author | rraghavan |
Wed, 14 Mar 2018 09:09:38 -0700 | |
changeset 49397 | d3a8aa01f26f |
parent 47216 | 71c04702a3d5 |
child 50380 | bec342339138 |
permissions | -rw-r--r-- |
42065 | 1 |
/* |
49397 | 2 |
* Copyright (c) 2016, 2018, Oracle and/or its affiliates. All rights reserved. |
42065 | 3 |
* Copyright (c) 2016 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 "c1/c1_MacroAssembler.hpp" |
|
28 |
#include "c1/c1_Runtime1.hpp" |
|
29 |
#include "classfile/systemDictionary.hpp" |
|
30 |
#include "gc/shared/collectedHeap.hpp" |
|
31 |
#include "interpreter/interpreter.hpp" |
|
32 |
#include "oops/arrayOop.hpp" |
|
33 |
#include "oops/markOop.hpp" |
|
34 |
#include "runtime/basicLock.hpp" |
|
35 |
#include "runtime/biasedLocking.hpp" |
|
36 |
#include "runtime/os.hpp" |
|
37 |
#include "runtime/sharedRuntime.hpp" |
|
38 |
#include "runtime/stubRoutines.hpp" |
|
39 |
||
40 |
void C1_MacroAssembler::inline_cache_check(Register receiver, Register iCache) { |
|
41 |
Label ic_miss, ic_hit; |
|
42 |
verify_oop(receiver); |
|
43 |
int klass_offset = oopDesc::klass_offset_in_bytes(); |
|
44 |
||
45 |
if (!ImplicitNullChecks || MacroAssembler::needs_explicit_null_check(klass_offset)) { |
|
46 |
if (VM_Version::has_CompareBranch()) { |
|
47 |
z_cgij(receiver, 0, Assembler::bcondEqual, ic_miss); |
|
48 |
} else { |
|
49 |
z_ltgr(receiver, receiver); |
|
50 |
z_bre(ic_miss); |
|
51 |
} |
|
52 |
} |
|
53 |
||
54 |
compare_klass_ptr(iCache, klass_offset, receiver, false); |
|
55 |
z_bre(ic_hit); |
|
56 |
||
57 |
// If icache check fails, then jump to runtime routine. |
|
58 |
// Note: RECEIVER must still contain the receiver! |
|
59 |
load_const_optimized(Z_R1_scratch, AddressLiteral(SharedRuntime::get_ic_miss_stub())); |
|
60 |
z_br(Z_R1_scratch); |
|
61 |
align(CodeEntryAlignment); |
|
62 |
bind(ic_hit); |
|
63 |
} |
|
64 |
||
65 |
void C1_MacroAssembler::explicit_null_check(Register base) { |
|
66 |
ShouldNotCallThis(); // unused |
|
67 |
} |
|
68 |
||
69 |
void C1_MacroAssembler::build_frame(int frame_size_in_bytes, int bang_size_in_bytes) { |
|
70 |
assert(bang_size_in_bytes >= frame_size_in_bytes, "stack bang size incorrect"); |
|
71 |
generate_stack_overflow_check(bang_size_in_bytes); |
|
72 |
save_return_pc(); |
|
46726
7801367e3cc9
8180659: [s390] micro-optimization in resize_frame_absolute()
lucy
parents:
42065
diff
changeset
|
73 |
push_frame(frame_size_in_bytes); |
42065 | 74 |
} |
75 |
||
76 |
void C1_MacroAssembler::verified_entry() { |
|
77 |
if (C1Breakpoint) z_illtrap(0xC1); |
|
78 |
} |
|
79 |
||
80 |
void C1_MacroAssembler::lock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { |
|
81 |
const int hdr_offset = oopDesc::mark_offset_in_bytes(); |
|
82 |
assert_different_registers(hdr, obj, disp_hdr); |
|
83 |
NearLabel done; |
|
84 |
||
85 |
verify_oop(obj); |
|
86 |
||
87 |
// Load object header. |
|
88 |
z_lg(hdr, Address(obj, hdr_offset)); |
|
89 |
||
90 |
// Save object being locked into the BasicObjectLock... |
|
91 |
z_stg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); |
|
92 |
||
93 |
if (UseBiasedLocking) { |
|
94 |
