Mon, 11 Oct 2010 04:18:58 -0700
6829194: JSR 292 needs to support compressed oops
Reviewed-by: kvn, jrose
1 /*
2 * Copyright (c) 1997, 2010, 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 "incls/_precompiled.incl"
26 #include "incls/_interpreter_sparc.cpp.incl"
30 // Generation of Interpreter
31 //
32 // The InterpreterGenerator generates the interpreter into Interpreter::_code.
35 #define __ _masm->
38 //----------------------------------------------------------------------------------------------------
43 int AbstractInterpreter::BasicType_as_index(BasicType type) {
44 int i = 0;
45 switch (type) {
46 case T_BOOLEAN: i = 0; break;
47 case T_CHAR : i = 1; break;
48 case T_BYTE : i = 2; break;
49 case T_SHORT : i = 3; break;
50 case T_INT : i = 4; break;
51 case T_LONG : i = 5; break;
52 case T_VOID : i = 6; break;
53 case T_FLOAT : i = 7; break;
54 case T_DOUBLE : i = 8; break;
55 case T_OBJECT : i = 9; break;
56 case T_ARRAY : i = 9; break;
57 default : ShouldNotReachHere();
58 }
59 assert(0 <= i && i < AbstractInterpreter::number_of_result_handlers, "index out of bounds");
60 return i;
61 }
64 #ifndef _LP64
65 address AbstractInterpreterGenerator::generate_slow_signature_handler() {
66 address entry = __ pc();
67 Argument argv(0, true);
69 // We are in the jni transition frame. Save the last_java_frame corresponding to the
70 // outer interpreter frame
71 //
72 __ set_last_Java_frame(FP, noreg);
73 // make sure the interpreter frame we've pushed has a valid return pc
74 __ mov(O7, I7);
75 __ mov(Lmethod, G3_scratch);
76 __ mov(Llocals, G4_scratch);
77 __ save_frame(0);
78 __ mov(G2_thread, L7_thread_cache);
79 __ add(argv.address_in_frame(), O3);
80 __ mov(G2_thread, O0);
81 __ mov(G3_scratch, O1);
82 __ call(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), relocInfo::runtime_call_type);
83 __ delayed()->mov(G4_scratch, O2);
84 __ mov(L7_thread_cache, G2_thread);
85 __ reset_last_Java_frame();
87 // load the register arguments (the C code packed them as varargs)
88 for (Argument ldarg = argv.successor(); ldarg.is_register(); ldarg = ldarg.successor()) {
89 __ ld_ptr(ldarg.address_in_frame(), ldarg.as_register());
90 }
91 __ ret();
92 __ delayed()->
93 restore(O0, 0, Lscratch); // caller's Lscratch gets the result handler
94 return entry;
95 }
98 #else
99 // LP64 passes floating point arguments in F1, F3, F5, etc. instead of
100 // O0, O1, O2 etc..
101 // Doubles are passed in D0, D2, D4
102 // We store the signature of the first 16 arguments in the first argument
103 // slot because it will be overwritten prior to calling the native
104 // function, with the pointer to the JNIEnv.
105 // If LP64 there can be up to 16 floating point arguments in registers
106 // or 6 integer registers.
107 address AbstractInterpreterGenerator::generate_slow_signature_handler() {
109 enum {
110 non_float = 0,
111 float_sig = 1,
112 double_sig = 2,
113 sig_mask = 3
114 };
116 address entry = __ pc();
117 Argument argv(0, true);
119 // We are in the jni transition frame. Save the last_java_frame corresponding to the
120 // outer interpreter frame
121 //
122 __ set_last_Java_frame(FP, noreg);
123 // make sure the interpreter frame we've pushed has a valid return pc
124 __ mov(O7, I7);
125 __ mov(Lmethod, G3_scratch);
126 __ mov(Llocals, G4_scratch);
127 __ save_frame(0);
128 __ mov(G2_thread, L7_thread_cache);
129 __ add(argv.address_in_frame(), O3);
130 __ mov(G2_thread, O0);
131 __ mov(G3_scratch, O1);
132 __ call(CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), relocInfo::runtime_call_type);
133 __ delayed()->mov(G4_scratch, O2);
134 __ mov(L7_thread_cache, G2_thread);
135 __ reset_last_Java_frame();
138 // load the register arguments (the C code packed them as varargs)
139 Address Sig = argv.address_in_frame(); // Argument 0 holds the signature
140 __ ld_ptr( Sig, G3_scratch ); // Get register argument signature word into G3_scratch
141 __ mov( G3_scratch, G4_scratch);
142 __ srl( G4_scratch, 2, G4_scratch); // Skip Arg 0
143 Label done;
144 for (Argument ldarg = argv.successor(); ldarg.is_float_register(); ldarg = ldarg.successor()) {
145 Label NonFloatArg;
146 Label LoadFloatArg;
147 Label LoadDoubleArg;
148 Label NextArg;
149 Address a = ldarg.address_in_frame();
150 __ andcc(G4_scratch, sig_mask, G3_scratch);
151 __ br(Assembler::zero, false, Assembler::pt, NonFloatArg);
152 __ delayed()->nop();
154 __ cmp(G3_scratch, float_sig );
155 __ br(Assembler::equal, false, Assembler::pt, LoadFloatArg);
156 __ delayed()->nop();
158 __ cmp(G3_scratch, double_sig );
159 __ br(Assembler::equal, false, Assembler::pt, LoadDoubleArg);
160 __ delayed()->nop();
162 __ bind(NonFloatArg);
163 // There are only 6 integer register arguments!
