1 //
2 // Copyright 2003-2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 // CA 95054 USA or visit www.sun.com if you need additional information or
21 // have any questions.
22 //
23 //
24
25 // AMD64 Architecture Description File
26
27 //----------REGISTER DEFINITION BLOCK------------------------------------------
28 // This information is used by the matcher and the register allocator to
29 // describe individual registers and classes of registers within the target
30 // archtecture.
31
32 register %{
33 //----------Architecture Description Register Definitions----------------------
34 // General Registers
35 // "reg_def" name ( register save type, C convention save type,
36 // ideal register type, encoding );
37 // Register Save Types:
38 //
39 // NS = No-Save: The register allocator assumes that these registers
40 // can be used without saving upon entry to the method, &
41 // that they do not need to be saved at call sites.
42 //
43 // SOC = Save-On-Call: The register allocator assumes that these registers
44 // can be used without saving upon entry to the method,
45 // but that they must be saved at call sites.
46 //
47 // SOE = Save-On-Entry: The register allocator assumes that these registers
48 // must be saved before using them upon entry to the
49 // method, but they do not need to be saved at call
50 // sites.
51 //
52 // AS = Always-Save: The register allocator assumes that these registers
53 // must be saved before using them upon entry to the
54 // method, & that they must be saved at call sites.
55 //
56 // Ideal Register Type is used to determine how to save & restore a
57 // register. Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
58 // spilled with LoadP/StoreP. If the register supports both, use Op_RegI.
59 //
60 // The encoding number is the actual bit-pattern placed into the opcodes.
61
62 // General Registers
63 // R8-R15 must be encoded with REX. (RSP, RBP, RSI, RDI need REX when
64 // used as byte registers)
65
66 // Previously set RBX, RSI, and RDI as save-on-entry for java code
67 // Turn off SOE in java-code due to frequent use of uncommon-traps.
68 // Now that allocator is better, turn on RSI and RDI as SOE registers.
69
70 reg_def RAX (SOC, SOC, Op_RegI, 0, rax->as_VMReg());
71 reg_def RAX_H(SOC, SOC, Op_RegI, 0, rax->as_VMReg()->next());
72
73 reg_def RCX (SOC, SOC, Op_RegI, 1, rcx->as_VMReg());
74 reg_def RCX_H(SOC, SOC, Op_RegI, 1, rcx->as_VMReg()->next());
75
76 reg_def RDX (SOC, SOC, Op_RegI, 2, rdx->as_VMReg());
77 reg_def RDX_H(SOC, SOC, Op_RegI, 2, rdx->as_VMReg()->next());
78
79 reg_def RBX (SOC, SOE, Op_RegI, 3, rbx->as_VMReg());
80 reg_def RBX_H(SOC, SOE, Op_RegI, 3, rbx->as_VMReg()->next());
81
82 reg_def RSP (NS, NS, Op_RegI, 4, rsp->as_VMReg());
83 reg_def RSP_H(NS, NS, Op_RegI, 4, rsp->as_VMReg()->next());
84
85 // now that adapter frames are gone RBP is always saved and restored by the prolog/epilog code
86 reg_def RBP (NS, SOE, Op_RegI, 5, rbp->as_VMReg());
87 reg_def RBP_H(NS, SOE, Op_RegI, 5, rbp->as_VMReg()->next());
88
89 #ifdef _WIN64
90
91 reg_def RSI (SOC, SOE, Op_RegI, 6, rsi->as_VMReg());
92 reg_def RSI_H(SOC, SOE, Op_RegI, 6, rsi->as_VMReg()->next());
93
94 reg_def RDI (SOC, SOE, Op_RegI, 7, rdi->as_VMReg());
95 reg_def RDI_H(SOC, SOE, Op_RegI, 7, rdi->as_VMReg()->next());
96
97 #else
98
99 reg_def RSI (SOC, SOC, Op_RegI, 6, rsi->as_VMReg());
100 reg_def RSI_H(SOC, SOC, Op_RegI, 6, rsi->as_VMReg()->next());
101
102 reg_def RDI (SOC, SOC, Op_RegI, 7, rdi->as_VMReg());
103 reg_def RDI_H(SOC, SOC, Op_RegI, 7, rdi->as_VMReg()->next());
104
105 #endif
106
107 reg_def R8 (SOC, SOC, Op_RegI, 8, r8->as_VMReg());
108 reg_def R8_H (SOC, SOC, Op_RegI, 8, r8->as_VMReg()->next());
109
110 reg_def R9 (SOC, SOC, Op_RegI, 9, r9->as_VMReg());
111 reg_def R9_H (SOC, SOC, Op_RegI, 9, r9->as_VMReg()->next());
112
113 reg_def R10 (SOC, SOC, Op_RegI, 10, r10->as_VMReg());
114 reg_def R10_H(SOC, SOC, Op_RegI, 10, r10->as_VMReg()->next());
115
116 reg_def R11 (SOC, SOC, Op_RegI, 11, r11->as_VMReg());
117 reg_def R11_H(SOC, SOC, Op_RegI, 11, r11->as_VMReg()->next());
118
119 reg_def R12 (SOC, SOE, Op_RegI, 12, r12->as_VMReg());
120 reg_def R12_H(SOC, SOE, Op_RegI, 12, r12->as_VMReg()->next());
121
122 reg_def R13 (SOC, SOE, Op_RegI, 13, r13->as_VMReg());
123 reg_def R13_H(SOC, SOE, Op_RegI, 13, r13->as_VMReg()->next());
124
125 reg_def R14 (SOC, SOE, Op_RegI, 14, r14->as_VMReg());
126 reg_def R14_H(SOC, SOE, Op_RegI, 14, r14->as_VMReg()->next());
127
128 reg_def R15 (SOC, SOE, Op_RegI, 15, r15->as_VMReg());
129 reg_def R15_H(SOC, SOE, Op_RegI, 15, r15->as_VMReg()->next());
130
131
132 // Floating Point Registers
133
134 // XMM registers. 128-bit registers or 4 words each, labeled (a)-d.
135 // Word a in each register holds a Float, words ab hold a Double. We
136 // currently do not use the SIMD capabilities, so registers cd are
137 // unused at the moment.
138 // XMM8-XMM15 must be encoded with REX.
139 // Linux ABI: No register preserved across function calls
140 // XMM0-XMM7 might hold parameters
141 // Windows ABI: XMM6-XMM15 preserved across function calls
142 // XMM0-XMM3 might hold parameters
143
144 reg_def XMM0 (SOC, SOC, Op_RegF, 0, xmm0->as_VMReg());
145 reg_def XMM0_H (SOC, SOC, Op_RegF, 0, xmm0->as_VMReg()->next());
146
147 reg_def XMM1 (SOC, SOC, Op_RegF, 1, xmm1->as_VMReg());
148 reg_def XMM1_H (SOC, SOC, Op_RegF, 1, xmm1->as_VMReg()->next());
149
150 reg_def XMM2 (SOC, SOC, Op_RegF, 2, xmm2->as_VMReg());
151 reg_def XMM2_H (SOC, SOC, Op_RegF, 2, xmm2->as_VMReg()->next());
152
153 reg_def XMM3 (SOC, SOC, Op_RegF, 3, xmm3->as_VMReg());
154 reg_def XMM3_H (SOC, SOC, Op_RegF, 3, xmm3->as_VMReg()->next());
155
156 reg_def XMM4 (SOC, SOC, Op_RegF, 4, xmm4->as_VMReg());
157 reg_def XMM4_H (SOC, SOC, Op_RegF, 4, xmm4->as_VMReg()->next());
158
159 reg_def XMM5 (SOC, SOC, Op_RegF, 5, xmm5->as_VMReg());
160 reg_def XMM5_H (SOC, SOC, Op_RegF, 5, xmm5->as_VMReg()->next());
161
162 #ifdef _WIN64
163
164 reg_def XMM6 (SOC, SOE, Op_RegF, 6, xmm6->as_VMReg());
165 reg_def XMM6_H (SOC, SOE, Op_RegF, 6, xmm6->as_VMReg()->next());
166
167 reg_def XMM7 (SOC, SOE, Op_RegF, 7, xmm7->as_VMReg());
168 reg_def XMM7_H (SOC, SOE, Op_RegF, 7, xmm7->as_VMReg()->next());
169
170 reg_def XMM8 (SOC, SOE, Op_RegF, 8, xmm8->as_VMReg());
171 reg_def XMM8_H (SOC, SOE, Op_RegF, 8, xmm8->as_VMReg()->next());
172
173 reg_def XMM9 (SOC, SOE, Op_RegF, 9, xmm9->as_VMReg());
174 reg_def XMM9_H (SOC, SOE, Op_RegF, 9, xmm9->as_VMReg()->next());
175
176 reg_def XMM10 (SOC, SOE, Op_RegF, 10, xmm10->as_VMReg());
177 reg_def XMM10_H(SOC, SOE, Op_RegF, 10, xmm10->as_VMReg()->next());
178
179 reg_def XMM11 (SOC, SOE, Op_RegF, 11, xmm11->as_VMReg());
180 reg_def XMM11_H(SOC, SOE, Op_RegF, 11, xmm11->as_VMReg()->next());
181
182 reg_def XMM12 (SOC, SOE, Op_RegF, 12, xmm12->as_VMReg());
183 reg_def XMM12_H(SOC, SOE, Op_RegF, 12, xmm12->as_VMReg()->next());
184
185 reg_def XMM13 (SOC, SOE, Op_RegF, 13, xmm13->as_VMReg());
186 reg_def XMM13_H(SOC, SOE, Op_RegF, 13, xmm13->as_VMReg()->next());
187
188 reg_def XMM14 (SOC, SOE, Op_RegF, 14, xmm14->as_VMReg());
189 reg_def XMM14_H(SOC, SOE, Op_RegF, 14, xmm14->as_VMReg()->next());
190
191 reg_def XMM15 (SOC, SOE, Op_RegF, 15, xmm15->as_VMReg());
192 reg_def XMM15_H(SOC, SOE, Op_RegF, 15, xmm15->as_VMReg()->next());
193
194 #else
195
196 reg_def XMM6 (SOC, SOC, Op_RegF, 6, xmm6->as_VMReg());
197 reg_def XMM6_H (SOC, SOC, Op_RegF, 6, xmm6->as_VMReg()->next());
198
199 reg_def XMM7 (SOC, SOC, Op_RegF, 7, xmm7->as_VMReg());
200 reg_def XMM7_H (SOC, SOC, Op_RegF, 7, xmm7->as_VMReg()->next());
201
202 reg_def XMM8 (SOC, SOC, Op_RegF, 8, xmm8->as_VMReg());
203 reg_def XMM8_H (SOC, SOC, Op_RegF, 8, xmm8->as_VMReg()->next());
204
205 reg_def XMM9 (SOC, SOC, Op_RegF, 9, xmm9->as_VMReg());
206 reg_def XMM9_H (SOC, SOC, Op_RegF, 9, xmm9->as_VMReg()->next());
207
208 reg_def XMM10 (SOC, SOC, Op_RegF, 10, xmm10->as_VMReg());
209 reg_def XMM10_H(SOC, SOC, Op_RegF, 10, xmm10->as_VMReg()->next());
210
211 reg_def XMM11 (SOC, SOC, Op_RegF, 11, xmm11->as_VMReg());
212 reg_def XMM11_H(SOC, SOC, Op_RegF, 11, xmm11->as_VMReg()->next());
213
214 reg_def XMM12 (SOC, SOC, Op_RegF, 12, xmm12->as_VMReg());
215 reg_def XMM12_H(SOC, SOC, Op_RegF, 12, xmm12->as_VMReg()->next());
216
217 reg_def XMM13 (SOC, SOC, Op_RegF, 13, xmm13->as_VMReg());
218 reg_def XMM13_H(SOC, SOC, Op_RegF, 13, xmm13->as_VMReg()->next());
219
220 reg_def XMM14 (SOC, SOC, Op_RegF, 14, xmm14->as_VMReg());
221 reg_def XMM14_H(SOC, SOC, Op_RegF, 14, xmm14->as_VMReg()->next());
222
223 reg_def XMM15 (SOC, SOC, Op_RegF, 15, xmm15->as_VMReg());
224 reg_def XMM15_H(SOC, SOC, Op_RegF, 15, xmm15->as_VMReg()->next());
225
226 #endif // _WIN64
227
228 reg_def RFLAGS(SOC, SOC, 0, 16, VMRegImpl::Bad());
229
230 // Specify priority of register selection within phases of register
231 // allocation. Highest priority is first. A useful heuristic is to
232 // give registers a low priority when they are required by machine
233 // instructions, like EAX and EDX on I486, and choose no-save registers
234 // before save-on-call, & save-on-call before save-on-entry. Registers
235 // which participate in fixed calling sequences should come last.
236 // Registers which are used as pairs must fall on an even boundary.
237
238 alloc_class chunk0(R10, R10_H,
239 R11, R11_H,
240 R8, R8_H,
241 R9, R9_H,
242 R12, R12_H,
243 RCX, RCX_H,
244 RBX, RBX_H,
245 RDI, RDI_H,
246 RDX, RDX_H,
247 RSI, RSI_H,
248 RAX, RAX_H,
249 RBP, RBP_H,
250 R13, R13_H,
251 R14, R14_H,
252 R15, R15_H,
253 RSP, RSP_H);
254
255 // XXX probably use 8-15 first on Linux
256 alloc_class chunk1(XMM0, XMM0_H,
257 XMM1, XMM1_H,
258 XMM2, XMM2_H,
259 XMM3, XMM3_H,
260 XMM4, XMM4_H,
261 XMM5, XMM5_H,
262 XMM6, XMM6_H,
263 XMM7, XMM7_H,
264 XMM8, XMM8_H,
265 XMM9, XMM9_H,
266 XMM10, XMM10_H,
267 XMM11, XMM11_H,
268 XMM12, XMM12_H,
269 XMM13, XMM13_H,
270 XMM14, XMM14_H,
271 XMM15, XMM15_H);
272
273 alloc_class chunk2(RFLAGS);
274
275
276 //----------Architecture Description Register Classes--------------------------
277 // Several register classes are automatically defined based upon information in
278 // this architecture description.
279 // 1) reg_class inline_cache_reg ( /* as def'd in frame section */ )
280 // 2) reg_class compiler_method_oop_reg ( /* as def'd in frame section */ )
281 // 2) reg_class interpreter_method_oop_reg ( /* as def'd in frame section */ )
282 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
283 //
284
285 // Class for all pointer registers (including RSP)
286 reg_class any_reg(RAX, RAX_H,
287 RDX, RDX_H,
288 RBP, RBP_H,
289 RDI, RDI_H,
290 RSI, RSI_H,
291 RCX, RCX_H,
292 RBX, RBX_H,
293 RSP, RSP_H,
294 R8, R8_H,
295 R9, R9_H,
296 R10, R10_H,
297 R11, R11_H,
298 R12, R12_H,
299 R13, R13_H,
300 R14, R14_H,
301 R15, R15_H);
302
303 // Class for all pointer registers except RSP
304 reg_class ptr_reg(RAX, RAX_H,
305 RDX, RDX_H,
306 RBP, RBP_H,
307 RDI, RDI_H,
308 RSI, RSI_H,
309 RCX, RCX_H,
310 RBX, RBX_H,
311 R8, R8_H,
312 R9, R9_H,
313 R10, R10_H,
314 R11, R11_H,
315 R13, R13_H,
316 R14, R14_H);
317
318 // Class for all pointer registers except RAX and RSP
319 reg_class ptr_no_rax_reg(RDX, RDX_H,
320 RBP, RBP_H,
321 RDI, RDI_H,
322 RSI, RSI_H,
323 RCX, RCX_H,
324 RBX, RBX_H,
325 R8, R8_H,
326 R9, R9_H,
327 R10, R10_H,
328 R11, R11_H,
329 R13, R13_H,
330 R14, R14_H);
331
332 reg_class ptr_no_rbp_reg(RDX, RDX_H,
333 RAX, RAX_H,
334 RDI, RDI_H,
335 RSI, RSI_H,
336 RCX, RCX_H,
337 RBX, RBX_H,
338 R8, R8_H,
339 R9, R9_H,
340 R10, R10_H,
341 R11, R11_H,
342 R13, R13_H,
343 R14, R14_H);
344
345 // Class for all pointer registers except RAX, RBX and RSP
346 reg_class ptr_no_rax_rbx_reg(RDX, RDX_H,
347 RBP, RBP_H,
348 RDI, RDI_H,
349 RSI, RSI_H,
350 RCX, RCX_H,
351 R8, R8_H,
352 R9, R9_H,
353 R10, R10_H,
354 R11, R11_H,
355 R13, R13_H,
356 R14, R14_H);
357
358 // Singleton class for RAX pointer register
359 reg_class ptr_rax_reg(RAX, RAX_H);
360
361 // Singleton class for RBX pointer register
362 reg_class ptr_rbx_reg(RBX, RBX_H);
363
364 // Singleton class for RSI pointer register
365 reg_class ptr_rsi_reg(RSI, RSI_H);
366
367 // Singleton class for RDI pointer register
368 reg_class ptr_rdi_reg(RDI, RDI_H);
369
370 // Singleton class for RBP pointer register
371 reg_class ptr_rbp_reg(RBP, RBP_H);
372
373 // Singleton class for stack pointer
374 reg_class ptr_rsp_reg(RSP, RSP_H);
375
376 // Singleton class for TLS pointer
377 reg_class ptr_r15_reg(R15, R15_H);
378
379 // Class for all long registers (except RSP)
380 reg_class long_reg(RAX, RAX_H,
381 RDX, RDX_H,
382 RBP, RBP_H,
383 RDI, RDI_H,
384 RSI, RSI_H,
385 RCX, RCX_H,
386 RBX, RBX_H,
387 R8, R8_H,
388 R9, R9_H,
389 R10, R10_H,
390 R11, R11_H,
391 R13, R13_H,
392 R14, R14_H);
393
394 // Class for all long registers except RAX, RDX (and RSP)
395 reg_class long_no_rax_rdx_reg(RBP, RBP_H,
396 RDI, RDI_H,
397 RSI, RSI_H,
398 RCX, RCX_H,
399 RBX, RBX_H,
400 R8, R8_H,
401 R9, R9_H,
402 R10, R10_H,
403 R11, R11_H,
404 R13, R13_H,
405 R14, R14_H);
406
407 // Class for all long registers except RCX (and RSP)
408 reg_class long_no_rcx_reg(RBP, RBP_H,
409 RDI, RDI_H,
410 RSI, RSI_H,
411 RAX, RAX_H,
412 RDX, RDX_H,
413 RBX, RBX_H,
414 R8, R8_H,
415 R9, R9_H,
416 R10, R10_H,
417 R11, R11_H,
418 R13, R13_H,
419 R14, R14_H);
420
421 // Class for all long registers except RAX (and RSP)
422 reg_class long_no_rax_reg(RBP, RBP_H,
423 RDX, RDX_H,
424 RDI, RDI_H,
425 RSI, RSI_H,
426 RCX, RCX_H,
427 RBX, RBX_H,
428 R8, R8_H,
429 R9, R9_H,
430 R10, R10_H,
431 R11, R11_H,
432 R13, R13_H,
433 R14, R14_H);
434
435 // Singleton class for RAX long register
436 reg_class long_rax_reg(RAX, RAX_H);
437
438 // Singleton class for RCX long register
439 reg_class long_rcx_reg(RCX, RCX_H);
440
441 // Singleton class for RDX long register
442 reg_class long_rdx_reg(RDX, RDX_H);
443
444 // Class for all int registers (except RSP)
445 reg_class int_reg(RAX,
446 RDX,
447 RBP,
448 RDI,
449 RSI,
450 RCX,
451 RBX,
452 R8,
453 R9,
454 R10,
455 R11,
456 R13,
457 R14);
458
459 // Class for all int registers except RCX (and RSP)
460 reg_class int_no_rcx_reg(RAX,
461 RDX,
462 RBP,
463 RDI,
464 RSI,
465 RBX,
466 R8,
467 R9,
468 R10,
469 R11,
470 R13,
471 R14);
472
473 // Class for all int registers except RAX, RDX (and RSP)
474 reg_class int_no_rax_rdx_reg(RBP,
475 RDI,
476 RSI,
477 RCX,
478 RBX,
479 R8,
480 R9,
481 R10,
482 R11,
483 R13,
484 R14);
485
486 // Singleton class for RAX int register
487 reg_class int_rax_reg(RAX);
488
489 // Singleton class for RBX int register
490 reg_class int_rbx_reg(RBX);
491
492 // Singleton class for RCX int register
493 reg_class int_rcx_reg(RCX);
494
495 // Singleton class for RCX int register
496 reg_class int_rdx_reg(RDX);
497
498 // Singleton class for RCX int register
499 reg_class int_rdi_reg(RDI);
500
501 // Singleton class for instruction pointer
502 // reg_class ip_reg(RIP);
503
504 // Singleton class for condition codes
505 reg_class int_flags(RFLAGS);
506
507 // Class for all float registers
508 reg_class float_reg(XMM0,
509 XMM1,
510 XMM2,
511 XMM3,
512 XMM4,
513 XMM5,
514 XMM6,
515 XMM7,
516 XMM8,
517 XMM9,
518 XMM10,
519 XMM11,
520 XMM12,
521 XMM13,
522 XMM14,
523 XMM15);
524
525 // Class for all double registers
526 reg_class double_reg(XMM0, XMM0_H,
527 XMM1, XMM1_H,
528 XMM2, XMM2_H,
529 XMM3, XMM3_H,
530 XMM4, XMM4_H,
531 XMM5, XMM5_H,
532 XMM6, XMM6_H,
533 XMM7, XMM7_H,
534 XMM8, XMM8_H,
535 XMM9, XMM9_H,
536 XMM10, XMM10_H,
537 XMM11, XMM11_H,
538 XMM12, XMM12_H,
539 XMM13, XMM13_H,
540 XMM14, XMM14_H,
541 XMM15, XMM15_H);
542 %}
543
544
545 //----------SOURCE BLOCK-------------------------------------------------------
546 // This is a block of C++ code which provides values, functions, and
547 // definitions necessary in the rest of the architecture description
548 source %{
549 #define RELOC_IMM64 Assembler::imm_operand
550 #define RELOC_DISP32 Assembler::disp32_operand
551
552 #define __ _masm.
