 |
|
2027 const Register t0 = R9; // scratch registers for crc calculation |
|
2028 const Register t1 = R10; |
|
2029 const Register t2 = R11; |
|
2030 const Register t3 = R12; |
|
2031 |
|
2032 const Register tc0 = R2; // registers to hold pre-calculated column addresses |
|
2033 const Register tc1 = R7; |
|
2034 const Register tc2 = R8; |
|
2035 const Register tc3 = table; // table address is reconstructed at the end of kernel_crc32_* emitters |
|
2036 |
|
2037 const Register tmp = t0; // Only used very locally to calculate byte buffer address. |
|
2038 |
|
2039 // Arguments are reversed on java expression stack. |
|
2040 // Calculate address of start element. |
|
2041 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { // Used for "updateDirectByteBuffer". |
|
2042 BLOCK_COMMENT("CRC32C_updateDirectByteBuffer {"); |
|
2043 // crc @ (SP + 5W) (32bit) |
|
2044 // buf @ (SP + 3W) (64bit ptr to long array) |
|
2045 // off @ (SP + 2W) (32bit) |
|
2046 // dataLen @ (SP + 1W) (32bit) |
|
2047 // data = buf + off |
|
2048 __ ld( data, 3*wordSize, argP); // start of byte buffer |
|
2049 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset |
|
2050 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process |
|
2051 __ lwz( crc, 5*wordSize, argP); // current crc state |
|
2052 __ add( data, data, tmp); // Add byte buffer offset. |
|
2053 __ sub( dataLen, dataLen, tmp); // (end_index - offset) |
|
2054 } else { // Used for "updateBytes update". |
|
2055 BLOCK_COMMENT("CRC32C_updateBytes {"); |
|
2056 // crc @ (SP + 4W) (32bit) |
|
2057 // buf @ (SP + 3W) (64bit ptr to byte array) |
|
2058 // off @ (SP + 2W) (32bit) |
|
2059 // dataLen @ (SP + 1W) (32bit) |
|
2060 // data = buf + off + base_offset |
|
2061 __ ld( data, 3*wordSize, argP); // start of byte buffer |
|
2062 __ lwa( tmp, 2*wordSize, argP); // byte buffer offset |
|
2063 __ lwa( dataLen, 1*wordSize, argP); // #bytes to process |
|
2064 __ add( data, data, tmp); // add byte buffer offset |
|
2065 __ sub( dataLen, dataLen, tmp); // (end_index - offset) |
|
2066 __ lwz( crc, 4*wordSize, argP); // current crc state |
|
2067 __ addi(data, data, arrayOopDesc::base_offset_in_bytes(T_BYTE)); |
|
2068 } |
|
2069 |
|
2070 StubRoutines::ppc64::generate_load_crc32c_table_addr(_masm, table); |
|
2071 |
|
2072 // Performance measurements show the 1word and 2word variants to be almost equivalent, |
|
2073 // with very light advantages for the 1word variant. We chose the 1word variant for |
|
2074 // code compactness. |
|
2075 __ kernel_crc32_1word(crc, data, dataLen, table, t0, t1, t2, t3, tc0, tc1, tc2, tc3, false); |
|
2076 |
|
2077 // Restore caller sp for c2i case and return. |
|
2078 __ mr(R1_SP, R21_sender_SP); // Cut the stack back to where the caller started. |
|
2079 __ blr(); |
|
2080 |
|
2081 BLOCK_COMMENT("} CRC32C_update{Bytes|DirectByteBuffer}"); |
|
2082 return start; |
|
2083 } |
|
2084 |
|
2085 return NULL; |
|
2086 } |
|
2087 |
|
2088 // ============================================================================= |
|
2089 // Exceptions |
|
2090 |
|
2091 void TemplateInterpreterGenerator::generate_throw_exception() { |
|
2092 Register Rexception = R17_tos, |
|
2093 Rcontinuation = R3_RET; |
|
2094 |
|
2095 // -------------------------------------------------------------------------- |
|
2096 // Entry point if an method returns with a pending exception (rethrow). |
|
