Mercurial > hg > graal-compiler
annotate src/cpu/sparc/vm/assembler_sparc.cpp @ 3946:0db80d8e77fc jdk8-b06
Added tag jdk8-b05 for changeset dce7d24674f4
author | schien |
---|---|
date | Thu, 15 Sep 2011 18:53:14 -0700 |
parents | 3d42f82cd811 |
children | 4fe626cbf0bf baf763f388e6 |
rev | line source |
---|---|
0 | 1 /* |
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
2 * Copyright (c) 1997, 2011, Oracle and/or its affiliates. All rights reserved. |
0 | 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 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1513
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1513
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1513
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1972 | 25 #include "precompiled.hpp" |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
26 #include "asm/assembler.hpp" |
1972 | 27 #include "assembler_sparc.inline.hpp" |
28 #include "gc_interface/collectedHeap.inline.hpp" | |
29 #include "interpreter/interpreter.hpp" | |
30 #include "memory/cardTableModRefBS.hpp" | |
31 #include "memory/resourceArea.hpp" | |
32 #include "prims/methodHandles.hpp" | |
33 #include "runtime/biasedLocking.hpp" | |
34 #include "runtime/interfaceSupport.hpp" | |
35 #include "runtime/objectMonitor.hpp" | |
36 #include "runtime/os.hpp" | |
37 #include "runtime/sharedRuntime.hpp" | |
38 #include "runtime/stubRoutines.hpp" | |
39 #ifndef SERIALGC | |
40 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp" | |
41 #include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp" | |
42 #include "gc_implementation/g1/heapRegion.hpp" | |
43 #endif | |
0 | 44 |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
45 #ifdef PRODUCT |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
46 #define BLOCK_COMMENT(str) /* nothing */ |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
47 #else |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
48 #define BLOCK_COMMENT(str) block_comment(str) |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
49 #endif |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
50 |
727 | 51 // Convert the raw encoding form into the form expected by the |
52 // constructor for Address. | |
53 Address Address::make_raw(int base, int index, int scale, int disp, bool disp_is_oop) { | |
54 assert(scale == 0, "not supported"); | |
55 RelocationHolder rspec; | |
56 if (disp_is_oop) { | |
57 rspec = Relocation::spec_simple(relocInfo::oop_type); | |
0 | 58 } |
727 | 59 |
60 Register rindex = as_Register(index); | |
61 if (rindex != G0) { | |
62 Address madr(as_Register(base), rindex); | |
63 madr._rspec = rspec; | |
64 return madr; | |
65 } else { | |
66 Address madr(as_Register(base), disp); | |
67 madr._rspec = rspec; | |
68 return madr; | |
69 } | |
70 } | |
71 | |
72 Address Argument::address_in_frame() const { | |
73 // Warning: In LP64 mode disp will occupy more than 10 bits, but | |
74 // op codes such as ld or ldx, only access disp() to get | |
75 // their simm13 argument. | |
76 int disp = ((_number - Argument::n_register_parameters + frame::memory_parameter_word_sp_offset) * BytesPerWord) + STACK_BIAS; | |
77 if (is_in()) | |
78 return Address(FP, disp); // In argument. | |
79 else | |
80 return Address(SP, disp); // Out argument. | |
0 | 81 } |
82 | |
83 static const char* argumentNames[][2] = { | |
84 {"A0","P0"}, {"A1","P1"}, {"A2","P2"}, {"A3","P3"}, {"A4","P4"}, | |
85 {"A5","P5"}, {"A6","P6"}, {"A7","P7"}, {"A8","P8"}, {"A9","P9"}, | |
86 {"A(n>9)","P(n>9)"} | |
87 }; | |
88 | |
89 const char* Argument::name() const { | |
90 int nofArgs = sizeof argumentNames / sizeof argumentNames[0]; | |
91 int num = number(); | |
92 if (num >= nofArgs) num = nofArgs - 1; | |
93 return argumentNames[num][is_in() ? 1 : 0]; | |
94 } | |
95 | |
96 void Assembler::print_instruction(int inst) { | |
97 const char* s; | |
98 switch (inv_op(inst)) { | |
99 default: s = "????"; break; | |
100 case call_op: s = "call"; break; | |
101 case branch_op: | |
102 switch (inv_op2(inst)) { | |
103 case fb_op2: s = "fb"; break; | |
104 case fbp_op2: s = "fbp"; break; | |
105 case br_op2: s = "br"; break; | |
106 case bp_op2: s = "bp"; break; | |
107 case cb_op2: s = "cb"; break; | |
3839 | 108 case bpr_op2: { |
109 if (is_cbcond(inst)) { | |
110 s = is_cxb(inst) ? "cxb" : "cwb"; | |
111 } else { | |
112 s = "bpr"; | |
113 } | |
114 break; | |
115 } | |
0 | 116 default: s = "????"; break; |
117 } | |
118 } | |
119 ::tty->print("%s", s); | |
120 } | |
121 | |
122 | |
123 // Patch instruction inst at offset inst_pos to refer to dest_pos | |
124 // and return the resulting instruction. | |
125 // We should have pcs, not offsets, but since all is relative, it will work out | |
126 // OK. | |
127 int Assembler::patched_branch(int dest_pos, int inst, int inst_pos) { | |
128 | |
129 int m; // mask for displacement field | |
130 int v; // new value for displacement field | |
131 const int word_aligned_ones = -4; | |
132 switch (inv_op(inst)) { | |
133 default: ShouldNotReachHere(); | |
134 case call_op: m = wdisp(word_aligned_ones, 0, 30); v = wdisp(dest_pos, inst_pos, 30); break; | |
135 case branch_op: | |
136 switch (inv_op2(inst)) { | |
137 case fbp_op2: m = wdisp( word_aligned_ones, 0, 19); v = wdisp( dest_pos, inst_pos, 19); break; | |
138 case bp_op2: m = wdisp( word_aligned_ones, 0, 19); v = wdisp( dest_pos, inst_pos, 19); break; | |
139 case fb_op2: m = wdisp( word_aligned_ones, 0, 22); v = wdisp( dest_pos, inst_pos, 22); break; | |
140 case br_op2: m = wdisp( word_aligned_ones, 0, 22); v = wdisp( dest_pos, inst_pos, 22); break; | |
141 case cb_op2: m = wdisp( word_aligned_ones, 0, 22); v = wdisp( dest_pos, inst_pos, 22); break; | |
3839 | 142 case bpr_op2: { |
143 if (is_cbcond(inst)) { | |
144 m = wdisp10(word_aligned_ones, 0); | |
145 v = wdisp10(dest_pos, inst_pos); | |
146 } else { | |
147 m = wdisp16(word_aligned_ones, 0); | |
148 v = wdisp16(dest_pos, inst_pos); | |
149 } | |
150 break; | |
151 } | |
0 | 152 default: ShouldNotReachHere(); |
153 } | |
154 } | |
155 return inst & ~m | v; | |
156 } | |
157 | |
158 // Return the offset of the branch destionation of instruction inst | |
159 // at offset pos. | |
160 // Should have pcs, but since all is relative, it works out. | |
161 int Assembler::branch_destination(int inst, int pos) { | |
162 int r; | |
163 switch (inv_op(inst)) { | |
164 default: ShouldNotReachHere(); | |
165 case call_op: r = inv_wdisp(inst, pos, 30); break; | |
166 case branch_op: | |
167 switch (inv_op2(inst)) { | |
168 case fbp_op2: r = inv_wdisp( inst, pos, 19); break; | |
169 case bp_op2: r = inv_wdisp( inst, pos, 19); break; | |
170 case fb_op2: r = inv_wdisp( inst, pos, 22); break; | |
171 case br_op2: r = inv_wdisp( inst, pos, 22); break; | |
172 case cb_op2: r = inv_wdisp( inst, pos, 22); break; | |
3839 | 173 case bpr_op2: { |
174 if (is_cbcond(inst)) { | |
175 r = inv_wdisp10(inst, pos); | |
176 } else { | |
177 r = inv_wdisp16(inst, pos); | |
178 } | |
179 break; | |
180 } | |
0 | 181 default: ShouldNotReachHere(); |
182 } | |
183 } | |
184 return r; | |
185 } | |
186 | |
187 int AbstractAssembler::code_fill_byte() { | |
188 return 0x00; // illegal instruction 0x00000000 | |
189 } | |
190 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
191 Assembler::Condition Assembler::reg_cond_to_cc_cond(Assembler::RCondition in) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
192 switch (in) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
193 case rc_z: return equal; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
194 case rc_lez: return lessEqual; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
195 case rc_lz: return less; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
196 case rc_nz: return notEqual; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
197 case rc_gz: return greater; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
198 case rc_gez: return greaterEqual; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
199 default: |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
200 ShouldNotReachHere(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
201 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
202 return equal; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
203 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
204 |
0 | 205 // Generate a bunch 'o stuff (including v9's |
206 #ifndef PRODUCT | |
207 void Assembler::test_v9() { | |
208 add( G0, G1, G2 ); | |
209 add( G3, 0, G4 ); | |
210 | |
211 addcc( G5, G6, G7 ); | |
212 addcc( I0, 1, I1 ); | |
213 addc( I2, I3, I4 ); | |
214 addc( I5, -1, I6 ); | |
215 addccc( I7, L0, L1 ); | |
216 addccc( L2, (1 << 12) - 2, L3 ); | |
217 | |
218 Label lbl1, lbl2, lbl3; | |
219 | |
220 bind(lbl1); | |
221 | |
222 bpr( rc_z, true, pn, L4, pc(), relocInfo::oop_type ); | |
223 delayed()->nop(); | |
224 bpr( rc_lez, false, pt, L5, lbl1); | |
225 delayed()->nop(); | |
226 | |
227 fb( f_never, true, pc() + 4, relocInfo::none); | |
228 delayed()->nop(); | |
229 fb( f_notEqual, false, lbl2 ); | |
230 delayed()->nop(); | |
231 | |
232 fbp( f_notZero, true, fcc0, pn, pc() - 4, relocInfo::none); | |
233 delayed()->nop(); | |
234 fbp( f_lessOrGreater, false, fcc1, pt, lbl3 ); | |
235 delayed()->nop(); | |
236 | |
237 br( equal, true, pc() + 1024, relocInfo::none); | |
238 delayed()->nop(); | |
239 br( lessEqual, false, lbl1 ); | |
240 delayed()->nop(); | |
241 br( never, false, lbl1 ); | |
242 delayed()->nop(); | |
243 | |
244 bp( less, true, icc, pn, pc(), relocInfo::none); | |
245 delayed()->nop(); | |
246 bp( lessEqualUnsigned, false, xcc, pt, lbl2 ); | |
247 delayed()->nop(); | |
248 | |
249 call( pc(), relocInfo::none); | |
250 delayed()->nop(); | |
251 call( lbl3 ); | |
252 delayed()->nop(); | |
253 | |
254 | |
255 casa( L6, L7, O0 ); | |
256 casxa( O1, O2, O3, 0 ); | |
257 | |
258 udiv( O4, O5, O7 ); | |
259 udiv( G0, (1 << 12) - 1, G1 ); | |
260 sdiv( G1, G2, G3 ); | |
261 sdiv( G4, -((1 << 12) - 1), G5 ); | |
262 udivcc( G6, G7, I0 ); | |
263 udivcc( I1, -((1 << 12) - 2), I2 ); | |
264 sdivcc( I3, I4, I5 ); | |
265 sdivcc( I6, -((1 << 12) - 0), I7 ); | |
266 | |
267 done(); | |
268 retry(); | |
269 | |
270 fadd( FloatRegisterImpl::S, F0, F1, F2 ); | |
271 fsub( FloatRegisterImpl::D, F34, F0, F62 ); | |
272 | |
273 fcmp( FloatRegisterImpl::Q, fcc0, F0, F60); | |
274 fcmpe( FloatRegisterImpl::S, fcc1, F31, F30); | |
275 | |
276 ftox( FloatRegisterImpl::D, F2, F4 ); | |
277 ftoi( FloatRegisterImpl::Q, F4, F8 ); | |
278 | |
279 ftof( FloatRegisterImpl::S, FloatRegisterImpl::Q, F3, F12 ); | |
280 | |
281 fxtof( FloatRegisterImpl::S, F4, F5 ); | |
282 fitof( FloatRegisterImpl::D, F6, F8 ); | |
283 | |
284 fmov( FloatRegisterImpl::Q, F16, F20 ); | |
285 fneg( FloatRegisterImpl::S, F6, F7 ); | |
286 fabs( FloatRegisterImpl::D, F10, F12 ); | |
287 | |
288 fmul( FloatRegisterImpl::Q, F24, F28, F32 ); | |
289 fmul( FloatRegisterImpl::S, FloatRegisterImpl::D, F8, F9, F14 ); | |
290 fdiv( FloatRegisterImpl::S, F10, F11, F12 ); | |
291 | |
292 fsqrt( FloatRegisterImpl::S, F13, F14 ); | |
293 | |
294 flush( L0, L1 ); | |
295 flush( L2, -1 ); | |
296 | |
297 flushw(); | |
298 | |
299 illtrap( (1 << 22) - 2); | |
300 | |
301 impdep1( 17, (1 << 19) - 1 ); | |
302 impdep2( 3, 0 ); | |
303 | |
304 jmpl( L3, L4, L5 ); | |
305 delayed()->nop(); | |
306 jmpl( L6, -1, L7, Relocation::spec_simple(relocInfo::none)); | |
307 delayed()->nop(); | |
308 | |
309 | |
310 ldf( FloatRegisterImpl::S, O0, O1, F15 ); | |
311 ldf( FloatRegisterImpl::D, O2, -1, F14 ); | |
312 | |
313 | |
314 ldfsr( O3, O4 ); | |
315 ldfsr( O5, -1 ); | |
316 ldxfsr( O6, O7 ); | |
317 ldxfsr( I0, -1 ); | |
318 | |
319 ldfa( FloatRegisterImpl::D, I1, I2, 1, F16 ); | |
320 ldfa( FloatRegisterImpl::Q, I3, -1, F36 ); | |
321 | |
322 ldsb( I4, I5, I6 ); | |
323 ldsb( I7, -1, G0 ); | |
324 ldsh( G1, G3, G4 ); | |
325 ldsh( G5, -1, G6 ); | |
326 ldsw( G7, L0, L1 ); | |
327 ldsw( L2, -1, L3 ); | |
328 ldub( L4, L5, L6 ); | |
329 ldub( L7, -1, O0 ); | |
330 lduh( O1, O2, O3 ); | |
331 lduh( O4, -1, O5 ); | |
332 lduw( O6, O7, G0 ); | |
333 lduw( G1, -1, G2 ); | |
334 ldx( G3, G4, G5 ); | |
335 ldx( G6, -1, G7 ); | |
336 ldd( I0, I1, I2 ); | |
337 ldd( I3, -1, I4 ); | |
338 | |
339 ldsba( I5, I6, 2, I7 ); | |
340 ldsba( L0, -1, L1 ); | |
341 ldsha( L2, L3, 3, L4 ); | |
342 ldsha( L5, -1, L6 ); | |
343 ldswa( L7, O0, (1 << 8) - 1, O1 ); | |
344 ldswa( O2, -1, O3 ); | |
345 lduba( O4, O5, 0, O6 ); | |
346 lduba( O7, -1, I0 ); | |
347 lduha( I1, I2, 1, I3 ); | |
348 lduha( I4, -1, I5 ); | |
349 lduwa( I6, I7, 2, L0 ); | |
350 lduwa( L1, -1, L2 ); | |
351 ldxa( L3, L4, 3, L5 ); | |
352 ldxa( L6, -1, L7 ); | |
353 ldda( G0, G1, 4, G2 ); | |
354 ldda( G3, -1, G4 ); | |
355 | |
356 ldstub( G5, G6, G7 ); | |
357 ldstub( O0, -1, O1 ); | |
358 | |
359 ldstuba( O2, O3, 5, O4 ); | |
360 ldstuba( O5, -1, O6 ); | |
361 | |
362 and3( I0, L0, O0 ); | |
363 and3( G7, -1, O7 ); | |
364 andcc( L2, I2, G2 ); | |
365 andcc( L4, -1, G4 ); | |
366 andn( I5, I6, I7 ); | |
367 andn( I6, -1, I7 ); | |
368 andncc( I5, I6, I7 ); | |
369 andncc( I7, -1, I6 ); | |
370 or3( I5, I6, I7 ); | |
371 or3( I7, -1, I6 ); | |
372 orcc( I5, I6, I7 ); | |
373 orcc( I7, -1, I6 ); | |
374 orn( I5, I6, I7 ); | |
375 orn( I7, -1, I6 ); | |
376 orncc( I5, I6, I7 ); | |
377 orncc( I7, -1, I6 ); | |
378 xor3( I5, I6, I7 ); | |
379 xor3( I7, -1, I6 ); | |
380 xorcc( I5, I6, I7 ); | |
381 xorcc( I7, -1, I6 ); | |
382 xnor( I5, I6, I7 ); | |
383 xnor( I7, -1, I6 ); | |
384 xnorcc( I5, I6, I7 ); | |
385 xnorcc( I7, -1, I6 ); | |
386 | |
387 membar( Membar_mask_bits(StoreStore | LoadStore | StoreLoad | LoadLoad | Sync | MemIssue | Lookaside ) ); | |
388 membar( StoreStore ); | |
389 membar( LoadStore ); | |
390 membar( StoreLoad ); | |
391 membar( LoadLoad ); | |
392 membar( Sync ); | |
393 membar( MemIssue ); | |
394 membar( Lookaside ); | |
395 | |
396 fmov( FloatRegisterImpl::S, f_ordered, true, fcc2, F16, F17 ); | |
397 fmov( FloatRegisterImpl::D, rc_lz, L5, F18, F20 ); | |
398 | |
399 movcc( overflowClear, false, icc, I6, L4 ); | |
400 movcc( f_unorderedOrEqual, true, fcc2, (1 << 10) - 1, O0 ); | |
401 | |
402 movr( rc_nz, I5, I6, I7 ); | |
403 movr( rc_gz, L1, -1, L2 ); | |
404 | |
405 mulx( I5, I6, I7 ); | |
406 mulx( I7, -1, I6 ); | |
407 sdivx( I5, I6, I7 ); | |
408 sdivx( I7, -1, I6 ); | |
409 udivx( I5, I6, I7 ); | |
410 udivx( I7, -1, I6 ); | |
411 | |
412 umul( I5, I6, I7 ); | |
413 umul( I7, -1, I6 ); | |
414 smul( I5, I6, I7 ); | |
415 smul( I7, -1, I6 ); | |
416 umulcc( I5, I6, I7 ); | |
417 umulcc( I7, -1, I6 ); | |
418 smulcc( I5, I6, I7 ); | |
419 smulcc( I7, -1, I6 ); | |
420 | |
421 mulscc( I5, I6, I7 ); | |
422 mulscc( I7, -1, I6 ); | |
423 | |
424 nop(); | |
425 | |
426 | |
427 popc( G0, G1); | |
428 popc( -1, G2); | |
429 | |
430 prefetch( L1, L2, severalReads ); | |
431 prefetch( L3, -1, oneRead ); | |
432 prefetcha( O3, O2, 6, severalWritesAndPossiblyReads ); | |
433 prefetcha( G2, -1, oneWrite ); | |
434 | |
435 rett( I7, I7); | |
436 delayed()->nop(); | |
437 rett( G0, -1, relocInfo::none); | |
438 delayed()->nop(); | |
439 | |
440 save( I5, I6, I7 ); | |
441 save( I7, -1, I6 ); | |
442 restore( I5, I6, I7 ); | |
443 restore( I7, -1, I6 ); | |
444 | |
445 saved(); | |
446 restored(); | |
447 | |
448 sethi( 0xaaaaaaaa, I3, Relocation::spec_simple(relocInfo::none)); | |
449 | |
450 sll( I5, I6, I7 ); | |
451 sll( I7, 31, I6 ); | |
452 srl( I5, I6, I7 ); | |
453 srl( I7, 0, I6 ); | |
454 sra( I5, I6, I7 ); | |
455 sra( I7, 30, I6 ); | |
456 sllx( I5, I6, I7 ); | |
457 sllx( I7, 63, I6 ); | |
458 srlx( I5, I6, I7 ); | |
459 srlx( I7, 0, I6 ); | |
460 srax( I5, I6, I7 ); | |
461 srax( I7, 62, I6 ); | |
462 | |
463 sir( -1 ); | |
464 | |
465 stbar(); | |
466 | |
467 stf( FloatRegisterImpl::Q, F40, G0, I7 ); | |
468 stf( FloatRegisterImpl::S, F18, I3, -1 ); | |
469 | |
470 stfsr( L1, L2 ); | |
471 stfsr( I7, -1 ); | |
472 stxfsr( I6, I5 ); | |
473 stxfsr( L4, -1 ); | |
474 | |
475 stfa( FloatRegisterImpl::D, F22, I6, I7, 7 ); | |
476 stfa( FloatRegisterImpl::Q, F44, G0, -1 ); | |
477 | |
478 stb( L5, O2, I7 ); | |
479 stb( I7, I6, -1 ); | |
480 sth( L5, O2, I7 ); | |
481 sth( I7, I6, -1 ); | |
482 stw( L5, O2, I7 ); | |
483 stw( I7, I6, -1 ); | |
484 stx( L5, O2, I7 ); | |
485 stx( I7, I6, -1 ); | |
486 std( L5, O2, I7 ); | |
487 std( I7, I6, -1 ); | |
488 | |
489 stba( L5, O2, I7, 8 ); | |
490 stba( I7, I6, -1 ); | |
491 stha( L5, O2, I7, 9 ); | |
492 stha( I7, I6, -1 ); | |
493 stwa( L5, O2, I7, 0 ); | |
494 stwa( I7, I6, -1 ); | |
495 stxa( L5, O2, I7, 11 ); | |
496 stxa( I7, I6, -1 ); | |
497 stda( L5, O2, I7, 12 ); | |
498 stda( I7, I6, -1 ); | |
499 | |
500 sub( I5, I6, I7 ); | |
501 sub( I7, -1, I6 ); | |
502 subcc( I5, I6, I7 ); | |
503 subcc( I7, -1, I6 ); | |
504 subc( I5, I6, I7 ); | |
505 subc( I7, -1, I6 ); | |
506 subccc( I5, I6, I7 ); | |
507 subccc( I7, -1, I6 ); | |
508 | |
509 swap( I5, I6, I7 ); | |
510 swap( I7, -1, I6 ); | |
511 | |
512 swapa( G0, G1, 13, G2 ); | |
513 swapa( I7, -1, I6 ); | |
514 | |
515 taddcc( I5, I6, I7 ); | |
516 taddcc( I7, -1, I6 ); | |
517 taddcctv( I5, I6, I7 ); | |
518 taddcctv( I7, -1, I6 ); | |
519 | |
520 tsubcc( I5, I6, I7 ); | |
521 tsubcc( I7, -1, I6 ); | |
522 tsubcctv( I5, I6, I7 ); | |
523 tsubcctv( I7, -1, I6 ); | |
524 | |
525 trap( overflowClear, xcc, G0, G1 ); | |
526 trap( lessEqual, icc, I7, 17 ); | |
527 | |
528 bind(lbl2); | |
529 bind(lbl3); | |
530 | |
531 code()->decode(); | |
532 } | |
533 | |
534 // Generate a bunch 'o stuff unique to V8 | |
535 void Assembler::test_v8_onlys() { | |
536 Label lbl1; | |
537 | |
538 cb( cp_0or1or2, false, pc() - 4, relocInfo::none); | |
539 delayed()->nop(); | |
540 cb( cp_never, true, lbl1); | |
541 delayed()->nop(); | |
542 | |
543 cpop1(1, 2, 3, 4); | |
544 cpop2(5, 6, 7, 8); | |
545 | |
546 ldc( I0, I1, 31); | |
547 ldc( I2, -1, 0); | |
548 | |
549 lddc( I4, I4, 30); | |
550 lddc( I6, 0, 1 ); | |
551 | |
552 ldcsr( L0, L1, 0); | |
553 ldcsr( L1, (1 << 12) - 1, 17 ); | |
554 | |
555 stc( 31, L4, L5); | |
556 stc( 30, L6, -(1 << 12) ); | |
557 | |
558 stdc( 0, L7, G0); | |
559 stdc( 1, G1, 0 ); | |
560 | |
561 stcsr( 16, G2, G3); | |
562 stcsr( 17, G4, 1 ); | |
563 | |
564 stdcq( 4, G5, G6); | |
565 stdcq( 5, G7, -1 ); | |
566 | |
567 bind(lbl1); | |
568 | |
569 code()->decode(); | |
570 } | |
571 #endif | |
572 | |
573 // Implementation of MacroAssembler | |
574 | |
575 void MacroAssembler::null_check(Register reg, int offset) { | |
576 if (needs_explicit_null_check((intptr_t)offset)) { | |
577 // provoke OS NULL exception if reg = NULL by | |
578 // accessing M[reg] w/o changing any registers | |
579 ld_ptr(reg, 0, G0); | |
580 } | |
581 else { | |
582 // nothing to do, (later) access of M[reg + offset] | |
583 // will provoke OS NULL exception if reg = NULL | |
584 } | |
585 } | |
586 | |
587 // Ring buffer jumps | |
588 | |
589 #ifndef PRODUCT | |
590 void MacroAssembler::ret( bool trace ) { if (trace) { | |
591 mov(I7, O7); // traceable register | |
592 JMP(O7, 2 * BytesPerInstWord); | |
593 } else { | |
594 jmpl( I7, 2 * BytesPerInstWord, G0 ); | |
595 } | |
596 } | |
597 | |
598 void MacroAssembler::retl( bool trace ) { if (trace) JMP(O7, 2 * BytesPerInstWord); | |
599 else jmpl( O7, 2 * BytesPerInstWord, G0 ); } | |
600 #endif /* PRODUCT */ | |
601 | |
602 | |
603 void MacroAssembler::jmp2(Register r1, Register r2, const char* file, int line ) { | |
604 assert_not_delayed(); | |
605 // This can only be traceable if r1 & r2 are visible after a window save | |
606 if (TraceJumps) { | |
607 #ifndef PRODUCT | |
608 save_frame(0); | |
609 verify_thread(); | |
610 ld(G2_thread, in_bytes(JavaThread::jmp_ring_index_offset()), O0); | |
611 add(G2_thread, in_bytes(JavaThread::jmp_ring_offset()), O1); | |
612 sll(O0, exact_log2(4*sizeof(intptr_t)), O2); | |
613 add(O2, O1, O1); | |
614 | |
615 add(r1->after_save(), r2->after_save(), O2); | |
616 set((intptr_t)file, O3); | |
617 set(line, O4); | |
618 Label L; | |
619 // get nearby pc, store jmp target | |
620 call(L, relocInfo::none); // No relocation for call to pc+0x8 | |
621 delayed()->st(O2, O1, 0); | |
622 bind(L); | |
623 | |
624 // store nearby pc | |
625 st(O7, O1, sizeof(intptr_t)); | |
626 // store file | |
627 st(O3, O1, 2*sizeof(intptr_t)); | |
628 // store line | |
629 st(O4, O1, 3*sizeof(intptr_t)); | |
630 add(O0, 1, O0); | |
631 and3(O0, JavaThread::jump_ring_buffer_size - 1, O0); | |
632 st(O0, G2_thread, in_bytes(JavaThread::jmp_ring_index_offset())); | |
633 restore(); | |
634 #endif /* PRODUCT */ | |
635 } | |
636 jmpl(r1, r2, G0); | |
637 } | |
638 void MacroAssembler::jmp(Register r1, int offset, const char* file, int line ) { | |
639 assert_not_delayed(); | |
640 // This can only be traceable if r1 is visible after a window save | |
641 if (TraceJumps) { | |
642 #ifndef PRODUCT | |
643 save_frame(0); | |
644 verify_thread(); | |
645 ld(G2_thread, in_bytes(JavaThread::jmp_ring_index_offset()), O0); | |
646 add(G2_thread, in_bytes(JavaThread::jmp_ring_offset()), O1); | |
647 sll(O0, exact_log2(4*sizeof(intptr_t)), O2); | |
648 add(O2, O1, O1); | |
649 | |
650 add(r1->after_save(), offset, O2); | |
651 set((intptr_t)file, O3); | |
652 set(line, O4); | |
653 Label L; | |
654 // get nearby pc, store jmp target | |
655 call(L, relocInfo::none); // No relocation for call to pc+0x8 | |
656 delayed()->st(O2, O1, 0); | |
657 bind(L); | |
658 | |
659 // store nearby pc | |
660 st(O7, O1, sizeof(intptr_t)); | |
661 // store file | |
662 st(O3, O1, 2*sizeof(intptr_t)); | |
663 // store line | |
664 st(O4, O1, 3*sizeof(intptr_t)); | |
665 add(O0, 1, O0); | |
666 and3(O0, JavaThread::jump_ring_buffer_size - 1, O0); | |
667 st(O0, G2_thread, in_bytes(JavaThread::jmp_ring_index_offset())); | |
668 restore(); | |
669 #endif /* PRODUCT */ | |
670 } | |
671 jmp(r1, offset); | |
672 } | |
673 | |
674 // This code sequence is relocatable to any address, even on LP64. | |
1680
a64438a2b7e8
6958465: Sparc aten build24.0: openjdk-7.ea-b96 failed Error: Formal argument ... requires an lvalue
coleenp
parents:
1552
diff
changeset
|
675 void MacroAssembler::jumpl(const AddressLiteral& addrlit, Register temp, Register d, int offset, const char* file, int line) { |
0 | 676 assert_not_delayed(); |
677 // Force fixed length sethi because NativeJump and NativeFarCall don't handle | |
678 // variable length instruction streams. | |
727 | 679 patchable_sethi(addrlit, temp); |
680 Address a(temp, addrlit.low10() + offset); // Add the offset to the displacement. | |
0 | 681 if (TraceJumps) { |
682 #ifndef PRODUCT | |
683 // Must do the add here so relocation can find the remainder of the | |
684 // value to be relocated. | |
727 | 685 add(a.base(), a.disp(), a.base(), addrlit.rspec(offset)); |
0 | 686 save_frame(0); |
687 verify_thread(); | |
688 ld(G2_thread, in_bytes(JavaThread::jmp_ring_index_offset()), O0); | |
689 add(G2_thread, in_bytes(JavaThread::jmp_ring_offset()), O1); | |
690 sll(O0, exact_log2(4*sizeof(intptr_t)), O2); | |
691 add(O2, O1, O1); | |
692 | |
693 set((intptr_t)file, O3); | |
694 set(line, O4); | |
695 Label L; | |
696 | |
697 // get nearby pc, store jmp target | |
698 call(L, relocInfo::none); // No relocation for call to pc+0x8 | |
699 delayed()->st(a.base()->after_save(), O1, 0); | |
700 bind(L); | |
701 | |
702 // store nearby pc | |
703 st(O7, O1, sizeof(intptr_t)); | |
704 // store file | |
705 st(O3, O1, 2*sizeof(intptr_t)); | |
706 // store line | |
707 st(O4, O1, 3*sizeof(intptr_t)); | |
708 add(O0, 1, O0); | |
709 and3(O0, JavaThread::jump_ring_buffer_size - 1, O0); | |
710 st(O0, G2_thread, in_bytes(JavaThread::jmp_ring_index_offset())); | |
711 restore(); | |
712 jmpl(a.base(), G0, d); | |
713 #else | |
727 | 714 jmpl(a.base(), a.disp(), d); |
0 | 715 #endif /* PRODUCT */ |
716 } else { | |
727 | 717 jmpl(a.base(), a.disp(), d); |
0 | 718 } |
719 } | |
720 | |
1680
a64438a2b7e8
6958465: Sparc aten build24.0: openjdk-7.ea-b96 failed Error: Formal argument ... requires an lvalue
coleenp
parents:
1552
diff
changeset
|
721 void MacroAssembler::jump(const AddressLiteral& addrlit, Register temp, int offset, const char* file, int line) { |
727 | 722 jumpl(addrlit, temp, G0, offset, file, line); |
0 | 723 } |
724 | |
725 | |
726 // Convert to C varargs format | |
727 void MacroAssembler::set_varargs( Argument inArg, Register d ) { | |
728 // spill register-resident args to their memory slots | |
729 // (SPARC calling convention requires callers to have already preallocated these) | |
730 // Note that the inArg might in fact be an outgoing argument, | |
731 // if a leaf routine or stub does some tricky argument shuffling. | |
732 // This routine must work even though one of the saved arguments | |
733 // is in the d register (e.g., set_varargs(Argument(0, false), O0)). | |
734 for (Argument savePtr = inArg; | |
735 savePtr.is_register(); | |
736 savePtr = savePtr.successor()) { | |
737 st_ptr(savePtr.as_register(), savePtr.address_in_frame()); | |
738 } | |
739 // return the address of the first memory slot | |
727 | 740 Address a = inArg.address_in_frame(); |
741 add(a.base(), a.disp(), d); | |
0 | 742 } |
743 | |
744 // Conditional breakpoint (for assertion checks in assembly code) | |
745 void MacroAssembler::breakpoint_trap(Condition c, CC cc) { | |
746 trap(c, cc, G0, ST_RESERVED_FOR_USER_0); | |
747 } | |
748 | |
749 // We want to use ST_BREAKPOINT here, but the debugger is confused by it. | |
750 void MacroAssembler::breakpoint_trap() { | |
751 trap(ST_RESERVED_FOR_USER_0); | |
752 } | |
753 | |
754 // flush windows (except current) using flushw instruction if avail. | |
755 void MacroAssembler::flush_windows() { | |
756 if (VM_Version::v9_instructions_work()) flushw(); | |
757 else flush_windows_trap(); | |
758 } | |
759 | |
760 // Write serialization page so VM thread can do a pseudo remote membar | |
761 // We use the current thread pointer to calculate a thread specific | |
762 // offset to write to within the page. This minimizes bus traffic | |
763 // due to cache line collision. | |
764 void MacroAssembler::serialize_memory(Register thread, Register tmp1, Register tmp2) { | |
765 srl(thread, os::get_serialize_page_shift_count(), tmp2); | |
766 if (Assembler::is_simm13(os::vm_page_size())) { | |
767 and3(tmp2, (os::vm_page_size() - sizeof(int)), tmp2); | |
768 } | |
769 else { | |
770 set((os::vm_page_size() - sizeof(int)), tmp1); | |
771 and3(tmp2, tmp1, tmp2); | |
772 } | |
727 | 773 set(os::get_memory_serialize_page(), tmp1); |
0 | 774 st(G0, tmp1, tmp2); |
775 } | |
776 | |
777 | |
778 | |
779 void MacroAssembler::enter() { | |
780 Unimplemented(); | |
781 } | |
782 | |
783 void MacroAssembler::leave() { | |
784 Unimplemented(); | |
785 } | |
786 | |
787 void MacroAssembler::mult(Register s1, Register s2, Register d) { | |
788 if(VM_Version::v9_instructions_work()) { | |
789 mulx (s1, s2, d); | |
790 } else { | |
791 smul (s1, s2, d); | |
792 } | |
793 } | |
794 | |
795 void MacroAssembler::mult(Register s1, int simm13a, Register d) { | |
796 if(VM_Version::v9_instructions_work()) { | |
797 mulx (s1, simm13a, d); | |
798 } else { | |
799 smul (s1, simm13a, d); | |
800 } | |
801 } | |
802 | |
803 | |
804 #ifdef ASSERT | |
805 void MacroAssembler::read_ccr_v8_assert(Register ccr_save) { | |
806 const Register s1 = G3_scratch; | |
807 const Register s2 = G4_scratch; | |
808 Label get_psr_test; | |
809 // Get the condition codes the V8 way. | |
810 read_ccr_trap(s1); | |
811 mov(ccr_save, s2); | |
812 // This is a test of V8 which has icc but not xcc | |
813 // so mask off the xcc bits | |
814 and3(s2, 0xf, s2); | |
815 // Compare condition codes from the V8 and V9 ways. | |
816 subcc(s2, s1, G0); | |
817 br(Assembler::notEqual, true, Assembler::pt, get_psr_test); | |
818 delayed()->breakpoint_trap(); | |
819 bind(get_psr_test); | |
820 } | |
821 | |
822 void MacroAssembler::write_ccr_v8_assert(Register ccr_save) { | |
823 const Register s1 = G3_scratch; | |
824 const Register s2 = G4_scratch; | |
825 Label set_psr_test; | |
826 // Write out the saved condition codes the V8 way | |
827 write_ccr_trap(ccr_save, s1, s2); | |
828 // Read back the condition codes using the V9 instruction | |
829 rdccr(s1); | |
830 mov(ccr_save, s2); | |
831 // This is a test of V8 which has icc but not xcc | |
832 // so mask off the xcc bits | |
833 and3(s2, 0xf, s2); | |
834 and3(s1, 0xf, s1); | |
835 // Compare the V8 way with the V9 way. | |
836 subcc(s2, s1, G0); | |
837 br(Assembler::notEqual, true, Assembler::pt, set_psr_test); | |
838 delayed()->breakpoint_trap(); | |
839 bind(set_psr_test); | |
840 } | |
841 #else | |
842 #define read_ccr_v8_assert(x) | |
843 #define write_ccr_v8_assert(x) | |
844 #endif // ASSERT | |
845 | |
846 void MacroAssembler::read_ccr(Register ccr_save) { | |
847 if (VM_Version::v9_instructions_work()) { | |
848 rdccr(ccr_save); | |
849 // Test code sequence used on V8. Do not move above rdccr. | |
850 read_ccr_v8_assert(ccr_save); | |
851 } else { | |
852 read_ccr_trap(ccr_save); | |
853 } | |
854 } | |
855 | |
856 void MacroAssembler::write_ccr(Register ccr_save) { | |
857 if (VM_Version::v9_instructions_work()) { | |
858 // Test code sequence used on V8. Do not move below wrccr. | |
859 write_ccr_v8_assert(ccr_save); | |
860 wrccr(ccr_save); | |
861 } else { | |
862 const Register temp_reg1 = G3_scratch; | |
863 const Register temp_reg2 = G4_scratch; | |
864 write_ccr_trap(ccr_save, temp_reg1, temp_reg2); | |
865 } | |
866 } | |
867 | |
868 | |
869 // Calls to C land | |
870 | |
871 #ifdef ASSERT | |
872 // a hook for debugging | |
873 static Thread* reinitialize_thread() { | |
874 return ThreadLocalStorage::thread(); | |
875 } | |
876 #else | |
877 #define reinitialize_thread ThreadLocalStorage::thread | |
878 #endif | |
879 | |
880 #ifdef ASSERT | |
881 address last_get_thread = NULL; | |
882 #endif | |
883 | |
884 // call this when G2_thread is not known to be valid | |
885 void MacroAssembler::get_thread() { | |
886 save_frame(0); // to avoid clobbering O0 | |
887 mov(G1, L0); // avoid clobbering G1 | |
888 mov(G5_method, L1); // avoid clobbering G5 | |
889 mov(G3, L2); // avoid clobbering G3 also | |
890 mov(G4, L5); // avoid clobbering G4 | |
891 #ifdef ASSERT | |
727 | 892 AddressLiteral last_get_thread_addrlit(&last_get_thread); |
893 set(last_get_thread_addrlit, L3); | |
0 | 894 inc(L4, get_pc(L4) + 2 * BytesPerInstWord); // skip getpc() code + inc + st_ptr to point L4 at call |
727 | 895 st_ptr(L4, L3, 0); |
0 | 896 #endif |
897 call(CAST_FROM_FN_PTR(address, reinitialize_thread), relocInfo::runtime_call_type); | |
898 delayed()->nop(); | |
899 mov(L0, G1); | |
900 mov(L1, G5_method); | |
901 mov(L2, G3); | |
902 mov(L5, G4); | |
903 restore(O0, 0, G2_thread); | |
904 } | |
905 | |
906 static Thread* verify_thread_subroutine(Thread* gthread_value) { | |
907 Thread* correct_value = ThreadLocalStorage::thread(); | |
908 guarantee(gthread_value == correct_value, "G2_thread value must be the thread"); | |
909 return correct_value; | |
910 } | |
911 | |
912 void MacroAssembler::verify_thread() { | |
913 if (VerifyThread) { | |
914 // NOTE: this chops off the heads of the 64-bit O registers. | |
915 #ifdef CC_INTERP | |
916 save_frame(0); | |
917 #else | |