553
554 static int preserve_SP_size() {
555 return LP64_ONLY(1 +) 2; // [rex,] op, rm(reg/reg)
556 }
557
558 // !!!!! Special hack to get all types of calls to specify the byte offset
559 // from the start of the call to the point where the return address
560 // will point.
561 int MachCallStaticJavaNode::ret_addr_offset()
562 {
563 int offset = 5; // 5 bytes from start of call to where return address points
564 if (_method_handle_invoke)
565 offset += preserve_SP_size();
566 return offset;
567 }
568
569 int MachCallDynamicJavaNode::ret_addr_offset()
570 {
571 return 15; // 15 bytes from start of call to where return address points
572 }
573
574 // In os_cpu .ad file
575 // int MachCallRuntimeNode::ret_addr_offset()
576
577 // Indicate if the safepoint node needs the polling page as an input.
578 // Since amd64 does not have absolute addressing but RIP-relative
579 // addressing and the polling page is within 2G, it doesn't.
580 bool SafePointNode::needs_polling_address_input()
581 {
582 return false;
583 }
584
585 //
586 // Compute padding required for nodes which need alignment
587 //
588
589 // The address of the call instruction needs to be 4-byte aligned to
590 // ensure that it does not span a cache line so that it can be patched.
591 int CallStaticJavaDirectNode::compute_padding(int current_offset) const
592 {
593 current_offset += 1; // skip call opcode byte
594 return round_to(current_offset, alignment_required()) - current_offset;
595 }
596
597 // The address of the call instruction needs to be 4-byte aligned to
598 // ensure that it does not span a cache line so that it can be patched.
599 int CallStaticJavaHandleNode::compute_padding(int current_offset) const
600 {
601 current_offset += preserve_SP_size(); // skip mov rbp, rsp
602 current_offset += 1; // skip call opcode byte
603 return round_to(current_offset, alignment_required()) - current_offset;
604 }
605
606 // The address of the call instruction needs to be 4-byte aligned to
607 // ensure that it does not span a cache line so that it can be patched.
608 int CallDynamicJavaDirectNode::compute_padding(int current_offset) const
609 {
610 current_offset += 11; // skip movq instruction + call opcode byte
611 return round_to(current_offset, alignment_required()) - current_offset;
612 }
613
614 #ifndef PRODUCT
615 void MachBreakpointNode::format(PhaseRegAlloc*, outputStream* st) const
616 {
617 st->print("INT3");
618 }
619 #endif
620
621 // EMIT_RM()
622 void emit_rm(CodeBuffer &cbuf, int f1, int f2, int f3)
623 {
624 unsigned char c = (unsigned char) ((f1 << 6) | (f2 << 3) | f3);
625 *(cbuf.code_end()) = c;
626 cbuf.set_code_end(cbuf.code_end() + 1);
627 }
628
629 // EMIT_CC()
630 void emit_cc(CodeBuffer &cbuf, int f1, int f2)
631 {
632 unsigned char c = (unsigned char) (f1 | f2);
633 *(cbuf.code_end()) = c;
634 cbuf.set_code_end(cbuf.code_end() + 1);
635 }
636
637 // EMIT_OPCODE()
638 void emit_opcode(CodeBuffer &cbuf, int code)
639 {
640 *(cbuf.code_end()) = (unsigned char) code;
641 cbuf.set_code_end(cbuf.code_end() + 1);
642 }
643
644 // EMIT_OPCODE() w/ relocation information
645 void emit_opcode(CodeBuffer &cbuf,
646 int code, relocInfo::relocType reloc, int offset, int format)
647 {
648 cbuf.relocate(cbuf.inst_mark() + offset, reloc, format);
649 emit_opcode(cbuf, code);
650 }
651
652 // EMIT_D8()
653 void emit_d8(CodeBuffer &cbuf, int d8)
654 {
655 *(cbuf.code_end()) = (unsigned char) d8;
656 cbuf.set_code_end(cbuf.code_end() + 1);
657 }
658
659 // EMIT_D16()
660 void emit_d16(CodeBuffer &cbuf, int d16)
661 {
662 *((short *)(cbuf.code_end())) = d16;
663 cbuf.set_code_end(cbuf.code_end() + 2);
664 }
665
666 // EMIT_D32()
667 void emit_d32(CodeBuffer &cbuf, int d32)
668 {
669 *((int *)(cbuf.code_end())) = d32;
670 cbuf.set_code_end(cbuf.code_end() + 4);
671 }
672
673 // EMIT_D64()
674 void emit_d64(CodeBuffer &cbuf, int64_t d64)
675 {
676 *((int64_t*) (cbuf.code_end())) = d64;
677 cbuf.set_code_end(cbuf.code_end() + 8);
678 }
679
680 // emit 32 bit value and construct relocation entry from relocInfo::relocType
681 void emit_d32_reloc(CodeBuffer& cbuf,
682 int d32,
683 relocInfo::relocType reloc,
684 int format)
685 {
686 assert(reloc != relocInfo::external_word_type, "use 2-arg emit_d32_reloc");
687 cbuf.relocate(cbuf.inst_mark(), reloc, format);
688
689 *((int*) (cbuf.code_end())) = d32;
690 cbuf.set_code_end(cbuf.code_end() + 4);
691 }
692
693 // emit 32 bit value and construct relocation entry from RelocationHolder
694 void emit_d32_reloc(CodeBuffer& cbuf,
695 int d32,
696 RelocationHolder const& rspec,
697 int format)
698 {
699 #ifdef ASSERT
700 if (rspec.reloc()->type() == relocInfo::oop_type &&
701 d32 != 0 && d32 != (intptr_t) Universe::non_oop_word()) {
702 assert(oop((intptr_t)d32)->is_oop() && (ScavengeRootsInCode || !oop((intptr_t)d32)->is_scavengable()), "cannot embed scavengable oops in code");
703 }
704 #endif
705 cbuf.relocate(cbuf.inst_mark(), rspec, format);
706
707 *((int* )(cbuf.code_end())) = d32;
708 cbuf.set_code_end(cbuf.code_end() + 4);
709 }
710
711 void emit_d32_reloc(CodeBuffer& cbuf, address addr) {
712 address next_ip = cbuf.code_end() + 4;
713 emit_d32_reloc(cbuf, (int) (addr - next_ip),
714 external_word_Relocation::spec(addr),
715 RELOC_DISP32);
716 }
717
718
719 // emit 64 bit value and construct relocation entry from relocInfo::relocType
720 void emit_d64_reloc(CodeBuffer& cbuf,
721 int64_t d64,
722 relocInfo::relocType reloc,
723 int format)
724 {
725 cbuf.relocate(cbuf.inst_mark(), reloc, format);
726
727 *((int64_t*) (cbuf.code_end())) = d64;
728 cbuf.set_code_end(cbuf.code_end() + 8);
729 }
730
731 // emit 64 bit value and construct relocation entry from RelocationHolder
732 void emit_d64_reloc(CodeBuffer& cbuf,
733 int64_t d64,
734 RelocationHolder const& rspec,
735 int format)
736 {
737 #ifdef ASSERT
738 if (rspec.reloc()->type() == relocInfo::oop_type &&
739 d64 != 0 && d64 != (int64_t) Universe::non_oop_word()) {
740 assert(oop(d64)->is_oop() && (ScavengeRootsInCode || !oop(d64)->is_scavengable()),
741 "cannot embed scavengable oops in code");
742 }
743 #endif
744 cbuf.relocate(cbuf.inst_mark(), rspec, format);
745
746 *((int64_t*) (cbuf.code_end())) = d64;
747 cbuf.set_code_end(cbuf.code_end() + 8);
748 }
749
750 // Access stack slot for load or store
751 void store_to_stackslot(CodeBuffer &cbuf, int opcode, int rm_field, int disp)
752 {
753 emit_opcode(cbuf, opcode); // (e.g., FILD [RSP+src])
754 if (-0x80 <= disp && disp < 0x80) {
755 emit_rm(cbuf, 0x01, rm_field, RSP_enc); // R/M byte
756 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
757 emit_d8(cbuf, disp); // Displacement // R/M byte
758 } else {
759 emit_rm(cbuf, 0x02, rm_field, RSP_enc); // R/M byte
760 emit_rm(cbuf, 0x00, RSP_enc, RSP_enc); // SIB byte
761 emit_d32(cbuf, disp); // Displacement // R/M byte
762 }
763 }
764
765 // rRegI ereg, memory mem) %{ // emit_reg_mem
766 void encode_RegMem(CodeBuffer &cbuf,
767 int reg,
768 int base, int index, int scale, int disp, bool disp_is_oop)
769 {
770 assert(!disp_is_oop, "cannot have disp");
771 int regenc = reg & 7;
772 int baseenc = base & 7;
773 int indexenc = index & 7;
774
775 // There is no index & no scale, use form without SIB byte
776 if (index == 0x4 && scale == 0 && base != RSP_enc && base != R12_enc) {
777 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
778 if (disp == 0 && base != RBP_enc && base != R13_enc) {
779 emit_rm(cbuf, 0x0, regenc, baseenc); // *
780 } else if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
781 // If 8-bit displacement, mode 0x1
782 emit_rm(cbuf, 0x1, regenc, baseenc); // *
783 emit_d8(cbuf, disp);
784 } else {
785 // If 32-bit displacement
786 if (base == -1) { // Special flag for absolute address
787 emit_rm(cbuf, 0x0, regenc, 0x5); // *
788 if (disp_is_oop) {
789 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
790 } else {
791 emit_d32(cbuf, disp);
792 }
793 } else {
794 // Normal base + offset
795 emit_rm(cbuf, 0x2, regenc, baseenc); // *
796 if (disp_is_oop) {
797 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
798 } else {
799 emit_d32(cbuf, disp);
800 }
801 }
802 }
803 } else {
804 // Else, encode with the SIB byte
805 // If no displacement, mode is 0x0; unless base is [RBP] or [R13]
806 if (disp == 0 && base != RBP_enc && base != R13_enc) {
807 // If no displacement
808 emit_rm(cbuf, 0x0, regenc, 0x4); // *
809 emit_rm(cbuf, scale, indexenc, baseenc);
810 } else {
811 if (-0x80 <= disp && disp < 0x80 && !disp_is_oop) {
812 // If 8-bit displacement, mode 0x1
813 emit_rm(cbuf, 0x1, regenc, 0x4); // *
814 emit_rm(cbuf, scale, indexenc, baseenc);
815 emit_d8(cbuf, disp);
816 } else {
817 // If 32-bit displacement
818 if (base == 0x04 ) {
819 emit_rm(cbuf, 0x2, regenc, 0x4);
820 emit_rm(cbuf, scale, indexenc, 0x04); // XXX is this valid???
821 } else {
822 emit_rm(cbuf, 0x2, regenc, 0x4);
823 emit_rm(cbuf, scale, indexenc, baseenc); // *
824 }
825 if (disp_is_oop) {
826 emit_d32_reloc(cbuf, disp, relocInfo::oop_type, RELOC_DISP32);
827 } else {
828 emit_d32(cbuf, disp);
829 }
830 }
831 }
832 }
833 }
834
835 void encode_copy(CodeBuffer &cbuf, int dstenc, int srcenc)
836 {
837 if (dstenc != srcenc) {
838 if (dstenc < 8) {
839 if (srcenc >= 8) {
840 emit_opcode(cbuf, Assembler::REX_B);
841 srcenc -= 8;
842 }
843 } else {
844 if (srcenc < 8) {
845 emit_opcode(cbuf, Assembler::REX_R);
846 } else {
847 emit_opcode(cbuf, Assembler::REX_RB);
848 srcenc -= 8;
849 }
850 dstenc -= 8;
851 }
852
853 emit_opcode(cbuf, 0x8B);
854 emit_rm(cbuf, 0x3, dstenc, srcenc);
855 }
856 }
857
858 void encode_CopyXD( CodeBuffer &cbuf, int dst_encoding, int src_encoding ) {
859 if( dst_encoding == src_encoding ) {
860 // reg-reg copy, use an empty encoding
861 } else {
862 MacroAssembler _masm(&cbuf);
863
864 __ movdqa(as_XMMRegister(dst_encoding), as_XMMRegister(src_encoding));
865 }
866 }
867
868
869 //=============================================================================
870 #ifndef PRODUCT
871 void MachPrologNode::format(PhaseRegAlloc* ra_, outputStream* st) const
872 {
873 Compile* C = ra_->C;
874
875 int framesize = C->frame_slots() << LogBytesPerInt;
876 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
877 // Remove wordSize for return adr already pushed
878 // and another for the RBP we are going to save
879 framesize -= 2*wordSize;
880 bool need_nop = true;
881
882 // Calls to C2R adapters often do not accept exceptional returns.
883 // We require that their callers must bang for them. But be
884 // careful, because some VM calls (such as call site linkage) can
885 // use several kilobytes of stack. But the stack safety zone should
886 // account for that. See bugs 4446381, 4468289, 4497237.
887 if (C->need_stack_bang(framesize)) {
888 st->print_cr("# stack bang"); st->print("\t");
889 need_nop = false;
890 }
891 st->print_cr("pushq rbp"); st->print("\t");
892
893 if (VerifyStackAtCalls) {
894 // Majik cookie to verify stack depth
895 st->print_cr("pushq 0xffffffffbadb100d"
896 "\t# Majik cookie for stack depth check");
897 st->print("\t");
898 framesize -= wordSize; // Remove 2 for cookie
899 need_nop = false;
900 }
901
902 if (framesize) {
903 st->print("subq rsp, #%d\t# Create frame", framesize);
904 if (framesize < 0x80 && need_nop) {
905 st->print("\n\tnop\t# nop for patch_verified_entry");
906 }
907 }
908 }
909 #endif
910
911 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
912 {
913 Compile* C = ra_->C;
914
915 // WARNING: Initial instruction MUST be 5 bytes or longer so that
916 // NativeJump::patch_verified_entry will be able to patch out the entry
917 // code safely. The fldcw is ok at 6 bytes, the push to verify stack
918 // depth is ok at 5 bytes, the frame allocation can be either 3 or
919 // 6 bytes. So if we don't do the fldcw or the push then we must
920 // use the 6 byte frame allocation even if we have no frame. :-(
921 // If method sets FPU control word do it now
922
923 int framesize = C->frame_slots() << LogBytesPerInt;
924 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
925 // Remove wordSize for return adr already pushed
926 // and another for the RBP we are going to save
927 framesize -= 2*wordSize;
928 bool need_nop = true;
929
930 // Calls to C2R adapters often do not accept exceptional returns.
931 // We require that their callers must bang for them. But be
932 // careful, because some VM calls (such as call site linkage) can
933 // use several kilobytes of stack. But the stack safety zone should
934 // account for that. See bugs 4446381, 4468289, 4497237.
935 if (C->need_stack_bang(framesize)) {
936 MacroAssembler masm(&cbuf);
937 masm.generate_stack_overflow_check(framesize);
938 need_nop = false;
939 }
940
941 // We always push rbp so that on return to interpreter rbp will be
942 // restored correctly and we can correct the stack.