2097 Interpreter::_rethrow_exception_entry = __ pc(); |
|
2098 { |
|
2099 __ restore_interpreter_state(R11_scratch1); // Sets R11_scratch1 = fp. |
|
2100 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); |
|
2101 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); |
|
2102 |
|
2103 // Compiled code destroys templateTableBase, reload. |
|
2104 __ load_const_optimized(R25_templateTableBase, (address)Interpreter::dispatch_table((TosState)0), R11_scratch1); |
|
2105 } |
|
2106 |
|
2107 // Entry point if a interpreted method throws an exception (throw). |
|
2108 Interpreter::_throw_exception_entry = __ pc(); |
|
2109 { |
|
2110 __ mr(Rexception, R3_RET); |
|
2111 |
|
2112 __ verify_thread(); |
|
2113 __ verify_oop(Rexception); |
|
2114 |
|
2115 // Expression stack must be empty before entering the VM in case of an exception. |
|
2116 __ empty_expression_stack(); |
|
2117 // Find exception handler address and preserve exception oop. |
|
2118 // Call C routine to find handler and jump to it. |
|
2119 __ call_VM(Rexception, CAST_FROM_FN_PTR(address, InterpreterRuntime::exception_handler_for_exception), Rexception); |
|
2120 __ mtctr(Rcontinuation); |
|
2121 // Push exception for exception handler bytecodes. |
|
2122 __ push_ptr(Rexception); |
|
2123 |
|
2124 // Jump to exception handler (may be remove activation entry!). |
|
2125 __ bctr(); |
|
2126 } |
|
2127 |
|
2128 // If the exception is not handled in the current frame the frame is |
|
2129 // removed and the exception is rethrown (i.e. exception |
|
2130 // continuation is _rethrow_exception). |
|
2131 // |
|
2132 // Note: At this point the bci is still the bxi for the instruction |
|
2133 // which caused the exception and the expression stack is |
|
2134 // empty. Thus, for any VM calls at this point, GC will find a legal |
|
2135 // oop map (with empty expression stack). |
|
2136 |
|
2137 // In current activation |
|
2138 // tos: exception |
|
2139 // bcp: exception bcp |
|
2140 |
|
2141 // -------------------------------------------------------------------------- |
|
2142 // JVMTI PopFrame support |
|
2143 |
|
2144 Interpreter::_remove_activation_preserving_args_entry = __ pc(); |
|
2145 { |
|
2146 // Set the popframe_processing bit in popframe_condition indicating that we are |
|
2147 // currently handling popframe, so that call_VMs that may happen later do not |
|
2148 // trigger new popframe handling cycles. |
|
2149 __ lwz(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); |
|
2150 __ ori(R11_scratch1, R11_scratch1, JavaThread::popframe_processing_bit); |
|
2151 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); |
|
2152 |
|
2153 // Empty the expression stack, as in normal exception handling. |
|
2154 __ empty_expression_stack(); |
|
2155 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, /* install_monitor_exception */ false); |
|
2156 |
|
2157 // Check to see whether we are returning to a deoptimized frame. |
|
2158 // (The PopFrame call ensures that the caller of the popped frame is |
|
2159 // either interpreted or compiled and deoptimizes it if compiled.) |
|
2160 // Note that we don't compare the return PC against the |
|
2161 // deoptimization blob's unpack entry because of the presence of |
|
2162 // adapter frames in C2. |
|
2163 Label Lcaller_not_deoptimized; |
|
2164 Register return_pc = R3_ARG1; |
|
2165 __ ld(return_pc, 0, R1_SP); |
|
2166 __ ld(return_pc, _abi(lr), return_pc); |
|