164 if ( ldarg.is_register() )
165 __ ld_ptr(ldarg.address_in_frame(), ldarg.as_register());
166 else {
167 // Optimization, see if there are any more args and get out prior to checking
168 // all 16 float registers. My guess is that this is rare.
169 // If is_register is false, then we are done the first six integer args.
170 __ tst(G4_scratch);
171 __ brx(Assembler::zero, false, Assembler::pt, done);
172 __ delayed()->nop();
174 }
175 __ ba(false, NextArg);
176 __ delayed()->srl( G4_scratch, 2, G4_scratch );
178 __ bind(LoadFloatArg);
179 __ ldf( FloatRegisterImpl::S, a, ldarg.as_float_register(), 4);
180 __ ba(false, NextArg);
181 __ delayed()->srl( G4_scratch, 2, G4_scratch );
183 __ bind(LoadDoubleArg);
184 __ ldf( FloatRegisterImpl::D, a, ldarg.as_double_register() );
185 __ ba(false, NextArg);
186 __ delayed()->srl( G4_scratch, 2, G4_scratch );
188 __ bind(NextArg);
190 }
192 __ bind(done);
193 __ ret();
194 __ delayed()->
195 restore(O0, 0, Lscratch); // caller's Lscratch gets the result handler
196 return entry;
197 }
198 #endif
200 void InterpreterGenerator::generate_counter_overflow(Label& Lcontinue) {
202 // Generate code to initiate compilation on the counter overflow.
204 // InterpreterRuntime::frequency_counter_overflow takes two arguments,
205 // the first indicates if the counter overflow occurs at a backwards branch (NULL bcp)
206 // and the second is only used when the first is true. We pass zero for both.
207 // The call returns the address of the verified entry point for the method or NULL
208 // if the compilation did not complete (either went background or bailed out).
209 __ set((int)false, O2);
210 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), O2, O2, true);
211 // returns verified_entry_point or NULL
212 // we ignore it in any case
213 __ ba(false, Lcontinue);
214 __ delayed()->nop();
216 }
219 // End of helpers
221 // Various method entries
223 // Abstract method entry
224 // Attempt to execute abstract method. Throw exception
225 //
226 address InterpreterGenerator::generate_abstract_entry(void) {
227 address entry = __ pc();
228 // abstract method entry
229 // throw exception
230 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_AbstractMethodError));
231 // the call_VM checks for exception, so we should never return here.
232 __ should_not_reach_here();
233 return entry;
235 }
238 // Method handle invoker
239 // Dispatch a method of the form java.dyn.MethodHandles::invoke(...)
240 address InterpreterGenerator::generate_method_handle_entry(void) {
241 if (!EnableMethodHandles) {
242 return generate_abstract_entry();
243 }
245 return MethodHandles::generate_method_handle_interpreter_entry(_masm);
246 }
249 //----------------------------------------------------------------------------------------------------
250 // Entry points & stack frame layout
251 //
252 // Here we generate the various kind of entries into the interpreter.
253 // The two main entry type are generic bytecode methods and native call method.
254 // These both come in synchronized and non-synchronized versions but the
255 // frame layout they create is very similar. The other method entry
256 // types are really just special purpose entries that are really entry
257 // and interpretation all in one. These are for trivial methods like
258 // accessor, empty, or special math methods.
259 //
260 // When control flow reaches any of the entry types for the interpreter
261 // the following holds ->
262 //
263 // C2 Calling Conventions:
264 //
265 // The entry code below assumes that the following registers are set
266 // when coming in:
267 // G5_method: holds the methodOop of the method to call
268 // Lesp: points to the TOS of the callers expression stack
269 // after having pushed all the parameters
270 //
271 // The entry code does the following to setup an interpreter frame
272 // pop parameters from the callers stack by adjusting Lesp
273 // set O0 to Lesp
274 // compute X = (max_locals - num_parameters)
275 // bump SP up by X to accomadate the extra locals
276 // compute X = max_expression_stack
277 // + vm_local_words
278 // + 16 words of register save area
279 // save frame doing a save sp, -X, sp growing towards lower addresses
280 // set Lbcp, Lmethod, LcpoolCache
281 // set Llocals to i0
282 // set Lmonitors to FP - rounded_vm_local_words
283 // set Lesp to Lmonitors - 4
284 //
285 // The frame has now been setup to do the rest of the entry code
287 // Try this optimization: Most method entries could live in a
288 // "one size fits all" stack frame without all the dynamic size
289 // calculations. It might be profitable to do all this calculation
290 // statically and approximately for "small enough" methods.