943 emit_opcode(cbuf, 0x50 | RBP_enc);
944
945 if (VerifyStackAtCalls) {
946 // Majik cookie to verify stack depth
947 emit_opcode(cbuf, 0x68); // pushq (sign-extended) 0xbadb100d
948 emit_d32(cbuf, 0xbadb100d);
949 framesize -= wordSize; // Remove 2 for cookie
950 need_nop = false;
951 }
952
953 if (framesize) {
954 emit_opcode(cbuf, Assembler::REX_W);
955 if (framesize < 0x80) {
956 emit_opcode(cbuf, 0x83); // sub SP,#framesize
957 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
958 emit_d8(cbuf, framesize);
959 if (need_nop) {
960 emit_opcode(cbuf, 0x90); // nop
961 }
962 } else {
963 emit_opcode(cbuf, 0x81); // sub SP,#framesize
964 emit_rm(cbuf, 0x3, 0x05, RSP_enc);
965 emit_d32(cbuf, framesize);
966 }
967 }
968
969 C->set_frame_complete(cbuf.code_end() - cbuf.code_begin());
970
971 #ifdef ASSERT
972 if (VerifyStackAtCalls) {
973 Label L;
974 MacroAssembler masm(&cbuf);
975 masm.push(rax);
976 masm.mov(rax, rsp);
977 masm.andptr(rax, StackAlignmentInBytes-1);
978 masm.cmpptr(rax, StackAlignmentInBytes-wordSize);
979 masm.pop(rax);
980 masm.jcc(Assembler::equal, L);
981 masm.stop("Stack is not properly aligned!");
982 masm.bind(L);
983 }
984 #endif
985 }
986
987 uint MachPrologNode::size(PhaseRegAlloc* ra_) const
988 {
989 return MachNode::size(ra_); // too many variables; just compute it
990 // the hard way
991 }
992
993 int MachPrologNode::reloc() const
994 {
995 return 0; // a large enough number
996 }
997
998 //=============================================================================
999 #ifndef PRODUCT
1000 void MachEpilogNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1001 {
1002 Compile* C = ra_->C;
1003 int framesize = C->frame_slots() << LogBytesPerInt;
1004 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1005 // Remove word for return adr already pushed
1006 // and RBP
1007 framesize -= 2*wordSize;
1008
1009 if (framesize) {
1010 st->print_cr("addq\trsp, %d\t# Destroy frame", framesize);
1011 st->print("\t");
1012 }
1013
1014 st->print_cr("popq\trbp");
1015 if (do_polling() && C->is_method_compilation()) {
1016 st->print_cr("\ttestl\trax, [rip + #offset_to_poll_page]\t"
1017 "# Safepoint: poll for GC");
1018 st->print("\t");
1019 }
1020 }
1021 #endif
1022
1023 void MachEpilogNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1024 {
1025 Compile* C = ra_->C;
1026 int framesize = C->frame_slots() << LogBytesPerInt;
1027 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1028 // Remove word for return adr already pushed
1029 // and RBP
1030 framesize -= 2*wordSize;
1031
1032 // Note that VerifyStackAtCalls' Majik cookie does not change the frame size popped here
1033
1034 if (framesize) {
1035 emit_opcode(cbuf, Assembler::REX_W);
1036 if (framesize < 0x80) {
1037 emit_opcode(cbuf, 0x83); // addq rsp, #framesize
1038 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1039 emit_d8(cbuf, framesize);
1040 } else {
1041 emit_opcode(cbuf, 0x81); // addq rsp, #framesize
1042 emit_rm(cbuf, 0x3, 0x00, RSP_enc);
1043 emit_d32(cbuf, framesize);
1044 }
1045 }
1046
1047 // popq rbp
1048 emit_opcode(cbuf, 0x58 | RBP_enc);
1049
1050 if (do_polling() && C->is_method_compilation()) {
1051 // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
1052 // XXX reg_mem doesn't support RIP-relative addressing yet
1053 cbuf.set_inst_mark();
1054 cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_return_type, 0); // XXX
1055 emit_opcode(cbuf, 0x85); // testl
1056 emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
1057 // cbuf.inst_mark() is beginning of instruction
1058 emit_d32_reloc(cbuf, os::get_polling_page());
1059 // relocInfo::poll_return_type,
1060 }
1061 }
1062
1063 uint MachEpilogNode::size(PhaseRegAlloc* ra_) const
1064 {
1065 Compile* C = ra_->C;
1066 int framesize = C->frame_slots() << LogBytesPerInt;
1067 assert((framesize & (StackAlignmentInBytes-1)) == 0, "frame size not aligned");
1068 // Remove word for return adr already pushed
1069 // and RBP
1070 framesize -= 2*wordSize;
1071
1072 uint size = 0;
1073
1074 if (do_polling() && C->is_method_compilation()) {
1075 size += 6;
1076 }
1077
1078 // count popq rbp
1079 size++;
1080
1081 if (framesize) {
1082 if (framesize < 0x80) {
1083 size += 4;
1084 } else if (framesize) {
1085 size += 7;
1086 }
1087 }
1088
1089 return size;
1090 }
1091
1092 int MachEpilogNode::reloc() const
1093 {
1094 return 2; // a large enough number
1095 }
1096
1097 const Pipeline* MachEpilogNode::pipeline() const
1098 {
1099 return MachNode::pipeline_class();
1100 }
1101
1102 int MachEpilogNode::safepoint_offset() const
1103 {
1104 return 0;
1105 }
1106
1107 //=============================================================================
1108
1109 enum RC {
1110 rc_bad,
1111 rc_int,
1112 rc_float,
1113 rc_stack
1114 };
1115
1116 static enum RC rc_class(OptoReg::Name reg)
1117 {
1118 if( !OptoReg::is_valid(reg) ) return rc_bad;
1119
1120 if (OptoReg::is_stack(reg)) return rc_stack;
1121
1122 VMReg r = OptoReg::as_VMReg(reg);
1123
1124 if (r->is_Register()) return rc_int;
1125
1126 assert(r->is_XMMRegister(), "must be");
1127 return rc_float;
1128 }
1129
1130 uint MachSpillCopyNode::implementation(CodeBuffer* cbuf,
1131 PhaseRegAlloc* ra_,
1132 bool do_size,
1133 outputStream* st) const
1134 {
1135
1136 // Get registers to move
1137 OptoReg::Name src_second = ra_->get_reg_second(in(1));
1138 OptoReg::Name src_first = ra_->get_reg_first(in(1));
1139 OptoReg::Name dst_second = ra_->get_reg_second(this);
1140 OptoReg::Name dst_first = ra_->get_reg_first(this);
1141
1142 enum RC src_second_rc = rc_class(src_second);
1143 enum RC src_first_rc = rc_class(src_first);
1144 enum RC dst_second_rc = rc_class(dst_second);
1145 enum RC dst_first_rc = rc_class(dst_first);
1146
1147 assert(OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first),
1148 "must move at least 1 register" );
1149
1150 if (src_first == dst_first && src_second == dst_second) {
1151 // Self copy, no move
1152 return 0;
1153 } else if (src_first_rc == rc_stack) {
1154 // mem ->
1155 if (dst_first_rc == rc_stack) {
1156 // mem -> mem
1157 assert(src_second != dst_first, "overlap");
1158 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1159 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1160 // 64-bit
1161 int src_offset = ra_->reg2offset(src_first);
1162 int dst_offset = ra_->reg2offset(dst_first);
1163 if (cbuf) {
1164 emit_opcode(*cbuf, 0xFF);
1165 encode_RegMem(*cbuf, RSI_enc, RSP_enc, 0x4, 0, src_offset, false);
1166
1167 emit_opcode(*cbuf, 0x8F);
1168 encode_RegMem(*cbuf, RAX_enc, RSP_enc, 0x4, 0, dst_offset, false);
1169
1170 #ifndef PRODUCT
1171 } else if (!do_size) {
1172 st->print("pushq [rsp + #%d]\t# 64-bit mem-mem spill\n\t"
1173 "popq [rsp + #%d]",
1174 src_offset,
1175 dst_offset);
1176 #endif
1177 }
1178 return
1179 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) +
1180 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4));
1181 } else {
1182 // 32-bit
1183 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1184 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1185 // No pushl/popl, so:
1186 int src_offset = ra_->reg2offset(src_first);
1187 int dst_offset = ra_->reg2offset(dst_first);
1188 if (cbuf) {
1189 emit_opcode(*cbuf, Assembler::REX_W);
1190 emit_opcode(*cbuf, 0x89);
1191 emit_opcode(*cbuf, 0x44);
1192 emit_opcode(*cbuf, 0x24);
1193 emit_opcode(*cbuf, 0xF8);
1194
1195 emit_opcode(*cbuf, 0x8B);
1196 encode_RegMem(*cbuf,
1197 RAX_enc,
1198 RSP_enc, 0x4, 0, src_offset,
1199 false);
1200
1201 emit_opcode(*cbuf, 0x89);
1202 encode_RegMem(*cbuf,
1203 RAX_enc,
1204 RSP_enc, 0x4, 0, dst_offset,
1205 false);
1206
1207 emit_opcode(*cbuf, Assembler::REX_W);
1208 emit_opcode(*cbuf, 0x8B);
1209 emit_opcode(*cbuf, 0x44);
1210 emit_opcode(*cbuf, 0x24);
1211 emit_opcode(*cbuf, 0xF8);
1212
1213 #ifndef PRODUCT
1214 } else if (!do_size) {
1215 st->print("movq [rsp - #8], rax\t# 32-bit mem-mem spill\n\t"
1216 "movl rax, [rsp + #%d]\n\t"
1217 "movl [rsp + #%d], rax\n\t"
1218 "movq rax, [rsp - #8]",
1219 src_offset,
1220 dst_offset);
1221 #endif
1222 }
1223 return
1224 5 + // movq
1225 3 + ((src_offset == 0) ? 0 : (src_offset < 0x80 ? 1 : 4)) + // movl
1226 3 + ((dst_offset == 0) ? 0 : (dst_offset < 0x80 ? 1 : 4)) + // movl
1227 5; // movq
1228 }
1229 } else if (dst_first_rc == rc_int) {
1230 // mem -> gpr
1231 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1232 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1233 // 64-bit
1234 int offset = ra_->reg2offset(src_first);
1235 if (cbuf) {
1236 if (Matcher::_regEncode[dst_first] < 8) {
1237 emit_opcode(*cbuf, Assembler::REX_W);
1238 } else {
1239 emit_opcode(*cbuf, Assembler::REX_WR);
1240 }
1241 emit_opcode(*cbuf, 0x8B);
1242 encode_RegMem(*cbuf,
1243 Matcher::_regEncode[dst_first],
1244 RSP_enc, 0x4, 0, offset,
1245 false);
1246 #ifndef PRODUCT
1247 } else if (!do_size) {
1248 st->print("movq %s, [rsp + #%d]\t# spill",
1249 Matcher::regName[dst_first],
1250 offset);
1251 #endif
1252 }
1253 return
1254 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1255 } else {
1256 // 32-bit
1257 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1258 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1259 int offset = ra_->reg2offset(src_first);
1260 if (cbuf) {
1261 if (Matcher::_regEncode[dst_first] >= 8) {
1262 emit_opcode(*cbuf, Assembler::REX_R);
1263 }
1264 emit_opcode(*cbuf, 0x8B);
1265 encode_RegMem(*cbuf,
1266 Matcher::_regEncode[dst_first],
1267 RSP_enc, 0x4, 0, offset,
1268 false);
1269 #ifndef PRODUCT
1270 } else if (!do_size) {
1271 st->print("movl %s, [rsp + #%d]\t# spill",
1272 Matcher::regName[dst_first],
1273 offset);
1274 #endif
1275 }
1276 return
1277 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1278 ((Matcher::_regEncode[dst_first] < 8)
1279 ? 3
1280 : 4); // REX
1281 }
1282 } else if (dst_first_rc == rc_float) {
1283 // mem-> xmm
1284 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1285 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1286 // 64-bit
1287 int offset = ra_->reg2offset(src_first);
1288 if (cbuf) {
1289 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
1290 if (Matcher::_regEncode[dst_first] >= 8) {
1291 emit_opcode(*cbuf, Assembler::REX_R);
1292 }
1293 emit_opcode(*cbuf, 0x0F);
1294 emit_opcode(*cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
1295 encode_RegMem(*cbuf,
1296 Matcher::_regEncode[dst_first],
1297 RSP_enc, 0x4, 0, offset,
1298 false);
1299 #ifndef PRODUCT
1300 } else if (!do_size) {
1301 st->print("%s %s, [rsp + #%d]\t# spill",
1302 UseXmmLoadAndClearUpper ? "movsd " : "movlpd",
1303 Matcher::regName[dst_first],
1304 offset);
1305 #endif
1306 }
1307 return
1308 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1309 ((Matcher::_regEncode[dst_first] < 8)
1310 ? 5
1311 : 6); // REX
1312 } else {
1313 // 32-bit
1314 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1315 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1316 int offset = ra_->reg2offset(src_first);
1317 if (cbuf) {
1318 emit_opcode(*cbuf, 0xF3);
1319 if (Matcher::_regEncode[dst_first] >= 8) {
1320 emit_opcode(*cbuf, Assembler::REX_R);
1321 }
1322 emit_opcode(*cbuf, 0x0F);
1323 emit_opcode(*cbuf, 0x10);
1324 encode_RegMem(*cbuf,
1325 Matcher::_regEncode[dst_first],
1326 RSP_enc, 0x4, 0, offset,
1327 false);
1328 #ifndef PRODUCT
1329 } else if (!do_size) {
1330 st->print("movss %s, [rsp + #%d]\t# spill",
1331 Matcher::regName[dst_first],
1332 offset);
1333 #endif
1334 }
1335 return
1336 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1337 ((Matcher::_regEncode[dst_first] < 8)
1338 ? 5
1339 : 6); // REX
1340 }
1341 }
1342 } else if (src_first_rc == rc_int) {
1343 // gpr ->
1344 if (dst_first_rc == rc_stack) {
1345 // gpr -> mem
1346 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1347 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1348 // 64-bit
1349 int offset = ra_->reg2offset(dst_first);
1350 if (cbuf) {
1351 if (Matcher::_regEncode[src_first] < 8) {
1352 emit_opcode(*cbuf, Assembler::REX_W);
1353 } else {
1354 emit_opcode(*cbuf, Assembler::REX_WR);
1355 }
1356 emit_opcode(*cbuf, 0x89);
1357 encode_RegMem(*cbuf,
1358 Matcher::_regEncode[src_first],
1359 RSP_enc, 0x4, 0, offset,
1360 false);
1361 #ifndef PRODUCT
1362 } else if (!do_size) {
1363 st->print("movq [rsp + #%d], %s\t# spill",
1364 offset,
1365 Matcher::regName[src_first]);
1366 #endif
1367 }
1368 return ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) + 4; // REX
1369 } else {
1370 // 32-bit
1371 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1372 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1373 int offset = ra_->reg2offset(dst_first);
1374 if (cbuf) {
1375 if (Matcher::_regEncode[src_first] >= 8) {
1376 emit_opcode(*cbuf, Assembler::REX_R);
1377 }
1378 emit_opcode(*cbuf, 0x89);
1379 encode_RegMem(*cbuf,
1380 Matcher::_regEncode[src_first],
1381 RSP_enc, 0x4, 0, offset,
1382 false);
1383 #ifndef PRODUCT
1384 } else if (!do_size) {
1385 st->print("movl [rsp + #%d], %s\t# spill",
1386 offset,
1387 Matcher::regName[src_first]);
1388 #endif
1389 }
1390 return
1391 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1392 ((Matcher::_regEncode[src_first] < 8)
1393 ? 3
1394 : 4); // REX
1395 }
1396 } else if (dst_first_rc == rc_int) {
1397 // gpr -> gpr
1398 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1399 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1400 // 64-bit
1401 if (cbuf) {
1402 if (Matcher::_regEncode[dst_first] < 8) {
1403 if (Matcher::_regEncode[src_first] < 8) {
1404 emit_opcode(*cbuf, Assembler::REX_W);
1405 } else {
1406 emit_opcode(*cbuf, Assembler::REX_WB);
1407 }
1408 } else {
1409 if (Matcher::_regEncode[src_first] < 8) {
1410 emit_opcode(*cbuf, Assembler::REX_WR);
1411 } else {
1412 emit_opcode(*cbuf, Assembler::REX_WRB);
1413 }
1414 }
1415 emit_opcode(*cbuf, 0x8B);
1416 emit_rm(*cbuf, 0x3,
1417 Matcher::_regEncode[dst_first] & 7,
1418 Matcher::_regEncode[src_first] & 7);
1419 #ifndef PRODUCT
1420 } else if (!do_size) {
1421 st->print("movq %s, %s\t# spill",
1422 Matcher::regName[dst_first],
1423 Matcher::regName[src_first]);
1424 #endif
1425 }
1426 return 3; // REX
1427 } else {
1428 // 32-bit
1429 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1430 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1431 if (cbuf) {
1432 if (Matcher::_regEncode[dst_first] < 8) {
1433 if (Matcher::_regEncode[src_first] >= 8) {
1434 emit_opcode(*cbuf, Assembler::REX_B);
1435 }
1436 } else {
1437 if (Matcher::_regEncode[src_first] < 8) {
1438 emit_opcode(*cbuf, Assembler::REX_R);
1439 } else {
1440 emit_opcode(*cbuf, Assembler::REX_RB);
1441 }
1442 }
1443 emit_opcode(*cbuf, 0x8B);
1444 emit_rm(*cbuf, 0x3,
1445 Matcher::_regEncode[dst_first] & 7,
1446 Matcher::_regEncode[src_first] & 7);
1447 #ifndef PRODUCT
1448 } else if (!do_size) {
1449 st->print("movl %s, %s\t# spill",
1450 Matcher::regName[dst_first],
1451 Matcher::regName[src_first]);
1452 #endif
1453 }
1454 return
1455 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1456 ? 2
1457 : 3; // REX
1458 }
1459 } else if (dst_first_rc == rc_float) {
1460 // gpr -> xmm
1461 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1462 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1463 // 64-bit
1464 if (cbuf) {
1465 emit_opcode(*cbuf, 0x66);
1466 if (Matcher::_regEncode[dst_first] < 8) {
1467 if (Matcher::_regEncode[src_first] < 8) {
1468 emit_opcode(*cbuf, Assembler::REX_W);
1469 } else {
1470 emit_opcode(*cbuf, Assembler::REX_WB);
1471 }
1472 } else {
1473 if (Matcher::_regEncode[src_first] < 8) {
1474 emit_opcode(*cbuf, Assembler::REX_WR);
1475 } else {
1476 emit_opcode(*cbuf, Assembler::REX_WRB);
1477 }
1478 }
1479 emit_opcode(*cbuf, 0x0F);
1480 emit_opcode(*cbuf, 0x6E);
1481 emit_rm(*cbuf, 0x3,
1482 Matcher::_regEncode[dst_first] & 7,
1483 Matcher::_regEncode[src_first] & 7);
1484 #ifndef PRODUCT
1485 } else if (!do_size) {
1486 st->print("movdq %s, %s\t# spill",
1487 Matcher::regName[dst_first],
1488 Matcher::regName[src_first]);
1489 #endif
1490 }
1491 return 5; // REX
1492 } else {
1493 // 32-bit
1494 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1495 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1496 if (cbuf) {
1497 emit_opcode(*cbuf, 0x66);
1498 if (Matcher::_regEncode[dst_first] < 8) {
1499 if (Matcher::_regEncode[src_first] >= 8) {
1500 emit_opcode(*cbuf, Assembler::REX_B);
1501 }
1502 } else {
1503 if (Matcher::_regEncode[src_first] < 8) {
1504 emit_opcode(*cbuf, Assembler::REX_R);
1505 } else {
1506 emit_opcode(*cbuf, Assembler::REX_RB);
1507 }
1508 }
1509 emit_opcode(*cbuf, 0x0F);
1510 emit_opcode(*cbuf, 0x6E);
1511 emit_rm(*cbuf, 0x3,
1512 Matcher::_regEncode[dst_first] & 7,
1513 Matcher::_regEncode[src_first] & 7);
1514 #ifndef PRODUCT
1515 } else if (!do_size) {
1516 st->print("movdl %s, %s\t# spill",
1517 Matcher::regName[dst_first],
1518 Matcher::regName[src_first]);
1519 #endif
1520 }
1521 return
1522 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1523 ? 4
1524 : 5; // REX
1525 }
1526 }
1527 } else if (src_first_rc == rc_float) {
1528 // xmm ->
1529 if (dst_first_rc == rc_stack) {
1530 // xmm -> mem
1531 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1532 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1533 // 64-bit
1534 int offset = ra_->reg2offset(dst_first);
1535 if (cbuf) {
1536 emit_opcode(*cbuf, 0xF2);
1537 if (Matcher::_regEncode[src_first] >= 8) {
1538 emit_opcode(*cbuf, Assembler::REX_R);
1539 }
1540 emit_opcode(*cbuf, 0x0F);
1541 emit_opcode(*cbuf, 0x11);
1542 encode_RegMem(*cbuf,
1543 Matcher::_regEncode[src_first],
1544 RSP_enc, 0x4, 0, offset,
1545 false);
1546 #ifndef PRODUCT
1547 } else if (!do_size) {
1548 st->print("movsd [rsp + #%d], %s\t# spill",
1549 offset,
1550 Matcher::regName[src_first]);
1551 #endif
1552 }
1553 return
1554 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1555 ((Matcher::_regEncode[src_first] < 8)
1556 ? 5
1557 : 6); // REX
1558 } else {
1559 // 32-bit
1560 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1561 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1562 int offset = ra_->reg2offset(dst_first);
1563 if (cbuf) {
1564 emit_opcode(*cbuf, 0xF3);
1565 if (Matcher::_regEncode[src_first] >= 8) {
1566 emit_opcode(*cbuf, Assembler::REX_R);
1567 }
1568 emit_opcode(*cbuf, 0x0F);
1569 emit_opcode(*cbuf, 0x11);
1570 encode_RegMem(*cbuf,
1571 Matcher::_regEncode[src_first],
1572 RSP_enc, 0x4, 0, offset,
1573 false);
1574 #ifndef PRODUCT
1575 } else if (!do_size) {
1576 st->print("movss [rsp + #%d], %s\t# spill",
1577 offset,
1578 Matcher::regName[src_first]);
1579 #endif
1580 }
1581 return
1582 ((offset == 0) ? 0 : (offset < 0x80 ? 1 : 4)) +
1583 ((Matcher::_regEncode[src_first] < 8)
1584 ? 5
1585 : 6); // REX
1586 }
1587 } else if (dst_first_rc == rc_int) {
1588 // xmm -> gpr
1589 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1590 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1591 // 64-bit
1592 if (cbuf) {
1593 emit_opcode(*cbuf, 0x66);
1594 if (Matcher::_regEncode[dst_first] < 8) {
1595 if (Matcher::_regEncode[src_first] < 8) {
1596 emit_opcode(*cbuf, Assembler::REX_W);
1597 } else {
1598 emit_opcode(*cbuf, Assembler::REX_WR); // attention!
1599 }
1600 } else {
1601 if (Matcher::_regEncode[src_first] < 8) {
1602 emit_opcode(*cbuf, Assembler::REX_WB); // attention!