918 // make sure G2_thread contains the right value | |
919 save_frame_and_mov(0, Lmethod, Lmethod); // to avoid clobbering O0 (and propagate Lmethod for -Xprof) | |
920 mov(G1, L1); // avoid clobbering G1 | |
921 // G2 saved below | |
922 mov(G3, L3); // avoid clobbering G3 | |
923 mov(G4, L4); // avoid clobbering G4 | |
924 mov(G5_method, L5); // avoid clobbering G5_method | |
925 #endif /* CC_INTERP */ | |
926 #if defined(COMPILER2) && !defined(_LP64) | |
927 // Save & restore possible 64-bit Long arguments in G-regs | |
928 srlx(G1,32,L0); | |
929 srlx(G4,32,L6); | |
930 #endif | |
931 call(CAST_FROM_FN_PTR(address,verify_thread_subroutine), relocInfo::runtime_call_type); | |
932 delayed()->mov(G2_thread, O0); | |
933 | |
934 mov(L1, G1); // Restore G1 | |
935 // G2 restored below | |
936 mov(L3, G3); // restore G3 | |
937 mov(L4, G4); // restore G4 | |
938 mov(L5, G5_method); // restore G5_method | |
939 #if defined(COMPILER2) && !defined(_LP64) | |
940 // Save & restore possible 64-bit Long arguments in G-regs | |
941 sllx(L0,32,G2); // Move old high G1 bits high in G2 | |
2002 | 942 srl(G1, 0,G1); // Clear current high G1 bits |
0 | 943 or3 (G1,G2,G1); // Recover 64-bit G1 |
944 sllx(L6,32,G2); // Move old high G4 bits high in G2 | |
2002 | 945 srl(G4, 0,G4); // Clear current high G4 bits |
0 | 946 or3 (G4,G2,G4); // Recover 64-bit G4 |
947 #endif | |
948 restore(O0, 0, G2_thread); | |
949 } | |
950 } | |
951 | |
952 | |
953 void MacroAssembler::save_thread(const Register thread_cache) { | |
954 verify_thread(); | |
955 if (thread_cache->is_valid()) { | |
956 assert(thread_cache->is_local() || thread_cache->is_in(), "bad volatile"); | |
957 mov(G2_thread, thread_cache); | |
958 } | |
959 if (VerifyThread) { | |
960 // smash G2_thread, as if the VM were about to anyway | |
961 set(0x67676767, G2_thread); | |
962 } | |
963 } | |
964 | |
965 | |
966 void MacroAssembler::restore_thread(const Register thread_cache) { | |
967 if (thread_cache->is_valid()) { | |
968 assert(thread_cache->is_local() || thread_cache->is_in(), "bad volatile"); | |
969 mov(thread_cache, G2_thread); | |
970 verify_thread(); | |
971 } else { | |
972 // do it the slow way | |
973 get_thread(); | |
974 } | |
975 } | |
976 | |
977 | |
978 // %%% maybe get rid of [re]set_last_Java_frame | |
979 void MacroAssembler::set_last_Java_frame(Register last_java_sp, Register last_Java_pc) { | |
980 assert_not_delayed(); | |
727 | 981 Address flags(G2_thread, JavaThread::frame_anchor_offset() + |
982 JavaFrameAnchor::flags_offset()); | |
983 Address pc_addr(G2_thread, JavaThread::last_Java_pc_offset()); | |
0 | 984 |
985 // Always set last_Java_pc and flags first because once last_Java_sp is visible | |
986 // has_last_Java_frame is true and users will look at the rest of the fields. | |
987 // (Note: flags should always be zero before we get here so doesn't need to be set.) | |
988 | |
989 #ifdef ASSERT | |
990 // Verify that flags was zeroed on return to Java | |
991 Label PcOk; | |
992 save_frame(0); // to avoid clobbering O0 | |
993 ld_ptr(pc_addr, L0); | |
3839 | 994 br_null_short(L0, Assembler::pt, PcOk); |
0 | 995 stop("last_Java_pc not zeroed before leaving Java"); |
996 bind(PcOk); | |
997 | |
998 // Verify that flags was zeroed on return to Java | |
999 Label FlagsOk; | |
1000 ld(flags, L0); | |
1001 tst(L0); | |
1002 br(Assembler::zero, false, Assembler::pt, FlagsOk); | |
1003 delayed() -> restore(); | |
1004 stop("flags not zeroed before leaving Java"); | |
1005 bind(FlagsOk); | |
1006 #endif /* ASSERT */ | |
1007 // | |
1008 // When returning from calling out from Java mode the frame anchor's last_Java_pc | |
1009 // will always be set to NULL. It is set here so that if we are doing a call to | |
1010 // native (not VM) that we capture the known pc and don't have to rely on the | |
1011 // native call having a standard frame linkage where we can find the pc. | |
1012 | |
1013 if (last_Java_pc->is_valid()) { | |
1014 st_ptr(last_Java_pc, pc_addr); | |
1015 } | |
1016 | |
1017 #ifdef _LP64 | |
1018 #ifdef ASSERT | |
1019 // Make sure that we have an odd stack | |
1020 Label StackOk; | |
1021 andcc(last_java_sp, 0x01, G0); | |
1022 br(Assembler::notZero, false, Assembler::pt, StackOk); | |
3839 | 1023 delayed()->nop(); |
0 | 1024 stop("Stack Not Biased in set_last_Java_frame"); |
1025 bind(StackOk); | |
1026 #endif // ASSERT | |
1027 assert( last_java_sp != G4_scratch, "bad register usage in set_last_Java_frame"); | |
1028 add( last_java_sp, STACK_BIAS, G4_scratch ); | |
727 | 1029 st_ptr(G4_scratch, G2_thread, JavaThread::last_Java_sp_offset()); |
0 | 1030 #else |
727 | 1031 st_ptr(last_java_sp, G2_thread, JavaThread::last_Java_sp_offset()); |
0 | 1032 #endif // _LP64 |
1033 } | |
1034 | |
1035 void MacroAssembler::reset_last_Java_frame(void) { | |
1036 assert_not_delayed(); | |
1037 | |
727 | 1038 Address sp_addr(G2_thread, JavaThread::last_Java_sp_offset()); |
1039 Address pc_addr(G2_thread, JavaThread::frame_anchor_offset() + JavaFrameAnchor::last_Java_pc_offset()); | |
1040 Address flags (G2_thread, JavaThread::frame_anchor_offset() + JavaFrameAnchor::flags_offset()); | |
0 | 1041 |
1042 #ifdef ASSERT | |
1043 // check that it WAS previously set | |
1044 #ifdef CC_INTERP | |
1045 save_frame(0); | |
1046 #else | |
1047 save_frame_and_mov(0, Lmethod, Lmethod); // Propagate Lmethod to helper frame for -Xprof | |
1048 #endif /* CC_INTERP */ | |
1049 ld_ptr(sp_addr, L0); | |
1050 tst(L0); | |
1051 breakpoint_trap(Assembler::zero, Assembler::ptr_cc); | |
1052 restore(); | |
1053 #endif // ASSERT | |
1054 | |
1055 st_ptr(G0, sp_addr); | |
1056 // Always return last_Java_pc to zero | |
1057 st_ptr(G0, pc_addr); | |
1058 // Always null flags after return to Java | |
1059 st(G0, flags); | |
1060 } | |
1061 | |
1062 | |
1063 void MacroAssembler::call_VM_base( | |
1064 Register oop_result, | |
1065 Register thread_cache, | |
1066 Register last_java_sp, | |
1067 address entry_point, | |
1068 int number_of_arguments, | |
1069 bool check_exceptions) | |
1070 { | |
1071 assert_not_delayed(); | |
1072 | |
1073 // determine last_java_sp register | |
1074 if (!last_java_sp->is_valid()) { | |
1075 last_java_sp = SP; | |
1076 } | |
1077 // debugging support | |
1078 assert(number_of_arguments >= 0 , "cannot have negative number of arguments"); | |
1079 | |
1080 // 64-bit last_java_sp is biased! | |
1081 set_last_Java_frame(last_java_sp, noreg); | |
1082 if (VerifyThread) mov(G2_thread, O0); // about to be smashed; pass early | |
1083 save_thread(thread_cache); | |
1084 // do the call | |
1085 call(entry_point, relocInfo::runtime_call_type); | |
1086 if (!VerifyThread) | |
1087 delayed()->mov(G2_thread, O0); // pass thread as first argument | |
1088 else | |
1089 delayed()->nop(); // (thread already passed) | |
1090 restore_thread(thread_cache); | |
1091 reset_last_Java_frame(); | |
1092 | |
1093 // check for pending exceptions. use Gtemp as scratch register. | |
1094 if (check_exceptions) { | |
1095 check_and_forward_exception(Gtemp); | |
1096 } | |
1097 | |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1098 #ifdef ASSERT |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1099 set(badHeapWordVal, G3); |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1100 set(badHeapWordVal, G4); |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1101 set(badHeapWordVal, G5); |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1102 #endif |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1103 |
0 | 1104 // get oop result if there is one and reset the value in the thread |
1105 if (oop_result->is_valid()) { | |
1106 get_vm_result(oop_result); | |
1107 } | |
1108 } | |
1109 | |
1110 void MacroAssembler::check_and_forward_exception(Register scratch_reg) | |
1111 { | |
1112 Label L; | |
1113 | |
1114 check_and_handle_popframe(scratch_reg); | |
1115 check_and_handle_earlyret(scratch_reg); | |
1116 | |
727 | 1117 Address exception_addr(G2_thread, Thread::pending_exception_offset()); |
0 | 1118 ld_ptr(exception_addr, scratch_reg); |
3839 | 1119 br_null_short(scratch_reg, pt, L); |
0 | 1120 // we use O7 linkage so that forward_exception_entry has the issuing PC |
1121 call(StubRoutines::forward_exception_entry(), relocInfo::runtime_call_type); | |
1122 delayed()->nop(); | |
1123 bind(L); | |
1124 } | |
1125 | |
1126 | |
1127 void MacroAssembler::check_and_handle_popframe(Register scratch_reg) { | |
1128 } | |
1129 | |
1130 | |
1131 void MacroAssembler::check_and_handle_earlyret(Register scratch_reg) { | |
1132 } | |
1133 | |
1134 | |
1135 void MacroAssembler::call_VM(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions) { | |
1136 call_VM_base(oop_result, noreg, noreg, entry_point, number_of_arguments, check_exceptions); | |
1137 } | |
1138 | |
1139 | |
1140 void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) { | |
1141 // O0 is reserved for the thread | |
1142 mov(arg_1, O1); | |
1143 call_VM(oop_result, entry_point, 1, check_exceptions); | |
1144 } | |
1145 | |
1146 | |
1147 void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) { | |
1148 // O0 is reserved for the thread | |
1149 mov(arg_1, O1); | |
1150 mov(arg_2, O2); assert(arg_2 != O1, "smashed argument"); | |
1151 call_VM(oop_result, entry_point, 2, check_exceptions); | |
1152 } | |
1153 | |
1154 | |
1155 void MacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) { | |
1156 // O0 is reserved for the thread | |
1157 mov(arg_1, O1); | |
1158 mov(arg_2, O2); assert(arg_2 != O1, "smashed argument"); | |
1159 mov(arg_3, O3); assert(arg_3 != O1 && arg_3 != O2, "smashed argument"); | |
1160 call_VM(oop_result, entry_point, 3, check_exceptions); | |
1161 } | |
1162 | |
1163 | |
1164 | |
1165 // Note: The following call_VM overloadings are useful when a "save" | |
1166 // has already been performed by a stub, and the last Java frame is | |
1167 // the previous one. In that case, last_java_sp must be passed as FP | |
1168 // instead of SP. | |
1169 | |
1170 | |
1171 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments, bool check_exceptions) { | |
1172 call_VM_base(oop_result, noreg, last_java_sp, entry_point, number_of_arguments, check_exceptions); | |
1173 } | |
1174 | |
1175 | |
1176 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions) { | |
1177 // O0 is reserved for the thread | |
1178 mov(arg_1, O1); | |
1179 call_VM(oop_result, last_java_sp, entry_point, 1, check_exceptions); | |
1180 } | |
1181 | |
1182 | |
1183 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) { | |
1184 // O0 is reserved for the thread | |
1185 mov(arg_1, O1); | |
1186 mov(arg_2, O2); assert(arg_2 != O1, "smashed argument"); | |
1187 call_VM(oop_result, last_java_sp, entry_point, 2, check_exceptions); | |
1188 } | |
1189 | |
1190 | |
1191 void MacroAssembler::call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) { | |
1192 // O0 is reserved for the thread | |
1193 mov(arg_1, O1); | |
1194 mov(arg_2, O2); assert(arg_2 != O1, "smashed argument"); | |
1195 mov(arg_3, O3); assert(arg_3 != O1 && arg_3 != O2, "smashed argument"); | |
1196 call_VM(oop_result, last_java_sp, entry_point, 3, check_exceptions); | |
1197 } | |
1198 | |
1199 | |
1200 | |
1201 void MacroAssembler::call_VM_leaf_base(Register thread_cache, address entry_point, int number_of_arguments) { | |
1202 assert_not_delayed(); | |
1203 save_thread(thread_cache); | |
1204 // do the call | |
1205 call(entry_point, relocInfo::runtime_call_type); | |
1206 delayed()->nop(); | |
1207 restore_thread(thread_cache); | |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1208 #ifdef ASSERT |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1209 set(badHeapWordVal, G3); |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1210 set(badHeapWordVal, G4); |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1211 set(badHeapWordVal, G5); |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1212 #endif |
0 | 1213 } |
1214 | |
1215 | |
1216 void MacroAssembler::call_VM_leaf(Register thread_cache, address entry_point, int number_of_arguments) { | |
1217 call_VM_leaf_base(thread_cache, entry_point, number_of_arguments); | |
1218 } | |
1219 | |
1220 | |
1221 void MacroAssembler::call_VM_leaf(Register thread_cache, address entry_point, Register arg_1) { | |
1222 mov(arg_1, O0); | |
1223 call_VM_leaf(thread_cache, entry_point, 1); | |
1224 } | |
1225 | |
1226 | |
1227 void MacroAssembler::call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2) { | |
1228 mov(arg_1, O0); | |
1229 mov(arg_2, O1); assert(arg_2 != O0, "smashed argument"); | |
1230 call_VM_leaf(thread_cache, entry_point, 2); | |
1231 } | |
1232 | |
1233 | |
1234 void MacroAssembler::call_VM_leaf(Register thread_cache, address entry_point, Register arg_1, Register arg_2, Register arg_3) { | |
1235 mov(arg_1, O0); | |
1236 mov(arg_2, O1); assert(arg_2 != O0, "smashed argument"); | |
1237 mov(arg_3, O2); assert(arg_3 != O0 && arg_3 != O1, "smashed argument"); | |
1238 call_VM_leaf(thread_cache, entry_point, 3); | |
1239 } | |
1240 | |
1241 | |
1242 void MacroAssembler::get_vm_result(Register oop_result) { | |
1243 verify_thread(); | |
727 | 1244 Address vm_result_addr(G2_thread, JavaThread::vm_result_offset()); |
0 | 1245 ld_ptr( vm_result_addr, oop_result); |
1246 st_ptr(G0, vm_result_addr); | |
1247 verify_oop(oop_result); | |
1248 } | |
1249 | |
1250 | |
1251 void MacroAssembler::get_vm_result_2(Register oop_result) { | |
1252 verify_thread(); | |
727 | 1253 Address vm_result_addr_2(G2_thread, JavaThread::vm_result_2_offset()); |
0 | 1254 ld_ptr(vm_result_addr_2, oop_result); |
1255 st_ptr(G0, vm_result_addr_2); | |
1256 verify_oop(oop_result); | |
1257 } | |
1258 | |
1259 | |
1260 // We require that C code which does not return a value in vm_result will | |
1261 // leave it undisturbed. | |
1262 void MacroAssembler::set_vm_result(Register oop_result) { | |
1263 verify_thread(); | |
727 | 1264 Address vm_result_addr(G2_thread, JavaThread::vm_result_offset()); |
0 | 1265 verify_oop(oop_result); |
1266 | |
1267 # ifdef ASSERT | |
1268 // Check that we are not overwriting any other oop. | |
1269 #ifdef CC_INTERP | |
1270 save_frame(0); | |
1271 #else | |
1272 save_frame_and_mov(0, Lmethod, Lmethod); // Propagate Lmethod for -Xprof | |
1273 #endif /* CC_INTERP */ | |
1274 ld_ptr(vm_result_addr, L0); | |
1275 tst(L0); | |
1276 restore(); | |
1277 breakpoint_trap(notZero, Assembler::ptr_cc); | |
1278 // } | |
1279 # endif | |
1280 | |
1281 st_ptr(oop_result, vm_result_addr); | |
1282 } | |
1283 | |
1284 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1285 void MacroAssembler::card_table_write(jbyte* byte_map_base, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1286 Register tmp, Register obj) { |
0 | 1287 #ifdef _LP64 |
1288 srlx(obj, CardTableModRefBS::card_shift, obj); | |
1289 #else | |
1290 srl(obj, CardTableModRefBS::card_shift, obj); | |
1291 #endif | |
727 | 1292 assert(tmp != obj, "need separate temp reg"); |
1293 set((address) byte_map_base, tmp); | |
1294 stb(G0, tmp, obj); | |
0 | 1295 } |
1296 | |
727 | 1297 |
1298 void MacroAssembler::internal_sethi(const AddressLiteral& addrlit, Register d, bool ForceRelocatable) { | |
0 | 1299 address save_pc; |
1300 int shiftcnt; | |
1301 #ifdef _LP64 | |
1302 # ifdef CHECK_DELAY | |
727 | 1303 assert_not_delayed((char*) "cannot put two instructions in delay slot"); |
0 | 1304 # endif |
1305 v9_dep(); | |
1306 save_pc = pc(); | |
727 | 1307 |
1308 int msb32 = (int) (addrlit.value() >> 32); | |
1309 int lsb32 = (int) (addrlit.value()); | |
1310 | |
1311 if (msb32 == 0 && lsb32 >= 0) { | |
1312 Assembler::sethi(lsb32, d, addrlit.rspec()); | |
0 | 1313 } |
727 | 1314 else if (msb32 == -1) { |
1315 Assembler::sethi(~lsb32, d, addrlit.rspec()); | |
1316 xor3(d, ~low10(~0), d); | |
0 | 1317 } |
1318 else { | |
727 | 1319 Assembler::sethi(msb32, d, addrlit.rspec()); // msb 22-bits |
1320 if (msb32 & 0x3ff) // Any bits? | |
1321 or3(d, msb32 & 0x3ff, d); // msb 32-bits are now in lsb 32 | |
1322 if (lsb32 & 0xFFFFFC00) { // done? | |
1323 if ((lsb32 >> 20) & 0xfff) { // Any bits set? | |
1324 sllx(d, 12, d); // Make room for next 12 bits | |
1325 or3(d, (lsb32 >> 20) & 0xfff, d); // Or in next 12 | |
1326 shiftcnt = 0; // We already shifted | |
0 | 1327 } |
1328 else | |
1329 shiftcnt = 12; | |
727 | 1330 if ((lsb32 >> 10) & 0x3ff) { |
1331 sllx(d, shiftcnt + 10, d); // Make room for last 10 bits | |
1332 or3(d, (lsb32 >> 10) & 0x3ff, d); // Or in next 10 | |
0 | 1333 shiftcnt = 0; |
1334 } | |
1335 else | |
1336 shiftcnt = 10; | |
727 | 1337 sllx(d, shiftcnt + 10, d); // Shift leaving disp field 0'd |
0 | 1338 } |
1339 else | |
727 | 1340 sllx(d, 32, d); |
0 | 1341 } |
727 | 1342 // Pad out the instruction sequence so it can be patched later. |
1343 if (ForceRelocatable || (addrlit.rtype() != relocInfo::none && | |
1344 addrlit.rtype() != relocInfo::runtime_call_type)) { | |
1345 while (pc() < (save_pc + (7 * BytesPerInstWord))) | |
0 | 1346 nop(); |
1347 } | |
1348 #else | |
727 | 1349 Assembler::sethi(addrlit.value(), d, addrlit.rspec()); |
0 | 1350 #endif |
727 | 1351 } |
1352 | |
1353 | |
1354 void MacroAssembler::sethi(const AddressLiteral& addrlit, Register d) { | |
1355 internal_sethi(addrlit, d, false); | |
0 | 1356 } |
1357 | |
727 | 1358 |
1359 void MacroAssembler::patchable_sethi(const AddressLiteral& addrlit, Register d) { | |
1360 internal_sethi(addrlit, d, true); | |
1361 } | |
1362 | |
1363 | |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1364 int MacroAssembler::insts_for_sethi(address a, bool worst_case) { |
0 | 1365 #ifdef _LP64 |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1366 if (worst_case) return 7; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1367 intptr_t iaddr = (intptr_t) a; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1368 int msb32 = (int) (iaddr >> 32); |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1369 int lsb32 = (int) (iaddr); |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1370 int count; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1371 if (msb32 == 0 && lsb32 >= 0) |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1372 count = 1; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1373 else if (msb32 == -1) |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1374 count = 2; |
0 | 1375 else { |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1376 count = 2; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1377 if (msb32 & 0x3ff) |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1378 count++; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1379 if (lsb32 & 0xFFFFFC00 ) { |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1380 if ((lsb32 >> 20) & 0xfff) count += 2; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1381 if ((lsb32 >> 10) & 0x3ff) count += 2; |
0 | 1382 } |
1383 } | |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1384 return count; |
0 | 1385 #else |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1386 return 1; |
0 | 1387 #endif |
1388 } | |
1389 | |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1390 int MacroAssembler::worst_case_insts_for_set() { |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1391 return insts_for_sethi(NULL, true) + 1; |
0 | 1392 } |
1393 | |
727 | 1394 |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1395 // Keep in sync with MacroAssembler::insts_for_internal_set |
727 | 1396 void MacroAssembler::internal_set(const AddressLiteral& addrlit, Register d, bool ForceRelocatable) { |
1397 intptr_t value = addrlit.value(); | |
1398 | |
1399 if (!ForceRelocatable && addrlit.rspec().type() == relocInfo::none) { | |
0 | 1400 // can optimize |
727 | 1401 if (-4096 <= value && value <= 4095) { |
0 | 1402 or3(G0, value, d); // setsw (this leaves upper 32 bits sign-extended) |
1403 return; | |
1404 } | |
1405 if (inv_hi22(hi22(value)) == value) { | |
727 | 1406 sethi(addrlit, d); |
0 | 1407 return; |
1408 } | |
1409 } | |
727 | 1410 assert_not_delayed((char*) "cannot put two instructions in delay slot"); |
1411 internal_sethi(addrlit, d, ForceRelocatable); | |
1412 if (ForceRelocatable || addrlit.rspec().type() != relocInfo::none || addrlit.low10() != 0) { | |
1413 add(d, addrlit.low10(), d, addrlit.rspec()); | |
0 | 1414 } |
1415 } | |
1416 | |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1417 // Keep in sync with MacroAssembler::internal_set |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1418 int MacroAssembler::insts_for_internal_set(intptr_t value) { |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1419 // can optimize |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1420 if (-4096 <= value && value <= 4095) { |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1421 return 1; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1422 } |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1423 if (inv_hi22(hi22(value)) == value) { |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1424 return insts_for_sethi((address) value); |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1425 } |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1426 int count = insts_for_sethi((address) value); |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1427 AddressLiteral al(value); |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1428 if (al.low10() != 0) { |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1429 count++; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1430 } |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1431 return count; |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1432 } |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1433 |
727 | 1434 void MacroAssembler::set(const AddressLiteral& al, Register d) { |
1435 internal_set(al, d, false); | |
1436 } | |
1437 | |
1438 void MacroAssembler::set(intptr_t value, Register d) { | |
1439 AddressLiteral al(value); | |
1440 internal_set(al, d, false); | |
0 | 1441 } |
1442 | |
727 | 1443 void MacroAssembler::set(address addr, Register d, RelocationHolder const& rspec) { |
1444 AddressLiteral al(addr, rspec); | |
1445 internal_set(al, d, false); | |
1446 } | |
1447 | |
1448 void MacroAssembler::patchable_set(const AddressLiteral& al, Register d) { | |
1449 internal_set(al, d, true); | |
1450 } | |
1451 | |
1452 void MacroAssembler::patchable_set(intptr_t value, Register d) { | |
1453 AddressLiteral al(value); | |
1454 internal_set(al, d, true); | |
1455 } | |
0 | 1456 |
1457 | |
1458 void MacroAssembler::set64(jlong value, Register d, Register tmp) { | |
1459 assert_not_delayed(); | |
1460 v9_dep(); | |
1461 | |
1462 int hi = (int)(value >> 32); | |
1463 int lo = (int)(value & ~0); | |
1464 // (Matcher::isSimpleConstant64 knows about the following optimizations.) | |
1465 if (Assembler::is_simm13(lo) && value == lo) { | |
1466 or3(G0, lo, d); | |
1467 } else if (hi == 0) { | |
1468 Assembler::sethi(lo, d); // hardware version zero-extends to upper 32 | |
1469 if (low10(lo) != 0) | |
1470 or3(d, low10(lo), d); | |
1471 } | |
1472 else if (hi == -1) { | |
1473 Assembler::sethi(~lo, d); // hardware version zero-extends to upper 32 | |
1474 xor3(d, low10(lo) ^ ~low10(~0), d); | |
1475 } | |
1476 else if (lo == 0) { | |
1477 if (Assembler::is_simm13(hi)) { | |
1478 or3(G0, hi, d); | |
1479 } else { | |
1480 Assembler::sethi(hi, d); // hardware version zero-extends to upper 32 | |
1481 if (low10(hi) != 0) | |
1482 or3(d, low10(hi), d); | |
1483 } | |
1484 sllx(d, 32, d); | |
1485 } | |
1486 else { | |
1487 Assembler::sethi(hi, tmp); | |
1488 Assembler::sethi(lo, d); // macro assembler version sign-extends | |
1489 if (low10(hi) != 0) | |
1490 or3 (tmp, low10(hi), tmp); | |
1491 if (low10(lo) != 0) | |
1492 or3 ( d, low10(lo), d); | |
1493 sllx(tmp, 32, tmp); | |
1494 or3 (d, tmp, d); | |
1495 } | |
1496 } | |
1497 | |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1498 int MacroAssembler::insts_for_set64(jlong value) { |
2008 | 1499 v9_dep(); |
1500 | |
2076
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1501 int hi = (int) (value >> 32); |
7737fa7ec2b5
7006044: materialize cheap non-oop pointers on 64-bit SPARC
twisti
parents:
2008
diff
changeset
|
1502 int lo = (int) (value & ~0); |
2008 | 1503 int count = 0; |
1504 | |
1505 // (Matcher::isSimpleConstant64 knows about the following optimizations.) | |
1506 if (Assembler::is_simm13(lo) && value == lo) { | |
1507 count++; | |
1508 } else if (hi == 0) { | |
1509 count++; | |
1510 if (low10(lo) != 0) | |
1511 count++; | |
1512 } | |
1513 else if (hi == -1) { | |
1514 count += 2; | |
1515 } | |
1516 else if (lo == 0) { | |
1517 if (Assembler::is_simm13(hi)) { | |
1518 count++; | |
1519 } else { | |
1520 count++; | |
1521 if (low10(hi) != 0) | |
1522 count++; | |
1523 } | |
1524 count++; | |
1525 } | |
1526 else { | |
1527 count += 2; | |
1528 if (low10(hi) != 0) | |
1529 count++; | |
1530 if (low10(lo) != 0) | |
1531 count++; | |
1532 count += 2; | |
1533 } | |
1534 return count; | |
1535 } | |
1536 | |
0 | 1537 // compute size in bytes of sparc frame, given |
1538 // number of extraWords | |
1539 int MacroAssembler::total_frame_size_in_bytes(int extraWords) { | |
1540 | |
1541 int nWords = frame::memory_parameter_word_sp_offset; | |
1542 | |
1543 nWords += extraWords; | |
1544 | |
1545 if (nWords & 1) ++nWords; // round up to double-word | |
1546 | |
1547 return nWords * BytesPerWord; | |
1548 } | |
1549 | |
1550 | |
1551 // save_frame: given number of "extra" words in frame, | |
1552 // issue approp. save instruction (p 200, v8 manual) | |
1553 | |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1554 void MacroAssembler::save_frame(int extraWords) { |
0 | 1555 int delta = -total_frame_size_in_bytes(extraWords); |
1556 if (is_simm13(delta)) { | |
1557 save(SP, delta, SP); | |
1558 } else { | |
1559 set(delta, G3_scratch); | |
1560 save(SP, G3_scratch, SP); | |
1561 } | |
1562 } | |
1563 | |
1564 | |
1565 void MacroAssembler::save_frame_c1(int size_in_bytes) { | |
1566 if (is_simm13(-size_in_bytes)) { | |
1567 save(SP, -size_in_bytes, SP); | |
1568 } else { | |
1569 set(-size_in_bytes, G3_scratch); | |
1570 save(SP, G3_scratch, SP); | |
1571 } | |
1572 } | |
1573 | |
1574 | |
1575 void MacroAssembler::save_frame_and_mov(int extraWords, | |
1576 Register s1, Register d1, | |
1577 Register s2, Register d2) { | |
1578 assert_not_delayed(); | |
1579 | |
1580 // The trick here is to use precisely the same memory word | |
1581 // that trap handlers also use to save the register. | |
1582 // This word cannot be used for any other purpose, but | |
1583 // it works fine to save the register's value, whether or not | |
1584 // an interrupt flushes register windows at any given moment! | |
1585 Address s1_addr; | |
1586 if (s1->is_valid() && (s1->is_in() || s1->is_local())) { | |
1587 s1_addr = s1->address_in_saved_window(); | |
1588 st_ptr(s1, s1_addr); | |
1589 } | |
1590 | |
1591 Address s2_addr; | |
1592 if (s2->is_valid() && (s2->is_in() || s2->is_local())) { | |
1593 s2_addr = s2->address_in_saved_window(); | |
1594 st_ptr(s2, s2_addr); | |
1595 } | |
1596 | |
1597 save_frame(extraWords); | |
1598 | |
1599 if (s1_addr.base() == SP) { | |
1600 ld_ptr(s1_addr.after_save(), d1); | |
1601 } else if (s1->is_valid()) { | |
1602 mov(s1->after_save(), d1); | |
1603 } | |
1604 | |
1605 if (s2_addr.base() == SP) { | |
1606 ld_ptr(s2_addr.after_save(), d2); | |
1607 } else if (s2->is_valid()) { | |
1608 mov(s2->after_save(), d2); | |
1609 } | |
1610 } | |
1611 | |
1612 | |
727 | 1613 AddressLiteral MacroAssembler::allocate_oop_address(jobject obj) { |
0 | 1614 assert(oop_recorder() != NULL, "this assembler needs an OopRecorder"); |
1615 int oop_index = oop_recorder()->allocate_index(obj); | |
727 | 1616 return AddressLiteral(obj, oop_Relocation::spec(oop_index)); |
0 | 1617 } |
1618 | |
1619 | |
727 | 1620 AddressLiteral MacroAssembler::constant_oop_address(jobject obj) { |
0 | 1621 assert(oop_recorder() != NULL, "this assembler needs an OopRecorder"); |
1622 int oop_index = oop_recorder()->find_index(obj); | |
727 | 1623 return AddressLiteral(obj, oop_Relocation::spec(oop_index)); |
0 | 1624 } |
1625 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1626 void MacroAssembler::set_narrow_oop(jobject obj, Register d) { |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1627 assert(oop_recorder() != NULL, "this assembler needs an OopRecorder"); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1628 int oop_index = oop_recorder()->find_index(obj); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1629 RelocationHolder rspec = oop_Relocation::spec(oop_index); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1630 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1631 assert_not_delayed(); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1632 // Relocation with special format (see relocInfo_sparc.hpp). |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1633 relocate(rspec, 1); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1634 // Assembler::sethi(0x3fffff, d); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1635 emit_long( op(branch_op) | rd(d) | op2(sethi_op2) | hi22(0x3fffff) ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1636 // Don't add relocation for 'add'. Do patching during 'sethi' processing. |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1637 add(d, 0x3ff, d); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1638 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1639 } |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
1640 |
0 | 1641 |
1642 void MacroAssembler::align(int modulus) { | |
1643 while (offset() % modulus != 0) nop(); | |
1644 } | |
1645 | |
1646 | |
1647 void MacroAssembler::safepoint() { | |
1648 relocate(breakpoint_Relocation::spec(breakpoint_Relocation::safepoint)); | |
1649 } | |
1650 | |
1651 | |
1652 void RegistersForDebugging::print(outputStream* s) { | |
1653 int j; | |
1654 for ( j = 0; j < 8; ++j ) | |
1655 if ( j != 6 ) s->print_cr("i%d = 0x%.16lx", j, i[j]); | |
1656 else s->print_cr( "fp = 0x%.16lx", i[j]); | |
1657 s->cr(); | |
1658 | |
1659 for ( j = 0; j < 8; ++j ) | |
1660 s->print_cr("l%d = 0x%.16lx", j, l[j]); | |
1661 s->cr(); | |
1662 | |
1663 for ( j = 0; j < 8; ++j ) | |
1664 if ( j != 6 ) s->print_cr("o%d = 0x%.16lx", j, o[j]); | |
1665 else s->print_cr( "sp = 0x%.16lx", o[j]); | |
1666 s->cr(); | |
1667 | |
1668 for ( j = 0; j < 8; ++j ) | |
1669 s->print_cr("g%d = 0x%.16lx", j, g[j]); | |
1670 s->cr(); | |
1671 | |
1672 // print out floats with compression | |
1673 for (j = 0; j < 32; ) { | |
1674 jfloat val = f[j]; | |
1675 int last = j; | |
1676 for ( ; last+1 < 32; ++last ) { | |
1677 char b1[1024], b2[1024]; | |
1678 sprintf(b1, "%f", val); | |
1679 sprintf(b2, "%f", f[last+1]); | |
1680 if (strcmp(b1, b2)) | |
1681 break; | |
1682 } | |
1683 s->print("f%d", j); | |
1684 if ( j != last ) s->print(" - f%d", last); | |
1685 s->print(" = %f", val); | |
1686 s->fill_to(25); | |
1687 s->print_cr(" (0x%x)", val); | |
1688 j = last + 1; | |
1689 } | |
1690 s->cr(); | |
1691 | |
1692 // and doubles (evens only) | |
1693 for (j = 0; j < 32; ) { | |
1694 jdouble val = d[j]; | |
1695 int last = j; | |
1696 for ( ; last+1 < 32; ++last ) { | |
1697 char b1[1024], b2[1024]; | |
1698 sprintf(b1, "%f", val); | |