biased_locking_enter(obj, hdr, Z_R1_scratch, Z_R0_scratch, done, &slow_case); |
|
95 |
} |
|
96 |
||
97 |
// and mark it as unlocked. |
|
98 |
z_oill(hdr, markOopDesc::unlocked_value); |
|
99 |
// Save unlocked object header into the displaced header location on the stack. |
|
100 |
z_stg(hdr, Address(disp_hdr, (intptr_t)0)); |
|
101 |
// Test if object header is still the same (i.e. unlocked), and if so, store the |
|
102 |
// displaced header address in the object header. If it is not the same, get the |
|
103 |
// object header instead. |
|
104 |
z_csg(hdr, disp_hdr, hdr_offset, obj); |
|
105 |
// If the object header was the same, we're done. |
|
106 |
if (PrintBiasedLockingStatistics) { |
|
107 |
Unimplemented(); |
|
108 |
#if 0 |
|
109 |
cond_inc32(Assembler::equal, |
|
110 |
ExternalAddress((address)BiasedLocking::fast_path_entry_count_addr())); |
|
111 |
#endif |
|
112 |
} |
|
113 |
branch_optimized(Assembler::bcondEqual, done); |
|
114 |
// If the object header was not the same, it is now in the hdr register. |
|
115 |
// => Test if it is a stack pointer into the same stack (recursive locking), i.e.: |
|
116 |
// |
|
117 |
// 1) (hdr & markOopDesc::lock_mask_in_place) == 0 |
|
118 |
// 2) rsp <= hdr |
|
119 |
// 3) hdr <= rsp + page_size |
|
120 |
// |
|
121 |
// These 3 tests can be done by evaluating the following expression: |
|
122 |
// |
|
123 |
// (hdr - Z_SP) & (~(page_size-1) | markOopDesc::lock_mask_in_place) |
|
124 |
// |
|
125 |
// assuming both the stack pointer and page_size have their least |
|
126 |
// significant 2 bits cleared and page_size is a power of 2 |
|
127 |
z_sgr(hdr, Z_SP); |
|
128 |
||
129 |
load_const_optimized(Z_R0_scratch, (~(os::vm_page_size()-1) | markOopDesc::lock_mask_in_place)); |
|
130 |
z_ngr(hdr, Z_R0_scratch); // AND sets CC (result eq/ne 0). |
|
131 |
// For recursive locking, the result is zero. => Save it in the displaced header |
|
132 |
// location (NULL in the displaced hdr location indicates recursive locking). |
|
133 |
z_stg(hdr, Address(disp_hdr, (intptr_t)0)); |
|
134 |
// Otherwise we don't care about the result and handle locking via runtime call. |
|
135 |
branch_optimized(Assembler::bcondNotZero, slow_case); |
|
136 |
// done |
|
137 |
bind(done); |
|
138 |
} |
|
139 |
||
140 |
void C1_MacroAssembler::unlock_object(Register hdr, Register obj, Register disp_hdr, Label& slow_case) { |
|
141 |
const int aligned_mask = BytesPerWord -1; |
|
142 |
const int hdr_offset = oopDesc::mark_offset_in_bytes(); |
|
143 |
assert_different_registers(hdr, obj, disp_hdr); |
|
144 |
NearLabel done; |
|
145 |
||
146 |
if (UseBiasedLocking) { |
|
147 |
// Load object. |
|
148 |
z_lg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); |
|
149 |
biased_locking_exit(obj, hdr, done); |
|
150 |
} |
|
151 |
||
152 |
// Load displaced header. |
|
153 |
z_ltg(hdr, Address(disp_hdr, (intptr_t)0)); |
|
154 |
// If the loaded hdr is NULL we had recursive locking, and we are done. |
|
155 |
z_bre(done); |
|
156 |
if (!UseBiasedLocking) { |
|
157 |
// Load object. |
|
158 |
z_lg(obj, Address(disp_hdr, BasicObjectLock::obj_offset_in_bytes())); |
|
159 |
} |
|
160 |