1603 } else {
1604 emit_opcode(*cbuf, Assembler::REX_WRB);
1605 }
1606 }
1607 emit_opcode(*cbuf, 0x0F);
1608 emit_opcode(*cbuf, 0x7E);
1609 emit_rm(*cbuf, 0x3,
1610 Matcher::_regEncode[dst_first] & 7,
1611 Matcher::_regEncode[src_first] & 7);
1612 #ifndef PRODUCT
1613 } else if (!do_size) {
1614 st->print("movdq %s, %s\t# spill",
1615 Matcher::regName[dst_first],
1616 Matcher::regName[src_first]);
1617 #endif
1618 }
1619 return 5; // REX
1620 } else {
1621 // 32-bit
1622 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1623 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1624 if (cbuf) {
1625 emit_opcode(*cbuf, 0x66);
1626 if (Matcher::_regEncode[dst_first] < 8) {
1627 if (Matcher::_regEncode[src_first] >= 8) {
1628 emit_opcode(*cbuf, Assembler::REX_R); // attention!
1629 }
1630 } else {
1631 if (Matcher::_regEncode[src_first] < 8) {
1632 emit_opcode(*cbuf, Assembler::REX_B); // attention!
1633 } else {
1634 emit_opcode(*cbuf, Assembler::REX_RB);
1635 }
1636 }
1637 emit_opcode(*cbuf, 0x0F);
1638 emit_opcode(*cbuf, 0x7E);
1639 emit_rm(*cbuf, 0x3,
1640 Matcher::_regEncode[dst_first] & 7,
1641 Matcher::_regEncode[src_first] & 7);
1642 #ifndef PRODUCT
1643 } else if (!do_size) {
1644 st->print("movdl %s, %s\t# spill",
1645 Matcher::regName[dst_first],
1646 Matcher::regName[src_first]);
1647 #endif
1648 }
1649 return
1650 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1651 ? 4
1652 : 5; // REX
1653 }
1654 } else if (dst_first_rc == rc_float) {
1655 // xmm -> xmm
1656 if ((src_first & 1) == 0 && src_first + 1 == src_second &&
1657 (dst_first & 1) == 0 && dst_first + 1 == dst_second) {
1658 // 64-bit
1659 if (cbuf) {
1660 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
1661 if (Matcher::_regEncode[dst_first] < 8) {
1662 if (Matcher::_regEncode[src_first] >= 8) {
1663 emit_opcode(*cbuf, Assembler::REX_B);
1664 }
1665 } else {
1666 if (Matcher::_regEncode[src_first] < 8) {
1667 emit_opcode(*cbuf, Assembler::REX_R);
1668 } else {
1669 emit_opcode(*cbuf, Assembler::REX_RB);
1670 }
1671 }
1672 emit_opcode(*cbuf, 0x0F);
1673 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1674 emit_rm(*cbuf, 0x3,
1675 Matcher::_regEncode[dst_first] & 7,
1676 Matcher::_regEncode[src_first] & 7);
1677 #ifndef PRODUCT
1678 } else if (!do_size) {
1679 st->print("%s %s, %s\t# spill",
1680 UseXmmRegToRegMoveAll ? "movapd" : "movsd ",
1681 Matcher::regName[dst_first],
1682 Matcher::regName[src_first]);
1683 #endif
1684 }
1685 return
1686 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1687 ? 4
1688 : 5; // REX
1689 } else {
1690 // 32-bit
1691 assert(!((src_first & 1) == 0 && src_first + 1 == src_second), "no transform");
1692 assert(!((dst_first & 1) == 0 && dst_first + 1 == dst_second), "no transform");
1693 if (cbuf) {
1694 if (!UseXmmRegToRegMoveAll)
1695 emit_opcode(*cbuf, 0xF3);
1696 if (Matcher::_regEncode[dst_first] < 8) {
1697 if (Matcher::_regEncode[src_first] >= 8) {
1698 emit_opcode(*cbuf, Assembler::REX_B);
1699 }
1700 } else {
1701 if (Matcher::_regEncode[src_first] < 8) {
1702 emit_opcode(*cbuf, Assembler::REX_R);
1703 } else {
1704 emit_opcode(*cbuf, Assembler::REX_RB);
1705 }
1706 }
1707 emit_opcode(*cbuf, 0x0F);
1708 emit_opcode(*cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
1709 emit_rm(*cbuf, 0x3,
1710 Matcher::_regEncode[dst_first] & 7,
1711 Matcher::_regEncode[src_first] & 7);
1712 #ifndef PRODUCT
1713 } else if (!do_size) {
1714 st->print("%s %s, %s\t# spill",
1715 UseXmmRegToRegMoveAll ? "movaps" : "movss ",
1716 Matcher::regName[dst_first],
1717 Matcher::regName[src_first]);
1718 #endif
1719 }
1720 return
1721 (Matcher::_regEncode[src_first] < 8 && Matcher::_regEncode[dst_first] < 8)
1722 ? (UseXmmRegToRegMoveAll ? 3 : 4)
1723 : (UseXmmRegToRegMoveAll ? 4 : 5); // REX
1724 }
1725 }
1726 }
1727
1728 assert(0," foo ");
1729 Unimplemented();
1730
1731 return 0;
1732 }
1733
1734 #ifndef PRODUCT
1735 void MachSpillCopyNode::format(PhaseRegAlloc *ra_, outputStream* st) const
1736 {
1737 implementation(NULL, ra_, false, st);
1738 }
1739 #endif
1740
1741 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const
1742 {
1743 implementation(&cbuf, ra_, false, NULL);
1744 }
1745
1746 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const
1747 {
1748 return implementation(NULL, ra_, true, NULL);
1749 }
1750
1751 //=============================================================================
1752 #ifndef PRODUCT
1753 void MachNopNode::format(PhaseRegAlloc*, outputStream* st) const
1754 {
1755 st->print("nop \t# %d bytes pad for loops and calls", _count);
1756 }
1757 #endif
1758
1759 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc*) const
1760 {
1761 MacroAssembler _masm(&cbuf);
1762 __ nop(_count);
1763 }
1764
1765 uint MachNopNode::size(PhaseRegAlloc*) const
1766 {
1767 return _count;
1768 }
1769
1770
1771 //=============================================================================
1772 #ifndef PRODUCT
1773 void BoxLockNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1774 {
1775 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1776 int reg = ra_->get_reg_first(this);
1777 st->print("leaq %s, [rsp + #%d]\t# box lock",
1778 Matcher::regName[reg], offset);
1779 }
1780 #endif
1781
1782 void BoxLockNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1783 {
1784 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1785 int reg = ra_->get_encode(this);
1786 if (offset >= 0x80) {
1787 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1788 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset]
1789 emit_rm(cbuf, 0x2, reg & 7, 0x04);
1790 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1791 emit_d32(cbuf, offset);
1792 } else {
1793 emit_opcode(cbuf, reg < 8 ? Assembler::REX_W : Assembler::REX_WR);
1794 emit_opcode(cbuf, 0x8D); // LEA reg,[SP+offset]
1795 emit_rm(cbuf, 0x1, reg & 7, 0x04);
1796 emit_rm(cbuf, 0x0, 0x04, RSP_enc);
1797 emit_d8(cbuf, offset);
1798 }
1799 }
1800
1801 uint BoxLockNode::size(PhaseRegAlloc *ra_) const
1802 {
1803 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1804 return (offset < 0x80) ? 5 : 8; // REX
1805 }
1806
1807 //=============================================================================
1808
1809 // emit call stub, compiled java to interpreter
1810 void emit_java_to_interp(CodeBuffer& cbuf)
1811 {
1812 // Stub is fixed up when the corresponding call is converted from
1813 // calling compiled code to calling interpreted code.
1814 // movq rbx, 0
1815 // jmp -5 # to self
1816
1817 address mark = cbuf.inst_mark(); // get mark within main instrs section
1818
1819 // Note that the code buffer's inst_mark is always relative to insts.
1820 // That's why we must use the macroassembler to generate a stub.
1821 MacroAssembler _masm(&cbuf);
1822
1823 address base =
1824 __ start_a_stub(Compile::MAX_stubs_size);
1825 if (base == NULL) return; // CodeBuffer::expand failed
1826 // static stub relocation stores the instruction address of the call
1827 __ relocate(static_stub_Relocation::spec(mark), RELOC_IMM64);
1828 // static stub relocation also tags the methodOop in the code-stream.
1829 __ movoop(rbx, (jobject) NULL); // method is zapped till fixup time
1830 // This is recognized as unresolved by relocs/nativeinst/ic code
1831 __ jump(RuntimeAddress(__ pc()));
1832
1833 // Update current stubs pointer and restore code_end.
1834 __ end_a_stub();
1835 }
1836
1837 // size of call stub, compiled java to interpretor
1838 uint size_java_to_interp()
1839 {
1840 return 15; // movq (1+1+8); jmp (1+4)
1841 }
1842
1843 // relocation entries for call stub, compiled java to interpretor
1844 uint reloc_java_to_interp()
1845 {
1846 return 4; // 3 in emit_java_to_interp + 1 in Java_Static_Call
1847 }
1848
1849 //=============================================================================
1850 #ifndef PRODUCT
1851 void MachUEPNode::format(PhaseRegAlloc* ra_, outputStream* st) const
1852 {
1853 if (UseCompressedOops) {
1854 st->print_cr("movl rscratch1, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t", oopDesc::klass_offset_in_bytes());
1855 if (Universe::narrow_oop_shift() != 0) {
1856 st->print_cr("leaq rscratch1, [r12_heapbase, r, Address::times_8, 0]");
1857 }
1858 st->print_cr("cmpq rax, rscratch1\t # Inline cache check");
1859 } else {
1860 st->print_cr("cmpq rax, [j_rarg0 + oopDesc::klass_offset_in_bytes() #%d]\t"
1861 "# Inline cache check", oopDesc::klass_offset_in_bytes());
1862 }
1863 st->print_cr("\tjne SharedRuntime::_ic_miss_stub");
1864 st->print_cr("\tnop");
1865 if (!OptoBreakpoint) {
1866 st->print_cr("\tnop");
1867 }
1868 }
1869 #endif
1870
1871 void MachUEPNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const
1872 {
1873 MacroAssembler masm(&cbuf);
1874 #ifdef ASSERT
1875 uint code_size = cbuf.code_size();
1876 #endif
1877 if (UseCompressedOops) {
1878 masm.load_klass(rscratch1, j_rarg0);
1879 masm.cmpptr(rax, rscratch1);
1880 } else {
1881 masm.cmpptr(rax, Address(j_rarg0, oopDesc::klass_offset_in_bytes()));
1882 }
1883
1884 masm.jump_cc(Assembler::notEqual, RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
1885
1886 /* WARNING these NOPs are critical so that verified entry point is properly
1887 aligned for patching by NativeJump::patch_verified_entry() */
1888 int nops_cnt = 1;
1889 if (!OptoBreakpoint) {
1890 // Leave space for int3
1891 nops_cnt += 1;
1892 }
1893 if (UseCompressedOops) {
1894 // ??? divisible by 4 is aligned?
1895 nops_cnt += 1;
1896 }
1897 masm.nop(nops_cnt);
1898
1899 assert(cbuf.code_size() - code_size == size(ra_),
1900 "checking code size of inline cache node");
1901 }
1902
1903 uint MachUEPNode::size(PhaseRegAlloc* ra_) const
1904 {
1905 if (UseCompressedOops) {
1906 if (Universe::narrow_oop_shift() == 0) {
1907 return OptoBreakpoint ? 15 : 16;
1908 } else {
1909 return OptoBreakpoint ? 19 : 20;
1910 }
1911 } else {
1912 return OptoBreakpoint ? 11 : 12;
1913 }
1914 }
1915
1916
1917 //=============================================================================
1918 uint size_exception_handler()
1919 {
1920 // NativeCall instruction size is the same as NativeJump.
1921 // Note that this value is also credited (in output.cpp) to
1922 // the size of the code section.
1923 return NativeJump::instruction_size;
1924 }
1925
1926 // Emit exception handler code.
1927 int emit_exception_handler(CodeBuffer& cbuf)
1928 {
1929
1930 // Note that the code buffer's inst_mark is always relative to insts.
1931 // That's why we must use the macroassembler to generate a handler.
1932 MacroAssembler _masm(&cbuf);
1933 address base =
1934 __ start_a_stub(size_exception_handler());
1935 if (base == NULL) return 0; // CodeBuffer::expand failed
1936 int offset = __ offset();
1937 __ jump(RuntimeAddress(OptoRuntime::exception_blob()->instructions_begin()));
1938 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1939 __ end_a_stub();
1940 return offset;
1941 }
1942
1943 uint size_deopt_handler()
1944 {
1945 // three 5 byte instructions
1946 return 15;
1947 }
1948
1949 // Emit deopt handler code.
1950 int emit_deopt_handler(CodeBuffer& cbuf)
1951 {
1952
1953 // Note that the code buffer's inst_mark is always relative to insts.
1954 // That's why we must use the macroassembler to generate a handler.
1955 MacroAssembler _masm(&cbuf);
1956 address base =
1957 __ start_a_stub(size_deopt_handler());
1958 if (base == NULL) return 0; // CodeBuffer::expand failed
1959 int offset = __ offset();
1960 address the_pc = (address) __ pc();
1961 Label next;
1962 // push a "the_pc" on the stack without destroying any registers
1963 // as they all may be live.
1964
1965 // push address of "next"
1966 __ call(next, relocInfo::none); // reloc none is fine since it is a disp32
1967 __ bind(next);
1968 // adjust it so it matches "the_pc"
1969 __ subptr(Address(rsp, 0), __ offset() - offset);
1970 __ jump(RuntimeAddress(SharedRuntime::deopt_blob()->unpack()));
1971 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1972 __ end_a_stub();
1973 return offset;
1974 }
1975
1976 static void emit_double_constant(CodeBuffer& cbuf, double x) {
1977 int mark = cbuf.insts()->mark_off();
1978 MacroAssembler _masm(&cbuf);
1979 address double_address = __ double_constant(x);
1980 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift
1981 emit_d32_reloc(cbuf,
1982 (int) (double_address - cbuf.code_end() - 4),
1983 internal_word_Relocation::spec(double_address),
1984 RELOC_DISP32);
1985 }
1986
1987 static void emit_float_constant(CodeBuffer& cbuf, float x) {
1988 int mark = cbuf.insts()->mark_off();
1989 MacroAssembler _masm(&cbuf);
1990 address float_address = __ float_constant(x);
1991 cbuf.insts()->set_mark_off(mark); // preserve mark across masm shift
1992 emit_d32_reloc(cbuf,
1993 (int) (float_address - cbuf.code_end() - 4),
1994 internal_word_Relocation::spec(float_address),
1995 RELOC_DISP32);
1996 }
1997
1998
1999 const bool Matcher::match_rule_supported(int opcode) {
2000 if (!has_match_rule(opcode))
2001 return false;
2002
2003 return true; // Per default match rules are supported.
2004 }
2005
2006 int Matcher::regnum_to_fpu_offset(int regnum)
2007 {
2008 return regnum - 32; // The FP registers are in the second chunk
2009 }
2010
2011 // This is UltraSparc specific, true just means we have fast l2f conversion
2012 const bool Matcher::convL2FSupported(void) {
2013 return true;
2014 }
2015
2016 // Vector width in bytes
2017 const uint Matcher::vector_width_in_bytes(void) {
2018 return 8;
2019 }
2020
2021 // Vector ideal reg
2022 const uint Matcher::vector_ideal_reg(void) {
2023 return Op_RegD;
2024 }
2025
2026 // Is this branch offset short enough that a short branch can be used?
2027 //
2028 // NOTE: If the platform does not provide any short branch variants, then
2029 // this method should return false for offset 0.
2030 bool Matcher::is_short_branch_offset(int rule, int offset) {
2031 // the short version of jmpConUCF2 contains multiple branches,
2032 // making the reach slightly less
2033 if (rule == jmpConUCF2_rule)
2034 return (-126 <= offset && offset <= 125);
2035 return (-128 <= offset && offset <= 127);
2036 }
2037
2038 const bool Matcher::isSimpleConstant64(jlong value) {
2039 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
2040 //return value == (int) value; // Cf. storeImmL and immL32.
2041
2042 // Probably always true, even if a temp register is required.
2043 return true;
2044 }
2045
2046 // The ecx parameter to rep stosq for the ClearArray node is in words.
2047 const bool Matcher::init_array_count_is_in_bytes = false;
2048
2049 // Threshold size for cleararray.
2050 const int Matcher::init_array_short_size = 8 * BytesPerLong;
2051
2052 // Should the Matcher clone shifts on addressing modes, expecting them
2053 // to be subsumed into complex addressing expressions or compute them
2054 // into registers? True for Intel but false for most RISCs
2055 const bool Matcher::clone_shift_expressions = true;
2056
2057 // Is it better to copy float constants, or load them directly from
2058 // memory? Intel can load a float constant from a direct address,
2059 // requiring no extra registers. Most RISCs will have to materialize
2060 // an address into a register first, so they would do better to copy
2061 // the constant from stack.
2062 const bool Matcher::rematerialize_float_constants = true; // XXX
2063
2064 // If CPU can load and store mis-aligned doubles directly then no
2065 // fixup is needed. Else we split the double into 2 integer pieces
2066 // and move it piece-by-piece. Only happens when passing doubles into
2067 // C code as the Java calling convention forces doubles to be aligned.
2068 const bool Matcher::misaligned_doubles_ok = true;
2069
2070 // No-op on amd64
2071 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {}
2072
2073 // Advertise here if the CPU requires explicit rounding operations to
2074 // implement the UseStrictFP mode.
2075 const bool Matcher::strict_fp_requires_explicit_rounding = true;
2076
2077 // Are floats conerted to double when stored to stack during deoptimization?
2078 // On x64 it is stored without convertion so we can use normal access.
2079 bool Matcher::float_in_double() { return false; }
2080
2081 // Do ints take an entire long register or just half?
2082 const bool Matcher::int_in_long = true;
2083
2084 // Return whether or not this register is ever used as an argument.
2085 // This function is used on startup to build the trampoline stubs in
2086 // generateOptoStub. Registers not mentioned will be killed by the VM
2087 // call in the trampoline, and arguments in those registers not be
2088 // available to the callee.
2089 bool Matcher::can_be_java_arg(int reg)
2090 {
2091 return
2092 reg == RDI_num || reg == RDI_H_num ||
2093 reg == RSI_num || reg == RSI_H_num ||
2094 reg == RDX_num || reg == RDX_H_num ||
2095 reg == RCX_num || reg == RCX_H_num ||
2096 reg == R8_num || reg == R8_H_num ||
2097 reg == R9_num || reg == R9_H_num ||
2098 reg == R12_num || reg == R12_H_num ||
2099 reg == XMM0_num || reg == XMM0_H_num ||
2100 reg == XMM1_num || reg == XMM1_H_num ||
2101 reg == XMM2_num || reg == XMM2_H_num ||
2102 reg == XMM3_num || reg == XMM3_H_num ||
2103 reg == XMM4_num || reg == XMM4_H_num ||
2104 reg == XMM5_num || reg == XMM5_H_num ||
2105 reg == XMM6_num || reg == XMM6_H_num ||
2106 reg == XMM7_num || reg == XMM7_H_num;
2107 }
2108
2109 bool Matcher::is_spillable_arg(int reg)
2110 {
2111 return can_be_java_arg(reg);
2112 }
2113
2114 // Register for DIVI projection of divmodI
2115 RegMask Matcher::divI_proj_mask() {
2116 return INT_RAX_REG_mask;
2117 }
2118
2119 // Register for MODI projection of divmodI
2120 RegMask Matcher::modI_proj_mask() {
2121 return INT_RDX_REG_mask;
2122 }
2123
2124 // Register for DIVL projection of divmodL
2125 RegMask Matcher::divL_proj_mask() {
2126 return LONG_RAX_REG_mask;
2127 }
2128
2129 // Register for MODL projection of divmodL
2130 RegMask Matcher::modL_proj_mask() {
2131 return LONG_RDX_REG_mask;
2132 }
2133
2134 const RegMask Matcher::method_handle_invoke_SP_save_mask() {
2135 return PTR_RBP_REG_mask;
2136 }
2137
2138 static Address build_address(int b, int i, int s, int d) {
2139 Register index = as_Register(i);
2140 Address::ScaleFactor scale = (Address::ScaleFactor)s;
2141 if (index == rsp) {
2142 index = noreg;
2143 scale = Address::no_scale;
2144 }
2145 Address addr(as_Register(b), index, scale, d);
2146 return addr;
2147 }
2148
2149 %}
2150
2151 //----------ENCODING BLOCK-----------------------------------------------------
2152 // This block specifies the encoding classes used by the compiler to
2153 // output byte streams. Encoding classes are parameterized macros
2154 // used by Machine Instruction Nodes in order to generate the bit
2155 // encoding of the instruction. Operands specify their base encoding
2156 // interface with the interface keyword. There are currently
2157 // supported four interfaces, REG_INTER, CONST_INTER, MEMORY_INTER, &
2158 // COND_INTER. REG_INTER causes an operand to generate a function
2159 // which returns its register number when queried. CONST_INTER causes
2160 // an operand to generate a function which returns the value of the
2161 // constant when queried. MEMORY_INTER causes an operand to generate
2162 // four functions which return the Base Register, the Index Register,
2163 // the Scale Value, and the Offset Value of the operand when queried.