292 //-----------------------------------------------------------------------------------------------
294 // C1 Calling conventions
295 //
296 // Upon method entry, the following registers are setup:
297 //
298 // g2 G2_thread: current thread
299 // g5 G5_method: method to activate
300 // g4 Gargs : pointer to last argument
301 //
302 //
303 // Stack:
304 //
305 // +---------------+ <--- sp
306 // | |
307 // : reg save area :
308 // | |
309 // +---------------+ <--- sp + 0x40
310 // | |
311 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later)
312 // | |
313 // +---------------+ <--- sp + 0x5c
314 // | |
315 // : free :
316 // | |
317 // +---------------+ <--- Gargs
318 // | |
319 // : arguments :
320 // | |
321 // +---------------+
322 // | |
323 //
324 //
325 //
326 // AFTER FRAME HAS BEEN SETUP for method interpretation the stack looks like:
327 //
328 // +---------------+ <--- sp
329 // | |
330 // : reg save area :
331 // | |
332 // +---------------+ <--- sp + 0x40
333 // | |
334 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later)
335 // | |
336 // +---------------+ <--- sp + 0x5c
337 // | |
338 // : :
339 // | | <--- Lesp
340 // +---------------+ <--- Lmonitors (fp - 0x18)
341 // | VM locals |
342 // +---------------+ <--- fp
343 // | |
344 // : reg save area :
345 // | |
346 // +---------------+ <--- fp + 0x40
347 // | |
348 // : extra 7 slots : note: these slots are not really needed for the interpreter (fix later)
349 // | |
350 // +---------------+ <--- fp + 0x5c
351 // | |
352 // : free :
353 // | |
354 // +---------------+
355 // | |
356 // : nonarg locals :
357 // | |
358 // +---------------+
359 // | |
360 // : arguments :
361 // | | <--- Llocals
362 // +---------------+ <--- Gargs
363 // | |
365 address AbstractInterpreterGenerator::generate_method_entry(AbstractInterpreter::MethodKind kind) {
366 // determine code generation flags
367 bool synchronized = false;
368 address entry_point = NULL;
370 switch (kind) {
371 case Interpreter::zerolocals : break;
372 case Interpreter::zerolocals_synchronized: synchronized = true; break;
373 case Interpreter::native : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(false); break;
374 case Interpreter::native_synchronized : entry_point = ((InterpreterGenerator*)this)->generate_native_entry(true); break;
375 case Interpreter::empty : entry_point = ((InterpreterGenerator*)this)->generate_empty_entry(); break;
376 case Interpreter::accessor : entry_point = ((InterpreterGenerator*)this)->generate_accessor_entry(); break;
377 case Interpreter::abstract : entry_point = ((InterpreterGenerator*)this)->generate_abstract_entry(); break;
378 case Interpreter::method_handle : entry_point = ((InterpreterGenerator*)this)->generate_method_handle_entry(); break;
379 case Interpreter::java_lang_math_sin : break;
380 case Interpreter::java_lang_math_cos : break;
381 case Interpreter::java_lang_math_tan : break;
382 case Interpreter::java_lang_math_sqrt : break;
383 case Interpreter::java_lang_math_abs : break;
384 case Interpreter::java_lang_math_log : break;
385 case Interpreter::java_lang_math_log10 : break;
386 default : ShouldNotReachHere(); break;
387 }
389 if (entry_point) return entry_point;
391 return ((InterpreterGenerator*)this)->generate_normal_entry(synchronized);
392 }
395 bool AbstractInterpreter::can_be_compiled(methodHandle m) {
396 // No special entry points that preclude compilation
397 return true;
398 }
400 // This method tells the deoptimizer how big an interpreted frame must be:
401 int AbstractInterpreter::size_activation(methodOop method,
402 int tempcount,
403 int popframe_extra_args,
404 int moncount,
405 int callee_param_count,
406 int callee_locals,
407 bool is_top_frame) {
408 return layout_activation(method,
409 tempcount,
410 popframe_extra_args,
411 moncount,
412 callee_param_count,
413 callee_locals,
414 (frame*)NULL,
415 (frame*)NULL,
416 is_top_frame);
417 }
419 void Deoptimization::unwind_callee_save_values(frame* f, vframeArray* vframe_array) {
421 // This code is sort of the equivalent of C2IAdapter::setup_stack_frame back in
422 // the days we had adapter frames. When we deoptimize a situation where a
423 // compiled caller calls a compiled caller will have registers it expects
424 // to survive the call to the callee. If we deoptimize the callee the only
425 // way we can restore these registers is to have the oldest interpreter
426 // frame that we create restore these values. That is what this routine
427 // will accomplish.
429 // At the moment we have modified c2 to not have any callee save registers
430 // so this problem does not exist and this routine is just a place holder.
432 assert(f->is_interpreted_frame(), "must be interpreted");
433 }
436 //----------------------------------------------------------------------------------------------------
437 // Exceptions