1699 sprintf(b2, "%f", d[last+1]); | |
1700 if (strcmp(b1, b2)) | |
1701 break; | |
1702 } | |
1703 s->print("d%d", 2 * j); | |
1704 if ( j != last ) s->print(" - d%d", last); | |
1705 s->print(" = %f", val); | |
1706 s->fill_to(30); | |
1707 s->print("(0x%x)", *(int*)&val); | |
1708 s->fill_to(42); | |
1709 s->print_cr("(0x%x)", *(1 + (int*)&val)); | |
1710 j = last + 1; | |
1711 } | |
1712 s->cr(); | |
1713 } | |
1714 | |
1715 void RegistersForDebugging::save_registers(MacroAssembler* a) { | |
1716 a->sub(FP, round_to(sizeof(RegistersForDebugging), sizeof(jdouble)) - STACK_BIAS, O0); | |
1717 a->flush_windows(); | |
1718 int i; | |
1719 for (i = 0; i < 8; ++i) { | |
1720 a->ld_ptr(as_iRegister(i)->address_in_saved_window().after_save(), L1); a->st_ptr( L1, O0, i_offset(i)); | |
1721 a->ld_ptr(as_lRegister(i)->address_in_saved_window().after_save(), L1); a->st_ptr( L1, O0, l_offset(i)); | |
1722 a->st_ptr(as_oRegister(i)->after_save(), O0, o_offset(i)); | |
1723 a->st_ptr(as_gRegister(i)->after_save(), O0, g_offset(i)); | |
1724 } | |
1725 for (i = 0; i < 32; ++i) { | |
1726 a->stf(FloatRegisterImpl::S, as_FloatRegister(i), O0, f_offset(i)); | |
1727 } | |
1728 for (i = 0; i < (VM_Version::v9_instructions_work() ? 64 : 32); i += 2) { | |
1729 a->stf(FloatRegisterImpl::D, as_FloatRegister(i), O0, d_offset(i)); | |
1730 } | |
1731 } | |
1732 | |
1733 void RegistersForDebugging::restore_registers(MacroAssembler* a, Register r) { | |
1734 for (int i = 1; i < 8; ++i) { | |
1735 a->ld_ptr(r, g_offset(i), as_gRegister(i)); | |
1736 } | |
1737 for (int j = 0; j < 32; ++j) { | |
1738 a->ldf(FloatRegisterImpl::S, O0, f_offset(j), as_FloatRegister(j)); | |
1739 } | |
1740 for (int k = 0; k < (VM_Version::v9_instructions_work() ? 64 : 32); k += 2) { | |
1741 a->ldf(FloatRegisterImpl::D, O0, d_offset(k), as_FloatRegister(k)); | |
1742 } | |
1743 } | |
1744 | |
1745 | |
1746 // pushes double TOS element of FPU stack on CPU stack; pops from FPU stack | |
1747 void MacroAssembler::push_fTOS() { | |
1748 // %%%%%% need to implement this | |
1749 } | |
1750 | |
1751 // pops double TOS element from CPU stack and pushes on FPU stack | |
1752 void MacroAssembler::pop_fTOS() { | |
1753 // %%%%%% need to implement this | |
1754 } | |
1755 | |
1756 void MacroAssembler::empty_FPU_stack() { | |
1757 // %%%%%% need to implement this | |
1758 } | |
1759 | |
1760 void MacroAssembler::_verify_oop(Register reg, const char* msg, const char * file, int line) { | |
1761 // plausibility check for oops | |
1762 if (!VerifyOops) return; | |
1763 | |
1764 if (reg == G0) return; // always NULL, which is always an oop | |
1765 | |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1766 BLOCK_COMMENT("verify_oop {"); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1767 char buffer[64]; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1768 #ifdef COMPILER1 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1769 if (CommentedAssembly) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1770 snprintf(buffer, sizeof(buffer), "verify_oop at %d", offset()); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1771 block_comment(buffer); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1772 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1773 #endif |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1774 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1775 int len = strlen(file) + strlen(msg) + 1 + 4; |
0 | 1776 sprintf(buffer, "%d", line); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1777 len += strlen(buffer); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1778 sprintf(buffer, " at offset %d ", offset()); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1779 len += strlen(buffer); |
0 | 1780 char * real_msg = new char[len]; |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
1781 sprintf(real_msg, "%s%s(%s:%d)", msg, buffer, file, line); |
0 | 1782 |
1783 // Call indirectly to solve generation ordering problem | |
727 | 1784 AddressLiteral a(StubRoutines::verify_oop_subroutine_entry_address()); |
0 | 1785 |
1786 // Make some space on stack above the current register window. | |
1787 // Enough to hold 8 64-bit registers. | |
1788 add(SP,-8*8,SP); | |
1789 | |
1790 // Save some 64-bit registers; a normal 'save' chops the heads off | |
1791 // of 64-bit longs in the 32-bit build. | |
1792 stx(O0,SP,frame::register_save_words*wordSize+STACK_BIAS+0*8); | |
1793 stx(O1,SP,frame::register_save_words*wordSize+STACK_BIAS+1*8); | |
1794 mov(reg,O0); // Move arg into O0; arg might be in O7 which is about to be crushed | |
1795 stx(O7,SP,frame::register_save_words*wordSize+STACK_BIAS+7*8); | |
1796 | |
1797 set((intptr_t)real_msg, O1); | |
1798 // Load address to call to into O7 | |
1799 load_ptr_contents(a, O7); | |
1800 // Register call to verify_oop_subroutine | |
1801 callr(O7, G0); | |
1802 delayed()->nop(); | |
1803 // recover frame size | |
1804 add(SP, 8*8,SP); | |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
1805 BLOCK_COMMENT("} verify_oop"); |
0 | 1806 } |
1807 | |
1808 void MacroAssembler::_verify_oop_addr(Address addr, const char* msg, const char * file, int line) { | |
1809 // plausibility check for oops | |
1810 if (!VerifyOops) return; | |
1811 | |
1812 char buffer[64]; | |
1813 sprintf(buffer, "%d", line); | |
1814 int len = strlen(file) + strlen(msg) + 1 + 4 + strlen(buffer); | |
1815 sprintf(buffer, " at SP+%d ", addr.disp()); | |
1816 len += strlen(buffer); | |
1817 char * real_msg = new char[len]; | |
1818 sprintf(real_msg, "%s at SP+%d (%s:%d)", msg, addr.disp(), file, line); | |
1819 | |
1820 // Call indirectly to solve generation ordering problem | |
727 | 1821 AddressLiteral a(StubRoutines::verify_oop_subroutine_entry_address()); |
0 | 1822 |
1823 // Make some space on stack above the current register window. | |
1824 // Enough to hold 8 64-bit registers. | |
1825 add(SP,-8*8,SP); | |
1826 | |
1827 // Save some 64-bit registers; a normal 'save' chops the heads off | |
1828 // of 64-bit longs in the 32-bit build. | |
1829 stx(O0,SP,frame::register_save_words*wordSize+STACK_BIAS+0*8); | |
1830 stx(O1,SP,frame::register_save_words*wordSize+STACK_BIAS+1*8); | |
1831 ld_ptr(addr.base(), addr.disp() + 8*8, O0); // Load arg into O0; arg might be in O7 which is about to be crushed | |
1832 stx(O7,SP,frame::register_save_words*wordSize+STACK_BIAS+7*8); | |
1833 | |
1834 set((intptr_t)real_msg, O1); | |
1835 // Load address to call to into O7 | |
1836 load_ptr_contents(a, O7); | |
1837 // Register call to verify_oop_subroutine | |
1838 callr(O7, G0); | |
1839 delayed()->nop(); | |
1840 // recover frame size | |
1841 add(SP, 8*8,SP); | |
1842 } | |
1843 | |
1844 // side-door communication with signalHandler in os_solaris.cpp | |
1845 address MacroAssembler::_verify_oop_implicit_branch[3] = { NULL }; | |
1846 | |
1847 // This macro is expanded just once; it creates shared code. Contract: | |
1848 // receives an oop in O0. Must restore O0 & O7 from TLS. Must not smash ANY | |
1849 // registers, including flags. May not use a register 'save', as this blows | |
1850 // the high bits of the O-regs if they contain Long values. Acts as a 'leaf' | |
1851 // call. | |
1852 void MacroAssembler::verify_oop_subroutine() { | |
1853 assert( VM_Version::v9_instructions_work(), "VerifyOops not supported for V8" ); | |
1854 | |
1855 // Leaf call; no frame. | |
1856 Label succeed, fail, null_or_fail; | |
1857 | |
1858 // O0 and O7 were saved already (O0 in O0's TLS home, O7 in O5's TLS home). | |
1859 // O0 is now the oop to be checked. O7 is the return address. | |
1860 Register O0_obj = O0; | |
1861 | |
1862 // Save some more registers for temps. | |
1863 stx(O2,SP,frame::register_save_words*wordSize+STACK_BIAS+2*8); | |
1864 stx(O3,SP,frame::register_save_words*wordSize+STACK_BIAS+3*8); | |
1865 stx(O4,SP,frame::register_save_words*wordSize+STACK_BIAS+4*8); | |
1866 stx(O5,SP,frame::register_save_words*wordSize+STACK_BIAS+5*8); | |
1867 | |
1868 // Save flags | |
1869 Register O5_save_flags = O5; | |
1870 rdccr( O5_save_flags ); | |
1871 | |
1872 { // count number of verifies | |
1873 Register O2_adr = O2; | |
1874 Register O3_accum = O3; | |
727 | 1875 inc_counter(StubRoutines::verify_oop_count_addr(), O2_adr, O3_accum); |
0 | 1876 } |
1877 | |
1878 Register O2_mask = O2; | |
1879 Register O3_bits = O3; | |
1880 Register O4_temp = O4; | |
1881 | |
1882 // mark lower end of faulting range | |
1883 assert(_verify_oop_implicit_branch[0] == NULL, "set once"); | |
1884 _verify_oop_implicit_branch[0] = pc(); | |
1885 | |
1886 // We can't check the mark oop because it could be in the process of | |
1887 // locking or unlocking while this is running. | |
1888 set(Universe::verify_oop_mask (), O2_mask); | |
1889 set(Universe::verify_oop_bits (), O3_bits); | |
1890 | |
1891 // assert((obj & oop_mask) == oop_bits); | |
1892 and3(O0_obj, O2_mask, O4_temp); | |
3839 | 1893 cmp_and_brx_short(O4_temp, O3_bits, notEqual, pn, null_or_fail); |
0 | 1894 |
1895 if ((NULL_WORD & Universe::verify_oop_mask()) == Universe::verify_oop_bits()) { | |
1896 // the null_or_fail case is useless; must test for null separately | |
3839 | 1897 br_null_short(O0_obj, pn, succeed); |
0 | 1898 } |
1899 | |
1900 // Check the klassOop of this object for being in the right area of memory. | |
1901 // Cannot do the load in the delay above slot in case O0 is null | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1902 load_klass(O0_obj, O0_obj); |
0 | 1903 // assert((klass & klass_mask) == klass_bits); |
1904 if( Universe::verify_klass_mask() != Universe::verify_oop_mask() ) | |
1905 set(Universe::verify_klass_mask(), O2_mask); | |
1906 if( Universe::verify_klass_bits() != Universe::verify_oop_bits() ) | |
1907 set(Universe::verify_klass_bits(), O3_bits); | |
1908 and3(O0_obj, O2_mask, O4_temp); | |
3839 | 1909 cmp_and_brx_short(O4_temp, O3_bits, notEqual, pn, fail); |
0 | 1910 // Check the klass's klass |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
1911 load_klass(O0_obj, O0_obj); |
0 | 1912 and3(O0_obj, O2_mask, O4_temp); |
1913 cmp(O4_temp, O3_bits); | |
1914 brx(notEqual, false, pn, fail); | |
1915 delayed()->wrccr( O5_save_flags ); // Restore CCR's | |
1916 | |
1917 // mark upper end of faulting range | |
1918 _verify_oop_implicit_branch[1] = pc(); | |
1919 | |
1920 //----------------------- | |
1921 // all tests pass | |
1922 bind(succeed); | |
1923 | |
1924 // Restore prior 64-bit registers | |
1925 ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+0*8,O0); | |
1926 ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+1*8,O1); | |
1927 ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+2*8,O2); | |
1928 ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+3*8,O3); | |
1929 ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+4*8,O4); | |
1930 ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+5*8,O5); | |
1931 | |
1932 retl(); // Leaf return; restore prior O7 in delay slot | |
1933 delayed()->ldx(SP,frame::register_save_words*wordSize+STACK_BIAS+7*8,O7); | |
1934 | |
1935 //----------------------- | |
1936 bind(null_or_fail); // nulls are less common but OK | |
1937 br_null(O0_obj, false, pt, succeed); | |
1938 delayed()->wrccr( O5_save_flags ); // Restore CCR's | |
1939 | |
1940 //----------------------- | |
1941 // report failure: | |
1942 bind(fail); | |
1943 _verify_oop_implicit_branch[2] = pc(); | |
1944 | |
1945 wrccr( O5_save_flags ); // Restore CCR's | |
1946 | |
1947 save_frame(::round_to(sizeof(RegistersForDebugging) / BytesPerWord, 2)); | |
1948 | |
1949 // stop_subroutine expects message pointer in I1. | |
1950 mov(I1, O1); | |
1951 | |
1952 // Restore prior 64-bit registers | |
1953 ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+0*8,I0); | |
1954 ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+1*8,I1); | |
1955 ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+2*8,I2); | |
1956 ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+3*8,I3); | |
1957 ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+4*8,I4); | |
1958 ldx(FP,frame::register_save_words*wordSize+STACK_BIAS+5*8,I5); | |
1959 | |
1960 // factor long stop-sequence into subroutine to save space | |
1961 assert(StubRoutines::Sparc::stop_subroutine_entry_address(), "hasn't been generated yet"); | |
1962 | |
1963 // call indirectly to solve generation ordering problem | |
727 | 1964 AddressLiteral al(StubRoutines::Sparc::stop_subroutine_entry_address()); |
1965 load_ptr_contents(al, O5); | |
0 | 1966 jmpl(O5, 0, O7); |
1967 delayed()->nop(); | |
1968 } | |
1969 | |
1970 | |
1971 void MacroAssembler::stop(const char* msg) { | |
1972 // save frame first to get O7 for return address | |
1973 // add one word to size in case struct is odd number of words long | |
1974 // It must be doubleword-aligned for storing doubles into it. | |
1975 | |
1976 save_frame(::round_to(sizeof(RegistersForDebugging) / BytesPerWord, 2)); | |
1977 | |
1978 // stop_subroutine expects message pointer in I1. | |
1979 set((intptr_t)msg, O1); | |
1980 | |
1981 // factor long stop-sequence into subroutine to save space | |
1982 assert(StubRoutines::Sparc::stop_subroutine_entry_address(), "hasn't been generated yet"); | |
1983 | |
1984 // call indirectly to solve generation ordering problem | |
727 | 1985 AddressLiteral a(StubRoutines::Sparc::stop_subroutine_entry_address()); |
0 | 1986 load_ptr_contents(a, O5); |
1987 jmpl(O5, 0, O7); | |
1988 delayed()->nop(); | |
1989 | |
1990 breakpoint_trap(); // make stop actually stop rather than writing | |
1991 // unnoticeable results in the output files. | |
1992 | |
1993 // restore(); done in callee to save space! | |
1994 } | |
1995 | |
1996 | |
1997 void MacroAssembler::warn(const char* msg) { | |
1998 save_frame(::round_to(sizeof(RegistersForDebugging) / BytesPerWord, 2)); | |
1999 RegistersForDebugging::save_registers(this); | |
2000 mov(O0, L0); | |
2001 set((intptr_t)msg, O0); | |
2002 call( CAST_FROM_FN_PTR(address, warning) ); | |
2003 delayed()->nop(); | |
2004 // ret(); | |
2005 // delayed()->restore(); | |
2006 RegistersForDebugging::restore_registers(this, L0); | |
2007 restore(); | |
2008 } | |
2009 | |
2010 | |
2011 void MacroAssembler::untested(const char* what) { | |
2012 // We must be able to turn interactive prompting off | |
2013 // in order to run automated test scripts on the VM | |
2014 // Use the flag ShowMessageBoxOnError | |
2015 | |
2016 char* b = new char[1024]; | |
2017 sprintf(b, "untested: %s", what); | |
2018 | |
2019 if ( ShowMessageBoxOnError ) stop(b); | |
2020 else warn(b); | |
2021 } | |
2022 | |
2023 | |
2024 void MacroAssembler::stop_subroutine() { | |
2025 RegistersForDebugging::save_registers(this); | |
2026 | |
2027 // for the sake of the debugger, stick a PC on the current frame | |
2028 // (this assumes that the caller has performed an extra "save") | |
2029 mov(I7, L7); | |
2030 add(O7, -7 * BytesPerInt, I7); | |
2031 | |
2032 save_frame(); // one more save to free up another O7 register | |
2033 mov(I0, O1); // addr of reg save area | |
2034 | |
2035 // We expect pointer to message in I1. Caller must set it up in O1 | |
2036 mov(I1, O0); // get msg | |
2037 call (CAST_FROM_FN_PTR(address, MacroAssembler::debug), relocInfo::runtime_call_type); | |
2038 delayed()->nop(); | |
2039 | |
2040 restore(); | |
2041 | |
2042 RegistersForDebugging::restore_registers(this, O0); | |
2043 | |
2044 save_frame(0); | |
2045 call(CAST_FROM_FN_PTR(address,breakpoint)); | |
2046 delayed()->nop(); | |
2047 restore(); | |
2048 | |
2049 mov(L7, I7); | |
2050 retl(); | |
2051 delayed()->restore(); // see stop above | |
2052 } | |
2053 | |
2054 | |
2055 void MacroAssembler::debug(char* msg, RegistersForDebugging* regs) { | |
2056 if ( ShowMessageBoxOnError ) { | |
2057 JavaThreadState saved_state = JavaThread::current()->thread_state(); | |
2058 JavaThread::current()->set_thread_state(_thread_in_vm); | |
2059 { | |
2060 // In order to get locks work, we need to fake a in_VM state | |
2061 ttyLocker ttyl; | |
2062 ::tty->print_cr("EXECUTION STOPPED: %s\n", msg); | |
2063 if (CountBytecodes || TraceBytecodes || StopInterpreterAt) { | |
2064 ::tty->print_cr("Interpreter::bytecode_counter = %d", BytecodeCounter::counter_value()); | |
2065 } | |
2066 if (os::message_box(msg, "Execution stopped, print registers?")) | |
2067 regs->print(::tty); | |
2068 } | |
2069 ThreadStateTransition::transition(JavaThread::current(), _thread_in_vm, saved_state); | |
2070 } | |
2071 else | |
2072 ::tty->print_cr("=============== DEBUG MESSAGE: %s ================\n", msg); | |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
2073 assert(false, err_msg("DEBUG MESSAGE: %s", msg)); |
0 | 2074 } |
2075 | |
2076 | |
2077 #ifndef PRODUCT | |
2078 void MacroAssembler::test() { | |
2079 ResourceMark rm; | |
2080 | |
2081 CodeBuffer cb("test", 10000, 10000); | |
2082 MacroAssembler* a = new MacroAssembler(&cb); | |
2083 VM_Version::allow_all(); | |
2084 a->test_v9(); | |
2085 a->test_v8_onlys(); | |
2086 VM_Version::revert(); | |
2087 | |
2088 StubRoutines::Sparc::test_stop_entry()(); | |
2089 } | |
2090 #endif | |
2091 | |
2092 | |
2093 void MacroAssembler::calc_mem_param_words(Register Rparam_words, Register Rresult) { | |
2094 subcc( Rparam_words, Argument::n_register_parameters, Rresult); // how many mem words? | |
2095 Label no_extras; | |
2096 br( negative, true, pt, no_extras ); // if neg, clear reg | |
727 | 2097 delayed()->set(0, Rresult); // annuled, so only if taken |
0 | 2098 bind( no_extras ); |
2099 } | |
2100 | |
2101 | |
2102 void MacroAssembler::calc_frame_size(Register Rextra_words, Register Rresult) { | |
2103 #ifdef _LP64 | |
2104 add(Rextra_words, frame::memory_parameter_word_sp_offset, Rresult); | |
2105 #else | |
2106 add(Rextra_words, frame::memory_parameter_word_sp_offset + 1, Rresult); | |
2107 #endif | |
2108 bclr(1, Rresult); | |
2109 sll(Rresult, LogBytesPerWord, Rresult); // Rresult has total frame bytes | |
2110 } | |
2111 | |
2112 | |
2113 void MacroAssembler::calc_frame_size_and_save(Register Rextra_words, Register Rresult) { | |
2114 calc_frame_size(Rextra_words, Rresult); | |
2115 neg(Rresult); | |
2116 save(SP, Rresult, SP); | |
2117 } | |
2118 | |
2119 | |
2120 // --------------------------------------------------------- | |
2121 Assembler::RCondition cond2rcond(Assembler::Condition c) { | |
2122 switch (c) { | |
2123 /*case zero: */ | |
2124 case Assembler::equal: return Assembler::rc_z; | |
2125 case Assembler::lessEqual: return Assembler::rc_lez; | |
2126 case Assembler::less: return Assembler::rc_lz; | |
2127 /*case notZero:*/ | |
2128 case Assembler::notEqual: return Assembler::rc_nz; | |
2129 case Assembler::greater: return Assembler::rc_gz; | |
2130 case Assembler::greaterEqual: return Assembler::rc_gez; | |
2131 } | |
2132 ShouldNotReachHere(); | |
2133 return Assembler::rc_z; | |
2134 } | |
2135 | |
3839 | 2136 // compares (32 bit) register with zero and branches. NOT FOR USE WITH 64-bit POINTERS |
2137 void MacroAssembler::cmp_zero_and_br(Condition c, Register s1, Label& L, bool a, Predict p) { | |
0 | 2138 tst(s1); |
2139 br (c, a, p, L); | |
2140 } | |
2141 | |
2142 // Compares a pointer register with zero and branches on null. | |
2143 // Does a test & branch on 32-bit systems and a register-branch on 64-bit. | |
2144 void MacroAssembler::br_null( Register s1, bool a, Predict p, Label& L ) { | |
2145 assert_not_delayed(); | |
2146 #ifdef _LP64 | |
2147 bpr( rc_z, a, p, s1, L ); | |
2148 #else | |
2149 tst(s1); | |
2150 br ( zero, a, p, L ); | |
2151 #endif | |
2152 } | |
2153 | |
2154 void MacroAssembler::br_notnull( Register s1, bool a, Predict p, Label& L ) { | |
2155 assert_not_delayed(); | |
2156 #ifdef _LP64 | |
2157 bpr( rc_nz, a, p, s1, L ); | |
2158 #else | |
2159 tst(s1); | |
2160 br ( notZero, a, p, L ); | |
2161 #endif | |
2162 } | |
2163 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2164 void MacroAssembler::br_on_reg_cond( RCondition rc, bool a, Predict p, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2165 Register s1, address d, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2166 relocInfo::relocType rt ) { |
3839 | 2167 assert_not_delayed(); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2168 if (VM_Version::v9_instructions_work()) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2169 bpr(rc, a, p, s1, d, rt); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2170 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2171 tst(s1); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2172 br(reg_cond_to_cc_cond(rc), a, p, d, rt); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2173 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2174 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2175 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2176 void MacroAssembler::br_on_reg_cond( RCondition rc, bool a, Predict p, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2177 Register s1, Label& L ) { |
3839 | 2178 assert_not_delayed(); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2179 if (VM_Version::v9_instructions_work()) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2180 bpr(rc, a, p, s1, L); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2181 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2182 tst(s1); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2183 br(reg_cond_to_cc_cond(rc), a, p, L); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2184 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2185 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
2186 |
3839 | 2187 // Compare registers and branch with nop in delay slot or cbcond without delay slot. |
2188 | |
2189 // Compare integer (32 bit) values (icc only). | |
2190 void MacroAssembler::cmp_and_br_short(Register s1, Register s2, Condition c, | |
2191 Predict p, Label& L) { | |
2192 assert_not_delayed(); | |
2193 if (use_cbcond(L)) { | |
2194 Assembler::cbcond(c, icc, s1, s2, L); | |
2195 } else { | |
2196 cmp(s1, s2); | |
2197 br(c, false, p, L); | |
2198 delayed()->nop(); | |
2199 } | |
2200 } | |
2201 | |
2202 // Compare integer (32 bit) values (icc only). | |
2203 void MacroAssembler::cmp_and_br_short(Register s1, int simm13a, Condition c, | |
2204 Predict p, Label& L) { | |
2205 assert_not_delayed(); | |
2206 if (is_simm(simm13a,5) && use_cbcond(L)) { | |
2207 Assembler::cbcond(c, icc, s1, simm13a, L); | |
2208 } else { | |
2209 cmp(s1, simm13a); | |
2210 br(c, false, p, L); | |
2211 delayed()->nop(); | |
2212 } | |
2213 } | |
2214 | |
2215 // Branch that tests xcc in LP64 and icc in !LP64 | |
2216 void MacroAssembler::cmp_and_brx_short(Register s1, Register s2, Condition c, | |
2217 Predict p, Label& L) { | |
2218 assert_not_delayed(); | |
2219 if (use_cbcond(L)) { | |
2220 Assembler::cbcond(c, ptr_cc, s1, s2, L); | |
2221 } else { | |
2222 cmp(s1, s2); | |
2223 brx(c, false, p, L); | |
2224 delayed()->nop(); | |
2225 } | |
2226 } | |
2227 | |
2228 // Branch that tests xcc in LP64 and icc in !LP64 | |
2229 void MacroAssembler::cmp_and_brx_short(Register s1, int simm13a, Condition c, | |
2230 Predict p, Label& L) { | |
2231 assert_not_delayed(); | |
2232 if (is_simm(simm13a,5) && use_cbcond(L)) { | |
2233 Assembler::cbcond(c, ptr_cc, s1, simm13a, L); | |
2234 } else { | |
2235 cmp(s1, simm13a); | |
2236 brx(c, false, p, L); | |
2237 delayed()->nop(); | |
2238 } | |
2239 } | |
2240 | |
2241 // Short branch version for compares a pointer with zero. | |
2242 | |
2243 void MacroAssembler::br_null_short(Register s1, Predict p, Label& L) { | |
2244 assert_not_delayed(); | |
2245 if (use_cbcond(L)) { | |
2246 Assembler::cbcond(zero, ptr_cc, s1, 0, L); | |
2247 return; | |
2248 } | |
2249 br_null(s1, false, p, L); | |
2250 delayed()->nop(); | |
2251 } | |
2252 | |
2253 void MacroAssembler::br_notnull_short(Register s1, Predict p, Label& L) { | |
2254 assert_not_delayed(); | |
2255 if (use_cbcond(L)) { | |
2256 Assembler::cbcond(notZero, ptr_cc, s1, 0, L); | |
2257 return; | |
2258 } | |
2259 br_notnull(s1, false, p, L); | |
2260 delayed()->nop(); | |
2261 } | |
2262 | |
2263 // Unconditional short branch | |
2264 void MacroAssembler::ba_short(Label& L) { | |
2265 if (use_cbcond(L)) { | |
2266 Assembler::cbcond(equal, icc, G0, G0, L); | |
2267 return; | |
2268 } | |
2269 br(always, false, pt, L); | |
2270 delayed()->nop(); | |
2271 } | |
0 | 2272 |
2273 // instruction sequences factored across compiler & interpreter | |
2274 | |
2275 | |
2276 void MacroAssembler::lcmp( Register Ra_hi, Register Ra_low, | |
2277 Register Rb_hi, Register Rb_low, | |
2278 Register Rresult) { | |
2279 | |
2280 Label check_low_parts, done; | |
2281 | |
2282 cmp(Ra_hi, Rb_hi ); // compare hi parts | |
2283 br(equal, true, pt, check_low_parts); | |
2284 delayed()->cmp(Ra_low, Rb_low); // test low parts | |
2285 | |
2286 // And, with an unsigned comparison, it does not matter if the numbers | |
2287 // are negative or not. | |
2288 // E.g., -2 cmp -1: the low parts are 0xfffffffe and 0xffffffff. | |
2289 // The second one is bigger (unsignedly). | |
2290 | |
2291 // Other notes: The first move in each triplet can be unconditional | |
2292 // (and therefore probably prefetchable). | |
2293 // And the equals case for the high part does not need testing, | |
2294 // since that triplet is reached only after finding the high halves differ. | |
2295 | |
2296 if (VM_Version::v9_instructions_work()) { | |
3839 | 2297 mov(-1, Rresult); |
2298 ba(done); delayed()-> movcc(greater, false, icc, 1, Rresult); | |
2299 } else { | |
0 | 2300 br(less, true, pt, done); delayed()-> set(-1, Rresult); |
2301 br(greater, true, pt, done); delayed()-> set( 1, Rresult); | |
2302 } | |
2303 | |
2304 bind( check_low_parts ); | |
2305 | |
2306 if (VM_Version::v9_instructions_work()) { | |
2307 mov( -1, Rresult); | |
2308 movcc(equal, false, icc, 0, Rresult); | |
2309 movcc(greaterUnsigned, false, icc, 1, Rresult); | |
3839 | 2310 } else { |
2311 set(-1, Rresult); | |
0 | 2312 br(equal, true, pt, done); delayed()->set( 0, Rresult); |
2313 br(greaterUnsigned, true, pt, done); delayed()->set( 1, Rresult); | |
2314 } | |
2315 bind( done ); | |
2316 } | |
2317 | |
2318 void MacroAssembler::lneg( Register Rhi, Register Rlow ) { | |
2319 subcc( G0, Rlow, Rlow ); | |
2320 subc( G0, Rhi, Rhi ); | |
2321 } | |
2322 | |
2323 void MacroAssembler::lshl( Register Rin_high, Register Rin_low, | |
2324 Register Rcount, | |
2325 Register Rout_high, Register Rout_low, | |
2326 Register Rtemp ) { | |
2327 | |
2328 | |
2329 Register Ralt_count = Rtemp; | |
2330 Register Rxfer_bits = Rtemp; | |
2331 | |
2332 assert( Ralt_count != Rin_high | |
2333 && Ralt_count != Rin_low | |
2334 && Ralt_count != Rcount | |
2335 && Rxfer_bits != Rin_low | |
2336 && Rxfer_bits != Rin_high | |
2337 && Rxfer_bits != Rcount | |
2338 && Rxfer_bits != Rout_low | |
2339 && Rout_low != Rin_high, | |
2340 "register alias checks"); | |
2341 | |
2342 Label big_shift, done; | |
2343 | |
2344 // This code can be optimized to use the 64 bit shifts in V9. | |
2345 // Here we use the 32 bit shifts. | |
2346 | |
3839 | 2347 and3( Rcount, 0x3f, Rcount); // take least significant 6 bits |
2348 subcc(Rcount, 31, Ralt_count); | |
0 | 2349 br(greater, true, pn, big_shift); |
3839 | 2350 delayed()->dec(Ralt_count); |
0 | 2351 |
2352 // shift < 32 bits, Ralt_count = Rcount-31 | |
2353 | |
2354 // We get the transfer bits by shifting right by 32-count the low | |
2355 // register. This is done by shifting right by 31-count and then by one | |
2356 // more to take care of the special (rare) case where count is zero | |
2357 // (shifting by 32 would not work). | |
2358 | |
3839 | 2359 neg(Ralt_count); |
0 | 2360 |
2361 // The order of the next two instructions is critical in the case where | |
2362 // Rin and Rout are the same and should not be reversed. | |
2363 | |
3839 | 2364 srl(Rin_low, Ralt_count, Rxfer_bits); // shift right by 31-count |
0 | 2365 if (Rcount != Rout_low) { |
3839 | 2366 sll(Rin_low, Rcount, Rout_low); // low half |
0 | 2367 } |
3839 | 2368 sll(Rin_high, Rcount, Rout_high); |
0 | 2369 if (Rcount == Rout_low) { |
3839 | 2370 sll(Rin_low, Rcount, Rout_low); // low half |
0 | 2371 } |
3839 | 2372 srl(Rxfer_bits, 1, Rxfer_bits ); // shift right by one more |
2373 ba(done); | |
2374 delayed()->or3(Rout_high, Rxfer_bits, Rout_high); // new hi value: or in shifted old hi part and xfer from low | |
0 | 2375 |
2376 // shift >= 32 bits, Ralt_count = Rcount-32 | |
2377 bind(big_shift); | |
3839 | 2378 sll(Rin_low, Ralt_count, Rout_high ); |
2379 clr(Rout_low); | |
0 | 2380 |
2381 bind(done); | |
2382 } | |
2383 | |
2384 | |
2385 void MacroAssembler::lshr( Register Rin_high, Register Rin_low, | |
2386 Register Rcount, | |
2387 Register Rout_high, Register Rout_low, | |
2388 Register Rtemp ) { | |
2389 | |
2390 Register Ralt_count = Rtemp; | |
2391 Register Rxfer_bits = Rtemp; | |
2392 | |
2393 assert( Ralt_count != Rin_high | |
2394 && Ralt_count != Rin_low | |
2395 && Ralt_count != Rcount | |
2396 && Rxfer_bits != Rin_low | |
2397 && Rxfer_bits != Rin_high | |
2398 && Rxfer_bits != Rcount | |
2399 && Rxfer_bits != Rout_high | |
2400 && Rout_high != Rin_low, | |
2401 "register alias checks"); | |
2402 | |
2403 Label big_shift, done; | |
2404 | |
2405 // This code can be optimized to use the 64 bit shifts in V9. | |
2406 // Here we use the 32 bit shifts. | |
2407 | |
3839 | 2408 and3( Rcount, 0x3f, Rcount); // take least significant 6 bits |
2409 subcc(Rcount, 31, Ralt_count); | |
0 | 2410 br(greater, true, pn, big_shift); |
2411 delayed()->dec(Ralt_count); | |
2412 | |
2413 // shift < 32 bits, Ralt_count = Rcount-31 | |
2414 | |
2415 // We get the transfer bits by shifting left by 32-count the high | |
2416 // register. This is done by shifting left by 31-count and then by one | |
2417 // more to take care of the special (rare) case where count is zero | |
2418 // (shifting by 32 would not work). | |
2419 | |
3839 | 2420 neg(Ralt_count); |
0 | 2421 if (Rcount != Rout_low) { |
3839 | 2422 srl(Rin_low, Rcount, Rout_low); |
0 | 2423 } |
2424 | |
2425 // The order of the next two instructions is critical in the case where | |
2426 // Rin and Rout are the same and should not be reversed. | |
2427 | |
3839 | 2428 sll(Rin_high, Ralt_count, Rxfer_bits); // shift left by 31-count |
2429 sra(Rin_high, Rcount, Rout_high ); // high half | |
2430 sll(Rxfer_bits, 1, Rxfer_bits); // shift left by one more | |
0 | 2431 if (Rcount == Rout_low) { |
3839 | 2432 srl(Rin_low, Rcount, Rout_low); |
0 | 2433 } |
3839 | 2434 ba(done); |
2435 delayed()->or3(Rout_low, Rxfer_bits, Rout_low); // new low value: or shifted old low part and xfer from high | |
0 | 2436 |
2437 // shift >= 32 bits, Ralt_count = Rcount-32 | |
2438 bind(big_shift); | |
2439 | |
3839 | 2440 sra(Rin_high, Ralt_count, Rout_low); |
2441 sra(Rin_high, 31, Rout_high); // sign into hi | |
0 | 2442 |
2443 bind( done ); | |
2444 } | |
2445 | |
2446 | |
2447 | |
2448 void MacroAssembler::lushr( Register Rin_high, Register Rin_low, | |
2449 Register Rcount, | |
2450 Register Rout_high, Register Rout_low, | |
2451 Register Rtemp ) { | |
2452 | |
2453 Register Ralt_count = Rtemp; | |
2454 Register Rxfer_bits = Rtemp; | |
2455 | |
2456 assert( Ralt_count != Rin_high | |
2457 && Ralt_count != Rin_low | |