2164 // COND_INTER causes an operand to generate six functions which return
2165 // the encoding code (ie - encoding bits for the instruction)
2166 // associated with each basic boolean condition for a conditional
2167 // instruction.
2168 //
2169 // Instructions specify two basic values for encoding. Again, a
2170 // function is available to check if the constant displacement is an
2171 // oop. They use the ins_encode keyword to specify their encoding
2172 // classes (which must be a sequence of enc_class names, and their
2173 // parameters, specified in the encoding block), and they use the
2174 // opcode keyword to specify, in order, their primary, secondary, and
2175 // tertiary opcode. Only the opcode sections which a particular
2176 // instruction needs for encoding need to be specified.
2177 encode %{
2178 // Build emit functions for each basic byte or larger field in the
2179 // intel encoding scheme (opcode, rm, sib, immediate), and call them
2180 // from C++ code in the enc_class source block. Emit functions will
2181 // live in the main source block for now. In future, we can
2182 // generalize this by adding a syntax that specifies the sizes of
2183 // fields in an order, so that the adlc can build the emit functions
2184 // automagically
2185
2186 // Emit primary opcode
2187 enc_class OpcP
2188 %{
2189 emit_opcode(cbuf, $primary);
2190 %}
2191
2192 // Emit secondary opcode
2193 enc_class OpcS
2194 %{
2195 emit_opcode(cbuf, $secondary);
2196 %}
2197
2198 // Emit tertiary opcode
2199 enc_class OpcT
2200 %{
2201 emit_opcode(cbuf, $tertiary);
2202 %}
2203
2204 // Emit opcode directly
2205 enc_class Opcode(immI d8)
2206 %{
2207 emit_opcode(cbuf, $d8$$constant);
2208 %}
2209
2210 // Emit size prefix
2211 enc_class SizePrefix
2212 %{
2213 emit_opcode(cbuf, 0x66);
2214 %}
2215
2216 enc_class reg(rRegI reg)
2217 %{
2218 emit_rm(cbuf, 0x3, 0, $reg$$reg & 7);
2219 %}
2220
2221 enc_class reg_reg(rRegI dst, rRegI src)
2222 %{
2223 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2224 %}
2225
2226 enc_class opc_reg_reg(immI opcode, rRegI dst, rRegI src)
2227 %{
2228 emit_opcode(cbuf, $opcode$$constant);
2229 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2230 %}
2231
2232 enc_class cmpfp_fixup()
2233 %{
2234 // jnp,s exit
2235 emit_opcode(cbuf, 0x7B);
2236 emit_d8(cbuf, 0x0A);
2237
2238 // pushfq
2239 emit_opcode(cbuf, 0x9C);
2240
2241 // andq $0xffffff2b, (%rsp)
2242 emit_opcode(cbuf, Assembler::REX_W);
2243 emit_opcode(cbuf, 0x81);
2244 emit_opcode(cbuf, 0x24);
2245 emit_opcode(cbuf, 0x24);
2246 emit_d32(cbuf, 0xffffff2b);
2247
2248 // popfq
2249 emit_opcode(cbuf, 0x9D);
2250
2251 // nop (target for branch to avoid branch to branch)
2252 emit_opcode(cbuf, 0x90);
2253 %}
2254
2255 enc_class cmpfp3(rRegI dst)
2256 %{
2257 int dstenc = $dst$$reg;
2258
2259 // movl $dst, -1
2260 if (dstenc >= 8) {
2261 emit_opcode(cbuf, Assembler::REX_B);
2262 }
2263 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
2264 emit_d32(cbuf, -1);
2265
2266 // jp,s done
2267 emit_opcode(cbuf, 0x7A);
2268 emit_d8(cbuf, dstenc < 4 ? 0x08 : 0x0A);
2269
2270 // jb,s done
2271 emit_opcode(cbuf, 0x72);
2272 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
2273
2274 // setne $dst
2275 if (dstenc >= 4) {
2276 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
2277 }
2278 emit_opcode(cbuf, 0x0F);
2279 emit_opcode(cbuf, 0x95);
2280 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
2281
2282 // movzbl $dst, $dst
2283 if (dstenc >= 4) {
2284 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
2285 }
2286 emit_opcode(cbuf, 0x0F);
2287 emit_opcode(cbuf, 0xB6);
2288 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
2289 %}
2290
2291 enc_class cdql_enc(no_rax_rdx_RegI div)
2292 %{
2293 // Full implementation of Java idiv and irem; checks for
2294 // special case as described in JVM spec., p.243 & p.271.
2295 //
2296 // normal case special case
2297 //
2298 // input : rax: dividend min_int
2299 // reg: divisor -1
2300 //
2301 // output: rax: quotient (= rax idiv reg) min_int
2302 // rdx: remainder (= rax irem reg) 0
2303 //
2304 // Code sequnce:
2305 //
2306 // 0: 3d 00 00 00 80 cmp $0x80000000,%eax
2307 // 5: 75 07/08 jne e <normal>
2308 // 7: 33 d2 xor %edx,%edx
2309 // [div >= 8 -> offset + 1]
2310 // [REX_B]
2311 // 9: 83 f9 ff cmp $0xffffffffffffffff,$div
2312 // c: 74 03/04 je 11 <done>
2313 // 000000000000000e <normal>:
2314 // e: 99 cltd
2315 // [div >= 8 -> offset + 1]
2316 // [REX_B]
2317 // f: f7 f9 idiv $div
2318 // 0000000000000011 <done>:
2319
2320 // cmp $0x80000000,%eax
2321 emit_opcode(cbuf, 0x3d);
2322 emit_d8(cbuf, 0x00);
2323 emit_d8(cbuf, 0x00);
2324 emit_d8(cbuf, 0x00);
2325 emit_d8(cbuf, 0x80);
2326
2327 // jne e <normal>
2328 emit_opcode(cbuf, 0x75);
2329 emit_d8(cbuf, $div$$reg < 8 ? 0x07 : 0x08);
2330
2331 // xor %edx,%edx
2332 emit_opcode(cbuf, 0x33);
2333 emit_d8(cbuf, 0xD2);
2334
2335 // cmp $0xffffffffffffffff,%ecx
2336 if ($div$$reg >= 8) {
2337 emit_opcode(cbuf, Assembler::REX_B);
2338 }
2339 emit_opcode(cbuf, 0x83);
2340 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2341 emit_d8(cbuf, 0xFF);
2342
2343 // je 11 <done>
2344 emit_opcode(cbuf, 0x74);
2345 emit_d8(cbuf, $div$$reg < 8 ? 0x03 : 0x04);
2346
2347 // <normal>
2348 // cltd
2349 emit_opcode(cbuf, 0x99);
2350
2351 // idivl (note: must be emitted by the user of this rule)
2352 // <done>
2353 %}
2354
2355 enc_class cdqq_enc(no_rax_rdx_RegL div)
2356 %{
2357 // Full implementation of Java ldiv and lrem; checks for
2358 // special case as described in JVM spec., p.243 & p.271.
2359 //
2360 // normal case special case
2361 //
2362 // input : rax: dividend min_long
2363 // reg: divisor -1
2364 //
2365 // output: rax: quotient (= rax idiv reg) min_long
2366 // rdx: remainder (= rax irem reg) 0
2367 //
2368 // Code sequnce:
2369 //
2370 // 0: 48 ba 00 00 00 00 00 mov $0x8000000000000000,%rdx
2371 // 7: 00 00 80
2372 // a: 48 39 d0 cmp %rdx,%rax
2373 // d: 75 08 jne 17 <normal>
2374 // f: 33 d2 xor %edx,%edx
2375 // 11: 48 83 f9 ff cmp $0xffffffffffffffff,$div
2376 // 15: 74 05 je 1c <done>
2377 // 0000000000000017 <normal>:
2378 // 17: 48 99 cqto
2379 // 19: 48 f7 f9 idiv $div
2380 // 000000000000001c <done>:
2381
2382 // mov $0x8000000000000000,%rdx
2383 emit_opcode(cbuf, Assembler::REX_W);
2384 emit_opcode(cbuf, 0xBA);
2385 emit_d8(cbuf, 0x00);
2386 emit_d8(cbuf, 0x00);
2387 emit_d8(cbuf, 0x00);
2388 emit_d8(cbuf, 0x00);
2389 emit_d8(cbuf, 0x00);
2390 emit_d8(cbuf, 0x00);
2391 emit_d8(cbuf, 0x00);
2392 emit_d8(cbuf, 0x80);
2393
2394 // cmp %rdx,%rax
2395 emit_opcode(cbuf, Assembler::REX_W);
2396 emit_opcode(cbuf, 0x39);
2397 emit_d8(cbuf, 0xD0);
2398
2399 // jne 17 <normal>
2400 emit_opcode(cbuf, 0x75);
2401 emit_d8(cbuf, 0x08);
2402
2403 // xor %edx,%edx
2404 emit_opcode(cbuf, 0x33);
2405 emit_d8(cbuf, 0xD2);
2406
2407 // cmp $0xffffffffffffffff,$div
2408 emit_opcode(cbuf, $div$$reg < 8 ? Assembler::REX_W : Assembler::REX_WB);
2409 emit_opcode(cbuf, 0x83);
2410 emit_rm(cbuf, 0x3, 0x7, $div$$reg & 7);
2411 emit_d8(cbuf, 0xFF);
2412
2413 // je 1e <done>
2414 emit_opcode(cbuf, 0x74);
2415 emit_d8(cbuf, 0x05);
2416
2417 // <normal>
2418 // cqto
2419 emit_opcode(cbuf, Assembler::REX_W);
2420 emit_opcode(cbuf, 0x99);
2421
2422 // idivq (note: must be emitted by the user of this rule)
2423 // <done>
2424 %}
2425
2426 // Opcde enc_class for 8/32 bit immediate instructions with sign-extension
2427 enc_class OpcSE(immI imm)
2428 %{
2429 // Emit primary opcode and set sign-extend bit
2430 // Check for 8-bit immediate, and set sign extend bit in opcode
2431 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2432 emit_opcode(cbuf, $primary | 0x02);
2433 } else {
2434 // 32-bit immediate
2435 emit_opcode(cbuf, $primary);
2436 }
2437 %}
2438
2439 enc_class OpcSErm(rRegI dst, immI imm)
2440 %{
2441 // OpcSEr/m
2442 int dstenc = $dst$$reg;
2443 if (dstenc >= 8) {
2444 emit_opcode(cbuf, Assembler::REX_B);
2445 dstenc -= 8;
2446 }
2447 // Emit primary opcode and set sign-extend bit
2448 // Check for 8-bit immediate, and set sign extend bit in opcode
2449 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2450 emit_opcode(cbuf, $primary | 0x02);
2451 } else {
2452 // 32-bit immediate
2453 emit_opcode(cbuf, $primary);
2454 }
2455 // Emit r/m byte with secondary opcode, after primary opcode.
2456 emit_rm(cbuf, 0x3, $secondary, dstenc);
2457 %}
2458
2459 enc_class OpcSErm_wide(rRegL dst, immI imm)
2460 %{
2461 // OpcSEr/m
2462 int dstenc = $dst$$reg;
2463 if (dstenc < 8) {
2464 emit_opcode(cbuf, Assembler::REX_W);
2465 } else {
2466 emit_opcode(cbuf, Assembler::REX_WB);
2467 dstenc -= 8;
2468 }
2469 // Emit primary opcode and set sign-extend bit
2470 // Check for 8-bit immediate, and set sign extend bit in opcode
2471 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2472 emit_opcode(cbuf, $primary | 0x02);
2473 } else {
2474 // 32-bit immediate
2475 emit_opcode(cbuf, $primary);
2476 }
2477 // Emit r/m byte with secondary opcode, after primary opcode.
2478 emit_rm(cbuf, 0x3, $secondary, dstenc);
2479 %}
2480
2481 enc_class Con8or32(immI imm)
2482 %{
2483 // Check for 8-bit immediate, and set sign extend bit in opcode
2484 if (-0x80 <= $imm$$constant && $imm$$constant < 0x80) {
2485 $$$emit8$imm$$constant;
2486 } else {
2487 // 32-bit immediate
2488 $$$emit32$imm$$constant;
2489 }
2490 %}
2491
2492 enc_class Lbl(label labl)
2493 %{
2494 // JMP, CALL
2495 Label* l = $labl$$label;
2496 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2497 %}
2498
2499 enc_class LblShort(label labl)
2500 %{
2501 // JMP, CALL
2502 Label* l = $labl$$label;
2503 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2504 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2505 emit_d8(cbuf, disp);
2506 %}
2507
2508 enc_class opc2_reg(rRegI dst)
2509 %{
2510 // BSWAP
2511 emit_cc(cbuf, $secondary, $dst$$reg);
2512 %}
2513
2514 enc_class opc3_reg(rRegI dst)
2515 %{
2516 // BSWAP
2517 emit_cc(cbuf, $tertiary, $dst$$reg);
2518 %}
2519
2520 enc_class reg_opc(rRegI div)
2521 %{
2522 // INC, DEC, IDIV, IMOD, JMP indirect, ...
2523 emit_rm(cbuf, 0x3, $secondary, $div$$reg & 7);
2524 %}
2525
2526 enc_class Jcc(cmpOp cop, label labl)
2527 %{
2528 // JCC
2529 Label* l = $labl$$label;
2530 $$$emit8$primary;
2531 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2532 emit_d32(cbuf, l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0);
2533 %}
2534
2535 enc_class JccShort (cmpOp cop, label labl)
2536 %{
2537 // JCC
2538 Label *l = $labl$$label;
2539 emit_cc(cbuf, $primary, $cop$$cmpcode);
2540 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
2541 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
2542 emit_d8(cbuf, disp);
2543 %}
2544
2545 enc_class enc_cmov(cmpOp cop)
2546 %{
2547 // CMOV
2548 $$$emit8$primary;
2549 emit_cc(cbuf, $secondary, $cop$$cmpcode);
2550 %}
2551
2552 enc_class enc_cmovf_branch(cmpOp cop, regF dst, regF src)
2553 %{
2554 // Invert sense of branch from sense of cmov
2555 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2556 emit_d8(cbuf, ($dst$$reg < 8 && $src$$reg < 8)
2557 ? (UseXmmRegToRegMoveAll ? 3 : 4)
2558 : (UseXmmRegToRegMoveAll ? 4 : 5) ); // REX
2559 // UseXmmRegToRegMoveAll ? movaps(dst, src) : movss(dst, src)
2560 if (!UseXmmRegToRegMoveAll) emit_opcode(cbuf, 0xF3);
2561 if ($dst$$reg < 8) {
2562 if ($src$$reg >= 8) {
2563 emit_opcode(cbuf, Assembler::REX_B);
2564 }
2565 } else {
2566 if ($src$$reg < 8) {
2567 emit_opcode(cbuf, Assembler::REX_R);
2568 } else {
2569 emit_opcode(cbuf, Assembler::REX_RB);
2570 }
2571 }
2572 emit_opcode(cbuf, 0x0F);
2573 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2574 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2575 %}
2576
2577 enc_class enc_cmovd_branch(cmpOp cop, regD dst, regD src)
2578 %{
2579 // Invert sense of branch from sense of cmov
2580 emit_cc(cbuf, 0x70, $cop$$cmpcode ^ 1);
2581 emit_d8(cbuf, $dst$$reg < 8 && $src$$reg < 8 ? 4 : 5); // REX
2582
2583 // UseXmmRegToRegMoveAll ? movapd(dst, src) : movsd(dst, src)
2584 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x66 : 0xF2);
2585 if ($dst$$reg < 8) {
2586 if ($src$$reg >= 8) {
2587 emit_opcode(cbuf, Assembler::REX_B);
2588 }
2589 } else {
2590 if ($src$$reg < 8) {
2591 emit_opcode(cbuf, Assembler::REX_R);
2592 } else {
2593 emit_opcode(cbuf, Assembler::REX_RB);
2594 }
2595 }
2596 emit_opcode(cbuf, 0x0F);
2597 emit_opcode(cbuf, UseXmmRegToRegMoveAll ? 0x28 : 0x10);
2598 emit_rm(cbuf, 0x3, $dst$$reg & 7, $src$$reg & 7);
2599 %}
2600
2601 enc_class enc_PartialSubtypeCheck()
2602 %{
2603 Register Rrdi = as_Register(RDI_enc); // result register
2604 Register Rrax = as_Register(RAX_enc); // super class
2605 Register Rrcx = as_Register(RCX_enc); // killed
2606 Register Rrsi = as_Register(RSI_enc); // sub class
2607 Label miss;
2608 const bool set_cond_codes = true;
2609
2610 MacroAssembler _masm(&cbuf);
2611 __ check_klass_subtype_slow_path(Rrsi, Rrax, Rrcx, Rrdi,
2612 NULL, &miss,
2613 /*set_cond_codes:*/ true);
2614 if ($primary) {
2615 __ xorptr(Rrdi, Rrdi);
2616 }
2617 __ bind(miss);
2618 %}
2619
2620 enc_class Java_To_Interpreter(method meth)
2621 %{
2622 // CALL Java_To_Interpreter
2623 // This is the instruction starting address for relocation info.
2624 cbuf.set_inst_mark();
2625 $$$emit8$primary;
2626 // CALL directly to the runtime
2627 emit_d32_reloc(cbuf,
2628 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2629 runtime_call_Relocation::spec(),
2630 RELOC_DISP32);
2631 %}
2632
2633 enc_class preserve_SP %{
2634 debug_only(int off0 = cbuf.code_size());
2635 MacroAssembler _masm(&cbuf);
2636 // RBP is preserved across all calls, even compiled calls.
2637 // Use it to preserve RSP in places where the callee might change the SP.
2638 __ movptr(rbp, rsp);
2639 debug_only(int off1 = cbuf.code_size());
2640 assert(off1 - off0 == preserve_SP_size(), "correct size prediction");
2641 %}
2642
2643 enc_class restore_SP %{
2644 MacroAssembler _masm(&cbuf);
2645 __ movptr(rsp, rbp);
2646 %}
2647
2648 enc_class Java_Static_Call(method meth)
2649 %{
2650 // JAVA STATIC CALL
2651 // CALL to fixup routine. Fixup routine uses ScopeDesc info to
2652 // determine who we intended to call.
2653 cbuf.set_inst_mark();
2654 $$$emit8$primary;
2655
2656 if (!_method) {
2657 emit_d32_reloc(cbuf,
2658 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2659 runtime_call_Relocation::spec(),
2660 RELOC_DISP32);
2661 } else if (_optimized_virtual) {
2662 emit_d32_reloc(cbuf,
2663 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2664 opt_virtual_call_Relocation::spec(),
2665 RELOC_DISP32);
2666 } else {
2667 emit_d32_reloc(cbuf,
2668 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2669 static_call_Relocation::spec(),
2670 RELOC_DISP32);
2671 }
2672 if (_method) {
2673 // Emit stub for static call
2674 emit_java_to_interp(cbuf);
2675 }
2676 %}
2677
2678 enc_class Java_Dynamic_Call(method meth)
2679 %{
2680 // JAVA DYNAMIC CALL
2681 // !!!!!
2682 // Generate "movq rax, -1", placeholder instruction to load oop-info
2683 // emit_call_dynamic_prologue( cbuf );
2684 cbuf.set_inst_mark();
2685
2686 // movq rax, -1
2687 emit_opcode(cbuf, Assembler::REX_W);
2688 emit_opcode(cbuf, 0xB8 | RAX_enc);
2689 emit_d64_reloc(cbuf,
2690 (int64_t) Universe::non_oop_word(),
2691 oop_Relocation::spec_for_immediate(), RELOC_IMM64);
2692 address virtual_call_oop_addr = cbuf.inst_mark();
2693 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine
2694 // who we intended to call.