2167 __ call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::interpreter_contains), return_pc); |
|
2168 __ cmpdi(CCR0, R3_RET, 0); |
|
2169 __ bne(CCR0, Lcaller_not_deoptimized); |
|
2170 |
|
2171 // The deoptimized case. |
|
2172 // In this case, we can't call dispatch_next() after the frame is |
|
2173 // popped, but instead must save the incoming arguments and restore |
|
2174 // them after deoptimization has occurred. |
|
2175 __ ld(R4_ARG2, in_bytes(Method::const_offset()), R19_method); |
|
2176 __ lhz(R4_ARG2 /* number of params */, in_bytes(ConstMethod::size_of_parameters_offset()), R4_ARG2); |
|
2177 __ slwi(R4_ARG2, R4_ARG2, Interpreter::logStackElementSize); |
|
2178 __ addi(R5_ARG3, R18_locals, Interpreter::stackElementSize); |
|
2179 __ subf(R5_ARG3, R4_ARG2, R5_ARG3); |
|
2180 // Save these arguments. |
|
2181 __ call_VM_leaf(CAST_FROM_FN_PTR(address, Deoptimization::popframe_preserve_args), R16_thread, R4_ARG2, R5_ARG3); |
|
2182 |
|
2183 // Inform deoptimization that it is responsible for restoring these arguments. |
|
2184 __ load_const_optimized(R11_scratch1, JavaThread::popframe_force_deopt_reexecution_bit); |
|
2185 __ stw(R11_scratch1, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); |
|
2186 |
|
2187 // Return from the current method into the deoptimization blob. Will eventually |
|
2188 // end up in the deopt interpeter entry, deoptimization prepared everything that |
|
2189 // we will reexecute the call that called us. |
|
2190 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*reload return_pc*/ return_pc, R11_scratch1, R12_scratch2); |
|
2191 __ mtlr(return_pc); |
|
2192 __ blr(); |
|
2193 |
|
2194 // The non-deoptimized case. |
|
2195 __ bind(Lcaller_not_deoptimized); |
|
2196 |
|
2197 // Clear the popframe condition flag. |
|
2198 __ li(R0, 0); |
|
2199 __ stw(R0, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); |
|
2200 |
|
2201 // Get out of the current method and re-execute the call that called us. |
|
2202 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); |
|
2203 __ restore_interpreter_state(R11_scratch1); |
|
2204 __ ld(R12_scratch2, _ijava_state_neg(top_frame_sp), R11_scratch1); |
|
2205 __ resize_frame_absolute(R12_scratch2, R11_scratch1, R0); |
|
2206 if (ProfileInterpreter) { |
|
2207 __ set_method_data_pointer_for_bcp(); |
|
2208 __ ld(R11_scratch1, 0, R1_SP); |
|
2209 __ std(R28_mdx, _ijava_state_neg(mdx), R11_scratch1); |
|
2210 } |
|
2211 #if INCLUDE_JVMTI |
|
2212 Label L_done; |
|
2213 |
|
2214 __ lbz(R11_scratch1, 0, R14_bcp); |
|
2215 __ cmpwi(CCR0, R11_scratch1, Bytecodes::_invokestatic); |
|
2216 __ bne(CCR0, L_done); |
|
2217 |
|
2218 // The member name argument must be restored if _invokestatic is re-executed after a PopFrame call. |
|
2219 // Detect such a case in the InterpreterRuntime function and return the member name argument, or NULL. |
|
2220 __ ld(R4_ARG2, 0, R18_locals); |
|
2221 __ MacroAssembler::call_VM(R4_ARG2, CAST_FROM_FN_PTR(address, InterpreterRuntime::member_name_arg_or_null), R4_ARG2, R19_method, R14_bcp, false); |
|
2222 __ restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); |
|
2223 __ cmpdi(CCR0, R4_ARG2, 0); |
|
2224 __ beq(CCR0, L_done); |
|
2225 __ std(R4_ARG2, wordSize, R15_esp); |
|
2226 __ bind(L_done); |
|
2227 #endif // INCLUDE_JVMTI |
|
2228 __ dispatch_next(vtos); |
|
2229 } |
|
2230 // end of JVMTI PopFrame support |
|
2231 |
|
