Mon, 26 Sep 2011 10:24:05 -0700
7081933: Use zeroing elimination optimization for large array
Summary: Don't zero new typeArray during runtime call if the allocation is followed by arraycopy into it.
Reviewed-by: twisti
1 /*
2 * Copyright (c) 1999, 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 "c1/c1_MacroAssembler.hpp"
27 #include "c1/c1_Runtime1.hpp"
28 #include "classfile/systemDictionary.hpp"
29 #include "gc_interface/collectedHeap.hpp"
30 #include "interpreter/interpreter.hpp"
31 #include "oops/arrayOop.hpp"
32 #include "oops/markOop.hpp"
33 #include "runtime/basicLock.hpp"
34 #include "runtime/biasedLocking.hpp"
35 #include "runtime/os.hpp"
36 #include "runtime/stubRoutines.hpp"
38 void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) {
39 Label L;
40 const Register temp_reg = G3_scratch;
41 // Note: needs more testing of out-of-line vs. inline slow case
42 verify_oop(receiver);
43 load_klass(receiver, temp_reg);
44 cmp_and_brx_short(temp_reg, iCache, Assembler::equal, Assembler::pt, L);
45 AddressLiteral ic_miss(SharedRuntime::get_ic_miss_stub());
46 jump_to(ic_miss, temp_reg);
47 delayed()->nop();
48 align(CodeEntryAlignment);
49 bind(L);
50 }
53 void C1_MacroAssembler::explicit_null_check(Register base) {
54 Unimplemented();
55 }
58 void C1_MacroAssembler::build_frame(int frame_size_in_bytes) {
60 generate_stack_overflow_check(frame_size_in_bytes);
61 // Create the frame.
62 save_frame_c1(frame_size_in_bytes);
63 }
66 void C1_MacroAssembler::unverified_entry(Register receiver, Register ic_klass) {
67 if (C1Breakpoint) breakpoint_trap();
68 inline_cache_check(receiver, ic_klass);
69 }
72 void C1_MacroAssembler::verified_entry() {
73 if (C1Breakpoint) breakpoint_trap();
74 // build frame
75 verify_FPU(0, "method_entry");
76 }
79 void C1_MacroAssembler::lock_object(Register Rmark, Register Roop, Register Rbox, Register Rscratch, Label& slow_case) {
80 assert_different_registers(Rmark, Roop, Rbox, Rscratch);
82 Label done;
84 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes());
86 // The following move must be the first instruction of emitted since debug
87 // information may be generated for it.
88 // Load object header
89 ld_ptr(mark_addr, Rmark);
91 verify_oop(Roop);
93 // save object being locked into the BasicObjectLock
94 st_ptr(Roop, Rbox, BasicObjectLock::obj_offset_in_bytes());
96 if (UseBiasedLocking) {
97 biased_locking_enter(Roop, Rmark, Rscratch, done, &slow_case);
98 }
100 // Save Rbox in Rscratch to be used for the cas operation
101 mov(Rbox, Rscratch);
103 // and mark it unlocked
104 or3(Rmark, markOopDesc::unlocked_value, Rmark);
106 // save unlocked object header into the displaced header location on the stack
107 st_ptr(Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes());
109 // compare object markOop with Rmark and if equal exchange Rscratch with object markOop
110 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
111 casx_under_lock(mark_addr.base(), Rmark, Rscratch, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
112 // if compare/exchange succeeded we found an unlocked object and we now have locked it
113 // hence we are done
114 cmp(Rmark, Rscratch);
115 brx(Assembler::equal, false, Assembler::pt, done);
116 delayed()->sub(Rscratch, SP, Rscratch); //pull next instruction into delay slot
117 // we did not find an unlocked object so see if this is a recursive case
118 // sub(Rscratch, SP, Rscratch);
119 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant");
120 andcc(Rscratch, 0xfffff003, Rscratch);
121 brx(Assembler::notZero, false, Assembler::pn, slow_case);
122 delayed()->st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes());
123 bind(done);
124 }
127 void C1_MacroAssembler::unlock_object(Register Rmark, Register Roop, Register Rbox, Label& slow_case) {
128 assert_different_registers(Rmark, Roop, Rbox);
130 Label done;
132 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes());
133 assert(mark_addr.disp() == 0, "cas must take a zero displacement");
135 if (UseBiasedLocking) {
136 // load the object out of the BasicObjectLock
137 ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop);
138 verify_oop(Roop);
139 biased_locking_exit(mark_addr, Rmark, done);
140 }
141 // Test first it it is a fast recursive unlock
142 ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark);
143 br_null_short(Rmark, Assembler::pt, done);
144 if (!UseBiasedLocking) {
145 // load object
146 ld_ptr(Rbox, BasicObjectLock::obj_offset_in_bytes(), Roop);
147 verify_oop(Roop);
148 }
150 // Check if it is still a light weight lock, this is is true if we see