2695 cbuf.set_inst_mark();
2696 $$$emit8$primary;
2697 emit_d32_reloc(cbuf,
2698 (int) ($meth$$method - ((intptr_t) cbuf.code_end()) - 4),
2699 virtual_call_Relocation::spec(virtual_call_oop_addr),
2700 RELOC_DISP32);
2701 %}
2702
2703 enc_class Java_Compiled_Call(method meth)
2704 %{
2705 // JAVA COMPILED CALL
2706 int disp = in_bytes(methodOopDesc:: from_compiled_offset());
2707
2708 // XXX XXX offset is 128 is 1.5 NON-PRODUCT !!!
2709 // assert(-0x80 <= disp && disp < 0x80, "compiled_code_offset isn't small");
2710
2711 // callq *disp(%rax)
2712 cbuf.set_inst_mark();
2713 $$$emit8$primary;
2714 if (disp < 0x80) {
2715 emit_rm(cbuf, 0x01, $secondary, RAX_enc); // R/M byte
2716 emit_d8(cbuf, disp); // Displacement
2717 } else {
2718 emit_rm(cbuf, 0x02, $secondary, RAX_enc); // R/M byte
2719 emit_d32(cbuf, disp); // Displacement
2720 }
2721 %}
2722
2723 enc_class reg_opc_imm(rRegI dst, immI8 shift)
2724 %{
2725 // SAL, SAR, SHR
2726 int dstenc = $dst$$reg;
2727 if (dstenc >= 8) {
2728 emit_opcode(cbuf, Assembler::REX_B);
2729 dstenc -= 8;
2730 }
2731 $$$emit8$primary;
2732 emit_rm(cbuf, 0x3, $secondary, dstenc);
2733 $$$emit8$shift$$constant;
2734 %}
2735
2736 enc_class reg_opc_imm_wide(rRegL dst, immI8 shift)
2737 %{
2738 // SAL, SAR, SHR
2739 int dstenc = $dst$$reg;
2740 if (dstenc < 8) {
2741 emit_opcode(cbuf, Assembler::REX_W);
2742 } else {
2743 emit_opcode(cbuf, Assembler::REX_WB);
2744 dstenc -= 8;
2745 }
2746 $$$emit8$primary;
2747 emit_rm(cbuf, 0x3, $secondary, dstenc);
2748 $$$emit8$shift$$constant;
2749 %}
2750
2751 enc_class load_immI(rRegI dst, immI src)
2752 %{
2753 int dstenc = $dst$$reg;
2754 if (dstenc >= 8) {
2755 emit_opcode(cbuf, Assembler::REX_B);
2756 dstenc -= 8;
2757 }
2758 emit_opcode(cbuf, 0xB8 | dstenc);
2759 $$$emit32$src$$constant;
2760 %}
2761
2762 enc_class load_immL(rRegL dst, immL src)
2763 %{
2764 int dstenc = $dst$$reg;
2765 if (dstenc < 8) {
2766 emit_opcode(cbuf, Assembler::REX_W);
2767 } else {
2768 emit_opcode(cbuf, Assembler::REX_WB);
2769 dstenc -= 8;
2770 }
2771 emit_opcode(cbuf, 0xB8 | dstenc);
2772 emit_d64(cbuf, $src$$constant);
2773 %}
2774
2775 enc_class load_immUL32(rRegL dst, immUL32 src)
2776 %{
2777 // same as load_immI, but this time we care about zeroes in the high word
2778 int dstenc = $dst$$reg;
2779 if (dstenc >= 8) {
2780 emit_opcode(cbuf, Assembler::REX_B);
2781 dstenc -= 8;
2782 }
2783 emit_opcode(cbuf, 0xB8 | dstenc);
2784 $$$emit32$src$$constant;
2785 %}
2786
2787 enc_class load_immL32(rRegL dst, immL32 src)
2788 %{
2789 int dstenc = $dst$$reg;
2790 if (dstenc < 8) {
2791 emit_opcode(cbuf, Assembler::REX_W);
2792 } else {
2793 emit_opcode(cbuf, Assembler::REX_WB);
2794 dstenc -= 8;
2795 }
2796 emit_opcode(cbuf, 0xC7);
2797 emit_rm(cbuf, 0x03, 0x00, dstenc);
2798 $$$emit32$src$$constant;
2799 %}
2800
2801 enc_class load_immP31(rRegP dst, immP32 src)
2802 %{
2803 // same as load_immI, but this time we care about zeroes in the high word
2804 int dstenc = $dst$$reg;
2805 if (dstenc >= 8) {
2806 emit_opcode(cbuf, Assembler::REX_B);
2807 dstenc -= 8;
2808 }
2809 emit_opcode(cbuf, 0xB8 | dstenc);
2810 $$$emit32$src$$constant;
2811 %}
2812
2813 enc_class load_immP(rRegP dst, immP src)
2814 %{
2815 int dstenc = $dst$$reg;
2816 if (dstenc < 8) {
2817 emit_opcode(cbuf, Assembler::REX_W);
2818 } else {
2819 emit_opcode(cbuf, Assembler::REX_WB);
2820 dstenc -= 8;
2821 }
2822 emit_opcode(cbuf, 0xB8 | dstenc);
2823 // This next line should be generated from ADLC
2824 if ($src->constant_is_oop()) {
2825 emit_d64_reloc(cbuf, $src$$constant, relocInfo::oop_type, RELOC_IMM64);
2826 } else {
2827 emit_d64(cbuf, $src$$constant);
2828 }
2829 %}
2830
2831 enc_class load_immF(regF dst, immF con)
2832 %{
2833 // XXX reg_mem doesn't support RIP-relative addressing yet
2834 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2835 emit_float_constant(cbuf, $con$$constant);
2836 %}
2837
2838 enc_class load_immD(regD dst, immD con)
2839 %{
2840 // XXX reg_mem doesn't support RIP-relative addressing yet
2841 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2842 emit_double_constant(cbuf, $con$$constant);
2843 %}
2844
2845 enc_class load_conF (regF dst, immF con) %{ // Load float constant
2846 emit_opcode(cbuf, 0xF3);
2847 if ($dst$$reg >= 8) {
2848 emit_opcode(cbuf, Assembler::REX_R);
2849 }
2850 emit_opcode(cbuf, 0x0F);
2851 emit_opcode(cbuf, 0x10);
2852 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2853 emit_float_constant(cbuf, $con$$constant);
2854 %}
2855
2856 enc_class load_conD (regD dst, immD con) %{ // Load double constant
2857 // UseXmmLoadAndClearUpper ? movsd(dst, con) : movlpd(dst, con)
2858 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
2859 if ($dst$$reg >= 8) {
2860 emit_opcode(cbuf, Assembler::REX_R);
2861 }
2862 emit_opcode(cbuf, 0x0F);
2863 emit_opcode(cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12);
2864 emit_rm(cbuf, 0x0, $dst$$reg & 7, 0x5); // 00 reg 101
2865 emit_double_constant(cbuf, $con$$constant);
2866 %}
2867
2868 // Encode a reg-reg copy. If it is useless, then empty encoding.
2869 enc_class enc_copy(rRegI dst, rRegI src)
2870 %{
2871 encode_copy(cbuf, $dst$$reg, $src$$reg);
2872 %}
2873
2874 // Encode xmm reg-reg copy. If it is useless, then empty encoding.
2875 enc_class enc_CopyXD( RegD dst, RegD src ) %{
2876 encode_CopyXD( cbuf, $dst$$reg, $src$$reg );
2877 %}
2878
2879 enc_class enc_copy_always(rRegI dst, rRegI src)
2880 %{
2881 int srcenc = $src$$reg;
2882 int dstenc = $dst$$reg;
2883
2884 if (dstenc < 8) {
2885 if (srcenc >= 8) {
2886 emit_opcode(cbuf, Assembler::REX_B);
2887 srcenc -= 8;
2888 }
2889 } else {
2890 if (srcenc < 8) {
2891 emit_opcode(cbuf, Assembler::REX_R);
2892 } else {
2893 emit_opcode(cbuf, Assembler::REX_RB);
2894 srcenc -= 8;
2895 }
2896 dstenc -= 8;
2897 }
2898
2899 emit_opcode(cbuf, 0x8B);
2900 emit_rm(cbuf, 0x3, dstenc, srcenc);
2901 %}
2902
2903 enc_class enc_copy_wide(rRegL dst, rRegL src)
2904 %{
2905 int srcenc = $src$$reg;
2906 int dstenc = $dst$$reg;
2907
2908 if (dstenc != srcenc) {
2909 if (dstenc < 8) {
2910 if (srcenc < 8) {
2911 emit_opcode(cbuf, Assembler::REX_W);
2912 } else {
2913 emit_opcode(cbuf, Assembler::REX_WB);
2914 srcenc -= 8;
2915 }
2916 } else {
2917 if (srcenc < 8) {
2918 emit_opcode(cbuf, Assembler::REX_WR);
2919 } else {
2920 emit_opcode(cbuf, Assembler::REX_WRB);
2921 srcenc -= 8;
2922 }
2923 dstenc -= 8;
2924 }
2925 emit_opcode(cbuf, 0x8B);
2926 emit_rm(cbuf, 0x3, dstenc, srcenc);
2927 }
2928 %}
2929
2930 enc_class Con32(immI src)
2931 %{
2932 // Output immediate
2933 $$$emit32$src$$constant;
2934 %}
2935
2936 enc_class Con64(immL src)
2937 %{
2938 // Output immediate
2939 emit_d64($src$$constant);
2940 %}
2941
2942 enc_class Con32F_as_bits(immF src)
2943 %{
2944 // Output Float immediate bits
2945 jfloat jf = $src$$constant;
2946 jint jf_as_bits = jint_cast(jf);
2947 emit_d32(cbuf, jf_as_bits);
2948 %}
2949
2950 enc_class Con16(immI src)
2951 %{
2952 // Output immediate
2953 $$$emit16$src$$constant;
2954 %}
2955
2956 // How is this different from Con32??? XXX
2957 enc_class Con_d32(immI src)
2958 %{
2959 emit_d32(cbuf,$src$$constant);
2960 %}
2961
2962 enc_class conmemref (rRegP t1) %{ // Con32(storeImmI)
2963 // Output immediate memory reference
2964 emit_rm(cbuf, 0x00, $t1$$reg, 0x05 );
2965 emit_d32(cbuf, 0x00);
2966 %}
2967
2968 enc_class jump_enc(rRegL switch_val, rRegI dest) %{
2969 MacroAssembler masm(&cbuf);
2970
2971 Register switch_reg = as_Register($switch_val$$reg);
2972 Register dest_reg = as_Register($dest$$reg);
2973 address table_base = masm.address_table_constant(_index2label);
2974
2975 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2976 // to do that and the compiler is using that register as one it can allocate.
2977 // So we build it all by hand.
2978 // Address index(noreg, switch_reg, Address::times_1);
2979 // ArrayAddress dispatch(table, index);
2980
2981 Address dispatch(dest_reg, switch_reg, Address::times_1);
2982
2983 masm.lea(dest_reg, InternalAddress(table_base));
2984 masm.jmp(dispatch);
2985 %}
2986
2987 enc_class jump_enc_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
2988 MacroAssembler masm(&cbuf);
2989
2990 Register switch_reg = as_Register($switch_val$$reg);
2991 Register dest_reg = as_Register($dest$$reg);
2992 address table_base = masm.address_table_constant(_index2label);
2993
2994 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
2995 // to do that and the compiler is using that register as one it can allocate.
2996 // So we build it all by hand.
2997 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
2998 // ArrayAddress dispatch(table, index);
2999
3000 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant, (int)$offset$$constant);
3001
3002 masm.lea(dest_reg, InternalAddress(table_base));
3003 masm.jmp(dispatch);
3004 %}
3005
3006 enc_class jump_enc_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
3007 MacroAssembler masm(&cbuf);
3008
3009 Register switch_reg = as_Register($switch_val$$reg);
3010 Register dest_reg = as_Register($dest$$reg);
3011 address table_base = masm.address_table_constant(_index2label);
3012
3013 // We could use jump(ArrayAddress) except that the macro assembler needs to use r10
3014 // to do that and the compiler is using that register as one it can allocate.
3015 // So we build it all by hand.
3016 // Address index(noreg, switch_reg, (Address::ScaleFactor)$shift$$constant);
3017 // ArrayAddress dispatch(table, index);
3018
3019 Address dispatch(dest_reg, switch_reg, (Address::ScaleFactor)$shift$$constant);
3020 masm.lea(dest_reg, InternalAddress(table_base));
3021 masm.jmp(dispatch);
3022
3023 %}
3024
3025 enc_class lock_prefix()
3026 %{
3027 if (os::is_MP()) {
3028 emit_opcode(cbuf, 0xF0); // lock
3029 }
3030 %}
3031
3032 enc_class REX_mem(memory mem)
3033 %{
3034 if ($mem$$base >= 8) {
3035 if ($mem$$index < 8) {
3036 emit_opcode(cbuf, Assembler::REX_B);
3037 } else {
3038 emit_opcode(cbuf, Assembler::REX_XB);
3039 }
3040 } else {
3041 if ($mem$$index >= 8) {
3042 emit_opcode(cbuf, Assembler::REX_X);
3043 }
3044 }
3045 %}
3046
3047 enc_class REX_mem_wide(memory mem)
3048 %{
3049 if ($mem$$base >= 8) {
3050 if ($mem$$index < 8) {
3051 emit_opcode(cbuf, Assembler::REX_WB);
3052 } else {
3053 emit_opcode(cbuf, Assembler::REX_WXB);
3054 }
3055 } else {
3056 if ($mem$$index < 8) {
3057 emit_opcode(cbuf, Assembler::REX_W);
3058 } else {
3059 emit_opcode(cbuf, Assembler::REX_WX);
3060 }
3061 }
3062 %}
3063
3064 // for byte regs
3065 enc_class REX_breg(rRegI reg)
3066 %{
3067 if ($reg$$reg >= 4) {
3068 emit_opcode(cbuf, $reg$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3069 }
3070 %}
3071
3072 // for byte regs
3073 enc_class REX_reg_breg(rRegI dst, rRegI src)
3074 %{
3075 if ($dst$$reg < 8) {
3076 if ($src$$reg >= 4) {
3077 emit_opcode(cbuf, $src$$reg < 8 ? Assembler::REX : Assembler::REX_B);
3078 }
3079 } else {
3080 if ($src$$reg < 8) {
3081 emit_opcode(cbuf, Assembler::REX_R);
3082 } else {
3083 emit_opcode(cbuf, Assembler::REX_RB);
3084 }
3085 }
3086 %}
3087
3088 // for byte regs
3089 enc_class REX_breg_mem(rRegI reg, memory mem)
3090 %{
3091 if ($reg$$reg < 8) {
3092 if ($mem$$base < 8) {
3093 if ($mem$$index >= 8) {
3094 emit_opcode(cbuf, Assembler::REX_X);
3095 } else if ($reg$$reg >= 4) {
3096 emit_opcode(cbuf, Assembler::REX);
3097 }
3098 } else {
3099 if ($mem$$index < 8) {
3100 emit_opcode(cbuf, Assembler::REX_B);
3101 } else {
3102 emit_opcode(cbuf, Assembler::REX_XB);
3103 }
3104 }
3105 } else {
3106 if ($mem$$base < 8) {
3107 if ($mem$$index < 8) {
3108 emit_opcode(cbuf, Assembler::REX_R);
3109 } else {
3110 emit_opcode(cbuf, Assembler::REX_RX);
3111 }
3112 } else {
3113 if ($mem$$index < 8) {
3114 emit_opcode(cbuf, Assembler::REX_RB);
3115 } else {
3116 emit_opcode(cbuf, Assembler::REX_RXB);
3117 }
3118 }
3119 }
3120 %}
3121
3122 enc_class REX_reg(rRegI reg)
3123 %{
3124 if ($reg$$reg >= 8) {
3125 emit_opcode(cbuf, Assembler::REX_B);
3126 }
3127 %}
3128
3129 enc_class REX_reg_wide(rRegI reg)
3130 %{
3131 if ($reg$$reg < 8) {
3132 emit_opcode(cbuf, Assembler::REX_W);
3133 } else {
3134 emit_opcode(cbuf, Assembler::REX_WB);
3135 }
3136 %}
3137
3138 enc_class REX_reg_reg(rRegI dst, rRegI src)
3139 %{
3140 if ($dst$$reg < 8) {
3141 if ($src$$reg >= 8) {
3142 emit_opcode(cbuf, Assembler::REX_B);
3143 }
3144 } else {
3145 if ($src$$reg < 8) {
3146 emit_opcode(cbuf, Assembler::REX_R);
3147 } else {
3148 emit_opcode(cbuf, Assembler::REX_RB);
3149 }
3150 }
3151 %}
3152
3153 enc_class REX_reg_reg_wide(rRegI dst, rRegI src)
3154 %{
3155 if ($dst$$reg < 8) {
3156 if ($src$$reg < 8) {
3157 emit_opcode(cbuf, Assembler::REX_W);
3158 } else {
3159 emit_opcode(cbuf, Assembler::REX_WB);
3160 }
3161 } else {
3162 if ($src$$reg < 8) {
3163 emit_opcode(cbuf, Assembler::REX_WR);
3164 } else {
3165 emit_opcode(cbuf, Assembler::REX_WRB);
3166 }
3167 }
3168 %}
3169
3170 enc_class REX_reg_mem(rRegI reg, memory mem)
3171 %{
3172 if ($reg$$reg < 8) {
3173 if ($mem$$base < 8) {
3174 if ($mem$$index >= 8) {
3175 emit_opcode(cbuf, Assembler::REX_X);
3176 }
3177 } else {
3178 if ($mem$$index < 8) {
3179 emit_opcode(cbuf, Assembler::REX_B);
3180 } else {
3181 emit_opcode(cbuf, Assembler::REX_XB);
3182 }
3183 }
3184 } else {
3185 if ($mem$$base < 8) {
3186 if ($mem$$index < 8) {
3187 emit_opcode(cbuf, Assembler::REX_R);
3188 } else {
3189 emit_opcode(cbuf, Assembler::REX_RX);
3190 }
3191 } else {
3192 if ($mem$$index < 8) {
3193 emit_opcode(cbuf, Assembler::REX_RB);
3194 } else {
3195 emit_opcode(cbuf, Assembler::REX_RXB);
3196 }
3197 }
3198 }
3199 %}
3200
3201 enc_class REX_reg_mem_wide(rRegL reg, memory mem)
3202 %{
3203 if ($reg$$reg < 8) {
3204 if ($mem$$base < 8) {
3205 if ($mem$$index < 8) {
3206 emit_opcode(cbuf, Assembler::REX_W);
3207 } else {
3208 emit_opcode(cbuf, Assembler::REX_WX);
3209 }
3210 } else {
3211 if ($mem$$index < 8) {
3212 emit_opcode(cbuf, Assembler::REX_WB);
3213 } else {
3214 emit_opcode(cbuf, Assembler::REX_WXB);
3215 }
3216 }
3217 } else {
3218 if ($mem$$base < 8) {
3219 if ($mem$$index < 8) {
3220 emit_opcode(cbuf, Assembler::REX_WR);
3221 } else {
3222 emit_opcode(cbuf, Assembler::REX_WRX);
3223 }
3224 } else {
3225 if ($mem$$index < 8) {
3226 emit_opcode(cbuf, Assembler::REX_WRB);
3227 } else {
3228 emit_opcode(cbuf, Assembler::REX_WRXB);
3229 }
3230 }
3231 }
3232 %}
3233
3234 enc_class reg_mem(rRegI ereg, memory mem)
3235 %{
3236 // High registers handle in encode_RegMem
3237 int reg = $ereg$$reg;
3238 int base = $mem$$base;
3239 int index = $mem$$index;
3240 int scale = $mem$$scale;
3241 int disp = $mem$$disp;
3242 bool disp_is_oop = $mem->disp_is_oop();
3243
3244 encode_RegMem(cbuf, reg, base, index, scale, disp, disp_is_oop);