2458 && Ralt_count != Rcount | |
2459 && Rxfer_bits != Rin_low | |
2460 && Rxfer_bits != Rin_high | |
2461 && Rxfer_bits != Rcount | |
2462 && Rxfer_bits != Rout_high | |
2463 && Rout_high != Rin_low, | |
2464 "register alias checks"); | |
2465 | |
2466 Label big_shift, done; | |
2467 | |
2468 // This code can be optimized to use the 64 bit shifts in V9. | |
2469 // Here we use the 32 bit shifts. | |
2470 | |
3839 | 2471 and3( Rcount, 0x3f, Rcount); // take least significant 6 bits |
2472 subcc(Rcount, 31, Ralt_count); | |
0 | 2473 br(greater, true, pn, big_shift); |
2474 delayed()->dec(Ralt_count); | |
2475 | |
2476 // shift < 32 bits, Ralt_count = Rcount-31 | |
2477 | |
2478 // We get the transfer bits by shifting left by 32-count the high | |
2479 // register. This is done by shifting left by 31-count and then by one | |
2480 // more to take care of the special (rare) case where count is zero | |
2481 // (shifting by 32 would not work). | |
2482 | |
3839 | 2483 neg(Ralt_count); |
0 | 2484 if (Rcount != Rout_low) { |
3839 | 2485 srl(Rin_low, Rcount, Rout_low); |
0 | 2486 } |
2487 | |
2488 // The order of the next two instructions is critical in the case where | |
2489 // Rin and Rout are the same and should not be reversed. | |
2490 | |
3839 | 2491 sll(Rin_high, Ralt_count, Rxfer_bits); // shift left by 31-count |
2492 srl(Rin_high, Rcount, Rout_high ); // high half | |
2493 sll(Rxfer_bits, 1, Rxfer_bits); // shift left by one more | |
0 | 2494 if (Rcount == Rout_low) { |
3839 | 2495 srl(Rin_low, Rcount, Rout_low); |
0 | 2496 } |
3839 | 2497 ba(done); |
2498 delayed()->or3(Rout_low, Rxfer_bits, Rout_low); // new low value: or shifted old low part and xfer from high | |
0 | 2499 |
2500 // shift >= 32 bits, Ralt_count = Rcount-32 | |
2501 bind(big_shift); | |
2502 | |
3839 | 2503 srl(Rin_high, Ralt_count, Rout_low); |
2504 clr(Rout_high); | |
0 | 2505 |
2506 bind( done ); | |
2507 } | |
2508 | |
2509 #ifdef _LP64 | |
2510 void MacroAssembler::lcmp( Register Ra, Register Rb, Register Rresult) { | |
2511 cmp(Ra, Rb); | |
3839 | 2512 mov(-1, Rresult); |
0 | 2513 movcc(equal, false, xcc, 0, Rresult); |
2514 movcc(greater, false, xcc, 1, Rresult); | |
2515 } | |
2516 #endif | |
2517 | |
2518 | |
2258
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2519 void MacroAssembler::load_sized_value(Address src, Register dst, size_t size_in_bytes, bool is_signed) { |
1503 | 2520 switch (size_in_bytes) { |
2258
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2521 case 8: ld_long(src, dst); break; |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2522 case 4: ld( src, dst); break; |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2523 case 2: is_signed ? ldsh(src, dst) : lduh(src, dst); break; |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2524 case 1: is_signed ? ldsb(src, dst) : ldub(src, dst); break; |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2525 default: ShouldNotReachHere(); |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2526 } |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2527 } |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2528 |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2529 void MacroAssembler::store_sized_value(Register src, Address dst, size_t size_in_bytes) { |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2530 switch (size_in_bytes) { |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2531 case 8: st_long(src, dst); break; |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2532 case 4: st( src, dst); break; |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2533 case 2: sth( src, dst); break; |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2534 case 1: stb( src, dst); break; |
28bf941f445e
7018378: JSR 292: _bound_int_mh produces wrong result on 64-bit SPARC
twisti
parents:
2127
diff
changeset
|
2535 default: ShouldNotReachHere(); |
1503 | 2536 } |
2537 } | |
2538 | |
2539 | |
0 | 2540 void MacroAssembler::float_cmp( bool is_float, int unordered_result, |
2541 FloatRegister Fa, FloatRegister Fb, | |
2542 Register Rresult) { | |
2543 | |
2544 fcmp(is_float ? FloatRegisterImpl::S : FloatRegisterImpl::D, fcc0, Fa, Fb); | |
2545 | |
2546 Condition lt = unordered_result == -1 ? f_unorderedOrLess : f_less; | |
2547 Condition eq = f_equal; | |
2548 Condition gt = unordered_result == 1 ? f_unorderedOrGreater : f_greater; | |
2549 | |
2550 if (VM_Version::v9_instructions_work()) { | |
2551 | |
3839 | 2552 mov(-1, Rresult); |
2553 movcc(eq, true, fcc0, 0, Rresult); | |
2554 movcc(gt, true, fcc0, 1, Rresult); | |
0 | 2555 |
2556 } else { | |
2557 Label done; | |
2558 | |
3839 | 2559 set( -1, Rresult ); |
0 | 2560 //fb(lt, true, pn, done); delayed()->set( -1, Rresult ); |
2561 fb( eq, true, pn, done); delayed()->set( 0, Rresult ); | |
2562 fb( gt, true, pn, done); delayed()->set( 1, Rresult ); | |
2563 | |
2564 bind (done); | |
2565 } | |
2566 } | |
2567 | |
2568 | |
2569 void MacroAssembler::fneg( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d) | |
2570 { | |
2571 if (VM_Version::v9_instructions_work()) { | |
2572 Assembler::fneg(w, s, d); | |
2573 } else { | |
2574 if (w == FloatRegisterImpl::S) { | |
2575 Assembler::fneg(w, s, d); | |
2576 } else if (w == FloatRegisterImpl::D) { | |
2577 // number() does a sanity check on the alignment. | |
2578 assert(((s->encoding(FloatRegisterImpl::D) & 1) == 0) && | |
2579 ((d->encoding(FloatRegisterImpl::D) & 1) == 0), "float register alignment check"); | |
2580 | |
2581 Assembler::fneg(FloatRegisterImpl::S, s, d); | |
2582 Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor()); | |
2583 } else { | |
2584 assert(w == FloatRegisterImpl::Q, "Invalid float register width"); | |
2585 | |
2586 // number() does a sanity check on the alignment. | |
2587 assert(((s->encoding(FloatRegisterImpl::D) & 3) == 0) && | |
2588 ((d->encoding(FloatRegisterImpl::D) & 3) == 0), "float register alignment check"); | |
2589 | |
2590 Assembler::fneg(FloatRegisterImpl::S, s, d); | |
2591 Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor()); | |
2592 Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor(), d->successor()->successor()); | |
2593 Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor()->successor(), d->successor()->successor()->successor()); | |
2594 } | |
2595 } | |
2596 } | |
2597 | |
2598 void MacroAssembler::fmov( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d) | |
2599 { | |
2600 if (VM_Version::v9_instructions_work()) { | |
2601 Assembler::fmov(w, s, d); | |
2602 } else { | |
2603 if (w == FloatRegisterImpl::S) { | |
2604 Assembler::fmov(w, s, d); | |
2605 } else if (w == FloatRegisterImpl::D) { | |
2606 // number() does a sanity check on the alignment. | |
2607 assert(((s->encoding(FloatRegisterImpl::D) & 1) == 0) && | |
2608 ((d->encoding(FloatRegisterImpl::D) & 1) == 0), "float register alignment check"); | |
2609 | |
2610 Assembler::fmov(FloatRegisterImpl::S, s, d); | |
2611 Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor()); | |
2612 } else { | |
2613 assert(w == FloatRegisterImpl::Q, "Invalid float register width"); | |
2614 | |
2615 // number() does a sanity check on the alignment. | |
2616 assert(((s->encoding(FloatRegisterImpl::D) & 3) == 0) && | |
2617 ((d->encoding(FloatRegisterImpl::D) & 3) == 0), "float register alignment check"); | |
2618 | |
2619 Assembler::fmov(FloatRegisterImpl::S, s, d); | |
2620 Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor()); | |
2621 Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor(), d->successor()->successor()); | |
2622 Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor()->successor(), d->successor()->successor()->successor()); | |
2623 } | |
2624 } | |
2625 } | |
2626 | |
2627 void MacroAssembler::fabs( FloatRegisterImpl::Width w, FloatRegister s, FloatRegister d) | |
2628 { | |
2629 if (VM_Version::v9_instructions_work()) { | |
2630 Assembler::fabs(w, s, d); | |
2631 } else { | |
2632 if (w == FloatRegisterImpl::S) { | |
2633 Assembler::fabs(w, s, d); | |
2634 } else if (w == FloatRegisterImpl::D) { | |
2635 // number() does a sanity check on the alignment. | |
2636 assert(((s->encoding(FloatRegisterImpl::D) & 1) == 0) && | |
2637 ((d->encoding(FloatRegisterImpl::D) & 1) == 0), "float register alignment check"); | |
2638 | |
2639 Assembler::fabs(FloatRegisterImpl::S, s, d); | |
2640 Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor()); | |
2641 } else { | |
2642 assert(w == FloatRegisterImpl::Q, "Invalid float register width"); | |
2643 | |
2644 // number() does a sanity check on the alignment. | |
2645 assert(((s->encoding(FloatRegisterImpl::D) & 3) == 0) && | |
2646 ((d->encoding(FloatRegisterImpl::D) & 3) == 0), "float register alignment check"); | |
2647 | |
2648 Assembler::fabs(FloatRegisterImpl::S, s, d); | |
2649 Assembler::fmov(FloatRegisterImpl::S, s->successor(), d->successor()); | |
2650 Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor(), d->successor()->successor()); | |
2651 Assembler::fmov(FloatRegisterImpl::S, s->successor()->successor()->successor(), d->successor()->successor()->successor()); | |
2652 } | |
2653 } | |
2654 } | |
2655 | |
2656 void MacroAssembler::save_all_globals_into_locals() { | |
2657 mov(G1,L1); | |
2658 mov(G2,L2); | |
2659 mov(G3,L3); | |
2660 mov(G4,L4); | |
2661 mov(G5,L5); | |
2662 mov(G6,L6); | |
2663 mov(G7,L7); | |
2664 } | |
2665 | |
2666 void MacroAssembler::restore_globals_from_locals() { | |
2667 mov(L1,G1); | |
2668 mov(L2,G2); | |
2669 mov(L3,G3); | |
2670 mov(L4,G4); | |
2671 mov(L5,G5); | |
2672 mov(L6,G6); | |
2673 mov(L7,G7); | |
2674 } | |
2675 | |
2676 // Use for 64 bit operation. | |
2677 void MacroAssembler::casx_under_lock(Register top_ptr_reg, Register top_reg, Register ptr_reg, address lock_addr, bool use_call_vm) | |
2678 { | |
2679 // store ptr_reg as the new top value | |
2680 #ifdef _LP64 | |
2681 casx(top_ptr_reg, top_reg, ptr_reg); | |
2682 #else | |
2683 cas_under_lock(top_ptr_reg, top_reg, ptr_reg, lock_addr, use_call_vm); | |
2684 #endif // _LP64 | |
2685 } | |
2686 | |
2687 // [RGV] This routine does not handle 64 bit operations. | |
2688 // use casx_under_lock() or casx directly!!! | |
2689 void MacroAssembler::cas_under_lock(Register top_ptr_reg, Register top_reg, Register ptr_reg, address lock_addr, bool use_call_vm) | |
2690 { | |
2691 // store ptr_reg as the new top value | |
2692 if (VM_Version::v9_instructions_work()) { | |
2693 cas(top_ptr_reg, top_reg, ptr_reg); | |
2694 } else { | |
2695 | |
2696 // If the register is not an out nor global, it is not visible | |
2697 // after the save. Allocate a register for it, save its | |
2698 // value in the register save area (the save may not flush | |
2699 // registers to the save area). | |
2700 | |
2701 Register top_ptr_reg_after_save; | |
2702 Register top_reg_after_save; | |
2703 Register ptr_reg_after_save; | |
2704 | |
2705 if (top_ptr_reg->is_out() || top_ptr_reg->is_global()) { | |
2706 top_ptr_reg_after_save = top_ptr_reg->after_save(); | |
2707 } else { | |
2708 Address reg_save_addr = top_ptr_reg->address_in_saved_window(); | |
2709 top_ptr_reg_after_save = L0; | |
2710 st(top_ptr_reg, reg_save_addr); | |
2711 } | |
2712 | |
2713 if (top_reg->is_out() || top_reg->is_global()) { | |
2714 top_reg_after_save = top_reg->after_save(); | |
2715 } else { | |
2716 Address reg_save_addr = top_reg->address_in_saved_window(); | |
2717 top_reg_after_save = L1; | |
2718 st(top_reg, reg_save_addr); | |
2719 } | |
2720 | |
2721 if (ptr_reg->is_out() || ptr_reg->is_global()) { | |
2722 ptr_reg_after_save = ptr_reg->after_save(); | |
2723 } else { | |
2724 Address reg_save_addr = ptr_reg->address_in_saved_window(); | |
2725 ptr_reg_after_save = L2; | |
2726 st(ptr_reg, reg_save_addr); | |
2727 } | |
2728 | |
2729 const Register& lock_reg = L3; | |
2730 const Register& lock_ptr_reg = L4; | |
2731 const Register& value_reg = L5; | |
2732 const Register& yield_reg = L6; | |
2733 const Register& yieldall_reg = L7; | |
2734 | |
2735 save_frame(); | |
2736 | |
2737 if (top_ptr_reg_after_save == L0) { | |
2738 ld(top_ptr_reg->address_in_saved_window().after_save(), top_ptr_reg_after_save); | |
2739 } | |
2740 | |
2741 if (top_reg_after_save == L1) { | |
2742 ld(top_reg->address_in_saved_window().after_save(), top_reg_after_save); | |
2743 } | |
2744 | |
2745 if (ptr_reg_after_save == L2) { | |
2746 ld(ptr_reg->address_in_saved_window().after_save(), ptr_reg_after_save); | |
2747 } | |
2748 | |
2749 Label(retry_get_lock); | |
2750 Label(not_same); | |
2751 Label(dont_yield); | |
2752 | |
2753 assert(lock_addr, "lock_address should be non null for v8"); | |
2754 set((intptr_t)lock_addr, lock_ptr_reg); | |
2755 // Initialize yield counter | |
2756 mov(G0,yield_reg); | |
2757 mov(G0, yieldall_reg); | |
2758 set(StubRoutines::Sparc::locked, lock_reg); | |
2759 | |
2760 bind(retry_get_lock); | |
3839 | 2761 cmp_and_br_short(yield_reg, V8AtomicOperationUnderLockSpinCount, Assembler::less, Assembler::pt, dont_yield); |
0 | 2762 |
2763 if(use_call_vm) { | |
2764 Untested("Need to verify global reg consistancy"); | |
2765 call_VM(noreg, CAST_FROM_FN_PTR(address, SharedRuntime::yield_all), yieldall_reg); | |
2766 } else { | |
2767 // Save the regs and make space for a C call | |
2768 save(SP, -96, SP); | |
2769 save_all_globals_into_locals(); | |
2770 call(CAST_FROM_FN_PTR(address,os::yield_all)); | |
2771 delayed()->mov(yieldall_reg, O0); | |
2772 restore_globals_from_locals(); | |
2773 restore(); | |
2774 } | |
2775 | |
2776 // reset the counter | |
2777 mov(G0,yield_reg); | |
2778 add(yieldall_reg, 1, yieldall_reg); | |
2779 | |
2780 bind(dont_yield); | |
2781 // try to get lock | |
2782 swap(lock_ptr_reg, 0, lock_reg); | |
2783 | |
2784 // did we get the lock? | |
2785 cmp(lock_reg, StubRoutines::Sparc::unlocked); | |
2786 br(Assembler::notEqual, true, Assembler::pn, retry_get_lock); | |
2787 delayed()->add(yield_reg,1,yield_reg); | |
2788 | |
2789 // yes, got lock. do we have the same top? | |
2790 ld(top_ptr_reg_after_save, 0, value_reg); | |
3839 | 2791 cmp_and_br_short(value_reg, top_reg_after_save, Assembler::notEqual, Assembler::pn, not_same); |
0 | 2792 |
2793 // yes, same top. | |
2794 st(ptr_reg_after_save, top_ptr_reg_after_save, 0); | |
2795 membar(Assembler::StoreStore); | |
2796 | |
2797 bind(not_same); | |
2798 mov(value_reg, ptr_reg_after_save); | |
2799 st(lock_reg, lock_ptr_reg, 0); // unlock | |
2800 | |
2801 restore(); | |
2802 } | |
2803 } | |
2804 | |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2805 RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_addr, |
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2806 Register tmp, |
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2807 int offset) { |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2808 intptr_t value = *delayed_value_addr; |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2809 if (value != 0) |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2810 return RegisterOrConstant(value + offset); |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2811 |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2812 // load indirectly to solve generation ordering problem |
727 | 2813 AddressLiteral a(delayed_value_addr); |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2814 load_ptr_contents(a, tmp); |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2815 |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2816 #ifdef ASSERT |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2817 tst(tmp); |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2818 breakpoint_trap(zero, xcc); |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2819 #endif |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2820 |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2821 if (offset != 0) |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2822 add(tmp, offset, tmp); |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2823 |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2824 return RegisterOrConstant(tmp); |
622
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2825 } |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2826 |
56aae7be60d4
6812678: macro assembler needs delayed binding of a few constants (for 6655638)
jrose
parents:
420
diff
changeset
|
2827 |
1503 | 2828 RegisterOrConstant MacroAssembler::regcon_andn_ptr(RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp) { |
2829 assert(d.register_or_noreg() != G0, "lost side effect"); | |
2830 if ((s2.is_constant() && s2.as_constant() == 0) || | |
2831 (s2.is_register() && s2.as_register() == G0)) { | |
2832 // Do nothing, just move value. | |
2833 if (s1.is_register()) { | |
2834 if (d.is_constant()) d = temp; | |
2835 mov(s1.as_register(), d.as_register()); | |
2836 return d; | |
2837 } else { | |
2838 return s1; | |
2839 } | |
2840 } | |
2841 | |
2842 if (s1.is_register()) { | |
2843 assert_different_registers(s1.as_register(), temp); | |
2844 if (d.is_constant()) d = temp; | |
2845 andn(s1.as_register(), ensure_simm13_or_reg(s2, temp), d.as_register()); | |
2846 return d; | |
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2847 } else { |
1503 | 2848 if (s2.is_register()) { |
2849 assert_different_registers(s2.as_register(), temp); | |
2850 if (d.is_constant()) d = temp; | |
2851 set(s1.as_constant(), temp); | |
2852 andn(temp, s2.as_register(), d.as_register()); | |
2853 return d; | |
2854 } else { | |
2855 intptr_t res = s1.as_constant() & ~s2.as_constant(); | |
2856 return res; | |
2857 } | |
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2858 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2859 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2860 |
1503 | 2861 RegisterOrConstant MacroAssembler::regcon_inc_ptr(RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp) { |
2862 assert(d.register_or_noreg() != G0, "lost side effect"); | |
2863 if ((s2.is_constant() && s2.as_constant() == 0) || | |
2864 (s2.is_register() && s2.as_register() == G0)) { | |
2865 // Do nothing, just move value. | |
2866 if (s1.is_register()) { | |
2867 if (d.is_constant()) d = temp; | |
2868 mov(s1.as_register(), d.as_register()); | |
2869 return d; | |
2870 } else { | |
2871 return s1; | |
2872 } | |
2873 } | |
2874 | |
2875 if (s1.is_register()) { | |
2876 assert_different_registers(s1.as_register(), temp); | |
2877 if (d.is_constant()) d = temp; | |
2878 add(s1.as_register(), ensure_simm13_or_reg(s2, temp), d.as_register()); | |
2879 return d; | |
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2880 } else { |
1503 | 2881 if (s2.is_register()) { |
2882 assert_different_registers(s2.as_register(), temp); | |
2883 if (d.is_constant()) d = temp; | |
2884 add(s2.as_register(), ensure_simm13_or_reg(s1, temp), d.as_register()); | |
2885 return d; | |
2886 } else { | |
2887 intptr_t res = s1.as_constant() + s2.as_constant(); | |
2888 return res; | |
2889 } | |
2890 } | |
2891 } | |
2892 | |
2893 RegisterOrConstant MacroAssembler::regcon_sll_ptr(RegisterOrConstant s1, RegisterOrConstant s2, RegisterOrConstant d, Register temp) { | |
2894 assert(d.register_or_noreg() != G0, "lost side effect"); | |
2895 if (!is_simm13(s2.constant_or_zero())) | |
2896 s2 = (s2.as_constant() & 0xFF); | |
2897 if ((s2.is_constant() && s2.as_constant() == 0) || | |
2898 (s2.is_register() && s2.as_register() == G0)) { | |
2899 // Do nothing, just move value. | |
2900 if (s1.is_register()) { | |
2901 if (d.is_constant()) d = temp; | |
2902 mov(s1.as_register(), d.as_register()); | |
2903 return d; | |
2904 } else { | |
2905 return s1; | |
2906 } | |
2907 } | |
2908 | |
2909 if (s1.is_register()) { | |
2910 assert_different_registers(s1.as_register(), temp); | |
2911 if (d.is_constant()) d = temp; | |
2912 sll_ptr(s1.as_register(), ensure_simm13_or_reg(s2, temp), d.as_register()); | |
2913 return d; | |
2914 } else { | |
2915 if (s2.is_register()) { | |
2916 assert_different_registers(s2.as_register(), temp); | |
2917 if (d.is_constant()) d = temp; | |
2918 set(s1.as_constant(), temp); | |
2919 sll_ptr(temp, s2.as_register(), d.as_register()); | |
2920 return d; | |
2921 } else { | |
2922 intptr_t res = s1.as_constant() << s2.as_constant(); | |
2923 return res; | |
2924 } | |
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2925 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2926 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2927 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2928 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2929 // Look up the method for a megamorphic invokeinterface call. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2930 // The target method is determined by <intf_klass, itable_index>. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2931 // The receiver klass is in recv_klass. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2932 // On success, the result will be in method_result, and execution falls through. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2933 // On failure, execution transfers to the given label. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2934 void MacroAssembler::lookup_interface_method(Register recv_klass, |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2935 Register intf_klass, |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2936 RegisterOrConstant itable_index, |
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2937 Register method_result, |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2938 Register scan_temp, |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2939 Register sethi_temp, |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2940 Label& L_no_such_interface) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2941 assert_different_registers(recv_klass, intf_klass, method_result, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2942 assert(itable_index.is_constant() || itable_index.as_register() == method_result, |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2943 "caller must use same register for non-constant itable index as for method"); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2944 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2945 // Compute start of first itableOffsetEntry (which is at the end of the vtable) |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2946 int vtable_base = instanceKlass::vtable_start_offset() * wordSize; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2947 int scan_step = itableOffsetEntry::size() * wordSize; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2948 int vte_size = vtableEntry::size() * wordSize; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2949 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2950 lduw(recv_klass, instanceKlass::vtable_length_offset() * wordSize, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2951 // %%% We should store the aligned, prescaled offset in the klassoop. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2952 // Then the next several instructions would fold away. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2953 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2954 int round_to_unit = ((HeapWordsPerLong > 1) ? BytesPerLong : 0); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2955 int itb_offset = vtable_base; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2956 if (round_to_unit != 0) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2957 // hoist first instruction of round_to(scan_temp, BytesPerLong): |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2958 itb_offset += round_to_unit - wordSize; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2959 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2960 int itb_scale = exact_log2(vtableEntry::size() * wordSize); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2961 sll(scan_temp, itb_scale, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2962 add(scan_temp, itb_offset, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2963 if (round_to_unit != 0) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2964 // Round up to align_object_offset boundary |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2965 // see code for instanceKlass::start_of_itable! |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2966 // Was: round_to(scan_temp, BytesPerLong); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2967 // Hoisted: add(scan_temp, BytesPerLong-1, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2968 and3(scan_temp, -round_to_unit, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2969 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2970 add(recv_klass, scan_temp, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2971 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2972 // Adjust recv_klass by scaled itable_index, so we can free itable_index. |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
2973 RegisterOrConstant itable_offset = itable_index; |
1503 | 2974 itable_offset = regcon_sll_ptr(itable_index, exact_log2(itableMethodEntry::size() * wordSize), itable_offset); |
2975 itable_offset = regcon_inc_ptr(itable_offset, itableMethodEntry::method_offset_in_bytes(), itable_offset); | |
1006
dcf03e02b020
6879902: CTW failure jdk6_18/hotspot/src/cpu/sparc/vm/assembler_sparc.hpp:845
twisti
parents:
986
diff
changeset
|
2976 add(recv_klass, ensure_simm13_or_reg(itable_offset, sethi_temp), recv_klass); |
623
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2977 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2978 // for (scan = klass->itable(); scan->interface() != NULL; scan += scan_step) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2979 // if (scan->interface() == intf) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2980 // result = (klass + scan->offset() + itable_index); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2981 // } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2982 // } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2983 Label search, found_method; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2984 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2985 for (int peel = 1; peel >= 0; peel--) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2986 // %%%% Could load both offset and interface in one ldx, if they were |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2987 // in the opposite order. This would save a load. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2988 ld_ptr(scan_temp, itableOffsetEntry::interface_offset_in_bytes(), method_result); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2989 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2990 // Check that this entry is non-null. A null entry means that |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2991 // the receiver class doesn't implement the interface, and wasn't the |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2992 // same as when the caller was compiled. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2993 bpr(Assembler::rc_z, false, Assembler::pn, method_result, L_no_such_interface); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2994 delayed()->cmp(method_result, intf_klass); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2995 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2996 if (peel) { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2997 brx(Assembler::equal, false, Assembler::pt, found_method); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2998 } else { |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
2999 brx(Assembler::notEqual, false, Assembler::pn, search); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3000 // (invert the test to fall through to found_method...) |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3001 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3002 delayed()->add(scan_temp, scan_step, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3003 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3004 if (!peel) break; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3005 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3006 bind(search); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3007 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3008 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3009 bind(found_method); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3010 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3011 // Got a hit. |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3012 int ito_offset = itableOffsetEntry::offset_offset_in_bytes(); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3013 // scan_temp[-scan_step] points to the vtable offset we need |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3014 ito_offset -= scan_step; |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3015 lduw(scan_temp, ito_offset, scan_temp); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3016 ld_ptr(recv_klass, scan_temp, method_result); |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3017 } |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3018 |
9adddb8c0fc8
6812831: factor duplicated assembly code for megamorphic invokeinterface (for 6655638)
jrose
parents:
622
diff
changeset
|
3019 |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3020 void MacroAssembler::check_klass_subtype(Register sub_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3021 Register super_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3022 Register temp_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3023 Register temp2_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3024 Label& L_success) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3025 Label L_failure, L_pop_to_failure; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3026 check_klass_subtype_fast_path(sub_klass, super_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3027 temp_reg, temp2_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3028 &L_success, &L_failure, NULL); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3029 Register sub_2 = sub_klass; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3030 Register sup_2 = super_klass; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3031 if (!sub_2->is_global()) sub_2 = L0; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3032 if (!sup_2->is_global()) sup_2 = L1; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3033 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3034 save_frame_and_mov(0, sub_klass, sub_2, super_klass, sup_2); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3035 check_klass_subtype_slow_path(sub_2, sup_2, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3036 L2, L3, L4, L5, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3037 NULL, &L_pop_to_failure); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3038 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3039 // on success: |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3040 restore(); |
