|
1 /* |
|
2 * Copyright (c) 1997, 2016, Oracle and/or its affiliates. All rights reserved. |
|
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
|
4 * |
|
5 * This code is free software; you can redistribute it and/or modify it |
|
6 * under the terms of the GNU General Public License version 2 only, as |
|
7 * published by the Free Software Foundation. |
|
8 * |
|
9 * This code is distributed in the hope that it will be useful, but WITHOUT |
|
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
12 * version 2 for more details (a copy is included in the LICENSE file that |
|
13 * accompanied this code). |
|
14 * |
|
15 * You should have received a copy of the GNU General Public License version |
|
16 * 2 along with this work; if not, write to the Free Software Foundation, |
|
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
18 * |
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
20 * or visit www.oracle.com if you need additional information or have any |
|
21 * questions. |
|
22 * |
|
23 */ |
|
24 |
|
25 #include "precompiled.hpp" |
|
26 #include "asm/macroAssembler.hpp" |
|
27 #include "interpreter/interp_masm.hpp" |
|
28 #include "interpreter/interpreter.hpp" |
|
29 #include "interpreter/interpreterRuntime.hpp" |
|
30 #include "interpreter/templateInterpreterGenerator.hpp" |
|
31 #include "runtime/arguments.hpp" |
|
32 #include "runtime/sharedRuntime.hpp" |
|
33 |
|
34 #define __ _masm-> |
|
35 |
|
36 |
|
37 address TemplateInterpreterGenerator::generate_slow_signature_handler() { |
|
38 address entry = __ pc(); |
|
39 // rbx,: method |
|
40 // rcx: temporary |
|
41 // rdi: pointer to locals |
|
42 // rsp: end of copied parameters area |
|
43 __ mov(rcx, rsp); |
|
44 __ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::slow_signature_handler), rbx, rdi, rcx); |
|
45 __ ret(0); |
|
46 return entry; |
|
47 } |
|
48 |
|
49 /** |
|
50 * Method entry for static native methods: |
|
51 * int java.util.zip.CRC32.update(int crc, int b) |
|
52 */ |
|
53 address TemplateInterpreterGenerator::generate_CRC32_update_entry() { |
|
54 if (UseCRC32Intrinsics) { |
|
55 address entry = __ pc(); |
|
56 |
|
57 // rbx: Method* |
|
58 // rsi: senderSP must preserved for slow path, set SP to it on fast path |
|
59 // rdx: scratch |
|
60 // rdi: scratch |
|
61 |
|
62 Label slow_path; |
|
63 // If we need a safepoint check, generate full interpreter entry. |
|
64 ExternalAddress state(SafepointSynchronize::address_of_state()); |
|
65 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), |
|
66 SafepointSynchronize::_not_synchronized); |
|
67 __ jcc(Assembler::notEqual, slow_path); |
|
68 |
|
69 // We don't generate local frame and don't align stack because |
|
70 // we call stub code and there is no safepoint on this path. |
|
71 |
|
72 // Load parameters |
|
73 const Register crc = rax; // crc |
|
74 const Register val = rdx; // source java byte value |
|
75 const Register tbl = rdi; // scratch |
|
76 |
|
77 // Arguments are reversed on java expression stack |
|
78 __ movl(val, Address(rsp, wordSize)); // byte value |
|
79 __ movl(crc, Address(rsp, 2*wordSize)); // Initial CRC |
|
80 |
|
81 __ lea(tbl, ExternalAddress(StubRoutines::crc_table_addr())); |
|
82 __ notl(crc); // ~crc |
|
83 __ update_byte_crc32(crc, val, tbl); |
|
84 __ notl(crc); // ~crc |
|
85 // result in rax |
|
86 |
|
87 // _areturn |
|
88 __ pop(rdi); // get return address |
|
89 __ mov(rsp, rsi); // set sp to sender sp |
|
90 __ jmp(rdi); |
|
91 |
|
92 // generate a vanilla native entry as the slow path |