3245 %}
3246
3247 enc_class RM_opc_mem(immI rm_opcode, memory mem)
3248 %{
3249 int rm_byte_opcode = $rm_opcode$$constant;
3250
3251 // High registers handle in encode_RegMem
3252 int base = $mem$$base;
3253 int index = $mem$$index;
3254 int scale = $mem$$scale;
3255 int displace = $mem$$disp;
3256
3257 bool disp_is_oop = $mem->disp_is_oop(); // disp-as-oop when
3258 // working with static
3259 // globals
3260 encode_RegMem(cbuf, rm_byte_opcode, base, index, scale, displace,
3261 disp_is_oop);
3262 %}
3263
3264 enc_class reg_lea(rRegI dst, rRegI src0, immI src1)
3265 %{
3266 int reg_encoding = $dst$$reg;
3267 int base = $src0$$reg; // 0xFFFFFFFF indicates no base
3268 int index = 0x04; // 0x04 indicates no index
3269 int scale = 0x00; // 0x00 indicates no scale
3270 int displace = $src1$$constant; // 0x00 indicates no displacement
3271 bool disp_is_oop = false;
3272 encode_RegMem(cbuf, reg_encoding, base, index, scale, displace,
3273 disp_is_oop);
3274 %}
3275
3276 enc_class neg_reg(rRegI dst)
3277 %{
3278 int dstenc = $dst$$reg;
3279 if (dstenc >= 8) {
3280 emit_opcode(cbuf, Assembler::REX_B);
3281 dstenc -= 8;
3282 }
3283 // NEG $dst
3284 emit_opcode(cbuf, 0xF7);
3285 emit_rm(cbuf, 0x3, 0x03, dstenc);
3286 %}
3287
3288 enc_class neg_reg_wide(rRegI dst)
3289 %{
3290 int dstenc = $dst$$reg;
3291 if (dstenc < 8) {
3292 emit_opcode(cbuf, Assembler::REX_W);
3293 } else {
3294 emit_opcode(cbuf, Assembler::REX_WB);
3295 dstenc -= 8;
3296 }
3297 // NEG $dst
3298 emit_opcode(cbuf, 0xF7);
3299 emit_rm(cbuf, 0x3, 0x03, dstenc);
3300 %}
3301
3302 enc_class setLT_reg(rRegI dst)
3303 %{
3304 int dstenc = $dst$$reg;
3305 if (dstenc >= 8) {
3306 emit_opcode(cbuf, Assembler::REX_B);
3307 dstenc -= 8;
3308 } else if (dstenc >= 4) {
3309 emit_opcode(cbuf, Assembler::REX);
3310 }
3311 // SETLT $dst
3312 emit_opcode(cbuf, 0x0F);
3313 emit_opcode(cbuf, 0x9C);
3314 emit_rm(cbuf, 0x3, 0x0, dstenc);
3315 %}
3316
3317 enc_class setNZ_reg(rRegI dst)
3318 %{
3319 int dstenc = $dst$$reg;
3320 if (dstenc >= 8) {
3321 emit_opcode(cbuf, Assembler::REX_B);
3322 dstenc -= 8;
3323 } else if (dstenc >= 4) {
3324 emit_opcode(cbuf, Assembler::REX);
3325 }
3326 // SETNZ $dst
3327 emit_opcode(cbuf, 0x0F);
3328 emit_opcode(cbuf, 0x95);
3329 emit_rm(cbuf, 0x3, 0x0, dstenc);
3330 %}
3331
3332 enc_class enc_cmpLTP(no_rcx_RegI p, no_rcx_RegI q, no_rcx_RegI y,
3333 rcx_RegI tmp)
3334 %{
3335 // cadd_cmpLT
3336
3337 int tmpReg = $tmp$$reg;
3338
3339 int penc = $p$$reg;
3340 int qenc = $q$$reg;
3341 int yenc = $y$$reg;
3342
3343 // subl $p,$q
3344 if (penc < 8) {
3345 if (qenc >= 8) {
3346 emit_opcode(cbuf, Assembler::REX_B);
3347 }
3348 } else {
3349 if (qenc < 8) {
3350 emit_opcode(cbuf, Assembler::REX_R);
3351 } else {
3352 emit_opcode(cbuf, Assembler::REX_RB);
3353 }
3354 }
3355 emit_opcode(cbuf, 0x2B);
3356 emit_rm(cbuf, 0x3, penc & 7, qenc & 7);
3357
3358 // sbbl $tmp, $tmp
3359 emit_opcode(cbuf, 0x1B);
3360 emit_rm(cbuf, 0x3, tmpReg, tmpReg);
3361
3362 // andl $tmp, $y
3363 if (yenc >= 8) {
3364 emit_opcode(cbuf, Assembler::REX_B);
3365 }
3366 emit_opcode(cbuf, 0x23);
3367 emit_rm(cbuf, 0x3, tmpReg, yenc & 7);
3368
3369 // addl $p,$tmp
3370 if (penc >= 8) {
3371 emit_opcode(cbuf, Assembler::REX_R);
3372 }
3373 emit_opcode(cbuf, 0x03);
3374 emit_rm(cbuf, 0x3, penc & 7, tmpReg);
3375 %}
3376
3377 // Compare the lonogs and set -1, 0, or 1 into dst
3378 enc_class cmpl3_flag(rRegL src1, rRegL src2, rRegI dst)
3379 %{
3380 int src1enc = $src1$$reg;
3381 int src2enc = $src2$$reg;
3382 int dstenc = $dst$$reg;
3383
3384 // cmpq $src1, $src2
3385 if (src1enc < 8) {
3386 if (src2enc < 8) {
3387 emit_opcode(cbuf, Assembler::REX_W);
3388 } else {
3389 emit_opcode(cbuf, Assembler::REX_WB);
3390 }
3391 } else {
3392 if (src2enc < 8) {
3393 emit_opcode(cbuf, Assembler::REX_WR);
3394 } else {
3395 emit_opcode(cbuf, Assembler::REX_WRB);
3396 }
3397 }
3398 emit_opcode(cbuf, 0x3B);
3399 emit_rm(cbuf, 0x3, src1enc & 7, src2enc & 7);
3400
3401 // movl $dst, -1
3402 if (dstenc >= 8) {
3403 emit_opcode(cbuf, Assembler::REX_B);
3404 }
3405 emit_opcode(cbuf, 0xB8 | (dstenc & 7));
3406 emit_d32(cbuf, -1);
3407
3408 // jl,s done
3409 emit_opcode(cbuf, 0x7C);
3410 emit_d8(cbuf, dstenc < 4 ? 0x06 : 0x08);
3411
3412 // setne $dst
3413 if (dstenc >= 4) {
3414 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_B);
3415 }
3416 emit_opcode(cbuf, 0x0F);
3417 emit_opcode(cbuf, 0x95);
3418 emit_opcode(cbuf, 0xC0 | (dstenc & 7));
3419
3420 // movzbl $dst, $dst
3421 if (dstenc >= 4) {
3422 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX : Assembler::REX_RB);
3423 }
3424 emit_opcode(cbuf, 0x0F);
3425 emit_opcode(cbuf, 0xB6);
3426 emit_rm(cbuf, 0x3, dstenc & 7, dstenc & 7);
3427 %}
3428
3429 enc_class Push_ResultXD(regD dst) %{
3430 int dstenc = $dst$$reg;
3431
3432 store_to_stackslot( cbuf, 0xDD, 0x03, 0 ); //FSTP [RSP]
3433
3434 // UseXmmLoadAndClearUpper ? movsd dst,[rsp] : movlpd dst,[rsp]
3435 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0xF2 : 0x66);
3436 if (dstenc >= 8) {
3437 emit_opcode(cbuf, Assembler::REX_R);
3438 }
3439 emit_opcode (cbuf, 0x0F );
3440 emit_opcode (cbuf, UseXmmLoadAndClearUpper ? 0x10 : 0x12 );
3441 encode_RegMem(cbuf, dstenc, RSP_enc, 0x4, 0, 0, false);
3442
3443 // add rsp,8
3444 emit_opcode(cbuf, Assembler::REX_W);
3445 emit_opcode(cbuf,0x83);
3446 emit_rm(cbuf,0x3, 0x0, RSP_enc);
3447 emit_d8(cbuf,0x08);
3448 %}
3449
3450 enc_class Push_SrcXD(regD src) %{
3451 int srcenc = $src$$reg;
3452
3453 // subq rsp,#8
3454 emit_opcode(cbuf, Assembler::REX_W);
3455 emit_opcode(cbuf, 0x83);
3456 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3457 emit_d8(cbuf, 0x8);
3458
3459 // movsd [rsp],src
3460 emit_opcode(cbuf, 0xF2);
3461 if (srcenc >= 8) {
3462 emit_opcode(cbuf, Assembler::REX_R);
3463 }
3464 emit_opcode(cbuf, 0x0F);
3465 emit_opcode(cbuf, 0x11);
3466 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false);
3467
3468 // fldd [rsp]
3469 emit_opcode(cbuf, 0x66);
3470 emit_opcode(cbuf, 0xDD);
3471 encode_RegMem(cbuf, 0x0, RSP_enc, 0x4, 0, 0, false);
3472 %}
3473
3474
3475 enc_class movq_ld(regD dst, memory mem) %{
3476 MacroAssembler _masm(&cbuf);
3477 __ movq($dst$$XMMRegister, $mem$$Address);
3478 %}
3479
3480 enc_class movq_st(memory mem, regD src) %{
3481 MacroAssembler _masm(&cbuf);
3482 __ movq($mem$$Address, $src$$XMMRegister);
3483 %}
3484
3485 enc_class pshufd_8x8(regF dst, regF src) %{
3486 MacroAssembler _masm(&cbuf);
3487
3488 encode_CopyXD(cbuf, $dst$$reg, $src$$reg);
3489 __ punpcklbw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg));
3490 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($dst$$reg), 0x00);
3491 %}
3492
3493 enc_class pshufd_4x16(regF dst, regF src) %{
3494 MacroAssembler _masm(&cbuf);
3495
3496 __ pshuflw(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), 0x00);
3497 %}
3498
3499 enc_class pshufd(regD dst, regD src, int mode) %{
3500 MacroAssembler _masm(&cbuf);
3501
3502 __ pshufd(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg), $mode);
3503 %}
3504
3505 enc_class pxor(regD dst, regD src) %{
3506 MacroAssembler _masm(&cbuf);
3507
3508 __ pxor(as_XMMRegister($dst$$reg), as_XMMRegister($src$$reg));
3509 %}
3510
3511 enc_class mov_i2x(regD dst, rRegI src) %{
3512 MacroAssembler _masm(&cbuf);
3513
3514 __ movdl(as_XMMRegister($dst$$reg), as_Register($src$$reg));
3515 %}
3516
3517 // obj: object to lock
3518 // box: box address (header location) -- killed
3519 // tmp: rax -- killed
3520 // scr: rbx -- killed
3521 //
3522 // What follows is a direct transliteration of fast_lock() and fast_unlock()
3523 // from i486.ad. See that file for comments.
3524 // TODO: where possible switch from movq (r, 0) to movl(r,0) and
3525 // use the shorter encoding. (Movl clears the high-order 32-bits).
3526
3527
3528 enc_class Fast_Lock(rRegP obj, rRegP box, rax_RegI tmp, rRegP scr)
3529 %{
3530 Register objReg = as_Register((int)$obj$$reg);
3531 Register boxReg = as_Register((int)$box$$reg);
3532 Register tmpReg = as_Register($tmp$$reg);
3533 Register scrReg = as_Register($scr$$reg);
3534 MacroAssembler masm(&cbuf);
3535
3536 // Verify uniqueness of register assignments -- necessary but not sufficient
3537 assert (objReg != boxReg && objReg != tmpReg &&
3538 objReg != scrReg && tmpReg != scrReg, "invariant") ;
3539
3540 if (_counters != NULL) {
3541 masm.atomic_incl(ExternalAddress((address) _counters->total_entry_count_addr()));
3542 }
3543 if (EmitSync & 1) {
3544 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3545 masm.movptr (Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3546 masm.cmpptr(rsp, (int32_t)NULL_WORD) ;
3547 } else
3548 if (EmitSync & 2) {
3549 Label DONE_LABEL;
3550 if (UseBiasedLocking) {
3551 // Note: tmpReg maps to the swap_reg argument and scrReg to the tmp_reg argument.
3552 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, false, DONE_LABEL, NULL, _counters);
3553 }
3554 // QQQ was movl...
3555 masm.movptr(tmpReg, 0x1);
3556 masm.orptr(tmpReg, Address(objReg, 0));
3557 masm.movptr(Address(boxReg, 0), tmpReg);
3558 if (os::is_MP()) {
3559 masm.lock();
3560 }
3561 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3562 masm.jcc(Assembler::equal, DONE_LABEL);
3563
3564 // Recursive locking
3565 masm.subptr(tmpReg, rsp);
3566 masm.andptr(tmpReg, 7 - os::vm_page_size());
3567 masm.movptr(Address(boxReg, 0), tmpReg);
3568
3569 masm.bind(DONE_LABEL);
3570 masm.nop(); // avoid branch to branch
3571 } else {
3572 Label DONE_LABEL, IsInflated, Egress;
3573
3574 masm.movptr(tmpReg, Address(objReg, 0)) ;
3575 masm.testl (tmpReg, 0x02) ; // inflated vs stack-locked|neutral|biased
3576 masm.jcc (Assembler::notZero, IsInflated) ;
3577
3578 // it's stack-locked, biased or neutral
3579 // TODO: optimize markword triage order to reduce the number of
3580 // conditional branches in the most common cases.
3581 // Beware -- there's a subtle invariant that fetch of the markword
3582 // at [FETCH], below, will never observe a biased encoding (*101b).
3583 // If this invariant is not held we'll suffer exclusion (safety) failure.
3584
3585 if (UseBiasedLocking && !UseOptoBiasInlining) {
3586 masm.biased_locking_enter(boxReg, objReg, tmpReg, scrReg, true, DONE_LABEL, NULL, _counters);
3587 masm.movptr(tmpReg, Address(objReg, 0)) ; // [FETCH]
3588 }
3589
3590 // was q will it destroy high?
3591 masm.orl (tmpReg, 1) ;
3592 masm.movptr(Address(boxReg, 0), tmpReg) ;
3593 if (os::is_MP()) { masm.lock(); }
3594 masm.cmpxchgptr(boxReg, Address(objReg, 0)); // Updates tmpReg
3595 if (_counters != NULL) {
3596 masm.cond_inc32(Assembler::equal,
3597 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3598 }
3599 masm.jcc (Assembler::equal, DONE_LABEL);
3600
3601 // Recursive locking
3602 masm.subptr(tmpReg, rsp);
3603 masm.andptr(tmpReg, 7 - os::vm_page_size());
3604 masm.movptr(Address(boxReg, 0), tmpReg);
3605 if (_counters != NULL) {
3606 masm.cond_inc32(Assembler::equal,
3607 ExternalAddress((address) _counters->fast_path_entry_count_addr()));
3608 }
3609 masm.jmp (DONE_LABEL) ;
3610
3611 masm.bind (IsInflated) ;
3612 // It's inflated
3613
3614 // TODO: someday avoid the ST-before-CAS penalty by
3615 // relocating (deferring) the following ST.
3616 // We should also think about trying a CAS without having
3617 // fetched _owner. If the CAS is successful we may
3618 // avoid an RTO->RTS upgrade on the $line.
3619 // Without cast to int32_t a movptr will destroy r10 which is typically obj
3620 masm.movptr(Address(boxReg, 0), (int32_t)intptr_t(markOopDesc::unused_mark())) ;
3621
3622 masm.mov (boxReg, tmpReg) ;
3623 masm.movptr (tmpReg, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3624 masm.testptr(tmpReg, tmpReg) ;
3625 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3626
3627 // It's inflated and appears unlocked
3628 if (os::is_MP()) { masm.lock(); }
3629 masm.cmpxchgptr(r15_thread, Address(boxReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3630 // Intentional fall-through into DONE_LABEL ...
3631
3632 masm.bind (DONE_LABEL) ;
3633 masm.nop () ; // avoid jmp to jmp
3634 }
3635 %}
3636
3637 // obj: object to unlock
3638 // box: box address (displaced header location), killed
3639 // RBX: killed tmp; cannot be obj nor box
3640 enc_class Fast_Unlock(rRegP obj, rax_RegP box, rRegP tmp)
3641 %{
3642
3643 Register objReg = as_Register($obj$$reg);
3644 Register boxReg = as_Register($box$$reg);
3645 Register tmpReg = as_Register($tmp$$reg);
3646 MacroAssembler masm(&cbuf);
3647
3648 if (EmitSync & 4) {
3649 masm.cmpptr(rsp, 0) ;
3650 } else
3651 if (EmitSync & 8) {
3652 Label DONE_LABEL;
3653 if (UseBiasedLocking) {
3654 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3655 }
3656
3657 // Check whether the displaced header is 0
3658 //(=> recursive unlock)
3659 masm.movptr(tmpReg, Address(boxReg, 0));
3660 masm.testptr(tmpReg, tmpReg);
3661 masm.jcc(Assembler::zero, DONE_LABEL);
3662
3663 // If not recursive lock, reset the header to displaced header
3664 if (os::is_MP()) {
3665 masm.lock();
3666 }
3667 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3668 masm.bind(DONE_LABEL);
3669 masm.nop(); // avoid branch to branch
3670 } else {
3671 Label DONE_LABEL, Stacked, CheckSucc ;
3672
3673 if (UseBiasedLocking && !UseOptoBiasInlining) {
3674 masm.biased_locking_exit(objReg, tmpReg, DONE_LABEL);
3675 }
3676
3677 masm.movptr(tmpReg, Address(objReg, 0)) ;
3678 masm.cmpptr(Address(boxReg, 0), (int32_t)NULL_WORD) ;
3679 masm.jcc (Assembler::zero, DONE_LABEL) ;
3680 masm.testl (tmpReg, 0x02) ;
3681 masm.jcc (Assembler::zero, Stacked) ;
3682
3683 // It's inflated
3684 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2)) ;
3685 masm.xorptr(boxReg, r15_thread) ;
3686 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::recursions_offset_in_bytes()-2)) ;
3687 masm.jcc (Assembler::notZero, DONE_LABEL) ;
3688 masm.movptr(boxReg, Address (tmpReg, ObjectMonitor::cxq_offset_in_bytes()-2)) ;
3689 masm.orptr (boxReg, Address (tmpReg, ObjectMonitor::EntryList_offset_in_bytes()-2)) ;
3690 masm.jcc (Assembler::notZero, CheckSucc) ;
3691 masm.movptr(Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3692 masm.jmp (DONE_LABEL) ;
3693
3694 if ((EmitSync & 65536) == 0) {
3695 Label LSuccess, LGoSlowPath ;
3696 masm.bind (CheckSucc) ;
3697 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3698 masm.jcc (Assembler::zero, LGoSlowPath) ;
3699
3700 // I'd much rather use lock:andl m->_owner, 0 as it's faster than the
3701 // the explicit ST;MEMBAR combination, but masm doesn't currently support
3702 // "ANDQ M,IMM". Don't use MFENCE here. lock:add to TOS, xchg, etc
3703 // are all faster when the write buffer is populated.