151 // the stack address of the basicLock in the markOop of the object
152 casx_under_lock(mark_addr.base(), Rbox, Rmark, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
153 cmp(Rbox, Rmark);
155 brx(Assembler::notEqual, false, Assembler::pn, slow_case);
156 delayed()->nop();
157 // Done
158 bind(done);
159 }
162 void C1_MacroAssembler::try_allocate(
163 Register obj, // result: pointer to object after successful allocation
164 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
165 int con_size_in_bytes, // object size in bytes if known at compile time
166 Register t1, // temp register, must be global register for incr_allocated_bytes
167 Register t2, // temp register
168 Label& slow_case // continuation point if fast allocation fails
169 ) {
170 RegisterOrConstant size_in_bytes = var_size_in_bytes->is_valid()
171 ? RegisterOrConstant(var_size_in_bytes) : RegisterOrConstant(con_size_in_bytes);
172 if (UseTLAB) {
173 tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case);
174 } else {
175 eden_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, t2, slow_case);
176 incr_allocated_bytes(size_in_bytes, t1, t2);
177 }
178 }
181 void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register t1, Register t2) {
182 assert_different_registers(obj, klass, len, t1, t2);
183 if (UseBiasedLocking && !len->is_valid()) {
184 ld_ptr(klass, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes(), t1);
185 } else {
186 set((intx)markOopDesc::prototype(), t1);
187 }
188 st_ptr(t1, obj, oopDesc::mark_offset_in_bytes());
189 if (UseCompressedOops) {
190 // Save klass
191 mov(klass, t1);
192 encode_heap_oop_not_null(t1);
193 stw(t1, obj, oopDesc::klass_offset_in_bytes());
194 } else {
195 st_ptr(klass, obj, oopDesc::klass_offset_in_bytes());
196 }
197 if (len->is_valid()) st(len, obj, arrayOopDesc::length_offset_in_bytes());
198 else if (UseCompressedOops) {
199 store_klass_gap(G0, obj);
200 }
201 }
204 void C1_MacroAssembler::initialize_body(Register base, Register index) {
205 assert_different_registers(base, index);
206 Label loop;
207 bind(loop);
208 subcc(index, HeapWordSize, index);
209 brx(Assembler::greaterEqual, true, Assembler::pt, loop);
210 delayed()->st_ptr(G0, base, index);
211 }
214 void C1_MacroAssembler::allocate_object(
215 Register obj, // result: pointer to object after successful allocation
216 Register t1, // temp register
217 Register t2, // temp register, must be a global register for try_allocate
218 Register t3, // temp register
219 int hdr_size, // object header size in words
220 int obj_size, // object size in words
221 Register klass, // object klass
222 Label& slow_case // continuation point if fast allocation fails
223 ) {
224 assert_different_registers(obj, t1, t2, t3, klass);
225 assert(klass == G5, "must be G5");
227 // allocate space & initialize header
228 if (!is_simm13(obj_size * wordSize)) {
229 // would need to use extra register to load
230 // object size => go the slow case for now
231 ba(slow_case);
232 delayed()->nop();
233 return;
234 }
235 try_allocate(obj, noreg, obj_size * wordSize, t2, t3, slow_case);
237 initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2);
238 }
240 void C1_MacroAssembler::initialize_object(
241 Register obj, // result: pointer to object after successful allocation
242 Register klass, // object klass
243 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
244 int con_size_in_bytes, // object size in bytes if known at compile time
245 Register t1, // temp register
246 Register t2 // temp register
247 ) {
248 const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize;
250 initialize_header(obj, klass, noreg, t1, t2);
252 #ifdef ASSERT
253 {
254 Label ok;
255 ld(klass, klassOopDesc::header_size() * HeapWordSize + Klass::layout_helper_offset_in_bytes(), t1);
256 if (var_size_in_bytes != noreg) {
257 cmp_and_brx_short(t1, var_size_in_bytes, Assembler::equal, Assembler::pt, ok);
258 } else {
259 cmp_and_brx_short(t1, con_size_in_bytes, Assembler::equal, Assembler::pt, ok);
260 }
261 stop("bad size in initialize_object");
262 should_not_reach_here();
264 bind(ok);
265 }
267 #endif
269 // initialize body
270 const int threshold = 5 * HeapWordSize; // approximate break even point for code size
271 if (var_size_in_bytes != noreg) {
272 // use a loop
273 add(obj, hdr_size_in_bytes, t1); // compute address of first element
274 sub(var_size_in_bytes, hdr_size_in_bytes, t2); // compute size of body
275 initialize_body(t1, t2);
276 #ifndef _LP64
277 } else if (VM_Version::v9_instructions_work() && con_size_in_bytes < threshold * 2) {
278 // on v9 we can do double word stores to fill twice as much space.