verify_oop(obj); |
|
161 |
// Test if object header is pointing to the displaced header, and if so, restore |
|
162 |
// the displaced header in the object. If the object header is not pointing to |
|
163 |
// the displaced header, get the object header instead. |
|
164 |
z_csg(disp_hdr, hdr, hdr_offset, obj); |
|
165 |
// If the object header was not pointing to the displaced header, |
|
166 |
// we do unlocking via runtime call. |
|
167 |
branch_optimized(Assembler::bcondNotEqual, slow_case); |
|
168 |
// done |
|
169 |
bind(done); |
|
170 |
} |
|
171 |
||
172 |
void C1_MacroAssembler::try_allocate( |
|
173 |
Register obj, // result: Pointer to object after successful allocation. |
|
174 |
Register var_size_in_bytes, // Object size in bytes if unknown at compile time; invalid otherwise. |
|
175 |
int con_size_in_bytes, // Object size in bytes if known at compile time. |
|
176 |
Register t1, // Temp register: Must be global register for incr_allocated_bytes. |
|
177 |
Label& slow_case // Continuation point if fast allocation fails. |
|
178 |
) { |
|
179 |
if (UseTLAB) { |
|
180 |
tlab_allocate(obj, var_size_in_bytes, con_size_in_bytes, t1, slow_case); |
|
181 |
} else { |
|
182 |
// Allocation in shared Eden not implemented, because sapjvm allocation trace does not allow it. |
|
183 |
z_brul(slow_case); |
|
184 |
} |
|
185 |
} |
|
186 |
||
187 |
void C1_MacroAssembler::initialize_header(Register obj, Register klass, Register len, Register Rzero, Register t1) { |
|
188 |
assert_different_registers(obj, klass, len, t1, Rzero); |
|
189 |
if (UseBiasedLocking && !len->is_valid()) { |
|
190 |
assert_different_registers(obj, klass, len, t1); |
|
191 |
z_lg(t1, Address(klass, Klass::prototype_header_offset())); |
|
192 |
} else { |
|
193 |
// This assumes that all prototype bits fit in an int32_t. |
|
194 |
load_const_optimized(t1, (intx)markOopDesc::prototype()); |
|
195 |
} |
|
196 |
z_stg(t1, Address(obj, oopDesc::mark_offset_in_bytes())); |
|
197 |
||
198 |
if (len->is_valid()) { |
|
199 |
// Length will be in the klass gap, if one exists. |
|
200 |
z_st(len, Address(obj, arrayOopDesc::length_offset_in_bytes())); |
|
201 |
} else if (UseCompressedClassPointers) { |
|
202 |
store_klass_gap(Rzero, obj); // Zero klass gap for compressed oops. |
|
203 |
} |
|
204 |
store_klass(klass, obj, t1); |
|
205 |
} |
|
206 |
||
207 |
void C1_MacroAssembler::initialize_body(Register objectFields, Register len_in_bytes, Register Rzero) { |
|
208 |
Label done; |
|
209 |
assert_different_registers(objectFields, len_in_bytes, Rzero); |
|
210 |
||
211 |
// Initialize object fields. |
|
212 |
// See documentation for MVCLE instruction!!! |
|
213 |
assert(objectFields->encoding()%2==0, "objectFields must be an even register"); |
|
214 |
assert(len_in_bytes->encoding() == (objectFields->encoding()+1), "objectFields and len_in_bytes must be a register pair"); |
|
215 |
assert(Rzero->encoding()%2==1, "Rzero must be an odd register"); |
|
216 |
||
217 |
// Use Rzero as src length, then mvcle will copy nothing |
|
218 |
// and fill the object with the padding value 0. |
|
219 |