2232 // -------------------------------------------------------------------------- |
|
2233 // Remove activation exception entry. |
|
2234 // This is jumped to if an interpreted method can't handle an exception itself |
|
2235 // (we come from the throw/rethrow exception entry above). We're going to call |
|
2236 // into the VM to find the exception handler in the caller, pop the current |
|
2237 // frame and return the handler we calculated. |
|
2238 Interpreter::_remove_activation_entry = __ pc(); |
|
2239 { |
|
2240 __ pop_ptr(Rexception); |
|
2241 __ verify_thread(); |
|
2242 __ verify_oop(Rexception); |
|
2243 __ std(Rexception, in_bytes(JavaThread::vm_result_offset()), R16_thread); |
|
2244 |
|
2245 __ unlock_if_synchronized_method(vtos, /* throw_monitor_exception */ false, true); |
|
2246 __ notify_method_exit(false, vtos, InterpreterMacroAssembler::SkipNotifyJVMTI, false); |
|
2247 |
|
2248 __ get_vm_result(Rexception); |
|
2249 |
|
2250 // We are done with this activation frame; find out where to go next. |
|
2251 // The continuation point will be an exception handler, which expects |
|
2252 // the following registers set up: |
|
2253 // |
|
2254 // RET: exception oop |
|
2255 // ARG2: Issuing PC (see generate_exception_blob()), only used if the caller is compiled. |
|
2256 |
|
2257 Register return_pc = R31; // Needs to survive the runtime call. |
|
2258 __ ld(return_pc, 0, R1_SP); |
|
2259 __ ld(return_pc, _abi(lr), return_pc); |
|
2260 __ call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::exception_handler_for_return_address), R16_thread, return_pc); |
|
2261 |
|
2262 // Remove the current activation. |
|
2263 __ merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ noreg, R11_scratch1, R12_scratch2); |
|
2264 |
|
2265 __ mr(R4_ARG2, return_pc); |
|
2266 __ mtlr(R3_RET); |
|
2267 __ mr(R3_RET, Rexception); |
|
2268 __ blr(); |
|
2269 } |
|
2270 } |
|
2271 |
|
2272 // JVMTI ForceEarlyReturn support. |
|
2273 // Returns "in the middle" of a method with a "fake" return value. |
|
2274 address TemplateInterpreterGenerator::generate_earlyret_entry_for(TosState state) { |
|
2275 |
|
2276 Register Rscratch1 = R11_scratch1, |
|
2277 Rscratch2 = R12_scratch2; |
|
2278 |
|
2279 address entry = __ pc(); |
|
2280 __ empty_expression_stack(); |
|
2281 |
|
2282 __ load_earlyret_value(state, Rscratch1); |
|
2283 |
|
2284 __ ld(Rscratch1, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); |
|
2285 // Clear the earlyret state. |
|
2286 __ li(R0, 0); |
|
2287 __ stw(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rscratch1); |
|
2288 |
|
2289 __ remove_activation(state, false, false); |
|
2290 // Copied from TemplateTable::_return. |
|
2291 // Restoration of lr done by remove_activation. |
|
2292 switch (state) { |
|
2293 // Narrow result if state is itos but result type is smaller. |
|
2294 case btos: |
|
2295 case ztos: |
|
2296 case ctos: |
|
2297 case stos: |
|
2298 case itos: __ narrow(R17_tos); /* fall through */ |
|
2299 case ltos: |
|
2300 case atos: __ mr(R3_RET, R17_tos); break; |
|
2301 case ftos: |
|
2302 case dtos: __ fmr(F1_RET, F15_ftos); break; |
|
2303 case vtos: // This might be a constructor. Final fields (and volatile fields on PPC64) need |
|
2304 // to get visible before the reference to the object gets stored anywhere. |
|
2305 __ membar(Assembler::StoreStore); break; |
|
2306 default : ShouldNotReachHere(); |
|
2307 } |
|
2308 __ blr(); |
