Tue, 24 Jul 2012 10:51:00 -0700
7023639: JSR 292 method handle invocation needs a fast path for compiled code
6984705: JSR 292 method handle creation should not go through JNI
Summary: remove assembly code for JDK 7 chained method handles
Reviewed-by: jrose, twisti, kvn, mhaupt
Contributed-by: John Rose <john.r.rose@oracle.com>, Christian Thalinger <christian.thalinger@oracle.com>, Michael Haupt <michael.haupt@oracle.com>
1 /*
2 * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "interpreter/bytecodeHistogram.hpp"
28 #include "interpreter/interpreter.hpp"
29 #include "interpreter/interpreterGenerator.hpp"
30 #include "interpreter/interpreterRuntime.hpp"
31 #include "interpreter/templateTable.hpp"
32 #include "oops/arrayOop.hpp"
33 #include "oops/methodDataOop.hpp"
34 #include "oops/methodOop.hpp"
35 #include "oops/oop.inline.hpp"
36 #include "prims/jvmtiExport.hpp"
37 #include "prims/jvmtiThreadState.hpp"
38 #include "prims/methodHandles.hpp"
39 #include "runtime/arguments.hpp"
40 #include "runtime/deoptimization.hpp"
41 #include "runtime/frame.inline.hpp"
42 #include "runtime/sharedRuntime.hpp"
43 #include "runtime/stubRoutines.hpp"
44 #include "runtime/synchronizer.hpp"
45 #include "runtime/timer.hpp"
46 #include "runtime/vframeArray.hpp"
47 #include "utilities/debug.hpp"
48 #ifdef COMPILER1
49 #include "c1/c1_Runtime1.hpp"
50 #endif
54 // Generation of Interpreter
55 //
56 // The InterpreterGenerator generates the interpreter into Interpreter::_code.
59 #define __ _masm->
62 //----------------------------------------------------------------------------------------------------
67 int AbstractInterpreter::BasicType_as_index(BasicType type) {
68 int i = 0;
69 switch (type) {
70 case T_BOOLEAN: i = 0; break;
71 case T_CHAR : i = 1; break;
72 case T_BYTE : i = 2; break;
73 case T_SHORT : i = 3; break;
74 case T_INT : i = 4; break;
75 case T_LONG : i = 5; break;
76 case T_VOID : i = 6; break;
77 case T_FLOAT : i = 7; break;
78 case T_DOUBLE : i = 8; break;
79 case T_OBJECT : i = 9; break;
80 case T_ARRAY : i = 9; break;
81 default : ShouldNotReachHere();
82 }
83 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
84 return i;
85 }
88 #ifndef _LP64
89 address AbstractInterpreterGenerator::generate_slow_signature_handler() {
90 address entry = __ pc();
91 Argument argv(0, true);
93 // We are in the jni transition frame. Save the last_java_frame corresponding to the
94 // outer interpreter frame
95 //
96 __ set_last_Java_frame(FP, noreg);
97 // make sure the interpreter frame we've pushed has a valid return pc
98 __ mov(O7, I7);
99 __ mov(Lmethod, G3_scratch);
100 __ mov(Llocals, G4_scratch);
101 __ save_frame(0);
102 __ mov(G2_thread, L7_thread_cache);
103 __ add(argv.address_in_frame(), O3);
104 __ mov(G2_thread, O0);
105 __ mov(G3_scratch, O1);
106 __ call(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), relocInfo::runtime_call_type);
107 __ delayed()->mov(G4_scratch, O2);
108 __ mov(L7_thread_cache, G2_thread);
109 __ reset_last_Java_frame();
111 // load the register arguments (the C code packed them as varargs)
112 for (Argument ldarg = argv.successor(); ldarg.is_register(); ldarg = ldarg.successor()) {
113 __ ld_ptr(ldarg.address_in_frame(), ldarg.as_register());
114 }
115 __ ret();
116 __ delayed()->
117 restore(O0, 0, Lscratch); // caller's Lscratch gets the result handler
118 return entry;
119 }
122 #else
123 // LP64 passes floating point arguments in F1, F3, F5, etc. instead of
124 // O0, O1, O2 etc..
125 // Doubles are passed in D0, D2, D4
126 // We store the signature of the first 16 arguments in the first argument
127 // slot because it will be overwritten prior to calling the native
128 // function, with the pointer to the JNIEnv.
129 // If LP64 there can be up to 16 floating point arguments in registers
130 // or 6 integer registers.