move_long_ext(objectFields, as_Register(Rzero->encoding()-1), 0); |
|
220 |
bind(done); |
|
221 |
} |
|
222 |
||
223 |
void C1_MacroAssembler::allocate_object( |
|
224 |
Register obj, // Result: pointer to object after successful allocation. |
|
225 |
Register t1, // temp register |
|
226 |
Register t2, // temp register: Must be a global register for try_allocate. |
|
227 |
int hdr_size, // object header size in words |
|
228 |
int obj_size, // object size in words |
|
229 |
Register klass, // object klass |
|
230 |
Label& slow_case // Continuation point if fast allocation fails. |
|
231 |
) { |
|
232 |
assert_different_registers(obj, t1, t2, klass); |
|
233 |
||
234 |
// Allocate space and initialize header. |
|
235 |
try_allocate(obj, noreg, obj_size * wordSize, t1, slow_case); |
|
236 |
||
237 |
initialize_object(obj, klass, noreg, obj_size * HeapWordSize, t1, t2); |
|
238 |
} |
|
239 |
||
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 |
assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, |
|
249 |
"con_size_in_bytes is not multiple of alignment"); |
|
250 |
assert(var_size_in_bytes == noreg, "not implemented"); |
|
251 |
const int hdr_size_in_bytes = instanceOopDesc::header_size() * HeapWordSize; |
|
252 |
||
253 |
const Register Rzero = t2; |
|
254 |
||
255 |
z_xgr(Rzero, Rzero); |
|
256 |
initialize_header(obj, klass, noreg, Rzero, t1); |
|
257 |
||
258 |
// Clear rest of allocated space. |
|
259 |
const int threshold = 4 * BytesPerWord; |
|
260 |
if (con_size_in_bytes <= threshold) { |
|
261 |
// Use explicit null stores. |
|
262 |
// code size = 6*n bytes (n = number of fields to clear) |
|
263 |
for (int i = hdr_size_in_bytes; i < con_size_in_bytes; i += BytesPerWord) |
|
264 |
z_stg(Rzero, Address(obj, i)); |
|
265 |
} else { |
|
266 |
// Code size generated by initialize_body() is 16. |
|
267 |
Register object_fields = Z_R0_scratch; |
|
268 |
Register len_in_bytes = Z_R1_scratch; |
|
269 |
z_la(object_fields, hdr_size_in_bytes, obj); |
|
270 |
load_const_optimized(len_in_bytes, con_size_in_bytes - hdr_size_in_bytes); |
|
271 |
initialize_body(object_fields, len_in_bytes, Rzero); |
|
272 |
} |
|
273 |
||
274 |
// Dtrace support is unimplemented. |
|
275 |
// if (CURRENT_ENV->dtrace_alloc_probes()) { |
|
276 |
// assert(obj == rax, "must be"); |
|
277 |
// call(RuntimeAddress(Runtime1::entry_for (Runtime1::dtrace_object_alloc_id))); |
|
278 |
// } |
|
279 |
||
280 |
verify_oop(obj); |
|
281 |
} |
|
282 |
||
283 |
void C1_MacroAssembler::allocate_array( |
|
284 |
Register obj, // result: Pointer to array after successful allocation. |
|
285 |
Register len, // array length |
|
286 |
Register t1, // temp register |
|
287 |
Register t2, // temp register |
|
288 |
int hdr_size, // object header size in words |
|
289 |
int elt_size, // element size in bytes |
|
290 |
Register klass, // object klass |
|
291 |
Label& slow_case // Continuation point if fast allocation fails. |
|
292 |
) { |
|
293 |
assert_different_registers(obj, len, t1, t2, klass); |
|
294 |
||
295 |
// Determine alignment mask. |
|
296 |
assert(!(BytesPerWord & 1), "must be a multiple of 2 for masking code to work"); |