|
2309 |
|
2310 return entry; |
|
2311 } // end of ForceEarlyReturn support |
|
2312 |
|
2313 //----------------------------------------------------------------------------- |
|
2314 // Helper for vtos entry point generation |
|
2315 |
|
2316 void TemplateInterpreterGenerator::set_vtos_entry_points(Template* t, |
|
2317 address& bep, |
|
2318 address& cep, |
|
2319 address& sep, |
|
2320 address& aep, |
|
2321 address& iep, |
|
2322 address& lep, |
|
2323 address& fep, |
|
2324 address& dep, |
|
2325 address& vep) { |
|
2326 assert(t->is_valid() && t->tos_in() == vtos, "illegal template"); |
|
2327 Label L; |
|
2328 |
|
2329 aep = __ pc(); __ push_ptr(); __ b(L); |
|
2330 fep = __ pc(); __ push_f(); __ b(L); |
|
2331 dep = __ pc(); __ push_d(); __ b(L); |
|
2332 lep = __ pc(); __ push_l(); __ b(L); |
|
2333 __ align(32, 12, 24); // align L |
|
2334 bep = cep = sep = |
|
2335 iep = __ pc(); __ push_i(); |
|
2336 vep = __ pc(); |
|
2337 __ bind(L); |
|
2338 generate_and_dispatch(t); |
|
2339 } |
|
2340 |
|
2341 //----------------------------------------------------------------------------- |
|
2342 |
|
2343 // Non-product code |
|
2344 #ifndef PRODUCT |
|
2345 address TemplateInterpreterGenerator::generate_trace_code(TosState state) { |
|
2346 //__ flush_bundle(); |
|
2347 address entry = __ pc(); |
|
2348 |
|
2349 const char *bname = NULL; |
|
2350 uint tsize = 0; |
|
2351 switch(state) { |
|
2352 case ftos: |
|
2353 bname = "trace_code_ftos {"; |
|
2354 tsize = 2; |
|
2355 break; |
|
2356 case btos: |
|
2357 bname = "trace_code_btos {"; |
|
2358 tsize = 2; |
|
2359 break; |
|
2360 case ztos: |
|
2361 bname = "trace_code_ztos {"; |
|
2362 tsize = 2; |
|
2363 break; |
|
2364 case ctos: |
|
2365 bname = "trace_code_ctos {"; |
|
2366 tsize = 2; |
|
2367 break; |
|
2368 case stos: |
|
2369 bname = "trace_code_stos {"; |
|
2370 tsize = 2; |
|
2371 break; |
|
2372 case itos: |
|
2373 bname = "trace_code_itos {"; |
|
2374 tsize = 2; |
|
2375 break; |
|
2376 case ltos: |
|
2377 bname = "trace_code_ltos {"; |
|
2378 tsize = 3; |
|
2379 break; |
|
2380 case atos: |
|
2381 bname = "trace_code_atos {"; |
|
2382 tsize = 2; |
|
2383 break; |
|
2384 case vtos: |
|
2385 // Note: In case of vtos, the topmost of stack value could be a int or doubl |
|
2386 // In case of a double (2 slots) we won't see the 2nd stack value. |
|
2387 // Maybe we simply should print the topmost 3 stack slots to cope with the problem. |
|
2388 bname = "trace_code_vtos {"; |
|
2389 tsize = 2; |
|
2390 |
|
2391 break; |
|
2392 case dtos: |
|
2393 bname = "trace_code_dtos {"; |
|
2394 tsize = 3; |
|
2395 break; |
|
2396 default: |
|
2397 ShouldNotReachHere(); |
|
2398 } |
|
2399 BLOCK_COMMENT(bname); |
|
2400 |
|
2401 // Support short-cut for TraceBytecodesAt. |
|
2402 // Don't call into the VM if we don't want to trace to speed up things. |
|
2403 Label Lskip_vm_call; |
|
2404 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { |
|
2405 int offs1 = __ load_const_optimized(R11_scratch1, (address) &TraceBytecodesAt, R0, true); |
|
2406 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); |
|
2407 __ ld(R11_scratch1, offs1, R11_scratch1); |
|
2408 __ lwa(R12_scratch2, offs2, R12_scratch2); |
|
2409 __ cmpd(CCR0, R12_scratch2, R11_scratch1); |
|
2410 __ blt(CCR0, Lskip_vm_call); |
|
2411 } |
|
2412 |
|
2413 __ push(state); |
|
2414 // Load 2 topmost expression stack values. |
|