3839 | 3041 ba_short(L_success); |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3042 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3043 // on failure: |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3044 bind(L_pop_to_failure); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3045 restore(); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3046 bind(L_failure); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3047 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3048 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3049 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3050 void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3051 Register super_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3052 Register temp_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3053 Register temp2_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3054 Label* L_success, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3055 Label* L_failure, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3056 Label* L_slow_path, |
3839 | 3057 RegisterOrConstant super_check_offset) { |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3058 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3059 Klass::secondary_super_cache_offset_in_bytes()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3060 int sco_offset = (klassOopDesc::header_size() * HeapWordSize + |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3061 Klass::super_check_offset_offset_in_bytes()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3062 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3063 bool must_load_sco = (super_check_offset.constant_or_zero() == -1); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3064 bool need_slow_path = (must_load_sco || |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3065 super_check_offset.constant_or_zero() == sco_offset); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3066 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3067 assert_different_registers(sub_klass, super_klass, temp_reg); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3068 if (super_check_offset.is_register()) { |
1503 | 3069 assert_different_registers(sub_klass, super_klass, temp_reg, |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3070 super_check_offset.as_register()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3071 } else if (must_load_sco) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3072 assert(temp2_reg != noreg, "supply either a temp or a register offset"); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3073 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3074 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3075 Label L_fallthrough; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3076 int label_nulls = 0; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3077 if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3078 if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3079 if (L_slow_path == NULL) { L_slow_path = &L_fallthrough; label_nulls++; } |
3839 | 3080 assert(label_nulls <= 1 || |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3081 (L_slow_path == &L_fallthrough && label_nulls <= 2 && !need_slow_path), |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3082 "at most one NULL in the batch, usually"); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3083 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3084 // If the pointers are equal, we are done (e.g., String[] elements). |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3085 // This self-check enables sharing of secondary supertype arrays among |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3086 // non-primary types such as array-of-interface. Otherwise, each such |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3087 // type would need its own customized SSA. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3088 // We move this check to the front of the fast path because many |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3089 // type checks are in fact trivially successful in this manner, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3090 // so we get a nicely predicted branch right at the start of the check. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3091 cmp(super_klass, sub_klass); |
3839 | 3092 brx(Assembler::equal, false, Assembler::pn, *L_success); |
3093 delayed()->nop(); | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3094 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3095 // Check the supertype display: |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3096 if (must_load_sco) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3097 // The super check offset is always positive... |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3098 lduw(super_klass, sco_offset, temp2_reg); |
665
c89f86385056
6814659: separable cleanups and subroutines for 6655638
jrose
parents:
644
diff
changeset
|
3099 super_check_offset = RegisterOrConstant(temp2_reg); |
1503 | 3100 // super_check_offset is register. |
3101 assert_different_registers(sub_klass, super_klass, temp_reg, super_check_offset.as_register()); | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3102 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3103 ld_ptr(sub_klass, super_check_offset, temp_reg); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3104 cmp(super_klass, temp_reg); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3105 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3106 // This check has worked decisively for primary supers. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3107 // Secondary supers are sought in the super_cache ('super_cache_addr'). |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3108 // (Secondary supers are interfaces and very deeply nested subtypes.) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3109 // This works in the same check above because of a tricky aliasing |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3110 // between the super_cache and the primary super display elements. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3111 // (The 'super_check_addr' can address either, as the case requires.) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3112 // Note that the cache is updated below if it does not help us find |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3113 // what we need immediately. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3114 // So if it was a primary super, we can just fail immediately. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3115 // Otherwise, it's the slow path for us (no success at this point). |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3116 |
3839 | 3117 // Hacked ba(), which may only be used just before L_fallthrough. |
3118 #define FINAL_JUMP(label) \ | |
3119 if (&(label) != &L_fallthrough) { \ | |
3120 ba(label); delayed()->nop(); \ | |
3121 } | |
3122 | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3123 if (super_check_offset.is_register()) { |
3839 | 3124 brx(Assembler::equal, false, Assembler::pn, *L_success); |
3125 delayed()->cmp(super_check_offset.as_register(), sc_offset); | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3126 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3127 if (L_failure == &L_fallthrough) { |
3839 | 3128 brx(Assembler::equal, false, Assembler::pt, *L_slow_path); |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3129 delayed()->nop(); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3130 } else { |
3839 | 3131 brx(Assembler::notEqual, false, Assembler::pn, *L_failure); |
3132 delayed()->nop(); | |
3133 FINAL_JUMP(*L_slow_path); | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3134 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3135 } else if (super_check_offset.as_constant() == sc_offset) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3136 // Need a slow path; fast failure is impossible. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3137 if (L_slow_path == &L_fallthrough) { |
3839 | 3138 brx(Assembler::equal, false, Assembler::pt, *L_success); |
3139 delayed()->nop(); | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3140 } else { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3141 brx(Assembler::notEqual, false, Assembler::pn, *L_slow_path); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3142 delayed()->nop(); |
3839 | 3143 FINAL_JUMP(*L_success); |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3144 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3145 } else { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3146 // No slow path; it's a fast decision. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3147 if (L_failure == &L_fallthrough) { |
3839 | 3148 brx(Assembler::equal, false, Assembler::pt, *L_success); |
3149 delayed()->nop(); | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3150 } else { |
3839 | 3151 brx(Assembler::notEqual, false, Assembler::pn, *L_failure); |
3152 delayed()->nop(); | |
3153 FINAL_JUMP(*L_success); | |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3154 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3155 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3156 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3157 bind(L_fallthrough); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3158 |
3839 | 3159 #undef FINAL_JUMP |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3160 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3161 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3162 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3163 void MacroAssembler::check_klass_subtype_slow_path(Register sub_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3164 Register super_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3165 Register count_temp, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3166 Register scan_temp, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3167 Register scratch_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3168 Register coop_reg, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3169 Label* L_success, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3170 Label* L_failure) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3171 assert_different_registers(sub_klass, super_klass, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3172 count_temp, scan_temp, scratch_reg, coop_reg); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3173 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3174 Label L_fallthrough, L_loop; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3175 int label_nulls = 0; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3176 if (L_success == NULL) { L_success = &L_fallthrough; label_nulls++; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3177 if (L_failure == NULL) { L_failure = &L_fallthrough; label_nulls++; } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3178 assert(label_nulls <= 1, "at most one NULL in the batch"); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3179 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3180 // a couple of useful fields in sub_klass: |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3181 int ss_offset = (klassOopDesc::header_size() * HeapWordSize + |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3182 Klass::secondary_supers_offset_in_bytes()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3183 int sc_offset = (klassOopDesc::header_size() * HeapWordSize + |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3184 Klass::secondary_super_cache_offset_in_bytes()); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3185 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3186 // Do a linear scan of the secondary super-klass chain. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3187 // This code is rarely used, so simplicity is a virtue here. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3188 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3189 #ifndef PRODUCT |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3190 int* pst_counter = &SharedRuntime::_partial_subtype_ctr; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3191 inc_counter((address) pst_counter, count_temp, scan_temp); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3192 #endif |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3193 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3194 // We will consult the secondary-super array. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3195 ld_ptr(sub_klass, ss_offset, scan_temp); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3196 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3197 // Compress superclass if necessary. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3198 Register search_key = super_klass; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3199 bool decode_super_klass = false; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3200 if (UseCompressedOops) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3201 if (coop_reg != noreg) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3202 encode_heap_oop_not_null(super_klass, coop_reg); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3203 search_key = coop_reg; |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3204 } else { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3205 encode_heap_oop_not_null(super_klass); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3206 decode_super_klass = true; // scarce temps! |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3207 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3208 // The superclass is never null; it would be a basic system error if a null |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3209 // pointer were to sneak in here. Note that we have already loaded the |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3210 // Klass::super_check_offset from the super_klass in the fast path, |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3211 // so if there is a null in that register, we are already in the afterlife. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3212 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3213 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3214 // Load the array length. (Positive movl does right thing on LP64.) |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3215 lduw(scan_temp, arrayOopDesc::length_offset_in_bytes(), count_temp); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3216 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3217 // Check for empty secondary super list |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3218 tst(count_temp); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3219 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3220 // Top of search loop |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3221 bind(L_loop); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3222 br(Assembler::equal, false, Assembler::pn, *L_failure); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3223 delayed()->add(scan_temp, heapOopSize, scan_temp); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3224 assert(heapOopSize != 0, "heapOopSize should be initialized"); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3225 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3226 // Skip the array header in all array accesses. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3227 int elem_offset = arrayOopDesc::base_offset_in_bytes(T_OBJECT); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3228 elem_offset -= heapOopSize; // the scan pointer was pre-incremented also |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3229 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3230 // Load next super to check |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3231 if (UseCompressedOops) { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3232 // Don't use load_heap_oop; we don't want to decode the element. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3233 lduw( scan_temp, elem_offset, scratch_reg ); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3234 } else { |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3235 ld_ptr( scan_temp, elem_offset, scratch_reg ); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3236 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3237 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3238 // Look for Rsuper_klass on Rsub_klass's secondary super-class-overflow list |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3239 cmp(scratch_reg, search_key); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3240 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3241 // A miss means we are NOT a subtype and need to keep looping |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3242 brx(Assembler::notEqual, false, Assembler::pn, L_loop); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3243 delayed()->deccc(count_temp); // decrement trip counter in delay slot |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3244 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3245 // Falling out the bottom means we found a hit; we ARE a subtype |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3246 if (decode_super_klass) decode_heap_oop(super_klass); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3247 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3248 // Success. Cache the super we found and proceed in triumph. |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3249 st_ptr(super_klass, sub_klass, sc_offset); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3250 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3251 if (L_success != &L_fallthrough) { |
3839 | 3252 ba(*L_success); |
644
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3253 delayed()->nop(); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3254 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3255 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3256 bind(L_fallthrough); |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3257 } |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3258 |
c517646eef23
6813212: factor duplicated assembly code for general subclass check (for 6655638)
jrose
parents:
642
diff
changeset
|
3259 |
710 | 3260 void MacroAssembler::check_method_handle_type(Register mtype_reg, Register mh_reg, |
3261 Register temp_reg, | |
3262 Label& wrong_method_type) { | |
3263 assert_different_registers(mtype_reg, mh_reg, temp_reg); | |
3264 // compare method type against that of the receiver | |
2357
8033953d67ff
7012648: move JSR 292 to package java.lang.invoke and adjust names
jrose
parents:
2258
diff
changeset
|
3265 RegisterOrConstant mhtype_offset = delayed_value(java_lang_invoke_MethodHandle::type_offset_in_bytes, temp_reg); |
1846 | 3266 load_heap_oop(mh_reg, mhtype_offset, temp_reg); |
3839 | 3267 cmp_and_brx_short(temp_reg, mtype_reg, Assembler::notEqual, Assembler::pn, wrong_method_type); |
710 | 3268 } |
3269 | |
3270 | |
1503 | 3271 // A method handle has a "vmslots" field which gives the size of its |
3272 // argument list in JVM stack slots. This field is either located directly | |
3273 // in every method handle, or else is indirectly accessed through the | |
3274 // method handle's MethodType. This macro hides the distinction. | |
3275 void MacroAssembler::load_method_handle_vmslots(Register vmslots_reg, Register mh_reg, | |
3276 Register temp_reg) { | |
3277 assert_different_registers(vmslots_reg, mh_reg, temp_reg); | |
3278 // load mh.type.form.vmslots | |
2357
8033953d67ff
7012648: move JSR 292 to package java.lang.invoke and adjust names
jrose
parents:
2258
diff
changeset
|
3279 if (java_lang_invoke_MethodHandle::vmslots_offset_in_bytes() != 0) { |
1503 | 3280 // hoist vmslots into every mh to avoid dependent load chain |
2357
8033953d67ff
7012648: move JSR 292 to package java.lang.invoke and adjust names
jrose
parents:
2258
diff
changeset
|
3281 ld( Address(mh_reg, delayed_value(java_lang_invoke_MethodHandle::vmslots_offset_in_bytes, temp_reg)), vmslots_reg); |
1503 | 3282 } else { |
3283 Register temp2_reg = vmslots_reg; | |
2357
8033953d67ff
7012648: move JSR 292 to package java.lang.invoke and adjust names
jrose
parents:
2258
diff
changeset
|
3284 load_heap_oop(Address(mh_reg, delayed_value(java_lang_invoke_MethodHandle::type_offset_in_bytes, temp_reg)), temp2_reg); |
8033953d67ff
7012648: move JSR 292 to package java.lang.invoke and adjust names
jrose
parents:
2258
diff
changeset
|
3285 load_heap_oop(Address(temp2_reg, delayed_value(java_lang_invoke_MethodType::form_offset_in_bytes, temp_reg)), temp2_reg); |
8033953d67ff
7012648: move JSR 292 to package java.lang.invoke and adjust names
jrose
parents:
2258
diff
changeset
|
3286 ld( Address(temp2_reg, delayed_value(java_lang_invoke_MethodTypeForm::vmslots_offset_in_bytes, temp_reg)), vmslots_reg); |
1503 | 3287 } |
3288 } | |
3289 | |
3290 | |
3291 void MacroAssembler::jump_to_method_handle_entry(Register mh_reg, Register temp_reg, bool emit_delayed_nop) { | |
710 | 3292 assert(mh_reg == G3_method_handle, "caller must put MH object in G3"); |
3293 assert_different_registers(mh_reg, temp_reg); | |
3294 | |
3295 // pick out the interpreted side of the handler | |
1846 | 3296 // NOTE: vmentry is not an oop! |
2357
8033953d67ff
7012648: move JSR 292 to package java.lang.invoke and adjust names
jrose
parents:
2258
diff
changeset
|
3297 ld_ptr(mh_reg, delayed_value(java_lang_invoke_MethodHandle::vmentry_offset_in_bytes, temp_reg), temp_reg); |
710 | 3298 |
3299 // off we go... | |
3300 ld_ptr(temp_reg, MethodHandleEntry::from_interpreted_entry_offset_in_bytes(), temp_reg); | |
3301 jmp(temp_reg, 0); | |
3302 | |
3303 // for the various stubs which take control at this point, | |
3304 // see MethodHandles::generate_method_handle_stub | |
3305 | |
1503 | 3306 // Some callers can fill the delay slot. |
3307 if (emit_delayed_nop) { | |
3308 delayed()->nop(); | |
3309 } | |
710 | 3310 } |
3311 | |
1503 | 3312 |
710 | 3313 RegisterOrConstant MacroAssembler::argument_offset(RegisterOrConstant arg_slot, |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
3314 Register temp_reg, |
710 | 3315 int extra_slot_offset) { |
3316 // cf. TemplateTable::prepare_invoke(), if (load_receiver). | |
1506 | 3317 int stackElementSize = Interpreter::stackElementSize; |
1503 | 3318 int offset = extra_slot_offset * stackElementSize; |
710 | 3319 if (arg_slot.is_constant()) { |
3320 offset += arg_slot.as_constant() * stackElementSize; | |
3321 return offset; | |
3322 } else { | |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
3323 assert(temp_reg != noreg, "must specify"); |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
3324 sll_ptr(arg_slot.as_register(), exact_log2(stackElementSize), temp_reg); |
710 | 3325 if (offset != 0) |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
3326 add(temp_reg, offset, temp_reg); |
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
3327 return temp_reg; |
710 | 3328 } |
3329 } | |
3330 | |
3331 | |
1503 | 3332 Address MacroAssembler::argument_address(RegisterOrConstant arg_slot, |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
3333 Register temp_reg, |
1503 | 3334 int extra_slot_offset) { |
3753
cba7b5c2d53f
7045514: SPARC assembly code for JSR 292 ricochet frames
never
parents:
3249
diff
changeset
|
3335 return Address(Gargs, argument_offset(arg_slot, temp_reg, extra_slot_offset)); |
1503 | 3336 } |
3337 | |
710 | 3338 |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3339 void MacroAssembler::biased_locking_enter(Register obj_reg, Register mark_reg, |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3340 Register temp_reg, |
0 | 3341 Label& done, Label* slow_case, |
3342 BiasedLockingCounters* counters) { | |
3343 assert(UseBiasedLocking, "why call this otherwise?"); | |
3344 | |
3345 if (PrintBiasedLockingStatistics) { | |
3346 assert_different_registers(obj_reg, mark_reg, temp_reg, O7); | |
3347 if (counters == NULL) | |
3348 counters = BiasedLocking::counters(); | |
3349 } | |
3350 | |
3351 Label cas_label; | |
3352 | |
3353 // Biased locking | |
3354 // See whether the lock is currently biased toward our thread and | |
3355 // whether the epoch is still valid | |
3356 // Note that the runtime guarantees sufficient alignment of JavaThread | |
3357 // pointers to allow age to be placed into low bits | |
3358 assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout"); | |
3359 and3(mark_reg, markOopDesc::biased_lock_mask_in_place, temp_reg); | |
3839 | 3360 cmp_and_brx_short(temp_reg, markOopDesc::biased_lock_pattern, Assembler::notEqual, Assembler::pn, cas_label); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
3361 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
3362 load_klass(obj_reg, temp_reg); |
727 | 3363 ld_ptr(Address(temp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg); |
0 | 3364 or3(G2_thread, temp_reg, temp_reg); |
3365 xor3(mark_reg, temp_reg, temp_reg); | |
3366 andcc(temp_reg, ~((int) markOopDesc::age_mask_in_place), temp_reg); | |
3367 if (counters != NULL) { | |
3368 cond_inc(Assembler::equal, (address) counters->biased_lock_entry_count_addr(), mark_reg, temp_reg); | |
3369 // Reload mark_reg as we may need it later | |
727 | 3370 ld_ptr(Address(obj_reg, oopDesc::mark_offset_in_bytes()), mark_reg); |
0 | 3371 } |
3372 brx(Assembler::equal, true, Assembler::pt, done); | |
3373 delayed()->nop(); | |
3374 | |
3375 Label try_revoke_bias; | |
3376 Label try_rebias; | |
727 | 3377 Address mark_addr = Address(obj_reg, oopDesc::mark_offset_in_bytes()); |
0 | 3378 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); |
3379 | |
3380 // At this point we know that the header has the bias pattern and | |
3381 // that we are not the bias owner in the current epoch. We need to | |
3382 // figure out more details about the state of the header in order to | |
3383 // know what operations can be legally performed on the object's | |
3384 // header. | |
3385 | |
3386 // If the low three bits in the xor result aren't clear, that means | |
3387 // the prototype header is no longer biased and we have to revoke | |
3388 // the bias on this object. | |
3389 btst(markOopDesc::biased_lock_mask_in_place, temp_reg); | |
3390 brx(Assembler::notZero, false, Assembler::pn, try_revoke_bias); | |
3391 | |
3392 // Biasing is still enabled for this data type. See whether the | |
3393 // epoch of the current bias is still valid, meaning that the epoch | |
3394 // bits of the mark word are equal to the epoch bits of the | |
3395 // prototype header. (Note that the prototype header's epoch bits | |
3396 // only change at a safepoint.) If not, attempt to rebias the object | |
3397 // toward the current thread. Note that we must be absolutely sure | |
3398 // that the current epoch is invalid in order to do this because | |
3399 // otherwise the manipulations it performs on the mark word are | |
3400 // illegal. | |
3401 delayed()->btst(markOopDesc::epoch_mask_in_place, temp_reg); | |
3402 brx(Assembler::notZero, false, Assembler::pn, try_rebias); | |
3403 | |
3404 // The epoch of the current bias is still valid but we know nothing | |
3405 // about the owner; it might be set or it might be clear. Try to | |
3406 // acquire the bias of the object using an atomic operation. If this | |
3407 // fails we will go in to the runtime to revoke the object's bias. | |
3408 // Note that we first construct the presumed unbiased header so we | |
3409 // don't accidentally blow away another thread's valid bias. | |
3410 delayed()->and3(mark_reg, | |
3411 markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place, | |
3412 mark_reg); | |
3413 or3(G2_thread, mark_reg, temp_reg); | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3414 casn(mark_addr.base(), mark_reg, temp_reg); |
0 | 3415 // If the biasing toward our thread failed, this means that |
3416 // another thread succeeded in biasing it toward itself and we | |
3417 // need to revoke that bias. The revocation will occur in the | |
3418 // interpreter runtime in the slow case. | |
3419 cmp(mark_reg, temp_reg); | |
3420 if (counters != NULL) { | |
3421 cond_inc(Assembler::zero, (address) counters->anonymously_biased_lock_entry_count_addr(), mark_reg, temp_reg); | |
3422 } | |
3423 if (slow_case != NULL) { | |
3424 brx(Assembler::notEqual, true, Assembler::pn, *slow_case); | |
3425 delayed()->nop(); | |
3426 } | |
3839 | 3427 ba_short(done); |
0 | 3428 |
3429 bind(try_rebias); | |
3430 // At this point we know the epoch has expired, meaning that the | |
3431 // current "bias owner", if any, is actually invalid. Under these | |
3432 // circumstances _only_, we are allowed to use the current header's | |
3433 // value as the comparison value when doing the cas to acquire the | |
3434 // bias in the current epoch. In other words, we allow transfer of | |
3435 // the bias from one thread to another directly in this situation. | |
3436 // | |
3437 // FIXME: due to a lack of registers we currently blow away the age | |
3438 // bits in this situation. Should attempt to preserve them. | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
3439 load_klass(obj_reg, temp_reg); |
727 | 3440 ld_ptr(Address(temp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg); |
0 | 3441 or3(G2_thread, temp_reg, temp_reg); |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3442 casn(mark_addr.base(), mark_reg, temp_reg); |
0 | 3443 // If the biasing toward our thread failed, this means that |
3444 // another thread succeeded in biasing it toward itself and we | |
3445 // need to revoke that bias. The revocation will occur in the | |
3446 // interpreter runtime in the slow case. | |
3447 cmp(mark_reg, temp_reg); | |
3448 if (counters != NULL) { | |
3449 cond_inc(Assembler::zero, (address) counters->rebiased_lock_entry_count_addr(), mark_reg, temp_reg); | |
3450 } | |
3451 if (slow_case != NULL) { | |
3452 brx(Assembler::notEqual, true, Assembler::pn, *slow_case); | |
3453 delayed()->nop(); | |
3454 } | |
3839 | 3455 ba_short(done); |
0 | 3456 |
3457 bind(try_revoke_bias); | |
3458 // The prototype mark in the klass doesn't have the bias bit set any | |
3459 // more, indicating that objects of this data type are not supposed | |
3460 // to be biased any more. We are going to try to reset the mark of | |
3461 // this object to the prototype value and fall through to the | |
3462 // CAS-based locking scheme. Note that if our CAS fails, it means | |
3463 // that another thread raced us for the privilege of revoking the | |
3464 // bias of this particular object, so it's okay to continue in the | |
3465 // normal locking code. | |
3466 // | |
3467 // FIXME: due to a lack of registers we currently blow away the age | |
3468 // bits in this situation. Should attempt to preserve them. | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
3469 load_klass(obj_reg, temp_reg); |
727 | 3470 ld_ptr(Address(temp_reg, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg); |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3471 casn(mark_addr.base(), mark_reg, temp_reg); |
0 | 3472 // Fall through to the normal CAS-based lock, because no matter what |
3473 // the result of the above CAS, some thread must have succeeded in | |
3474 // removing the bias bit from the object's header. | |