|
93 __ bind(slow_path); |
|
94 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native)); |
|
95 return entry; |
|
96 } |
|
97 return NULL; |
|
98 } |
|
99 |
|
100 /** |
|
101 * Method entry for static native methods: |
|
102 * int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len) |
|
103 * int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) |
|
104 */ |
|
105 address TemplateInterpreterGenerator::generate_CRC32_updateBytes_entry(AbstractInterpreter::MethodKind kind) { |
|
106 if (UseCRC32Intrinsics) { |
|
107 address entry = __ pc(); |
|
108 |
|
109 // rbx,: Method* |
|
110 // rsi: senderSP must preserved for slow path, set SP to it on fast path |
|
111 // rdx: scratch |
|
112 // rdi: scratch |
|
113 |
|
114 Label slow_path; |
|
115 // If we need a safepoint check, generate full interpreter entry. |
|
116 ExternalAddress state(SafepointSynchronize::address_of_state()); |
|
117 __ cmp32(ExternalAddress(SafepointSynchronize::address_of_state()), |
|
118 SafepointSynchronize::_not_synchronized); |
|
119 __ jcc(Assembler::notEqual, slow_path); |
|
120 |
|
121 // We don't generate local frame and don't align stack because |
|
122 // we call stub code and there is no safepoint on this path. |
|
123 |
|
124 // Load parameters |
|
125 const Register crc = rax; // crc |
|
126 const Register buf = rdx; // source java byte array address |
|
127 const Register len = rdi; // length |
|
128 |
|
129 // value x86_32 |
|
130 // interp. arg ptr ESP + 4 |
|
131 // int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int len) |
|
132 // 3 2 1 0 |
|
133 // int java.util.zip.CRC32.updateByteBuffer(int crc, long buf, int off, int len) |
|
134 // 4 2,3 1 0 |
|
135 |
|
136 // Arguments are reversed on java expression stack |
|
137 __ movl(len, Address(rsp, 4 + 0)); // Length |
|
138 // Calculate address of start element |
|
139 if (kind == Interpreter::java_util_zip_CRC32_updateByteBuffer) { |
|
140 __ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // long buf |
|
141 __ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset |
|
142 __ movl(crc, Address(rsp, 4 + 4 * wordSize)); // Initial CRC |
|
143 } else { |
|
144 __ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // byte[] array |
|
145 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size |
|
146 __ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset |
|
147 __ movl(crc, Address(rsp, 4 + 3 * wordSize)); // Initial CRC |
|
148 } |
|
149 |
|
150 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32()), crc, buf, len); |
|
151 // result in rax |
|
152 |
|
153 // _areturn |
|
154 __ pop(rdi); // get return address |
|
155 __ mov(rsp, rsi); // set sp to sender sp |
|
156 __ jmp(rdi); |
|
157 |
|
158 // generate a vanilla native entry as the slow path |
|
159 __ bind(slow_path); |
|
160 __ jump_to_entry(Interpreter::entry_for_kind(Interpreter::native)); |
|
161 return entry; |
|
162 } |
|
163 return NULL; |
|
164 } |
|
165 |
|
166 /** |
|
167 * Method entry for static native methods: |
|
168 * int java.util.zip.CRC32C.updateBytes(int crc, byte[] b, int off, int end) |
|
169 * int java.util.zip.CRC32C.updateByteBuffer(int crc, long address, int off, int end) |
|
170 */ |
|
171 address TemplateInterpreterGenerator::generate_CRC32C_updateBytes_entry(AbstractInterpreter::MethodKind kind) { |
|
172 if (UseCRC32CIntrinsics) { |
|
173 address entry = __ pc(); |
|
174 // Load parameters |
|
175 const Register crc = rax; // crc |
|
176 const Register buf = rcx; // source java byte array address |