|
297 |
||
298 |
// Check for negative or excessive length. |
|
299 |
compareU64_and_branch(len, (int32_t)max_array_allocation_length, bcondHigh, slow_case); |
|
300 |
||
301 |
// Compute array size. |
|
302 |
// Note: If 0 <= len <= max_length, len*elt_size + header + alignment is |
|
303 |
// smaller or equal to the largest integer. Also, since top is always |
|
304 |
// aligned, we can do the alignment here instead of at the end address |
|
305 |
// computation. |
|
306 |
const Register arr_size = t2; |
|
307 |
switch (elt_size) { |
|
308 |
case 1: lgr_if_needed(arr_size, len); break; |
|
309 |
case 2: z_sllg(arr_size, len, 1); break; |
|
310 |
case 4: z_sllg(arr_size, len, 2); break; |
|
311 |
case 8: z_sllg(arr_size, len, 3); break; |
|
312 |
default: ShouldNotReachHere(); |
|
313 |
} |
|
314 |
add2reg(arr_size, hdr_size * wordSize + MinObjAlignmentInBytesMask); // Add space for header & alignment. |
|
315 |
z_nill(arr_size, (~MinObjAlignmentInBytesMask) & 0xffff); // Align array size. |
|
316 |
||
317 |
try_allocate(obj, arr_size, 0, t1, slow_case); |
|
318 |
||
319 |
initialize_header(obj, klass, len, noreg, t1); |
|
320 |
||
321 |
// Clear rest of allocated space. |
|
322 |
Label done; |
|
323 |
Register object_fields = t1; |
|
324 |
Register Rzero = Z_R1_scratch; |
|
325 |
z_aghi(arr_size, -(hdr_size * BytesPerWord)); |
|
326 |
z_bre(done); // Jump if size of fields is zero. |
|
327 |
z_la(object_fields, hdr_size * BytesPerWord, obj); |
|
328 |
z_xgr(Rzero, Rzero); |
|
329 |
initialize_body(object_fields, arr_size, Rzero); |
|
330 |
bind(done); |
|
331 |
||
332 |
// Dtrace support is unimplemented. |
|
333 |
// if (CURRENT_ENV->dtrace_alloc_probes()) { |
|
334 |
// assert(obj == rax, "must be"); |
|
335 |
// call(RuntimeAddress(Runtime1::entry_for (Runtime1::dtrace_object_alloc_id))); |
|
336 |
// } |
|
337 |
||
338 |
verify_oop(obj); |
|
339 |
} |
|
340 |
||
341 |
||
342 |
#ifndef PRODUCT |
|
343 |
||
344 |
void C1_MacroAssembler::verify_stack_oop(int stack_offset) { |
|
345 |
Unimplemented(); |
|
346 |
// if (!VerifyOops) return; |
|
347 |
// verify_oop_addr(Address(SP, stack_offset + STACK_BIAS)); |
|
348 |
} |
|
349 |
||
350 |
void C1_MacroAssembler::verify_not_null_oop(Register r) { |
|
351 |
if (!VerifyOops) return; |
|
352 |
NearLabel not_null; |
|
353 |
compareU64_and_branch(r, (intptr_t)0, bcondNotEqual, not_null); |
|
354 |
stop("non-null oop required"); |
|
355 |
bind(not_null); |
|
356 |
verify_oop(r); |
|
357 |
} |
|
358 |
||
359 |
void C1_MacroAssembler::invalidate_registers(Register preserve1, |
|
360 |
Register preserve2, |
|
361 |
Register preserve3) { |
|
362 |
Register dead_value = noreg; |
|
363 |
for (int i = 0; i < FrameMap::nof_cpu_regs; i++) { |
|
364 |
Register r = as_Register(i); |
|
365 |
if (r != preserve1 && r != preserve2 && r != preserve3 && r != Z_SP && r != Z_thread) { |
|
366 |
if (dead_value == noreg) { |
|
367 |
load_const_optimized(r, 0xc1dead); |
|
368 |
dead_value = r; |
|
369 |
} else { |
|
370 |
z_lgr(r, dead_value); |
|
371 |
} |
|
372 |
} |
|
373 |
} |
|
374 |
} |
|
375 |
||
376 |
#endif // !PRODUCT |