3475 if (counters != NULL) { | |
3476 cmp(mark_reg, temp_reg); | |
3477 cond_inc(Assembler::zero, (address) counters->revoked_lock_entry_count_addr(), mark_reg, temp_reg); | |
3478 } | |
3479 | |
3480 bind(cas_label); | |
3481 } | |
3482 | |
3483 void MacroAssembler::biased_locking_exit (Address mark_addr, Register temp_reg, Label& done, | |
3484 bool allow_delay_slot_filling) { | |
3485 // Check for biased locking unlock case, which is a no-op | |
3486 // Note: we do not have to check the thread ID for two reasons. | |
3487 // First, the interpreter checks for IllegalMonitorStateException at | |
3488 // a higher level. Second, if the bias was revoked while we held the | |
3489 // lock, the object could not be rebiased toward another thread, so | |
3490 // the bias bit would be clear. | |
3491 ld_ptr(mark_addr, temp_reg); | |
3492 and3(temp_reg, markOopDesc::biased_lock_mask_in_place, temp_reg); | |
3493 cmp(temp_reg, markOopDesc::biased_lock_pattern); | |
3494 brx(Assembler::equal, allow_delay_slot_filling, Assembler::pt, done); | |
3495 delayed(); | |
3496 if (!allow_delay_slot_filling) { | |
3497 nop(); | |
3498 } | |
3499 } | |
3500 | |
3501 | |
3502 // CASN -- 32-64 bit switch hitter similar to the synthetic CASN provided by | |
3503 // Solaris/SPARC's "as". Another apt name would be cas_ptr() | |
3504 | |
3505 void MacroAssembler::casn (Register addr_reg, Register cmp_reg, Register set_reg ) { | |
3839 | 3506 casx_under_lock (addr_reg, cmp_reg, set_reg, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); |
0 | 3507 } |
3508 | |
3509 | |
3510 | |
3511 // compiler_lock_object() and compiler_unlock_object() are direct transliterations | |
3512 // of i486.ad fast_lock() and fast_unlock(). See those methods for detailed comments. | |
3513 // The code could be tightened up considerably. | |
3514 // | |
3515 // box->dhw disposition - post-conditions at DONE_LABEL. | |
3516 // - Successful inflated lock: box->dhw != 0. | |
3517 // Any non-zero value suffices. | |
3518 // Consider G2_thread, rsp, boxReg, or unused_mark() | |
3519 // - Successful Stack-lock: box->dhw == mark. | |
3520 // box->dhw must contain the displaced mark word value | |
3521 // - Failure -- icc.ZFlag == 0 and box->dhw is undefined. | |
3522 // The slow-path fast_enter() and slow_enter() operators | |
3523 // are responsible for setting box->dhw = NonZero (typically ::unused_mark). | |
3524 // - Biased: box->dhw is undefined | |
3525 // | |
3526 // SPARC refworkload performance - specifically jetstream and scimark - are | |
3527 // extremely sensitive to the size of the code emitted by compiler_lock_object | |
3528 // and compiler_unlock_object. Critically, the key factor is code size, not path | |
3529 // length. (Simply experiments to pad CLO with unexecuted NOPs demonstrte the | |
3530 // effect). | |
3531 | |
3532 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3533 void MacroAssembler::compiler_lock_object(Register Roop, Register Rmark, |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3534 Register Rbox, Register Rscratch, |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3535 BiasedLockingCounters* counters, |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3536 bool try_bias) { |
727 | 3537 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes()); |
0 | 3538 |
3539 verify_oop(Roop); | |
3540 Label done ; | |
3541 | |
3542 if (counters != NULL) { | |
3543 inc_counter((address) counters->total_entry_count_addr(), Rmark, Rscratch); | |
3544 } | |
3545 | |
3546 if (EmitSync & 1) { | |
3839 | 3547 mov(3, Rscratch); |
3548 st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
3549 cmp(SP, G0); | |
0 | 3550 return ; |
3551 } | |
3552 | |
3553 if (EmitSync & 2) { | |
3554 | |
3555 // Fetch object's markword | |
3556 ld_ptr(mark_addr, Rmark); | |
3557 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3558 if (try_bias) { |
0 | 3559 biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters); |
3560 } | |
3561 | |
3562 // Save Rbox in Rscratch to be used for the cas operation | |
3563 mov(Rbox, Rscratch); | |
3564 | |
3565 // set Rmark to markOop | markOopDesc::unlocked_value | |
3566 or3(Rmark, markOopDesc::unlocked_value, Rmark); | |
3567 | |
3568 // Initialize the box. (Must happen before we update the object mark!) | |
3569 st_ptr(Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
3570 | |
3571 // compare object markOop with Rmark and if equal exchange Rscratch with object markOop | |
3572 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
3573 casx_under_lock(mark_addr.base(), Rmark, Rscratch, | |
3574 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
3575 | |
3576 // if compare/exchange succeeded we found an unlocked object and we now have locked it | |
3577 // hence we are done | |
3578 cmp(Rmark, Rscratch); | |
3579 #ifdef _LP64 | |
3580 sub(Rscratch, STACK_BIAS, Rscratch); | |
3581 #endif | |
3582 brx(Assembler::equal, false, Assembler::pt, done); | |
3583 delayed()->sub(Rscratch, SP, Rscratch); //pull next instruction into delay slot | |
3584 | |
3585 // we did not find an unlocked object so see if this is a recursive case | |
3586 // sub(Rscratch, SP, Rscratch); | |
3587 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); | |
3588 andcc(Rscratch, 0xfffff003, Rscratch); | |
3589 st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
3839 | 3590 bind (done); |
0 | 3591 return ; |
3592 } | |
3593 | |
3594 Label Egress ; | |
3595 | |
3596 if (EmitSync & 256) { | |
3597 Label IsInflated ; | |
3598 | |
3839 | 3599 ld_ptr(mark_addr, Rmark); // fetch obj->mark |
0 | 3600 // Triage: biased, stack-locked, neutral, inflated |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3601 if (try_bias) { |
0 | 3602 biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters); |
3603 // Invariant: if control reaches this point in the emitted stream | |
3604 // then Rmark has not been modified. | |
3605 } | |
3606 | |
3607 // Store mark into displaced mark field in the on-stack basic-lock "box" | |
3608 // Critically, this must happen before the CAS | |
3609 // Maximize the ST-CAS distance to minimize the ST-before-CAS penalty. | |
3839 | 3610 st_ptr(Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes()); |
3611 andcc(Rmark, 2, G0); | |
3612 brx(Assembler::notZero, false, Assembler::pn, IsInflated); | |
3613 delayed()-> | |
0 | 3614 |
3615 // Try stack-lock acquisition. | |
3616 // Beware: the 1st instruction is in a delay slot | |
3839 | 3617 mov(Rbox, Rscratch); |
3618 or3(Rmark, markOopDesc::unlocked_value, Rmark); | |
3619 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
3620 casn(mark_addr.base(), Rmark, Rscratch); | |
3621 cmp(Rmark, Rscratch); | |
3622 brx(Assembler::equal, false, Assembler::pt, done); | |
0 | 3623 delayed()->sub(Rscratch, SP, Rscratch); |
3624 | |
3625 // Stack-lock attempt failed - check for recursive stack-lock. | |
3626 // See the comments below about how we might remove this case. | |
3627 #ifdef _LP64 | |
3839 | 3628 sub(Rscratch, STACK_BIAS, Rscratch); |
0 | 3629 #endif |
3630 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); | |
3839 | 3631 andcc(Rscratch, 0xfffff003, Rscratch); |
3632 br(Assembler::always, false, Assembler::pt, done); | |
3633 delayed()-> st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
3634 | |
3635 bind(IsInflated); | |
0 | 3636 if (EmitSync & 64) { |
3637 // If m->owner != null goto IsLocked | |
3638 // Pessimistic form: Test-and-CAS vs CAS | |
3639 // The optimistic form avoids RTS->RTO cache line upgrades. | |
3839 | 3640 ld_ptr(Rmark, ObjectMonitor::owner_offset_in_bytes() - 2, Rscratch); |
3641 andcc(Rscratch, Rscratch, G0); | |
3642 brx(Assembler::notZero, false, Assembler::pn, done); | |
3643 delayed()->nop(); | |
0 | 3644 // m->owner == null : it's unlocked. |
3645 } | |
3646 | |
3647 // Try to CAS m->owner from null to Self | |
3648 // Invariant: if we acquire the lock then _recursions should be 0. | |
3839 | 3649 add(Rmark, ObjectMonitor::owner_offset_in_bytes()-2, Rmark); |
3650 mov(G2_thread, Rscratch); | |
3651 casn(Rmark, G0, Rscratch); | |
3652 cmp(Rscratch, G0); | |
0 | 3653 // Intentional fall-through into done |
3654 } else { | |
3655 // Aggressively avoid the Store-before-CAS penalty | |
3656 // Defer the store into box->dhw until after the CAS | |
3657 Label IsInflated, Recursive ; | |
3658 | |
3659 // Anticipate CAS -- Avoid RTS->RTO upgrade | |
3839 | 3660 // prefetch (mark_addr, Assembler::severalWritesAndPossiblyReads); |
3661 | |
3662 ld_ptr(mark_addr, Rmark); // fetch obj->mark | |
0 | 3663 // Triage: biased, stack-locked, neutral, inflated |
3664 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3665 if (try_bias) { |
0 | 3666 biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters); |
3667 // Invariant: if control reaches this point in the emitted stream | |
3668 // then Rmark has not been modified. | |
3669 } | |
3839 | 3670 andcc(Rmark, 2, G0); |
3671 brx(Assembler::notZero, false, Assembler::pn, IsInflated); | |
0 | 3672 delayed()-> // Beware - dangling delay-slot |
3673 | |
3674 // Try stack-lock acquisition. | |
3675 // Transiently install BUSY (0) encoding in the mark word. | |
3676 // if the CAS of 0 into the mark was successful then we execute: | |
3677 // ST box->dhw = mark -- save fetched mark in on-stack basiclock box | |
3678 // ST obj->mark = box -- overwrite transient 0 value | |
3679 // This presumes TSO, of course. | |
3680 | |
3839 | 3681 mov(0, Rscratch); |
3682 or3(Rmark, markOopDesc::unlocked_value, Rmark); | |
3683 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
3684 casn(mark_addr.base(), Rmark, Rscratch); | |
3685 // prefetch (mark_addr, Assembler::severalWritesAndPossiblyReads); | |
3686 cmp(Rscratch, Rmark); | |
3687 brx(Assembler::notZero, false, Assembler::pn, Recursive); | |
3688 delayed()->st_ptr(Rmark, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
0 | 3689 if (counters != NULL) { |
3690 cond_inc(Assembler::equal, (address) counters->fast_path_entry_count_addr(), Rmark, Rscratch); | |
3691 } | |
3839 | 3692 ba(done); |
3693 delayed()->st_ptr(Rbox, mark_addr); | |
3694 | |
3695 bind(Recursive); | |
0 | 3696 // Stack-lock attempt failed - check for recursive stack-lock. |
3697 // Tests show that we can remove the recursive case with no impact | |
3698 // on refworkload 0.83. If we need to reduce the size of the code | |
3699 // emitted by compiler_lock_object() the recursive case is perfect | |
3700 // candidate. | |
3701 // | |
3702 // A more extreme idea is to always inflate on stack-lock recursion. | |
3703 // This lets us eliminate the recursive checks in compiler_lock_object | |
3704 // and compiler_unlock_object and the (box->dhw == 0) encoding. | |
3705 // A brief experiment - requiring changes to synchronizer.cpp, interpreter, | |
3706 // and showed a performance *increase*. In the same experiment I eliminated | |
3707 // the fast-path stack-lock code from the interpreter and always passed | |
3708 // control to the "slow" operators in synchronizer.cpp. | |
3709 | |
3710 // RScratch contains the fetched obj->mark value from the failed CASN. | |
3711 #ifdef _LP64 | |
3839 | 3712 sub(Rscratch, STACK_BIAS, Rscratch); |
0 | 3713 #endif |
3714 sub(Rscratch, SP, Rscratch); | |
3715 assert(os::vm_page_size() > 0xfff, "page size too small - change the constant"); | |
3839 | 3716 andcc(Rscratch, 0xfffff003, Rscratch); |
0 | 3717 if (counters != NULL) { |
3718 // Accounting needs the Rscratch register | |
3839 | 3719 st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes()); |
0 | 3720 cond_inc(Assembler::equal, (address) counters->fast_path_entry_count_addr(), Rmark, Rscratch); |
3839 | 3721 ba_short(done); |
0 | 3722 } else { |
3839 | 3723 ba(done); |
3724 delayed()->st_ptr(Rscratch, Rbox, BasicLock::displaced_header_offset_in_bytes()); | |
0 | 3725 } |
3726 | |
3839 | 3727 bind (IsInflated); |
0 | 3728 if (EmitSync & 64) { |
3729 // If m->owner != null goto IsLocked | |
3730 // Test-and-CAS vs CAS | |
3731 // Pessimistic form avoids futile (doomed) CAS attempts | |
3732 // The optimistic form avoids RTS->RTO cache line upgrades. | |
3839 | 3733 ld_ptr(Rmark, ObjectMonitor::owner_offset_in_bytes() - 2, Rscratch); |
3734 andcc(Rscratch, Rscratch, G0); | |
3735 brx(Assembler::notZero, false, Assembler::pn, done); | |
3736 delayed()->nop(); | |
0 | 3737 // m->owner == null : it's unlocked. |
3738 } | |
3739 | |
3740 // Try to CAS m->owner from null to Self | |
3741 // Invariant: if we acquire the lock then _recursions should be 0. | |
3839 | 3742 add(Rmark, ObjectMonitor::owner_offset_in_bytes()-2, Rmark); |
3743 mov(G2_thread, Rscratch); | |
3744 casn(Rmark, G0, Rscratch); | |
3745 cmp(Rscratch, G0); | |
0 | 3746 // ST box->displaced_header = NonZero. |
3747 // Any non-zero value suffices: | |
3748 // unused_mark(), G2_thread, RBox, RScratch, rsp, etc. | |
3839 | 3749 st_ptr(Rbox, Rbox, BasicLock::displaced_header_offset_in_bytes()); |
0 | 3750 // Intentional fall-through into done |
3751 } | |
3752 | |
3839 | 3753 bind (done); |
0 | 3754 } |
3755 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3756 void MacroAssembler::compiler_unlock_object(Register Roop, Register Rmark, |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3757 Register Rbox, Register Rscratch, |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3758 bool try_bias) { |
727 | 3759 Address mark_addr(Roop, oopDesc::mark_offset_in_bytes()); |
0 | 3760 |
3761 Label done ; | |
3762 | |
3763 if (EmitSync & 4) { | |
3839 | 3764 cmp(SP, G0); |
0 | 3765 return ; |
3766 } | |
3767 | |
3768 if (EmitSync & 8) { | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3769 if (try_bias) { |
0 | 3770 biased_locking_exit(mark_addr, Rscratch, done); |
3771 } | |
3772 | |
3773 // Test first if it is a fast recursive unlock | |
3774 ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark); | |
3839 | 3775 br_null_short(Rmark, Assembler::pt, done); |
0 | 3776 |
3777 // Check if it is still a light weight lock, this is is true if we see | |
3778 // the stack address of the basicLock in the markOop of the object | |
3779 assert(mark_addr.disp() == 0, "cas must take a zero displacement"); | |
3780 casx_under_lock(mark_addr.base(), Rbox, Rmark, | |
3781 (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); | |
3839 | 3782 ba(done); |
0 | 3783 delayed()->cmp(Rbox, Rmark); |
3839 | 3784 bind(done); |
0 | 3785 return ; |
3786 } | |
3787 | |
3788 // Beware ... If the aggregate size of the code emitted by CLO and CUO is | |
3789 // is too large performance rolls abruptly off a cliff. | |
3790 // This could be related to inlining policies, code cache management, or | |
3791 // I$ effects. | |
3792 Label LStacked ; | |
3793 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
356
diff
changeset
|
3794 if (try_bias) { |
0 | 3795 // TODO: eliminate redundant LDs of obj->mark |
3796 biased_locking_exit(mark_addr, Rscratch, done); | |
3797 } | |
3798 | |
3839 | 3799 ld_ptr(Roop, oopDesc::mark_offset_in_bytes(), Rmark); |
3800 ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rscratch); | |
3801 andcc(Rscratch, Rscratch, G0); | |
3802 brx(Assembler::zero, false, Assembler::pn, done); | |
3803 delayed()->nop(); // consider: relocate fetch of mark, above, into this DS | |
3804 andcc(Rmark, 2, G0); | |
3805 brx(Assembler::zero, false, Assembler::pt, LStacked); | |
3806 delayed()->nop(); | |
0 | 3807 |
3808 // It's inflated | |
3809 // Conceptually we need a #loadstore|#storestore "release" MEMBAR before | |
3810 // the ST of 0 into _owner which releases the lock. This prevents loads | |
3811 // and stores within the critical section from reordering (floating) | |
3812 // past the store that releases the lock. But TSO is a strong memory model | |
3813 // and that particular flavor of barrier is a noop, so we can safely elide it. | |
3814 // Note that we use 1-0 locking by default for the inflated case. We | |
3815 // close the resultant (and rare) race by having contented threads in | |
3816 // monitorenter periodically poll _owner. | |
3839 | 3817 ld_ptr(Rmark, ObjectMonitor::owner_offset_in_bytes() - 2, Rscratch); |
3818 ld_ptr(Rmark, ObjectMonitor::recursions_offset_in_bytes() - 2, Rbox); | |
3819 xor3(Rscratch, G2_thread, Rscratch); | |
3820 orcc(Rbox, Rscratch, Rbox); | |
3821 brx(Assembler::notZero, false, Assembler::pn, done); | |
0 | 3822 delayed()-> |
3839 | 3823 ld_ptr(Rmark, ObjectMonitor::EntryList_offset_in_bytes() - 2, Rscratch); |
3824 ld_ptr(Rmark, ObjectMonitor::cxq_offset_in_bytes() - 2, Rbox); | |
3825 orcc(Rbox, Rscratch, G0); | |
0 | 3826 if (EmitSync & 65536) { |
3827 Label LSucc ; | |
3839 | 3828 brx(Assembler::notZero, false, Assembler::pn, LSucc); |
3829 delayed()->nop(); | |
3830 ba(done); | |
3831 delayed()->st_ptr(G0, Rmark, ObjectMonitor::owner_offset_in_bytes() - 2); | |
3832 | |
3833 bind(LSucc); | |
3834 st_ptr(G0, Rmark, ObjectMonitor::owner_offset_in_bytes() - 2); | |
3835 if (os::is_MP()) { membar (StoreLoad); } | |
3836 ld_ptr(Rmark, ObjectMonitor::succ_offset_in_bytes() - 2, Rscratch); | |
3837 andcc(Rscratch, Rscratch, G0); | |
3838 brx(Assembler::notZero, false, Assembler::pt, done); | |
3839 delayed()->andcc(G0, G0, G0); | |
3840 add(Rmark, ObjectMonitor::owner_offset_in_bytes()-2, Rmark); | |
3841 mov(G2_thread, Rscratch); | |
3842 casn(Rmark, G0, Rscratch); | |
0 | 3843 // invert icc.zf and goto done |
3839 | 3844 br_notnull(Rscratch, false, Assembler::pt, done); |
3845 delayed()->cmp(G0, G0); | |
3846 ba(done); | |
3847 delayed()->cmp(G0, 1); | |
0 | 3848 } else { |
3839 | 3849 brx(Assembler::notZero, false, Assembler::pn, done); |
3850 delayed()->nop(); | |
3851 ba(done); | |
3852 delayed()->st_ptr(G0, Rmark, ObjectMonitor::owner_offset_in_bytes() - 2); | |
0 | 3853 } |
3854 | |
3839 | 3855 bind (LStacked); |
0 | 3856 // Consider: we could replace the expensive CAS in the exit |
3857 // path with a simple ST of the displaced mark value fetched from | |
3858 // the on-stack basiclock box. That admits a race where a thread T2 | |
3859 // in the slow lock path -- inflating with monitor M -- could race a | |
3860 // thread T1 in the fast unlock path, resulting in a missed wakeup for T2. | |
3861 // More precisely T1 in the stack-lock unlock path could "stomp" the | |
3862 // inflated mark value M installed by T2, resulting in an orphan | |
3863 // object monitor M and T2 becoming stranded. We can remedy that situation | |
3864 // by having T2 periodically poll the object's mark word using timed wait | |
3865 // operations. If T2 discovers that a stomp has occurred it vacates | |
3866 // the monitor M and wakes any other threads stranded on the now-orphan M. | |
3867 // In addition the monitor scavenger, which performs deflation, | |
3868 // would also need to check for orpan monitors and stranded threads. | |
3869 // | |
3870 // Finally, inflation is also used when T2 needs to assign a hashCode | |
3871 // to O and O is stack-locked by T1. The "stomp" race could cause | |
3872 // an assigned hashCode value to be lost. We can avoid that condition | |
3873 // and provide the necessary hashCode stability invariants by ensuring | |
3874 // that hashCode generation is idempotent between copying GCs. | |
3875 // For example we could compute the hashCode of an object O as | |
3876 // O's heap address XOR some high quality RNG value that is refreshed | |
3877 // at GC-time. The monitor scavenger would install the hashCode | |
3878 // found in any orphan monitors. Again, the mechanism admits a | |
3879 // lost-update "stomp" WAW race but detects and recovers as needed. | |
3880 // | |
3881 // A prototype implementation showed excellent results, although | |
3882 // the scavenger and timeout code was rather involved. | |
3883 | |
3839 | 3884 casn(mark_addr.base(), Rbox, Rscratch); |
3885 cmp(Rbox, Rscratch); | |
0 | 3886 // Intentional fall through into done ... |
3887 | |
3839 | 3888 bind(done); |
0 | 3889 } |
3890 | |
3891 | |
3892 | |
3893 void MacroAssembler::print_CPU_state() { | |
3894 // %%%%% need to implement this | |
3895 } | |
3896 | |
3897 void MacroAssembler::verify_FPU(int stack_depth, const char* s) { | |
3898 // %%%%% need to implement this | |
3899 } | |
3900 | |
3901 void MacroAssembler::push_IU_state() { | |
3902 // %%%%% need to implement this | |
3903 } | |
3904 | |
3905 | |
3906 void MacroAssembler::pop_IU_state() { | |
3907 // %%%%% need to implement this | |
3908 } | |
3909 | |
3910 | |
3911 void MacroAssembler::push_FPU_state() { | |
3912 // %%%%% need to implement this | |
3913 } | |
3914 | |
3915 | |
3916 void MacroAssembler::pop_FPU_state() { | |
3917 // %%%%% need to implement this | |
3918 } | |
3919 | |
3920 | |
3921 void MacroAssembler::push_CPU_state() { | |
3922 // %%%%% need to implement this | |
3923 } | |
3924 | |
3925 | |
3926 void MacroAssembler::pop_CPU_state() { | |
3927 // %%%%% need to implement this | |
3928 } | |
3929 | |
3930 | |
3931 | |
3932 void MacroAssembler::verify_tlab() { | |
3933 #ifdef ASSERT | |
3934 if (UseTLAB && VerifyOops) { | |
3935 Label next, next2, ok; | |
3936 Register t1 = L0; | |
3937 Register t2 = L1; | |
3938 Register t3 = L2; | |
3939 | |
3940 save_frame(0); | |
3941 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), t1); | |
3942 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_start_offset()), t2); | |
3943 or3(t1, t2, t3); | |
3839 | 3944 cmp_and_br_short(t1, t2, Assembler::greaterEqual, Assembler::pn, next); |
0 | 3945 stop("assert(top >= start)"); |
3946 should_not_reach_here(); | |
3947 | |
3948 bind(next); | |
3949 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), t1); | |
3950 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), t2); | |
3951 or3(t3, t2, t3); | |
3839 | 3952 cmp_and_br_short(t1, t2, Assembler::lessEqual, Assembler::pn, next2); |
0 | 3953 stop("assert(top <= end)"); |
3954 should_not_reach_here(); | |
3955 | |
3956 bind(next2); | |
3957 and3(t3, MinObjAlignmentInBytesMask, t3); | |
3839 | 3958 cmp_and_br_short(t3, 0, Assembler::lessEqual, Assembler::pn, ok); |
0 | 3959 stop("assert(aligned)"); |
3960 should_not_reach_here(); | |
3961 | |
3962 bind(ok); | |
3963 restore(); | |
3964 } | |
3965 #endif | |
3966 } | |
3967 | |
3968 | |
3969 void MacroAssembler::eden_allocate( | |
3970 Register obj, // result: pointer to object after successful allocation | |
3971 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
3972 int con_size_in_bytes, // object size in bytes if known at compile time | |
3973 Register t1, // temp register | |
3974 Register t2, // temp register | |
3975 Label& slow_case // continuation point if fast allocation fails | |
3976 ){ | |
3977 // make sure arguments make sense | |
3978 assert_different_registers(obj, var_size_in_bytes, t1, t2); | |
3979 assert(0 <= con_size_in_bytes && Assembler::is_simm13(con_size_in_bytes), "illegal object size"); | |
3980 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, "object size is not multiple of alignment"); | |
3981 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3982 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3983 // No allocation in the shared eden. |
3839 | 3984 ba_short(slow_case); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3985 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3986 // get eden boundaries |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3987 // note: we need both top & top_addr! |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3988 const Register top_addr = t1; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3989 const Register end = t2; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3990 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3991 CollectedHeap* ch = Universe::heap(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3992 set((intx)ch->top_addr(), top_addr); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3993 intx delta = (intx)ch->end_addr() - (intx)ch->top_addr(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3994 ld_ptr(top_addr, delta, end); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3995 ld_ptr(top_addr, 0, obj); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3996 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3997 // try to allocate |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3998 Label retry; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
3999 bind(retry); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4000 #ifdef ASSERT |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4001 // make sure eden top is properly aligned |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4002 { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4003 Label L; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4004 btst(MinObjAlignmentInBytesMask, obj); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4005 br(Assembler::zero, false, Assembler::pt, L); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4006 delayed()->nop(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4007 stop("eden top is not properly aligned"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4008 bind(L); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4009 } |
0 | 4010 #endif // ASSERT |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4011 const Register free = end; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4012 sub(end, obj, free); // compute amount of free space |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4013 if (var_size_in_bytes->is_valid()) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4014 // size is unknown at compile time |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4015 cmp(free, var_size_in_bytes); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4016 br(Assembler::lessUnsigned, false, Assembler::pn, slow_case); // if there is not enough space go the slow case |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4017 delayed()->add(obj, var_size_in_bytes, end); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4018 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4019 // size is known at compile time |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4020 cmp(free, con_size_in_bytes); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4021 br(Assembler::lessUnsigned, false, Assembler::pn, slow_case); // if there is not enough space go the slow case |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4022 delayed()->add(obj, con_size_in_bytes, end); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4023 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4024 // Compare obj with the value at top_addr; if still equal, swap the value of |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4025 // end with the value at top_addr. If not equal, read the value at top_addr |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4026 // into end. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4027 casx_under_lock(top_addr, obj, end, (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr()); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4028 // if someone beat us on the allocation, try again, otherwise continue |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4029 cmp(obj, end); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4030 brx(Assembler::notEqual, false, Assembler::pn, retry); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4031 delayed()->mov(end, obj); // nop if successfull since obj == end |
0 | 4032 |
4033 #ifdef ASSERT | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4034 // make sure eden top is properly aligned |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4035 { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4036 Label L; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4037 const Register top_addr = t1; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4038 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4039 set((intx)ch->top_addr(), top_addr); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4040 ld_ptr(top_addr, 0, top_addr); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4041 btst(MinObjAlignmentInBytesMask, top_addr); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4042 br(Assembler::zero, false, Assembler::pt, L); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4043 delayed()->nop(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4044 stop("eden top is not properly aligned"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4045 bind(L); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4046 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4047 #endif // ASSERT |
0 | 4048 } |
4049 } | |
4050 | |
4051 | |
4052 void MacroAssembler::tlab_allocate( | |
4053 Register obj, // result: pointer to object after successful allocation | |
4054 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise | |
4055 int con_size_in_bytes, // object size in bytes if known at compile time | |
4056 Register t1, // temp register | |
4057 Label& slow_case // continuation point if fast allocation fails | |
4058 ){ | |
4059 // make sure arguments make sense | |
4060 assert_different_registers(obj, var_size_in_bytes, t1); | |
4061 assert(0 <= con_size_in_bytes && is_simm13(con_size_in_bytes), "illegal object size"); | |
4062 assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, "object size is not multiple of alignment"); | |
4063 | |
4064 const Register free = t1; | |
4065 | |
4066 verify_tlab(); | |
4067 | |
4068 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), obj); | |
4069 | |
4070 // calculate amount of free space | |
4071 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), free); | |
4072 sub(free, obj, free); | |
4073 | |
4074 Label done; | |
4075 if (var_size_in_bytes == noreg) { | |
4076 cmp(free, con_size_in_bytes); | |
4077 } else { | |
4078 cmp(free, var_size_in_bytes); | |
4079 } | |
4080 br(Assembler::less, false, Assembler::pn, slow_case); | |
4081 // calculate the new top pointer | |
4082 if (var_size_in_bytes == noreg) { | |
4083 delayed()->add(obj, con_size_in_bytes, free); | |
4084 } else { | |
4085 delayed()->add(obj, var_size_in_bytes, free); | |
4086 } | |
4087 | |
4088 bind(done); | |
4089 | |
4090 #ifdef ASSERT | |
4091 // make sure new free pointer is properly aligned | |
4092 { | |
4093 Label L; | |
4094 btst(MinObjAlignmentInBytesMask, free); | |
4095 br(Assembler::zero, false, Assembler::pt, L); | |
4096 delayed()->nop(); | |
4097 stop("updated TLAB free is not properly aligned"); | |
4098 bind(L); | |
4099 } | |
4100 #endif // ASSERT | |
4101 | |
4102 // update the tlab top pointer | |
4103 st_ptr(free, G2_thread, in_bytes(JavaThread::tlab_top_offset())); | |
4104 verify_tlab(); | |
4105 } | |
4106 | |
4107 | |
4108 void MacroAssembler::tlab_refill(Label& retry, Label& try_eden, Label& slow_case) { | |
4109 Register top = O0; | |
4110 Register t1 = G1; | |
4111 Register t2 = G3; | |
4112 Register t3 = O1; | |
4113 assert_different_registers(top, t1, t2, t3, G4, G5 /* preserve G4 and G5 */); | |
4114 Label do_refill, discard_tlab; | |
4115 | |
4116 if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) { | |
4117 // No allocation in the shared eden. | |
3839 | 4118 ba_short(slow_case); |
0 | 4119 } |
4120 | |
4121 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_top_offset()), top); | |