|
177 const Register len = rdx; // length |
|
178 const Register end = len; |
|
179 |
|
180 // value x86_32 |
|
181 // interp. arg ptr ESP + 4 |
|
182 // int java.util.zip.CRC32.updateBytes(int crc, byte[] b, int off, int end) |
|
183 // 3 2 1 0 |
|
184 // int java.util.zip.CRC32.updateByteBuffer(int crc, long address, int off, int end) |
|
185 // 4 2,3 1 0 |
|
186 |
|
187 // Arguments are reversed on java expression stack |
|
188 __ movl(end, Address(rsp, 4 + 0)); // end |
|
189 __ subl(len, Address(rsp, 4 + 1 * wordSize)); // end - offset == length |
|
190 // Calculate address of start element |
|
191 if (kind == Interpreter::java_util_zip_CRC32C_updateDirectByteBuffer) { |
|
192 __ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // long address |
|
193 __ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset |
|
194 __ movl(crc, Address(rsp, 4 + 4 * wordSize)); // Initial CRC |
|
195 } else { |
|
196 __ movptr(buf, Address(rsp, 4 + 2 * wordSize)); // byte[] array |
|
197 __ addptr(buf, arrayOopDesc::base_offset_in_bytes(T_BYTE)); // + header size |
|
198 __ addptr(buf, Address(rsp, 4 + 1 * wordSize)); // + offset |
|
199 __ movl(crc, Address(rsp, 4 + 3 * wordSize)); // Initial CRC |
|
200 } |
|
201 __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, StubRoutines::updateBytesCRC32C()), crc, buf, len); |
|
202 // result in rax |
|
203 // _areturn |
|
204 __ pop(rdi); // get return address |
|
205 __ mov(rsp, rsi); // set sp to sender sp |
|
206 __ jmp(rdi); |
|
207 |
|
208 return entry; |
|
209 } |
|
210 return NULL; |
|
211 } |
|
212 |
|
213 /** |
|
214 * Method entry for static native method: |
|
215 * java.lang.Float.intBitsToFloat(int bits) |
|
216 */ |
|
217 address TemplateInterpreterGenerator::generate_Float_intBitsToFloat_entry() { |
|
218 if (UseSSE >= 1) { |
|
219 address entry = __ pc(); |
|
220 |
|
221 // rsi: the sender's SP |
|
222 |
|
223 // Skip safepoint check (compiler intrinsic versions of this method |
|
224 // do not perform safepoint checks either). |
|
225 |
|
226 // Load 'bits' into xmm0 (interpreter returns results in xmm0) |
|
227 __ movflt(xmm0, Address(rsp, wordSize)); |
|
228 |
|
229 // Return |
|
230 __ pop(rdi); // get return address |
|
231 __ mov(rsp, rsi); // set rsp to the sender's SP |
|
232 __ jmp(rdi); |
|
233 return entry; |
|
234 } |
|
235 |
|
236 return NULL; |
|
237 } |
|
238 |
|
239 /** |
|
240 * Method entry for static native method: |
|
241 * java.lang.Float.floatToRawIntBits(float value) |
|
242 */ |
|
243 address TemplateInterpreterGenerator::generate_Float_floatToRawIntBits_entry() { |
|
244 if (UseSSE >= 1) { |
|
245 address entry = __ pc(); |
|
246 |
|
247 // rsi: the sender's SP |
|
248 |
|
249 // Skip safepoint check (compiler intrinsic versions of this method |
|
250 // do not perform safepoint checks either). |
|
251 |
|
252 // Load the parameter (a floating-point value) into rax. |
|
253 __ movl(rax, Address(rsp, wordSize)); |
|
254 |
|
255 // Return |
|
256 __ pop(rdi); // get return address |
|
257 __ mov(rsp, rsi); // set rsp to the sender's SP |
|
258 __ jmp(rdi); |
|
259 return entry; |
|
260 } |
|
261 |
|
262 return NULL; |
|
263 } |
|
264 |
|
265 |
|
266 /** |
|
267 * Method entry for static native method: |
|
268 * java.lang.Double.longBitsToDouble(long bits) |
|
269 */ |
|
270 address TemplateInterpreterGenerator::generate_Double_longBitsToDouble_entry() { |
|
271 if (UseSSE >= 2) { |
|
272 address entry = __ pc(); |
|
273 |
|
274 // rsi: the sender's SP |
|
275 |