2415 __ ld(R6_ARG4, tsize*Interpreter::stackElementSize, R15_esp); |
|
2416 __ ld(R5_ARG3, Interpreter::stackElementSize, R15_esp); |
|
2417 __ mflr(R31); |
|
2418 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::trace_bytecode), /* unused */ R4_ARG2, R5_ARG3, R6_ARG4, false); |
|
2419 __ mtlr(R31); |
|
2420 __ pop(state); |
|
2421 |
|
2422 if (TraceBytecodesAt > 0 && TraceBytecodesAt < max_intx) { |
|
2423 __ bind(Lskip_vm_call); |
|
2424 } |
|
2425 __ blr(); |
|
2426 BLOCK_COMMENT("} trace_code"); |
|
2427 return entry; |
|
2428 } |
|
2429 |
|
2430 void TemplateInterpreterGenerator::count_bytecode() { |
|
2431 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeCounter::_counter_value, R12_scratch2, true); |
|
2432 __ lwz(R12_scratch2, offs, R11_scratch1); |
|
2433 __ addi(R12_scratch2, R12_scratch2, 1); |
|
2434 __ stw(R12_scratch2, offs, R11_scratch1); |
|
2435 } |
|
2436 |
|
2437 void TemplateInterpreterGenerator::histogram_bytecode(Template* t) { |
|
2438 int offs = __ load_const_optimized(R11_scratch1, (address) &BytecodeHistogram::_counters[t->bytecode()], R12_scratch2, true); |
|
2439 __ lwz(R12_scratch2, offs, R11_scratch1); |
|
2440 __ addi(R12_scratch2, R12_scratch2, 1); |
|
2441 __ stw(R12_scratch2, offs, R11_scratch1); |
|
2442 } |
|
2443 |
|
2444 void TemplateInterpreterGenerator::histogram_bytecode_pair(Template* t) { |
|
2445 const Register addr = R11_scratch1, |
|
2446 tmp = R12_scratch2; |
|
2447 // Get index, shift out old bytecode, bring in new bytecode, and store it. |
|
2448 // _index = (_index >> log2_number_of_codes) | |
|
2449 // (bytecode << log2_number_of_codes); |
|
2450 int offs1 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_index, tmp, true); |
|
2451 __ lwz(tmp, offs1, addr); |
|
2452 __ srwi(tmp, tmp, BytecodePairHistogram::log2_number_of_codes); |
|
2453 __ ori(tmp, tmp, ((int) t->bytecode()) << BytecodePairHistogram::log2_number_of_codes); |
|
2454 __ stw(tmp, offs1, addr); |
|
2455 |
|
2456 // Bump bucket contents. |
|
2457 // _counters[_index] ++; |
|
2458 int offs2 = __ load_const_optimized(addr, (address)&BytecodePairHistogram::_counters, R0, true); |
|
2459 __ sldi(tmp, tmp, LogBytesPerInt); |
|
2460 __ add(addr, tmp, addr); |
|
2461 __ lwz(tmp, offs2, addr); |
|
2462 __ addi(tmp, tmp, 1); |
|
2463 __ stw(tmp, offs2, addr); |
|
2464 } |
|
2465 |
|
2466 void TemplateInterpreterGenerator::trace_bytecode(Template* t) { |
|
2467 // Call a little run-time stub to avoid blow-up for each bytecode. |
|
2468 // The run-time runtime saves the right registers, depending on |
|
2469 // the tosca in-state for the given template. |
|
2470 |
|
2471 assert(Interpreter::trace_code(t->tos_in()) != NULL, |
|
2472 "entry must have been generated"); |
|
2473 |
|
2474 // Note: we destroy LR here. |
|
2475 __ bl(Interpreter::trace_code(t->tos_in())); |
|
2476 } |
|
2477 |
|
2478 void TemplateInterpreterGenerator::stop_interpreter_at() { |
|
2479 Label L; |
|
2480 int offs1 = __ load_const_optimized(R11_scratch1, (address) &StopInterpreterAt, R0, true); |
|
2481 int offs2 = __ load_const_optimized(R12_scratch2, (address) &BytecodeCounter::_counter_value, R0, true); |
|
2482 __ ld(R11_scratch1, offs1, R11_scratch1); |
|
2483 __ lwa(R12_scratch2, offs2, R12_scratch2); |
|
2484 __ cmpd(CCR0, R12_scratch2, R11_scratch1); |
|
2485 __ bne(CCR0, L); |
|
2486 __ illtrap(); |
|
2487 __ bind(L); |
|
2488 } |
|
2489 |
|
2490 #endif // !PRODUCT |