4122 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_end_offset()), t1); | |
4123 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset()), t2); | |
4124 | |
4125 // calculate amount of free space | |
4126 sub(t1, top, t1); | |
4127 srl_ptr(t1, LogHeapWordSize, t1); | |
4128 | |
4129 // Retain tlab and allocate object in shared space if | |
4130 // the amount free in the tlab is too large to discard. | |
4131 cmp(t1, t2); | |
4132 brx(Assembler::lessEqual, false, Assembler::pt, discard_tlab); | |
4133 | |
4134 // increment waste limit to prevent getting stuck on this slow path | |
4135 delayed()->add(t2, ThreadLocalAllocBuffer::refill_waste_limit_increment(), t2); | |
4136 st_ptr(t2, G2_thread, in_bytes(JavaThread::tlab_refill_waste_limit_offset())); | |
4137 if (TLABStats) { | |
4138 // increment number of slow_allocations | |
4139 ld(G2_thread, in_bytes(JavaThread::tlab_slow_allocations_offset()), t2); | |
4140 add(t2, 1, t2); | |
4141 stw(t2, G2_thread, in_bytes(JavaThread::tlab_slow_allocations_offset())); | |
4142 } | |
3839 | 4143 ba_short(try_eden); |
0 | 4144 |
4145 bind(discard_tlab); | |
4146 if (TLABStats) { | |
4147 // increment number of refills | |
4148 ld(G2_thread, in_bytes(JavaThread::tlab_number_of_refills_offset()), t2); | |
4149 add(t2, 1, t2); | |
4150 stw(t2, G2_thread, in_bytes(JavaThread::tlab_number_of_refills_offset())); | |
4151 // accumulate wastage | |
4152 ld(G2_thread, in_bytes(JavaThread::tlab_fast_refill_waste_offset()), t2); | |
4153 add(t2, t1, t2); | |
4154 stw(t2, G2_thread, in_bytes(JavaThread::tlab_fast_refill_waste_offset())); | |
4155 } | |
4156 | |
4157 // if tlab is currently allocated (top or end != null) then | |
4158 // fill [top, end + alignment_reserve) with array object | |
3839 | 4159 br_null_short(top, Assembler::pn, do_refill); |
0 | 4160 |
4161 set((intptr_t)markOopDesc::prototype()->copy_set_hash(0x2), t2); | |
4162 st_ptr(t2, top, oopDesc::mark_offset_in_bytes()); // set up the mark word | |
4163 // set klass to intArrayKlass | |
4164 sub(t1, typeArrayOopDesc::header_size(T_INT), t1); | |
4165 add(t1, ThreadLocalAllocBuffer::alignment_reserve(), t1); | |
4166 sll_ptr(t1, log2_intptr(HeapWordSize/sizeof(jint)), t1); | |
4167 st(t1, top, arrayOopDesc::length_offset_in_bytes()); | |
167
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4168 set((intptr_t)Universe::intArrayKlassObj_addr(), t2); |
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4169 ld_ptr(t2, 0, t2); |
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4170 // store klass last. concurrent gcs assumes klass length is valid if |
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4171 // klass field is not null. |
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4172 store_klass(t2, top); |
0 | 4173 verify_oop(top); |
4174 | |
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4175 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_start_offset()), t1); |
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4176 sub(top, t1, t1); // size of tlab's allocated portion |
2127
5577848f5923
7011463: Sparc MacroAssembler::incr_allocated_bytes() needs a RegisterOrConstant argument
phh
parents:
2102
diff
changeset
|
4177 incr_allocated_bytes(t1, t2, t3); |
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4178 |
0 | 4179 // refill the tlab with an eden allocation |
4180 bind(do_refill); | |
4181 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_size_offset()), t1); | |
4182 sll_ptr(t1, LogHeapWordSize, t1); | |
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4183 // allocate new tlab, address returned in top |
0 | 4184 eden_allocate(top, t1, 0, t2, t3, slow_case); |
4185 | |
4186 st_ptr(top, G2_thread, in_bytes(JavaThread::tlab_start_offset())); | |
4187 st_ptr(top, G2_thread, in_bytes(JavaThread::tlab_top_offset())); | |
4188 #ifdef ASSERT | |
4189 // check that tlab_size (t1) is still valid | |
4190 { | |
4191 Label ok; | |
4192 ld_ptr(G2_thread, in_bytes(JavaThread::tlab_size_offset()), t2); | |
4193 sll_ptr(t2, LogHeapWordSize, t2); | |
3839 | 4194 cmp_and_br_short(t1, t2, Assembler::equal, Assembler::pt, ok); |
0 | 4195 stop("assert(t1 == tlab_size)"); |
4196 should_not_reach_here(); | |
4197 | |
4198 bind(ok); | |
4199 } | |
4200 #endif // ASSERT | |
4201 add(top, t1, top); // t1 is tlab_size | |
4202 sub(top, ThreadLocalAllocBuffer::alignment_reserve_in_bytes(), top); | |
4203 st_ptr(top, G2_thread, in_bytes(JavaThread::tlab_end_offset())); | |
4204 verify_tlab(); | |
3839 | 4205 ba_short(retry); |
0 | 4206 } |
4207 | |
2127
5577848f5923
7011463: Sparc MacroAssembler::incr_allocated_bytes() needs a RegisterOrConstant argument
phh
parents:
2102
diff
changeset
|
4208 void MacroAssembler::incr_allocated_bytes(RegisterOrConstant size_in_bytes, |
5577848f5923
7011463: Sparc MacroAssembler::incr_allocated_bytes() needs a RegisterOrConstant argument
phh
parents:
2102
diff
changeset
|
4209 Register t1, Register t2) { |
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4210 // Bump total bytes allocated by this thread |
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4211 assert(t1->is_global(), "must be global reg"); // so all 64 bits are saved on a context switch |
2127
5577848f5923
7011463: Sparc MacroAssembler::incr_allocated_bytes() needs a RegisterOrConstant argument
phh
parents:
2102
diff
changeset
|
4212 assert_different_registers(size_in_bytes.register_or_noreg(), t1, t2); |
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4213 // v8 support has gone the way of the dodo |
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4214 ldx(G2_thread, in_bytes(JavaThread::allocated_bytes_offset()), t1); |
2127
5577848f5923
7011463: Sparc MacroAssembler::incr_allocated_bytes() needs a RegisterOrConstant argument
phh
parents:
2102
diff
changeset
|
4215 add(t1, ensure_simm13_or_reg(size_in_bytes, t2), t1); |
2100
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4216 stx(t1, G2_thread, in_bytes(JavaThread::allocated_bytes_offset())); |
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4217 } |
b1a2afa37ec4
7003271: Hotspot should track cumulative Java heap bytes allocated on a per-thread basis
phh
parents:
2008
diff
changeset
|
4218 |
0 | 4219 Assembler::Condition MacroAssembler::negate_condition(Assembler::Condition cond) { |
4220 switch (cond) { | |
4221 // Note some conditions are synonyms for others | |
4222 case Assembler::never: return Assembler::always; | |
4223 case Assembler::zero: return Assembler::notZero; | |
4224 case Assembler::lessEqual: return Assembler::greater; | |
4225 case Assembler::less: return Assembler::greaterEqual; | |
4226 case Assembler::lessEqualUnsigned: return Assembler::greaterUnsigned; | |
4227 case Assembler::lessUnsigned: return Assembler::greaterEqualUnsigned; | |
4228 case Assembler::negative: return Assembler::positive; | |
4229 case Assembler::overflowSet: return Assembler::overflowClear; | |
4230 case Assembler::always: return Assembler::never; | |
4231 case Assembler::notZero: return Assembler::zero; | |
4232 case Assembler::greater: return Assembler::lessEqual; | |
4233 case Assembler::greaterEqual: return Assembler::less; | |
4234 case Assembler::greaterUnsigned: return Assembler::lessEqualUnsigned; | |
4235 case Assembler::greaterEqualUnsigned: return Assembler::lessUnsigned; | |
4236 case Assembler::positive: return Assembler::negative; | |
4237 case Assembler::overflowClear: return Assembler::overflowSet; | |
4238 } | |
4239 | |
4240 ShouldNotReachHere(); return Assembler::overflowClear; | |
4241 } | |
4242 | |
4243 void MacroAssembler::cond_inc(Assembler::Condition cond, address counter_ptr, | |
4244 Register Rtmp1, Register Rtmp2 /*, Register Rtmp3, Register Rtmp4 */) { | |
4245 Condition negated_cond = negate_condition(cond); | |
4246 Label L; | |
4247 brx(negated_cond, false, Assembler::pt, L); | |
4248 delayed()->nop(); | |
4249 inc_counter(counter_ptr, Rtmp1, Rtmp2); | |
4250 bind(L); | |
4251 } | |
4252 | |
727 | 4253 void MacroAssembler::inc_counter(address counter_addr, Register Rtmp1, Register Rtmp2) { |
4254 AddressLiteral addrlit(counter_addr); | |
4255 sethi(addrlit, Rtmp1); // Move hi22 bits into temporary register. | |
4256 Address addr(Rtmp1, addrlit.low10()); // Build an address with low10 bits. | |
4257 ld(addr, Rtmp2); | |
0 | 4258 inc(Rtmp2); |
727 | 4259 st(Rtmp2, addr); |
4260 } | |
4261 | |
4262 void MacroAssembler::inc_counter(int* counter_addr, Register Rtmp1, Register Rtmp2) { | |
4263 inc_counter((address) counter_addr, Rtmp1, Rtmp2); | |
0 | 4264 } |
4265 | |
4266 SkipIfEqual::SkipIfEqual( | |
4267 MacroAssembler* masm, Register temp, const bool* flag_addr, | |
4268 Assembler::Condition condition) { | |
4269 _masm = masm; | |
727 | 4270 AddressLiteral flag(flag_addr); |
4271 _masm->sethi(flag, temp); | |
4272 _masm->ldub(temp, flag.low10(), temp); | |
0 | 4273 _masm->tst(temp); |
4274 _masm->br(condition, false, Assembler::pt, _label); | |
4275 _masm->delayed()->nop(); | |
4276 } | |
4277 | |
4278 SkipIfEqual::~SkipIfEqual() { | |
4279 _masm->bind(_label); | |
4280 } | |
4281 | |
4282 | |
4283 // Writes to stack successive pages until offset reached to check for | |
4284 // stack overflow + shadow pages. This clobbers tsp and scratch. | |
4285 void MacroAssembler::bang_stack_size(Register Rsize, Register Rtsp, | |
4286 Register Rscratch) { | |
4287 // Use stack pointer in temp stack pointer | |
4288 mov(SP, Rtsp); | |
4289 | |
4290 // Bang stack for total size given plus stack shadow page size. | |
4291 // Bang one page at a time because a large size can overflow yellow and | |
4292 // red zones (the bang will fail but stack overflow handling can't tell that | |
4293 // it was a stack overflow bang vs a regular segv). | |
4294 int offset = os::vm_page_size(); | |
4295 Register Roffset = Rscratch; | |
4296 | |
4297 Label loop; | |
4298 bind(loop); | |
4299 set((-offset)+STACK_BIAS, Rscratch); | |
4300 st(G0, Rtsp, Rscratch); | |
4301 set(offset, Roffset); | |
4302 sub(Rsize, Roffset, Rsize); | |
4303 cmp(Rsize, G0); | |
4304 br(Assembler::greater, false, Assembler::pn, loop); | |
4305 delayed()->sub(Rtsp, Roffset, Rtsp); | |
4306 | |
4307 // Bang down shadow pages too. | |
4308 // The -1 because we already subtracted 1 page. | |
4309 for (int i = 0; i< StackShadowPages-1; i++) { | |
4310 set((-i*offset)+STACK_BIAS, Rscratch); | |
4311 st(G0, Rtsp, Rscratch); | |
4312 } | |
4313 } | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4314 |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4315 /////////////////////////////////////////////////////////////////////////////////// |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4316 #ifndef SERIALGC |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4317 |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4318 static address satb_log_enqueue_with_frame = NULL; |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4319 static u_char* satb_log_enqueue_with_frame_end = NULL; |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4320 |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4321 static address satb_log_enqueue_frameless = NULL; |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4322 static u_char* satb_log_enqueue_frameless_end = NULL; |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4323 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4324 static int EnqueueCodeSize = 128 DEBUG_ONLY( + 256); // Instructions? |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4325 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4326 static void generate_satb_log_enqueue(bool with_frame) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4327 BufferBlob* bb = BufferBlob::create("enqueue_with_frame", EnqueueCodeSize); |
1748 | 4328 CodeBuffer buf(bb); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4329 MacroAssembler masm(&buf); |
3839 | 4330 |
4331 #define __ masm. | |
4332 | |
4333 address start = __ pc(); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4334 Register pre_val; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4335 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4336 Label refill, restart; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4337 if (with_frame) { |
3839 | 4338 __ save_frame(0); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4339 pre_val = I0; // Was O0 before the save. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4340 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4341 pre_val = O0; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4342 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4343 int satb_q_index_byte_offset = |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4344 in_bytes(JavaThread::satb_mark_queue_offset() + |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4345 PtrQueue::byte_offset_of_index()); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4346 int satb_q_buf_byte_offset = |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4347 in_bytes(JavaThread::satb_mark_queue_offset() + |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4348 PtrQueue::byte_offset_of_buf()); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4349 assert(in_bytes(PtrQueue::byte_width_of_index()) == sizeof(intptr_t) && |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4350 in_bytes(PtrQueue::byte_width_of_buf()) == sizeof(intptr_t), |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4351 "check sizes in assembly below"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4352 |
3839 | 4353 __ bind(restart); |
4354 __ ld_ptr(G2_thread, satb_q_index_byte_offset, L0); | |
4355 | |
4356 __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn, L0, refill); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4357 // If the branch is taken, no harm in executing this in the delay slot. |
3839 | 4358 __ delayed()->ld_ptr(G2_thread, satb_q_buf_byte_offset, L1); |
4359 __ sub(L0, oopSize, L0); | |
4360 | |
4361 __ st_ptr(pre_val, L1, L0); // [_buf + index] := I0 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4362 if (!with_frame) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4363 // Use return-from-leaf |
3839 | 4364 __ retl(); |
4365 __ delayed()->st_ptr(L0, G2_thread, satb_q_index_byte_offset); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4366 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4367 // Not delayed. |
3839 | 4368 __ st_ptr(L0, G2_thread, satb_q_index_byte_offset); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4369 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4370 if (with_frame) { |
3839 | 4371 __ ret(); |
4372 __ delayed()->restore(); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4373 } |
3839 | 4374 __ bind(refill); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4375 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4376 address handle_zero = |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4377 CAST_FROM_FN_PTR(address, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4378 &SATBMarkQueueSet::handle_zero_index_for_thread); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4379 // This should be rare enough that we can afford to save all the |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4380 // scratch registers that the calling context might be using. |
3839 | 4381 __ mov(G1_scratch, L0); |
4382 __ mov(G3_scratch, L1); | |
4383 __ mov(G4, L2); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4384 // We need the value of O0 above (for the write into the buffer), so we |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4385 // save and restore it. |
3839 | 4386 __ mov(O0, L3); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4387 // Since the call will overwrite O7, we save and restore that, as well. |
3839 | 4388 __ mov(O7, L4); |
4389 __ call_VM_leaf(L5, handle_zero, G2_thread); | |
4390 __ mov(L0, G1_scratch); | |
4391 __ mov(L1, G3_scratch); | |
4392 __ mov(L2, G4); | |
4393 __ mov(L3, O0); | |
4394 __ br(Assembler::always, /*annul*/false, Assembler::pt, restart); | |
4395 __ delayed()->mov(L4, O7); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4396 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4397 if (with_frame) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4398 satb_log_enqueue_with_frame = start; |
3839 | 4399 satb_log_enqueue_with_frame_end = __ pc(); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4400 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4401 satb_log_enqueue_frameless = start; |
3839 | 4402 satb_log_enqueue_frameless_end = __ pc(); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4403 } |
3839 | 4404 |
4405 #undef __ | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4406 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4407 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4408 static inline void generate_satb_log_enqueue_if_necessary(bool with_frame) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4409 if (with_frame) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4410 if (satb_log_enqueue_with_frame == 0) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4411 generate_satb_log_enqueue(with_frame); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4412 assert(satb_log_enqueue_with_frame != 0, "postcondition."); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4413 if (G1SATBPrintStubs) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4414 tty->print_cr("Generated with-frame satb enqueue:"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4415 Disassembler::decode((u_char*)satb_log_enqueue_with_frame, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4416 satb_log_enqueue_with_frame_end, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4417 tty); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4418 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4419 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4420 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4421 if (satb_log_enqueue_frameless == 0) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4422 generate_satb_log_enqueue(with_frame); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4423 assert(satb_log_enqueue_frameless != 0, "postcondition."); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4424 if (G1SATBPrintStubs) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4425 tty->print_cr("Generated frameless satb enqueue:"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4426 Disassembler::decode((u_char*)satb_log_enqueue_frameless, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4427 satb_log_enqueue_frameless_end, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4428 tty); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4429 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4430 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4431 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4432 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4433 |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4434 void MacroAssembler::g1_write_barrier_pre(Register obj, |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4435 Register index, |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4436 int offset, |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4437 Register pre_val, |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4438 Register tmp, |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4439 bool preserve_o_regs) { |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4440 Label filtered; |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4441 |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4442 if (obj == noreg) { |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4443 // We are not loading the previous value so make |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4444 // sure that we don't trash the value in pre_val |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4445 // with the code below. |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4446 assert_different_registers(pre_val, tmp); |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4447 } else { |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4448 // We will be loading the previous value |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4449 // in this code so... |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4450 assert(offset == 0 || index == noreg, "choose one"); |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4451 assert(pre_val == noreg, "check this code"); |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4452 } |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4453 |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4454 // Is marking active? |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4455 if (in_bytes(PtrQueue::byte_width_of_active()) == 4) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4456 ld(G2, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4457 in_bytes(JavaThread::satb_mark_queue_offset() + |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4458 PtrQueue::byte_offset_of_active()), |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4459 tmp); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4460 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4461 guarantee(in_bytes(PtrQueue::byte_width_of_active()) == 1, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4462 "Assumption"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4463 ldsb(G2, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4464 in_bytes(JavaThread::satb_mark_queue_offset() + |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4465 PtrQueue::byte_offset_of_active()), |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4466 tmp); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4467 } |
845
df6caf649ff7
6700789: G1: Enable use of compressed oops with G1 heaps
ysr
parents:
794
diff
changeset
|
4468 |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4469 // Check on whether to annul. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4470 br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, tmp, filtered); |
3839 | 4471 delayed()->nop(); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4472 |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4473 // Do we need to load the previous value? |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4474 if (obj != noreg) { |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4475 // Load the previous value... |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4476 if (index == noreg) { |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4477 if (Assembler::is_simm13(offset)) { |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4478 load_heap_oop(obj, offset, tmp); |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4479 } else { |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4480 set(offset, tmp); |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4481 load_heap_oop(obj, tmp, tmp); |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4482 } |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4483 } else { |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4484 load_heap_oop(obj, index, tmp); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4485 } |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4486 // Previous value has been loaded into tmp |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4487 pre_val = tmp; |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4488 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4489 |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4490 assert(pre_val != noreg, "must have a real register"); |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4491 |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4492 // Is the previous value null? |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4493 // Check on whether to annul. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4494 br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, pre_val, filtered); |
3839 | 4495 delayed()->nop(); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4496 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4497 // OK, it's not filtered, so we'll need to call enqueue. In the normal |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4498 // case, pre_val will be a scratch G-reg, but there are some cases in |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4499 // which it's an O-reg. In the first case, do a normal call. In the |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4500 // latter, do a save here and call the frameless version. |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4501 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4502 guarantee(pre_val->is_global() || pre_val->is_out(), |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4503 "Or we need to think harder."); |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4504 |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4505 if (pre_val->is_global() && !preserve_o_regs) { |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4506 generate_satb_log_enqueue_if_necessary(true); // with frame |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4507 |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4508 call(satb_log_enqueue_with_frame); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4509 delayed()->mov(pre_val, O0); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4510 } else { |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4511 generate_satb_log_enqueue_if_necessary(false); // frameless |
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4512 |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4513 save_frame(0); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4514 call(satb_log_enqueue_frameless); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4515 delayed()->mov(pre_val->after_save(), O0); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4516 restore(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4517 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4518 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4519 bind(filtered); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4520 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4521 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4522 static jint num_ct_writes = 0; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4523 static jint num_ct_writes_filtered_in_hr = 0; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4524 static jint num_ct_writes_filtered_null = 0; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4525 static G1CollectedHeap* g1 = NULL; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4526 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4527 static Thread* count_ct_writes(void* filter_val, void* new_val) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4528 Atomic::inc(&num_ct_writes); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4529 if (filter_val == NULL) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4530 Atomic::inc(&num_ct_writes_filtered_in_hr); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4531 } else if (new_val == NULL) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4532 Atomic::inc(&num_ct_writes_filtered_null); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4533 } else { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4534 if (g1 == NULL) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4535 g1 = G1CollectedHeap::heap(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4536 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4537 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4538 if ((num_ct_writes % 1000000) == 0) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4539 jint num_ct_writes_filtered = |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4540 num_ct_writes_filtered_in_hr + |
677 | 4541 num_ct_writes_filtered_null; |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4542 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4543 tty->print_cr("%d potential CT writes: %5.2f%% filtered\n" |
677 | 4544 " (%5.2f%% intra-HR, %5.2f%% null).", |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4545 num_ct_writes, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4546 100.0*(float)num_ct_writes_filtered/(float)num_ct_writes, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4547 100.0*(float)num_ct_writes_filtered_in_hr/ |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4548 (float)num_ct_writes, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4549 100.0*(float)num_ct_writes_filtered_null/ |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4550 (float)num_ct_writes); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4551 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4552 return Thread::current(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4553 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4554 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4555 static address dirty_card_log_enqueue = 0; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4556 static u_char* dirty_card_log_enqueue_end = 0; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4557 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4558 // This gets to assume that o0 contains the object address. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4559 static void generate_dirty_card_log_enqueue(jbyte* byte_map_base) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4560 BufferBlob* bb = BufferBlob::create("dirty_card_enqueue", EnqueueCodeSize*2); |
1748 | 4561 CodeBuffer buf(bb); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4562 MacroAssembler masm(&buf); |
3839 | 4563 #define __ masm. |
4564 address start = __ pc(); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4565 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4566 Label not_already_dirty, restart, refill; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4567 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4568 #ifdef _LP64 |
3839 | 4569 __ srlx(O0, CardTableModRefBS::card_shift, O0); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4570 #else |
3839 | 4571 __ srl(O0, CardTableModRefBS::card_shift, O0); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4572 #endif |
727 | 4573 AddressLiteral addrlit(byte_map_base); |
3839 | 4574 __ set(addrlit, O1); // O1 := <card table base> |
4575 __ ldub(O0, O1, O2); // O2 := [O0 + O1] | |
4576 | |
4577 __ br_on_reg_cond(Assembler::rc_nz, /*annul*/false, Assembler::pt, | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4578 O2, not_already_dirty); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4579 // Get O1 + O2 into a reg by itself -- useful in the take-the-branch |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4580 // case, harmless if not. |
3839 | 4581 __ delayed()->add(O0, O1, O3); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4582 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4583 // We didn't take the branch, so we're already dirty: return. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4584 // Use return-from-leaf |
3839 | 4585 __ retl(); |