|
276 // Skip safepoint check (compiler intrinsic versions of this method |
|
277 // do not perform safepoint checks either). |
|
278 |
|
279 // Load 'bits' into xmm0 (interpreter returns results in xmm0) |
|
280 __ movdbl(xmm0, Address(rsp, wordSize)); |
|
281 |
|
282 // Return |
|
283 __ pop(rdi); // get return address |
|
284 __ mov(rsp, rsi); // set rsp to the sender's SP |
|
285 __ jmp(rdi); |
|
286 return entry; |
|
287 } |
|
288 |
|
289 return NULL; |
|
290 } |
|
291 |
|
292 /** |
|
293 * Method entry for static native method: |
|
294 * java.lang.Double.doubleToRawLongBits(double value) |
|
295 */ |
|
296 address TemplateInterpreterGenerator::generate_Double_doubleToRawLongBits_entry() { |
|
297 if (UseSSE >= 2) { |
|
298 address entry = __ pc(); |
|
299 |
|
300 // rsi: the sender's SP |
|
301 |
|
302 // Skip safepoint check (compiler intrinsic versions of this method |
|
303 // do not perform safepoint checks either). |
|
304 |
|
305 // Load the parameter (a floating-point value) into rax. |
|
306 __ movl(rdx, Address(rsp, 2*wordSize)); |
|
307 __ movl(rax, Address(rsp, wordSize)); |
|
308 |
|
309 // Return |
|
310 __ pop(rdi); // get return address |
|
311 __ mov(rsp, rsi); // set rsp to the sender's SP |
|
312 __ jmp(rdi); |
|
313 return entry; |
|
314 } |
|
315 |
|
316 return NULL; |
|
317 } |
|
318 |
|
319 address TemplateInterpreterGenerator::generate_math_entry(AbstractInterpreter::MethodKind kind) { |
|
320 |
|
321 // rbx,: Method* |
|
322 // rcx: scratrch |
|
323 // rsi: sender sp |
|
324 |
|
325 if (!InlineIntrinsics) return NULL; // Generate a vanilla entry |
|
326 |
|
327 address entry_point = __ pc(); |
|
328 |
|
329 // These don't need a safepoint check because they aren't virtually |
|
330 // callable. We won't enter these intrinsics from compiled code. |
|
331 // If in the future we added an intrinsic which was virtually callable |
|
332 // we'd have to worry about how to safepoint so that this code is used. |
|
333 |
|
334 // mathematical functions inlined by compiler |
|
335 // (interpreter must provide identical implementation |
|
336 // in order to avoid monotonicity bugs when switching |
|
337 // from interpreter to compiler in the middle of some |
|
338 // computation) |
|
339 // |
|
340 // stack: [ ret adr ] <-- rsp |
|
341 // [ lo(arg) ] |
|
342 // [ hi(arg) ] |
|
343 // |
|
344 if (kind == Interpreter::java_lang_math_fmaD) { |
|
345 if (!UseFMA) { |
|
346 return NULL; // Generate a vanilla entry |
|
347 } |
|
348 __ movdbl(xmm2, Address(rsp, 5 * wordSize)); |
|
349 __ movdbl(xmm1, Address(rsp, 3 * wordSize)); |
|
350 __ movdbl(xmm0, Address(rsp, 1 * wordSize)); |
|
351 __ fmad(xmm0, xmm1, xmm2, xmm0); |
|
352 __ pop(rdi); // get return address |
|
353 __ mov(rsp, rsi); // set sp to sender sp |
|
354 __ jmp(rdi); |
|
355 |
|
356 return entry_point; |
|
357 } else if (kind == Interpreter::java_lang_math_fmaF) { |
|
358 if (!UseFMA) { |
|
359 return NULL; // Generate a vanilla entry |
|
360 } |
|
361 __ movflt(xmm2, Address(rsp, 3 * wordSize)); |
|
362 __ movflt(xmm1, Address(rsp, 2 * wordSize)); |
|
363 __ movflt(xmm0, Address(rsp, 1 * wordSize)); |
|
364 __ fmaf(xmm0, xmm1, xmm2, xmm0); |
|
365 __ pop(rdi); // get return address |
|
366 __ mov(rsp, rsi); // set sp to sender sp |
|
367 __ jmp(rdi); |
|
368 |
|
369 return entry_point; |
|
370 } |
|
371 |
|
372 __ fld_d(Address(rsp, 1*wordSize)); |
|
373 switch (kind) { |
|
374 case Interpreter::java_lang_math_sin : |
|
375 __ subptr(rsp, 2 * wordSize); |
|
376 __ fstp_d(Address(rsp, 0)); |