131 address AbstractInterpreterGenerator::generate_slow_signature_handler() {
133 enum {
134 non_float = 0,
135 float_sig = 1,
136 double_sig = 2,
137 sig_mask = 3
138 };
140 address entry = __ pc();
141 Argument argv(0, true);
143 // We are in the jni transition frame. Save the last_java_frame corresponding to the
144 // outer interpreter frame
145 //
146 __ set_last_Java_frame(FP, noreg);
147 // make sure the interpreter frame we've pushed has a valid return pc
148 __ mov(O7, I7);
149 __ mov(Lmethod, G3_scratch);
150 __ mov(Llocals, G4_scratch);
151 __ save_frame(0);
152 __ mov(G2_thread, L7_thread_cache);
153 __ add(argv.address_in_frame(), O3);
154 __ mov(G2_thread, O0);
155 __ mov(G3_scratch, O1);
156 __ call(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), relocInfo::runtime_call_type);
157 __ delayed()->mov(G4_scratch, O2);
158 __ mov(L7_thread_cache, G2_thread);
159 __ reset_last_Java_frame();
162 // load the register arguments (the C code packed them as varargs)
163 Address Sig = argv.address_in_frame(); // Argument 0 holds the signature
164 __ ld_ptr( Sig, G3_scratch ); // Get register argument signature word into G3_scratch
165 __ mov( G3_scratch, G4_scratch);
166 __ srl( G4_scratch, 2, G4_scratch); // Skip Arg 0
167 Label done;
168 for (Argument ldarg = argv.successor(); ldarg.is_float_register(); ldarg = ldarg.successor()) {
169 Label NonFloatArg;
170 Label LoadFloatArg;
171 Label LoadDoubleArg;
172 Label NextArg;
173 Address a = ldarg.address_in_frame();
174 __ andcc(G4_scratch, sig_mask, G3_scratch);
175 __ br(Assembler::zero, false, Assembler::pt, NonFloatArg);
176 __ delayed()->nop();
178 __ cmp(G3_scratch, float_sig );
179 __ br(Assembler::equal, false, Assembler::pt, LoadFloatArg);
180 __ delayed()->nop();
182 __ cmp(G3_scratch, double_sig );
183 __ br(Assembler::equal, false, Assembler::pt, LoadDoubleArg);
184 __ delayed()->nop();
186 __ bind(NonFloatArg);
187 // There are only 6 integer register arguments!
188 if ( ldarg.is_register() )
189 __ ld_ptr(ldarg.address_in_frame(), ldarg.as_register());
190 else {
191 // Optimization, see if there are any more args and get out prior to checking
192 // all 16 float registers. My guess is that this is rare.
193 // If is_register is false, then we are done the first six integer args.
194 __ br_null_short(G4_scratch, Assembler::pt, done);
195 }
196 __ ba(NextArg);
197 __ delayed()->srl( G4_scratch, 2, G4_scratch );
199 __ bind(LoadFloatArg);
200 __ ldf( FloatRegisterImpl::S, a, ldarg.as_float_register(), 4);
201 __ ba(NextArg);
202 __ delayed()->srl( G4_scratch, 2, G4_scratch );
204 __ bind(LoadDoubleArg);
205 __ ldf( FloatRegisterImpl::D, a, ldarg.as_double_register() );
206 __ ba(NextArg);
207 __ delayed()->srl( G4_scratch, 2, G4_scratch );
209 __ bind(NextArg);
211 }
213 __ bind(done);
214 __ ret();
215 __ delayed()->
216 restore(O0, 0, Lscratch); // caller's Lscratch gets the result handler
217 return entry;
218 }
219 #endif
221 void InterpreterGenerator::generate_counter_overflow(Label& Lcontinue) {
223 // Generate code to initiate compilation on the counter overflow.
225 // InterpreterRuntime::frequency_counter_overflow takes two arguments,
226 // the first indicates if the counter overflow occurs at a backwards branch (NULL bcp)
227 // and the second is only used when the first is true. We pass zero for both.
228 // The call returns the address of the verified entry point for the method or NULL
229 // if the compilation did not complete (either went background or bailed out).
230 __ set((int)false, O2);
231 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), O2, O2, true);
232 // returns verified_entry_point or NULL
233 // we ignore it in any case
234 __ ba_short(Lcontinue);
236 }
239 // End of helpers
241 // Various method entries
243 // Abstract method entry
244 // Attempt to execute abstract method. Throw exception
245 //
246 address InterpreterGenerator::generate_abstract_entry(void) {
247 address entry = __ pc();
248 // abstract method entry
249 // throw exception
250 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
251 // the call_VM checks for exception, so we should never return here.
252 __ should_not_reach_here();
253 return entry;
255 }
258 //----------------------------------------------------------------------------------------------------
259 // Entry points & stack frame layout
260 //
261 // Here we generate the various kind of entries into the interpreter.
262 // The two main entry type are generic bytecode methods and native call method.
263 // These both come in synchronized and non-synchronized versions but the
264 // frame layout they create is very similar. The other method entry
265 // types are really just special purpose entries that are really entry
266 // and interpretation all in one. These are for trivial methods like
267 // accessor, empty, or special math methods.