4586 __ delayed()->nop(); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4587 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4588 // Not dirty. |
3839 | 4589 __ bind(not_already_dirty); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4590 // First, dirty it. |
3839 | 4591 __ stb(G0, O3, G0); // [cardPtr] := 0 (i.e., dirty). |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4592 int dirty_card_q_index_byte_offset = |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4593 in_bytes(JavaThread::dirty_card_queue_offset() + |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4594 PtrQueue::byte_offset_of_index()); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4595 int dirty_card_q_buf_byte_offset = |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4596 in_bytes(JavaThread::dirty_card_queue_offset() + |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4597 PtrQueue::byte_offset_of_buf()); |
3839 | 4598 __ bind(restart); |
4599 __ ld_ptr(G2_thread, dirty_card_q_index_byte_offset, L0); | |
4600 | |
4601 __ br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn, | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4602 L0, refill); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4603 // If the branch is taken, no harm in executing this in the delay slot. |
3839 | 4604 __ delayed()->ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, L1); |
4605 __ sub(L0, oopSize, L0); | |
4606 | |
4607 __ st_ptr(O3, L1, L0); // [_buf + index] := I0 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4608 // Use return-from-leaf |
3839 | 4609 __ retl(); |
4610 __ delayed()->st_ptr(L0, G2_thread, dirty_card_q_index_byte_offset); | |
4611 | |
4612 __ bind(refill); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4613 address handle_zero = |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4614 CAST_FROM_FN_PTR(address, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4615 &DirtyCardQueueSet::handle_zero_index_for_thread); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4616 // This should be rare enough that we can afford to save all the |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4617 // scratch registers that the calling context might be using. |
3839 | 4618 __ mov(G1_scratch, L3); |
4619 __ mov(G3_scratch, L5); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4620 // We need the value of O3 above (for the write into the buffer), so we |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4621 // save and restore it. |
3839 | 4622 __ mov(O3, L6); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4623 // Since the call will overwrite O7, we save and restore that, as well. |
3839 | 4624 __ mov(O7, L4); |
4625 | |
4626 __ call_VM_leaf(L7_thread_cache, handle_zero, G2_thread); | |
4627 __ mov(L3, G1_scratch); | |
4628 __ mov(L5, G3_scratch); | |
4629 __ mov(L6, O3); | |
4630 __ br(Assembler::always, /*annul*/false, Assembler::pt, restart); | |
4631 __ delayed()->mov(L4, O7); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4632 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4633 dirty_card_log_enqueue = start; |
3839 | 4634 dirty_card_log_enqueue_end = __ pc(); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4635 // XXX Should have a guarantee here about not going off the end! |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4636 // Does it already do so? Do an experiment... |
3839 | 4637 |
4638 #undef __ | |
4639 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4640 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4641 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4642 static inline void |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4643 generate_dirty_card_log_enqueue_if_necessary(jbyte* byte_map_base) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4644 if (dirty_card_log_enqueue == 0) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4645 generate_dirty_card_log_enqueue(byte_map_base); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4646 assert(dirty_card_log_enqueue != 0, "postcondition."); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4647 if (G1SATBPrintStubs) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4648 tty->print_cr("Generated dirty_card enqueue:"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4649 Disassembler::decode((u_char*)dirty_card_log_enqueue, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4650 dirty_card_log_enqueue_end, |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4651 tty); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4652 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4653 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4654 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4655 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4656 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4657 void MacroAssembler::g1_write_barrier_post(Register store_addr, Register new_val, Register tmp) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4658 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4659 Label filtered; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4660 MacroAssembler* post_filter_masm = this; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4661 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4662 if (new_val == G0) return; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4663 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4664 G1SATBCardTableModRefBS* bs = (G1SATBCardTableModRefBS*) Universe::heap()->barrier_set(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4665 assert(bs->kind() == BarrierSet::G1SATBCT || |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4666 bs->kind() == BarrierSet::G1SATBCTLogging, "wrong barrier"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4667 if (G1RSBarrierRegionFilter) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4668 xor3(store_addr, new_val, tmp); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4669 #ifdef _LP64 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4670 srlx(tmp, HeapRegion::LogOfHRGrainBytes, tmp); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4671 #else |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4672 srl(tmp, HeapRegion::LogOfHRGrainBytes, tmp); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4673 #endif |
3249
e1162778c1c8
7009266: G1: assert(obj->is_oop_or_null(true )) failed: Error
johnc
parents:
2357
diff
changeset
|
4674 |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4675 if (G1PrintCTFilterStats) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4676 guarantee(tmp->is_global(), "Or stats won't work..."); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4677 // This is a sleazy hack: I'm temporarily hijacking G2, which I |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4678 // promise to restore. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4679 mov(new_val, G2); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4680 save_frame(0); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4681 mov(tmp, O0); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4682 mov(G2, O1); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4683 // Save G-regs that target may use. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4684 mov(G1, L1); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4685 mov(G2, L2); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4686 mov(G3, L3); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4687 mov(G4, L4); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4688 mov(G5, L5); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4689 call(CAST_FROM_FN_PTR(address, &count_ct_writes)); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4690 delayed()->nop(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4691 mov(O0, G2); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4692 // Restore G-regs that target may have used. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4693 mov(L1, G1); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4694 mov(L3, G3); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4695 mov(L4, G4); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4696 mov(L5, G5); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4697 restore(G0, G0, G0); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4698 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4699 // XXX Should I predict this taken or not? Does it mattern? |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4700 br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, tmp, filtered); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4701 delayed()->nop(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4702 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4703 |
794 | 4704 // If the "store_addr" register is an "in" or "local" register, move it to |
4705 // a scratch reg so we can pass it as an argument. | |
4706 bool use_scr = !(store_addr->is_global() || store_addr->is_out()); | |
4707 // Pick a scratch register different from "tmp". | |
4708 Register scr = (tmp == G1_scratch ? G3_scratch : G1_scratch); | |
4709 // Make sure we use up the delay slot! | |
4710 if (use_scr) { | |
4711 post_filter_masm->mov(store_addr, scr); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4712 } else { |
794 | 4713 post_filter_masm->nop(); |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4714 } |
794 | 4715 generate_dirty_card_log_enqueue_if_necessary(bs->byte_map_base); |
4716 save_frame(0); | |
4717 call(dirty_card_log_enqueue); | |
4718 if (use_scr) { | |
4719 delayed()->mov(scr, O0); | |
4720 } else { | |
4721 delayed()->mov(store_addr->after_save(), O0); | |
4722 } | |
4723 restore(); | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4724 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4725 bind(filtered); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4726 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4727 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4728 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4729 #endif // SERIALGC |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4730 /////////////////////////////////////////////////////////////////////////////////// |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4731 |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4732 void MacroAssembler::card_write_barrier_post(Register store_addr, Register new_val, Register tmp) { |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4733 // If we're writing constant NULL, we can skip the write barrier. |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4734 if (new_val == G0) return; |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4735 CardTableModRefBS* bs = (CardTableModRefBS*) Universe::heap()->barrier_set(); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4736 assert(bs->kind() == BarrierSet::CardTableModRef || |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4737 bs->kind() == BarrierSet::CardTableExtension, "wrong barrier"); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4738 card_table_write(bs->byte_map_base, tmp, store_addr); |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4739 } |
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
124
diff
changeset
|
4740 |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
4741 void MacroAssembler::load_klass(Register src_oop, Register klass) { |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4742 // The number of bytes in this code is used by |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4743 // MachCallDynamicJavaNode::ret_addr_offset() |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4744 // if this changes, change that. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4745 if (UseCompressedOops) { |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
4746 lduw(src_oop, oopDesc::klass_offset_in_bytes(), klass); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
4747 decode_heap_oop_not_null(klass); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4748 } else { |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
4749 ld_ptr(src_oop, oopDesc::klass_offset_in_bytes(), klass); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4750 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4751 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4752 |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
4753 void MacroAssembler::store_klass(Register klass, Register dst_oop) { |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4754 if (UseCompressedOops) { |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
4755 assert(dst_oop != klass, "not enough registers"); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
4756 encode_heap_oop_not_null(klass); |
167
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4757 st(klass, dst_oop, oopDesc::klass_offset_in_bytes()); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4758 } else { |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
124
diff
changeset
|
4759 st_ptr(klass, dst_oop, oopDesc::klass_offset_in_bytes()); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4760 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4761 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4762 |
167
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4763 void MacroAssembler::store_klass_gap(Register s, Register d) { |
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4764 if (UseCompressedOops) { |
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4765 assert(s != d, "not enough registers"); |
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
164
diff
changeset
|
4766 st(s, d, oopDesc::klass_gap_offset_in_bytes()); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4767 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4768 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4769 |
727 | 4770 void MacroAssembler::load_heap_oop(const Address& s, Register d) { |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4771 if (UseCompressedOops) { |
727 | 4772 lduw(s, d); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4773 decode_heap_oop(d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4774 } else { |
727 | 4775 ld_ptr(s, d); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4776 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4777 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4778 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4779 void MacroAssembler::load_heap_oop(Register s1, Register s2, Register d) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4780 if (UseCompressedOops) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4781 lduw(s1, s2, d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4782 decode_heap_oop(d, d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4783 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4784 ld_ptr(s1, s2, d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4785 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4786 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4787 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4788 void MacroAssembler::load_heap_oop(Register s1, int simm13a, Register d) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4789 if (UseCompressedOops) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4790 lduw(s1, simm13a, d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4791 decode_heap_oop(d, d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4792 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4793 ld_ptr(s1, simm13a, d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4794 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4795 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4796 |
1846 | 4797 void MacroAssembler::load_heap_oop(Register s1, RegisterOrConstant s2, Register d) { |
4798 if (s2.is_constant()) load_heap_oop(s1, s2.as_constant(), d); | |
4799 else load_heap_oop(s1, s2.as_register(), d); | |
4800 } | |
4801 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4802 void MacroAssembler::store_heap_oop(Register d, Register s1, Register s2) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4803 if (UseCompressedOops) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4804 assert(s1 != d && s2 != d, "not enough registers"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4805 encode_heap_oop(d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4806 st(d, s1, s2); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4807 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4808 st_ptr(d, s1, s2); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4809 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4810 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4811 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4812 void MacroAssembler::store_heap_oop(Register d, Register s1, int simm13a) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4813 if (UseCompressedOops) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4814 assert(s1 != d, "not enough registers"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4815 encode_heap_oop(d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4816 st(d, s1, simm13a); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4817 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4818 st_ptr(d, s1, simm13a); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4819 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4820 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4821 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4822 void MacroAssembler::store_heap_oop(Register d, const Address& a, int offset) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4823 if (UseCompressedOops) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4824 assert(a.base() != d, "not enough registers"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4825 encode_heap_oop(d); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4826 st(d, a, offset); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4827 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4828 st_ptr(d, a, offset); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4829 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4830 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4831 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4832 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4833 void MacroAssembler::encode_heap_oop(Register src, Register dst) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4834 assert (UseCompressedOops, "must be compressed"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4835 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4836 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
178
6d172e3548cb
6695819: verify_oopx rax: broken oop in decode_heap_oop
coleenp
parents:
124
diff
changeset
|
4837 verify_oop(src); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4838 if (Universe::narrow_oop_base() == NULL) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4839 srlx(src, LogMinObjAlignmentInBytes, dst); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4840 return; |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4841 } |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4842 Label done; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4843 if (src == dst) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4844 // optimize for frequent case src == dst |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4845 bpr(rc_nz, true, Assembler::pt, src, done); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4846 delayed() -> sub(src, G6_heapbase, dst); // annuled if not taken |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4847 bind(done); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4848 srlx(src, LogMinObjAlignmentInBytes, dst); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4849 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4850 bpr(rc_z, false, Assembler::pn, src, done); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4851 delayed() -> mov(G0, dst); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4852 // could be moved before branch, and annulate delay, |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4853 // but may add some unneeded work decoding null |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4854 sub(src, G6_heapbase, dst); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4855 srlx(dst, LogMinObjAlignmentInBytes, dst); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4856 bind(done); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4857 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4858 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4859 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4860 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4861 void MacroAssembler::encode_heap_oop_not_null(Register r) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4862 assert (UseCompressedOops, "must be compressed"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4863 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4864 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
178
6d172e3548cb
6695819: verify_oopx rax: broken oop in decode_heap_oop
coleenp
parents:
124
diff
changeset
|
4865 verify_oop(r); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4866 if (Universe::narrow_oop_base() != NULL) |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4867 sub(r, G6_heapbase, r); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4868 srlx(r, LogMinObjAlignmentInBytes, r); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4869 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4870 |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4871 void MacroAssembler::encode_heap_oop_not_null(Register src, Register dst) { |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4872 assert (UseCompressedOops, "must be compressed"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4873 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4874 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
178
6d172e3548cb
6695819: verify_oopx rax: broken oop in decode_heap_oop
coleenp
parents:
124
diff
changeset
|
4875 verify_oop(src); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4876 if (Universe::narrow_oop_base() == NULL) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4877 srlx(src, LogMinObjAlignmentInBytes, dst); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4878 } else { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4879 sub(src, G6_heapbase, dst); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4880 srlx(dst, LogMinObjAlignmentInBytes, dst); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4881 } |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4882 } |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4883 |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4884 // Same algorithm as oops.inline.hpp decode_heap_oop. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4885 void MacroAssembler::decode_heap_oop(Register src, Register dst) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4886 assert (UseCompressedOops, "must be compressed"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4887 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4888 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4889 sllx(src, LogMinObjAlignmentInBytes, dst); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4890 if (Universe::narrow_oop_base() != NULL) { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4891 Label done; |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4892 bpr(rc_nz, true, Assembler::pt, dst, done); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4893 delayed() -> add(dst, G6_heapbase, dst); // annuled if not taken |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4894 bind(done); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4895 } |
178
6d172e3548cb
6695819: verify_oopx rax: broken oop in decode_heap_oop
coleenp
parents:
124
diff
changeset
|
4896 verify_oop(dst); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4897 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4898 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4899 void MacroAssembler::decode_heap_oop_not_null(Register r) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4900 // Do not add assert code to this unless you change vtableStubs_sparc.cpp |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4901 // pd_code_size_limit. |
178
6d172e3548cb
6695819: verify_oopx rax: broken oop in decode_heap_oop
coleenp
parents:
124
diff
changeset
|
4902 // Also do not verify_oop as this is called by verify_oop. |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4903 assert (UseCompressedOops, "must be compressed"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4904 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4905 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4906 sllx(r, LogMinObjAlignmentInBytes, r); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4907 if (Universe::narrow_oop_base() != NULL) |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4908 add(r, G6_heapbase, r); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4909 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4910 |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4911 void MacroAssembler::decode_heap_oop_not_null(Register src, Register dst) { |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4912 // Do not add assert code to this unless you change vtableStubs_sparc.cpp |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4913 // pd_code_size_limit. |
178
6d172e3548cb
6695819: verify_oopx rax: broken oop in decode_heap_oop
coleenp
parents:
124
diff
changeset
|
4914 // Also do not verify_oop as this is called by verify_oop. |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4915 assert (UseCompressedOops, "must be compressed"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4916 assert (Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4917 assert (LogMinObjAlignmentInBytes == Universe::narrow_oop_shift(), "decode alg wrong"); |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4918 sllx(src, LogMinObjAlignmentInBytes, dst); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4919 if (Universe::narrow_oop_base() != NULL) |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
623
diff
changeset
|
4920 add(dst, G6_heapbase, dst); |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4921 } |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
4922 |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4923 void MacroAssembler::reinit_heapbase() { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4924 if (UseCompressedOops) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4925 // call indirectly to solve generation ordering problem |
727 | 4926 AddressLiteral base(Universe::narrow_oop_base_addr()); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4927 load_ptr_contents(base, G6_heapbase); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4928 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
4929 } |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4930 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4931 // Compare char[] arrays aligned to 4 bytes. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4932 void MacroAssembler::char_arrays_equals(Register ary1, Register ary2, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4933 Register limit, Register result, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4934 Register chr1, Register chr2, Label& Ldone) { |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4935 Label Lvector, Lloop; |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4936 assert(chr1 == result, "should be the same"); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4937 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4938 // Note: limit contains number of bytes (2*char_elements) != 0. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4939 andcc(limit, 0x2, chr1); // trailing character ? |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4940 br(Assembler::zero, false, Assembler::pt, Lvector); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4941 delayed()->nop(); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4942 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4943 // compare the trailing char |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4944 sub(limit, sizeof(jchar), limit); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4945 lduh(ary1, limit, chr1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4946 lduh(ary2, limit, chr2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4947 cmp(chr1, chr2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4948 br(Assembler::notEqual, true, Assembler::pt, Ldone); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4949 delayed()->mov(G0, result); // not equal |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4950 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4951 // only one char ? |
3839 | 4952 cmp_zero_and_br(zero, limit, Ldone, true, Assembler::pn); |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4953 delayed()->add(G0, 1, result); // zero-length arrays are equal |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4954 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4955 // word by word compare, dont't need alignment check |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4956 bind(Lvector); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4957 // Shift ary1 and ary2 to the end of the arrays, negate limit |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4958 add(ary1, limit, ary1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4959 add(ary2, limit, ary2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4960 neg(limit, limit); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4961 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4962 lduw(ary1, limit, chr1); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4963 bind(Lloop); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4964 lduw(ary2, limit, chr2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4965 cmp(chr1, chr2); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4966 br(Assembler::notEqual, true, Assembler::pt, Ldone); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4967 delayed()->mov(G0, result); // not equal |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4968 inccc(limit, 2*sizeof(jchar)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4969 // annul LDUW if branch is not taken to prevent access past end of array |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4970 br(Assembler::notZero, true, Assembler::pt, Lloop); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4971 delayed()->lduw(ary1, limit, chr1); // hoisted |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4972 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4973 // Caller should set it: |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4974 // add(G0, 1, result); // equals |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
845
diff
changeset
|
4975 } |