|
377 if (VM_Version::supports_sse2() && StubRoutines::dsin() != NULL) { |
|
378 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::dsin()))); |
|
379 } else { |
|
380 __ call_VM_leaf0(CAST_FROM_FN_PTR(address, SharedRuntime::dsin)); |
|
381 } |
|
382 __ addptr(rsp, 2 * wordSize); |
|
383 break; |
|
384 case Interpreter::java_lang_math_cos : |
|
385 __ subptr(rsp, 2 * wordSize); |
|
386 __ fstp_d(Address(rsp, 0)); |
|
387 if (VM_Version::supports_sse2() && StubRoutines::dcos() != NULL) { |
|
388 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::dcos()))); |
|
389 } else { |
|
390 __ call_VM_leaf0(CAST_FROM_FN_PTR(address, SharedRuntime::dcos)); |
|
391 } |
|
392 __ addptr(rsp, 2 * wordSize); |
|
393 break; |
|
394 case Interpreter::java_lang_math_tan : |
|
395 __ subptr(rsp, 2 * wordSize); |
|
396 __ fstp_d(Address(rsp, 0)); |
|
397 if (StubRoutines::dtan() != NULL) { |
|
398 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::dtan()))); |
|
399 } else { |
|
400 __ call_VM_leaf0(CAST_FROM_FN_PTR(address, SharedRuntime::dtan)); |
|
401 } |
|
402 __ addptr(rsp, 2 * wordSize); |
|
403 break; |
|
404 case Interpreter::java_lang_math_sqrt: |
|
405 __ fsqrt(); |
|
406 break; |
|
407 case Interpreter::java_lang_math_abs: |
|
408 __ fabs(); |
|
409 break; |
|
410 case Interpreter::java_lang_math_log: |
|
411 __ subptr(rsp, 2 * wordSize); |
|
412 __ fstp_d(Address(rsp, 0)); |
|
413 if (StubRoutines::dlog() != NULL) { |
|
414 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::dlog()))); |
|
415 } else { |
|
416 __ call_VM_leaf0(CAST_FROM_FN_PTR(address, SharedRuntime::dlog)); |
|
417 } |
|
418 __ addptr(rsp, 2 * wordSize); |
|
419 break; |
|
420 case Interpreter::java_lang_math_log10: |
|
421 __ subptr(rsp, 2 * wordSize); |
|
422 __ fstp_d(Address(rsp, 0)); |
|
423 if (StubRoutines::dlog10() != NULL) { |
|
424 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::dlog10()))); |
|
425 } else { |
|
426 __ call_VM_leaf0(CAST_FROM_FN_PTR(address, SharedRuntime::dlog10)); |
|
427 } |
|
428 __ addptr(rsp, 2 * wordSize); |
|
429 break; |
|
430 case Interpreter::java_lang_math_pow: |
|
431 __ fld_d(Address(rsp, 3*wordSize)); // second argument |
|
432 __ subptr(rsp, 4 * wordSize); |
|
433 __ fstp_d(Address(rsp, 0)); |
|
434 __ fstp_d(Address(rsp, 2 * wordSize)); |
|
435 if (StubRoutines::dpow() != NULL) { |
|
436 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::dpow()))); |
|
437 } else { |
|
438 __ call_VM_leaf0(CAST_FROM_FN_PTR(address, SharedRuntime::dpow)); |
|
439 } |
|
440 __ addptr(rsp, 4 * wordSize); |
|
441 break; |
|
442 case Interpreter::java_lang_math_exp: |
|
443 __ subptr(rsp, 2*wordSize); |
|
444 __ fstp_d(Address(rsp, 0)); |
|
445 if (StubRoutines::dexp() != NULL) { |
|
446 __ call(RuntimeAddress(CAST_FROM_FN_PTR(address, StubRoutines::dexp()))); |
|
447 } else { |
|
448 __ call_VM_leaf0(CAST_FROM_FN_PTR(address, SharedRuntime::dexp)); |
|
449 } |
|
450 __ addptr(rsp, 2*wordSize); |
|
451 break; |
|
452 default : |
|
453 ShouldNotReachHere(); |
|
454 } |
|
455 |
|
456 // return double result in xmm0 for interpreter and compilers. |
|
457 if (UseSSE >= 2) { |
|
458 __ subptr(rsp, 2*wordSize); |
|
459 __ fstp_d(Address(rsp, 0)); |
|
460 __ movdbl(xmm0, Address(rsp, 0)); |
|
461 __ addptr(rsp, 2*wordSize); |
|
462 } |
|
463 |
|
464 // done, result in FPU ST(0) or XMM0 |
|
465 __ pop(rdi); // get return address |
|
466 __ mov(rsp, rsi); // set sp to sender sp |
|
467 __ jmp(rdi); |
|
468 |
|
469 return entry_point; |
|
470 } |