268 //
269 // When control flow reaches any of the entry types for the interpreter
270 // the following holds ->
271 //
272 // C2 Calling Conventions:
273 //
274 // The entry code below assumes that the following registers are set
275 // when coming in:
276 // G5_method: holds the methodOop of the method to call
277 // Lesp: points to the TOS of the callers expression stack
278 // after having pushed all the parameters
279 //
280 // The entry code does the following to setup an interpreter frame
281 // pop parameters from the callers stack by adjusting Lesp
282 // set O0 to Lesp
283 // compute X = (max_locals - num_parameters)
284 // bump SP up by X to accomadate the extra locals
285 // compute X = max_expression_stack
286 // + vm_local_words
287 // + 16 words of register save area
288 // save frame doing a save sp, -X, sp growing towards lower addresses
289 // set Lbcp, Lmethod, LcpoolCache
290 // set Llocals to i0
291 // set Lmonitors to FP - rounded_vm_local_words
292 // set Lesp to Lmonitors - 4
293 //
294 // The frame has now been setup to do the rest of the entry code
296 // Try this optimization: Most method entries could live in a
297 // "one size fits all" stack frame without all the dynamic size
298 // calculations. It might be profitable to do all this calculation
299 // statically and approximately for "small enough" methods.
301 //-----------------------------------------------------------------------------------------------
303 // C1 Calling conventions
304 //
305 // Upon method entry, the following registers are setup:
306 //
307 // g2 G2_thread: current thread
308 // g5 G5_method: method to activate
309 // g4 Gargs : pointer to last argument
310 //
311 //
312 // Stack:
313 //
314 // +---------------+ <--- sp
315 // | |
316 // : reg save area :
317 // | |
318 // +---------------+ <--- sp + 0x40
319 // | |
320 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later)
321 // | |
322 // +---------------+ <--- sp + 0x5c
323 // | |
324 // : free :
325 // | |
326 // +---------------+ <--- Gargs
327 // | |
328 // : arguments :
329 // | |
330 // +---------------+
331 // | |
332 //
333 //
334 //
335 // AFTER FRAME HAS BEEN SETUP for method interpretation the stack looks like:
336 //
337 // +---------------+ <--- sp
338 // | |
339 // : reg save area :
340 // | |
341 // +---------------+ <--- sp + 0x40
342 // | |
343 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later)
344 // | |
345 // +---------------+ <--- sp + 0x5c
346 // | |
347 // : :
348 // | | <--- Lesp
349 // +---------------+ <--- Lmonitors (fp - 0x18)
350 // | VM locals |
351 // +---------------+ <--- fp
352 // | |
353 // : reg save area :
354 // | |
355 // +---------------+ <--- fp + 0x40
356 // | |
357 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later)
358 // | |
359 // +---------------+ <--- fp + 0x5c
360 // | |
361 // : free :
362 // | |
363 // +---------------+
364 // | |
365 // : nonarg locals :
366 // | |
367 // +---------------+
368 // | |
369 // : arguments :
370 // | | <--- Llocals
371 // +---------------+ <--- Gargs
372 // | |
374 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
375 // determine code generation flags
376 bool synchronized = false;
377 address entry_point = NULL;
379 switch (kind) {
380 case Interpreter::zerolocals : break;
381 case Interpreter::zerolocals_synchronized: synchronized = true; break;
382 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
383 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
384 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
385 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
386 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
388 case Interpreter::java_lang_math_sin : break;
389 case Interpreter::java_lang_math_cos : break;
390 case Interpreter::java_lang_math_tan : break;
391 case Interpreter::java_lang_math_sqrt : break;
392 case Interpreter::java_lang_math_abs : break;
393 case Interpreter::java_lang_math_log : break;
394 case Interpreter::java_lang_math_log10 : break;
395 case Interpreter::java_lang_math_pow : break;
396 case Interpreter::java_lang_math_exp : break;
397 case Interpreter::java_lang_ref_reference_get
398 : entry_point = ((InterpreterGenerator*)this)->generate_Reference_get_entry(); break;
399 default:
400 fatal(err_msg("unexpected method kind: %d", kind));
401 break;
402 }
404 if (entry_point) return entry_point;
406 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
407 }
410 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
411 // No special entry points that preclude compilation
412 return true;
413 }
415 void Deoptimization::unwind_callee_save_values(frame* f, vframeArray* vframe_array) {
417 // This code is sort of the equivalent of C2IAdapter::setup_stack_frame back in
418 // the days we had adapter frames. When we deoptimize a situation where a
419 // compiled caller calls a compiled caller will have registers it expects
420 // to survive the call to the callee. If we deoptimize the callee the only
421 // way we can restore these registers is to have the oldest interpreter
422 // frame that we create restore these values. That is what this routine
423 // will accomplish.
425 // At the moment we have modified c2 to not have any callee save registers
426 // so this problem does not exist and this routine is just a place holder.
428 assert(f->is_interpreted_frame(), "must be interpreted");
429 }
432 //----------------------------------------------------------------------------------------------------
433 // Exceptions