279 assert(hdr_size_in_bytes % 8 == 0, "double word aligned");
280 assert(con_size_in_bytes % 8 == 0, "double word aligned");
281 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += 2 * HeapWordSize) stx(G0, obj, i);
282 #endif
283 } else if (con_size_in_bytes <= threshold) {
284 // use explicit NULL stores
285 for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += HeapWordSize) st_ptr(G0, obj, i);
286 } else if (con_size_in_bytes > hdr_size_in_bytes) {
287 // use a loop
288 const Register base = t1;
289 const Register index = t2;
290 add(obj, hdr_size_in_bytes, base); // compute address of first element
291 // compute index = number of words to clear
292 set(con_size_in_bytes - hdr_size_in_bytes, index);
293 initialize_body(base, index);
294 }
296 if (CURRENT_ENV->dtrace_alloc_probes()) {
297 assert(obj == O0, "must be");
298 call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
299 relocInfo::runtime_call_type);
300 delayed()->nop();
301 }
303 verify_oop(obj);
304 }
307 void C1_MacroAssembler::allocate_array(
308 Register obj, // result: pointer to array after successful allocation
309 Register len, // array length
310 Register t1, // temp register
311 Register t2, // temp register
312 Register t3, // temp register
313 int hdr_size, // object header size in words
314 int elt_size, // element size in bytes
315 Register klass, // object klass
316 Label& slow_case // continuation point if fast allocation fails
317 ) {
318 assert_different_registers(obj, len, t1, t2, t3, klass);
319 assert(klass == G5, "must be G5");
320 assert(t1 == G1, "must be G1");
322 // determine alignment mask
323 assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work");
325 // check for negative or excessive length
326 // note: the maximum length allowed is chosen so that arrays of any
327 // element size with this length are always smaller or equal
328 // to the largest integer (i.e., array size computation will
329 // not overflow)
330 set(max_array_allocation_length, t1);
331 cmp(len, t1);
332 br(Assembler::greaterUnsigned, false, Assembler::pn, slow_case);
334 // compute array size
335 // note: if 0 <= len <= max_length, len*elt_size + header + alignment is
336 // smaller or equal to the largest integer; also, since top is always
337 // aligned, we can do the alignment here instead of at the end address
338 // computation
339 const Register arr_size = t1;
340 switch (elt_size) {
341 case 1: delayed()->mov(len, arr_size); break;
342 case 2: delayed()->sll(len, 1, arr_size); break;
343 case 4: delayed()->sll(len, 2, arr_size); break;
344 case 8: delayed()->sll(len, 3, arr_size); break;
345 default: ShouldNotReachHere();
346 }
347 add(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask, arr_size); // add space for header & alignment
348 and3(arr_size, ~MinObjAlignmentInBytesMask, arr_size); // align array size
350 // allocate space & initialize header
351 if (UseTLAB) {
352 tlab_allocate(obj, arr_size, 0, t2, slow_case);
353 } else {
354 eden_allocate(obj, arr_size, 0, t2, t3, slow_case);
355 }
356 initialize_header(obj, klass, len, t2, t3);
358 // initialize body
359 const Register base = t2;
360 const Register index = t3;
361 add(obj, hdr_size * wordSize, base); // compute address of first element
362 sub(arr_size, hdr_size * wordSize, index); // compute index = number of words to clear
363 initialize_body(base, index);
365 if (CURRENT_ENV->dtrace_alloc_probes()) {
366 assert(obj == O0, "must be");
367 call(CAST_FROM_FN_PTR(address, Runtime1::entry_for(Runtime1::dtrace_object_alloc_id)),
368 relocInfo::runtime_call_type);
369 delayed()->nop();
370 }
372 verify_oop(obj);
373 }
376 #ifndef PRODUCT
378 void C1_MacroAssembler::verify_stack_oop(int stack_offset) {
379 if (!VerifyOops) return;
380 verify_oop_addr(Address(SP, stack_offset + STACK_BIAS));
381 }
383 void C1_MacroAssembler::verify_not_null_oop(Register r) {
384 Label not_null;
385 br_notnull_short(r, Assembler::pt, not_null);
386 stop("non-null oop required");
387 bind(not_null);
388 if (!VerifyOops) return;
389 verify_oop(r);
390 }
392 void C1_MacroAssembler::invalidate_registers(bool iregisters, bool lregisters, bool oregisters,
393 Register preserve1, Register preserve2) {
394 if (iregisters) {
395 for (int i = 0; i < 6; i++) {
396 Register r = as_iRegister(i);
397 if (r != preserve1 && r != preserve2) set(0xdead, r);
398 }
399 }
400 if (oregisters) {
401 for (int i = 0; i < 6; i++) {
402 Register r = as_oRegister(i);
403 if (r != preserve1 && r != preserve2) set(0xdead, r);
404 }
405 }
406 if (lregisters) {
407 for (int i = 0; i < 8; i++) {
408 Register r = as_lRegister(i);
409 if (r != preserve1 && r != preserve2) set(0xdead, r);
410 }
411 }
412 }
415 #endif