3704 masm.movptr (Address (tmpReg, ObjectMonitor::owner_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3705 if (os::is_MP()) {
3706 masm.lock () ; masm.addl (Address(rsp, 0), 0) ;
3707 }
3708 masm.cmpptr(Address (tmpReg, ObjectMonitor::succ_offset_in_bytes()-2), (int32_t)NULL_WORD) ;
3709 masm.jcc (Assembler::notZero, LSuccess) ;
3710
3711 masm.movptr (boxReg, (int32_t)NULL_WORD) ; // box is really EAX
3712 if (os::is_MP()) { masm.lock(); }
3713 masm.cmpxchgptr(r15_thread, Address(tmpReg, ObjectMonitor::owner_offset_in_bytes()-2));
3714 masm.jcc (Assembler::notEqual, LSuccess) ;
3715 // Intentional fall-through into slow-path
3716
3717 masm.bind (LGoSlowPath) ;
3718 masm.orl (boxReg, 1) ; // set ICC.ZF=0 to indicate failure
3719 masm.jmp (DONE_LABEL) ;
3720
3721 masm.bind (LSuccess) ;
3722 masm.testl (boxReg, 0) ; // set ICC.ZF=1 to indicate success
3723 masm.jmp (DONE_LABEL) ;
3724 }
3725
3726 masm.bind (Stacked) ;
3727 masm.movptr(tmpReg, Address (boxReg, 0)) ; // re-fetch
3728 if (os::is_MP()) { masm.lock(); }
3729 masm.cmpxchgptr(tmpReg, Address(objReg, 0)); // Uses RAX which is box
3730
3731 if (EmitSync & 65536) {
3732 masm.bind (CheckSucc) ;
3733 }
3734 masm.bind(DONE_LABEL);
3735 if (EmitSync & 32768) {
3736 masm.nop(); // avoid branch to branch
3737 }
3738 }
3739 %}
3740
3741
3742 enc_class enc_rethrow()
3743 %{
3744 cbuf.set_inst_mark();
3745 emit_opcode(cbuf, 0xE9); // jmp entry
3746 emit_d32_reloc(cbuf,
3747 (int) (OptoRuntime::rethrow_stub() - cbuf.code_end() - 4),
3748 runtime_call_Relocation::spec(),
3749 RELOC_DISP32);
3750 %}
3751
3752 enc_class absF_encoding(regF dst)
3753 %{
3754 int dstenc = $dst$$reg;
3755 address signmask_address = (address) StubRoutines::x86::float_sign_mask();
3756
3757 cbuf.set_inst_mark();
3758 if (dstenc >= 8) {
3759 emit_opcode(cbuf, Assembler::REX_R);
3760 dstenc -= 8;
3761 }
3762 // XXX reg_mem doesn't support RIP-relative addressing yet
3763 emit_opcode(cbuf, 0x0F);
3764 emit_opcode(cbuf, 0x54);
3765 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3766 emit_d32_reloc(cbuf, signmask_address);
3767 %}
3768
3769 enc_class absD_encoding(regD dst)
3770 %{
3771 int dstenc = $dst$$reg;
3772 address signmask_address = (address) StubRoutines::x86::double_sign_mask();
3773
3774 cbuf.set_inst_mark();
3775 emit_opcode(cbuf, 0x66);
3776 if (dstenc >= 8) {
3777 emit_opcode(cbuf, Assembler::REX_R);
3778 dstenc -= 8;
3779 }
3780 // XXX reg_mem doesn't support RIP-relative addressing yet
3781 emit_opcode(cbuf, 0x0F);
3782 emit_opcode(cbuf, 0x54);
3783 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3784 emit_d32_reloc(cbuf, signmask_address);
3785 %}
3786
3787 enc_class negF_encoding(regF dst)
3788 %{
3789 int dstenc = $dst$$reg;
3790 address signflip_address = (address) StubRoutines::x86::float_sign_flip();
3791
3792 cbuf.set_inst_mark();
3793 if (dstenc >= 8) {
3794 emit_opcode(cbuf, Assembler::REX_R);
3795 dstenc -= 8;
3796 }
3797 // XXX reg_mem doesn't support RIP-relative addressing yet
3798 emit_opcode(cbuf, 0x0F);
3799 emit_opcode(cbuf, 0x57);
3800 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3801 emit_d32_reloc(cbuf, signflip_address);
3802 %}
3803
3804 enc_class negD_encoding(regD dst)
3805 %{
3806 int dstenc = $dst$$reg;
3807 address signflip_address = (address) StubRoutines::x86::double_sign_flip();
3808
3809 cbuf.set_inst_mark();
3810 emit_opcode(cbuf, 0x66);
3811 if (dstenc >= 8) {
3812 emit_opcode(cbuf, Assembler::REX_R);
3813 dstenc -= 8;
3814 }
3815 // XXX reg_mem doesn't support RIP-relative addressing yet
3816 emit_opcode(cbuf, 0x0F);
3817 emit_opcode(cbuf, 0x57);
3818 emit_rm(cbuf, 0x0, dstenc, 0x5); // 00 reg 101
3819 emit_d32_reloc(cbuf, signflip_address);
3820 %}
3821
3822 enc_class f2i_fixup(rRegI dst, regF src)
3823 %{
3824 int dstenc = $dst$$reg;
3825 int srcenc = $src$$reg;
3826
3827 // cmpl $dst, #0x80000000
3828 if (dstenc >= 8) {
3829 emit_opcode(cbuf, Assembler::REX_B);
3830 }
3831 emit_opcode(cbuf, 0x81);
3832 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
3833 emit_d32(cbuf, 0x80000000);
3834
3835 // jne,s done
3836 emit_opcode(cbuf, 0x75);
3837 if (srcenc < 8 && dstenc < 8) {
3838 emit_d8(cbuf, 0xF);
3839 } else if (srcenc >= 8 && dstenc >= 8) {
3840 emit_d8(cbuf, 0x11);
3841 } else {
3842 emit_d8(cbuf, 0x10);
3843 }
3844
3845 // subq rsp, #8
3846 emit_opcode(cbuf, Assembler::REX_W);
3847 emit_opcode(cbuf, 0x83);
3848 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3849 emit_d8(cbuf, 8);
3850
3851 // movss [rsp], $src
3852 emit_opcode(cbuf, 0xF3);
3853 if (srcenc >= 8) {
3854 emit_opcode(cbuf, Assembler::REX_R);
3855 }
3856 emit_opcode(cbuf, 0x0F);
3857 emit_opcode(cbuf, 0x11);
3858 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3859
3860 // call f2i_fixup
3861 cbuf.set_inst_mark();
3862 emit_opcode(cbuf, 0xE8);
3863 emit_d32_reloc(cbuf,
3864 (int)
3865 (StubRoutines::x86::f2i_fixup() - cbuf.code_end() - 4),
3866 runtime_call_Relocation::spec(),
3867 RELOC_DISP32);
3868
3869 // popq $dst
3870 if (dstenc >= 8) {
3871 emit_opcode(cbuf, Assembler::REX_B);
3872 }
3873 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3874
3875 // done:
3876 %}
3877
3878 enc_class f2l_fixup(rRegL dst, regF src)
3879 %{
3880 int dstenc = $dst$$reg;
3881 int srcenc = $src$$reg;
3882 address const_address = (address) StubRoutines::x86::double_sign_flip();
3883
3884 // cmpq $dst, [0x8000000000000000]
3885 cbuf.set_inst_mark();
3886 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
3887 emit_opcode(cbuf, 0x39);
3888 // XXX reg_mem doesn't support RIP-relative addressing yet
3889 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
3890 emit_d32_reloc(cbuf, const_address);
3891
3892
3893 // jne,s done
3894 emit_opcode(cbuf, 0x75);
3895 if (srcenc < 8 && dstenc < 8) {
3896 emit_d8(cbuf, 0xF);
3897 } else if (srcenc >= 8 && dstenc >= 8) {
3898 emit_d8(cbuf, 0x11);
3899 } else {
3900 emit_d8(cbuf, 0x10);
3901 }
3902
3903 // subq rsp, #8
3904 emit_opcode(cbuf, Assembler::REX_W);
3905 emit_opcode(cbuf, 0x83);
3906 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3907 emit_d8(cbuf, 8);
3908
3909 // movss [rsp], $src
3910 emit_opcode(cbuf, 0xF3);
3911 if (srcenc >= 8) {
3912 emit_opcode(cbuf, Assembler::REX_R);
3913 }
3914 emit_opcode(cbuf, 0x0F);
3915 emit_opcode(cbuf, 0x11);
3916 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3917
3918 // call f2l_fixup
3919 cbuf.set_inst_mark();
3920 emit_opcode(cbuf, 0xE8);
3921 emit_d32_reloc(cbuf,
3922 (int)
3923 (StubRoutines::x86::f2l_fixup() - cbuf.code_end() - 4),
3924 runtime_call_Relocation::spec(),
3925 RELOC_DISP32);
3926
3927 // popq $dst
3928 if (dstenc >= 8) {
3929 emit_opcode(cbuf, Assembler::REX_B);
3930 }
3931 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3932
3933 // done:
3934 %}
3935
3936 enc_class d2i_fixup(rRegI dst, regD src)
3937 %{
3938 int dstenc = $dst$$reg;
3939 int srcenc = $src$$reg;
3940
3941 // cmpl $dst, #0x80000000
3942 if (dstenc >= 8) {
3943 emit_opcode(cbuf, Assembler::REX_B);
3944 }
3945 emit_opcode(cbuf, 0x81);
3946 emit_rm(cbuf, 0x3, 0x7, dstenc & 7);
3947 emit_d32(cbuf, 0x80000000);
3948
3949 // jne,s done
3950 emit_opcode(cbuf, 0x75);
3951 if (srcenc < 8 && dstenc < 8) {
3952 emit_d8(cbuf, 0xF);
3953 } else if (srcenc >= 8 && dstenc >= 8) {
3954 emit_d8(cbuf, 0x11);
3955 } else {
3956 emit_d8(cbuf, 0x10);
3957 }
3958
3959 // subq rsp, #8
3960 emit_opcode(cbuf, Assembler::REX_W);
3961 emit_opcode(cbuf, 0x83);
3962 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
3963 emit_d8(cbuf, 8);
3964
3965 // movsd [rsp], $src
3966 emit_opcode(cbuf, 0xF2);
3967 if (srcenc >= 8) {
3968 emit_opcode(cbuf, Assembler::REX_R);
3969 }
3970 emit_opcode(cbuf, 0x0F);
3971 emit_opcode(cbuf, 0x11);
3972 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
3973
3974 // call d2i_fixup
3975 cbuf.set_inst_mark();
3976 emit_opcode(cbuf, 0xE8);
3977 emit_d32_reloc(cbuf,
3978 (int)
3979 (StubRoutines::x86::d2i_fixup() - cbuf.code_end() - 4),
3980 runtime_call_Relocation::spec(),
3981 RELOC_DISP32);
3982
3983 // popq $dst
3984 if (dstenc >= 8) {
3985 emit_opcode(cbuf, Assembler::REX_B);
3986 }
3987 emit_opcode(cbuf, 0x58 | (dstenc & 7));
3988
3989 // done:
3990 %}
3991
3992 enc_class d2l_fixup(rRegL dst, regD src)
3993 %{
3994 int dstenc = $dst$$reg;
3995 int srcenc = $src$$reg;
3996 address const_address = (address) StubRoutines::x86::double_sign_flip();
3997
3998 // cmpq $dst, [0x8000000000000000]
3999 cbuf.set_inst_mark();
4000 emit_opcode(cbuf, dstenc < 8 ? Assembler::REX_W : Assembler::REX_WR);
4001 emit_opcode(cbuf, 0x39);
4002 // XXX reg_mem doesn't support RIP-relative addressing yet
4003 emit_rm(cbuf, 0x0, dstenc & 7, 0x5); // 00 reg 101
4004 emit_d32_reloc(cbuf, const_address);
4005
4006
4007 // jne,s done
4008 emit_opcode(cbuf, 0x75);
4009 if (srcenc < 8 && dstenc < 8) {
4010 emit_d8(cbuf, 0xF);
4011 } else if (srcenc >= 8 && dstenc >= 8) {
4012 emit_d8(cbuf, 0x11);
4013 } else {
4014 emit_d8(cbuf, 0x10);
4015 }
4016
4017 // subq rsp, #8
4018 emit_opcode(cbuf, Assembler::REX_W);
4019 emit_opcode(cbuf, 0x83);
4020 emit_rm(cbuf, 0x3, 0x5, RSP_enc);
4021 emit_d8(cbuf, 8);
4022
4023 // movsd [rsp], $src
4024 emit_opcode(cbuf, 0xF2);
4025 if (srcenc >= 8) {
4026 emit_opcode(cbuf, Assembler::REX_R);
4027 }
4028 emit_opcode(cbuf, 0x0F);
4029 emit_opcode(cbuf, 0x11);
4030 encode_RegMem(cbuf, srcenc, RSP_enc, 0x4, 0, 0, false); // 2 bytes
4031
4032 // call d2l_fixup
4033 cbuf.set_inst_mark();
4034 emit_opcode(cbuf, 0xE8);
4035 emit_d32_reloc(cbuf,
4036 (int)
4037 (StubRoutines::x86::d2l_fixup() - cbuf.code_end() - 4),
4038 runtime_call_Relocation::spec(),
4039 RELOC_DISP32);
4040
4041 // popq $dst
4042 if (dstenc >= 8) {
4043 emit_opcode(cbuf, Assembler::REX_B);
4044 }
4045 emit_opcode(cbuf, 0x58 | (dstenc & 7));
4046
4047 // done:
4048 %}
4049
4050 // Safepoint Poll. This polls the safepoint page, and causes an
4051 // exception if it is not readable. Unfortunately, it kills
4052 // RFLAGS in the process.
4053 enc_class enc_safepoint_poll
4054 %{
4055 // testl %rax, off(%rip) // Opcode + ModRM + Disp32 == 6 bytes
4056 // XXX reg_mem doesn't support RIP-relative addressing yet
4057 cbuf.set_inst_mark();
4058 cbuf.relocate(cbuf.inst_mark(), relocInfo::poll_type, 0); // XXX
4059 emit_opcode(cbuf, 0x85); // testl
4060 emit_rm(cbuf, 0x0, RAX_enc, 0x5); // 00 rax 101 == 0x5
4061 // cbuf.inst_mark() is beginning of instruction
4062 emit_d32_reloc(cbuf, os::get_polling_page());
4063 // relocInfo::poll_type,
4064 %}
4065 %}
4066
4067
4068
4069 //----------FRAME--------------------------------------------------------------
4070 // Definition of frame structure and management information.
4071 //
4072 // S T A C K L A Y O U T Allocators stack-slot number
4073 // | (to get allocators register number
4074 // G Owned by | | v add OptoReg::stack0())
4075 // r CALLER | |
4076 // o | +--------+ pad to even-align allocators stack-slot
4077 // w V | pad0 | numbers; owned by CALLER
4078 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned
4079 // h ^ | in | 5
4080 // | | args | 4 Holes in incoming args owned by SELF
4081 // | | | | 3
4082 // | | +--------+
4083 // V | | old out| Empty on Intel, window on Sparc
4084 // | old |preserve| Must be even aligned.
4085 // | SP-+--------+----> Matcher::_old_SP, even aligned
4086 // | | in | 3 area for Intel ret address
4087 // Owned by |preserve| Empty on Sparc.
4088 // SELF +--------+
4089 // | | pad2 | 2 pad to align old SP
4090 // | +--------+ 1
4091 // | | locks | 0
4092 // | +--------+----> OptoReg::stack0(), even aligned
4093 // | | pad1 | 11 pad to align new SP
4094 // | +--------+
4095 // | | | 10
4096 // | | spills | 9 spills
4097 // V | | 8 (pad0 slot for callee)
4098 // -----------+--------+----> Matcher::_out_arg_limit, unaligned
4099 // ^ | out | 7
4100 // | | args | 6 Holes in outgoing args owned by CALLEE
4101 // Owned by +--------+
4102 // CALLEE | new out| 6 Empty on Intel, window on Sparc
4103 // | new |preserve| Must be even-aligned.
4104 // | SP-+--------+----> Matcher::_new_SP, even aligned
4105 // | | |
4106 //
4107 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is
4108 // known from SELF's arguments and the Java calling convention.
4109 // Region 6-7 is determined per call site.
4110 // Note 2: If the calling convention leaves holes in the incoming argument
4111 // area, those holes are owned by SELF. Holes in the outgoing area
4112 // are owned by the CALLEE. Holes should not be nessecary in the
4113 // incoming area, as the Java calling convention is completely under
4114 // the control of the AD file. Doubles can be sorted and packed to
4115 // avoid holes. Holes in the outgoing arguments may be nessecary for
4116 // varargs C calling conventions.
4117 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is
4118 // even aligned with pad0 as needed.
4119 // Region 6 is even aligned. Region 6-7 is NOT even aligned;
4120 // region 6-11 is even aligned; it may be padded out more so that
4121 // the region from SP to FP meets the minimum stack alignment.
4122 // Note 4: For I2C adapters, the incoming FP may not meet the minimum stack
4123 // alignment. Region 11, pad1, may be dynamically extended so that
4124 // SP meets the minimum alignment.
4125
4126 frame
4127 %{
4128 // What direction does stack grow in (assumed to be same for C & Java)
4129 stack_direction(TOWARDS_LOW);
4130
4131 // These three registers define part of the calling convention
4132 // between compiled code and the interpreter.
4133 inline_cache_reg(RAX); // Inline Cache Register
4134 interpreter_method_oop_reg(RBX); // Method Oop Register when
4135 // calling interpreter
4136
4137 // Optional: name the operand used by cisc-spilling to access
4138 // [stack_pointer + offset]
4139 cisc_spilling_operand_name(indOffset32);
4140
4141 // Number of stack slots consumed by locking an object
4142 sync_stack_slots(2);
4143
4144 // Compiled code's Frame Pointer
4145 frame_pointer(RSP);
4146
4147 // Interpreter stores its frame pointer in a register which is
4148 // stored to the stack by I2CAdaptors.
4149 // I2CAdaptors convert from interpreted java to compiled java.
4150 interpreter_frame_pointer(RBP);
4151
4152 // Stack alignment requirement
4153 stack_alignment(StackAlignmentInBytes); // Alignment size in bytes (128-bit -> 16 bytes)
4154
4155 // Number of stack slots between incoming argument block and the start of
4156 // a new frame. The PROLOG must add this many slots to the stack. The
4157 // EPILOG must remove this many slots. amd64 needs two slots for
4158 // return address.
4159 in_preserve_stack_slots(4 + 2 * VerifyStackAtCalls);
4160
4161 // Number of outgoing stack slots killed above the out_preserve_stack_slots
4162 // for calls to C. Supports the var-args backing area for register parms.
4163 varargs_C_out_slots_killed(frame::arg_reg_save_area_bytes/BytesPerInt);
4164
4165 // The after-PROLOG location of the return address. Location of
4166 // return address specifies a type (REG or STACK) and a number
4167 // representing the register number (i.e. - use a register name) or
4168 // stack slot.
4169 // Ret Addr is on stack in slot 0 if no locks or verification or alignment.
4170 // Otherwise, it is above the locks and verification slot and alignment word
4171 return_addr(STACK - 2 +
4172 round_to(2 + 2 * VerifyStackAtCalls +
4173 Compile::current()->fixed_slots(),
4174 WordsPerLong * 2));
4175
4176 // Body of function which returns an integer array locating
4177 // arguments either in registers or in stack slots. Passed an array
4178 // of ideal registers called "sig" and a "length" count. Stack-slot
4179 // offsets are based on outgoing arguments, i.e. a CALLER setting up
4180 // arguments for a CALLEE. Incoming stack arguments are
4181 // automatically biased by the preserve_stack_slots field above.
4182
4183 calling_convention
4184 %{
4185 // No difference between ingoing/outgoing just pass false
4186 SharedRuntime::java_calling_convention(sig_bt, regs, length, false);
4187 %}
4188
4189 c_calling_convention
4190 %{
4191 // This is obviously always outgoing
4192 (void) SharedRuntime::c_calling_convention(sig_bt, regs, length);
4193 %}
4194
4195 // Location of compiled Java return values. Same as C for now.
4196 return_value
4197 %{
4198 assert(ideal_reg >= Op_RegI && ideal_reg <= Op_RegL,
4199 "only return normal values");
4200
4201 static const int lo[Op_RegL + 1] = {
4202 0,
4203 0,
4204 RAX_num, // Op_RegN
4205 RAX_num, // Op_RegI
4206 RAX_num, // Op_RegP
4207 XMM0_num, // Op_RegF
4208 XMM0_num, // Op_RegD
4209 RAX_num // Op_RegL
4210 };
4211 static const int hi[Op_RegL + 1] = {
4212 0,
4213 0,
4214 OptoReg::Bad, // Op_RegN
4215 OptoReg::Bad, // Op_RegI
4216 RAX_H_num, // Op_RegP
4217 OptoReg::Bad, // Op_RegF
4218 XMM0_H_num, // Op_RegD
4219 RAX_H_num // Op_RegL
4220 };
4221 assert(ARRAY_SIZE(hi) == _last_machine_leaf - 1, "missing type");
4222 return OptoRegPair(hi[ideal_reg], lo[ideal_reg]);
4223 %}
4224 %}
4225
4226 //----------ATTRIBUTES---------------------------------------------------------
4227 //----------Operand Attributes-------------------------------------------------
4228 op_attrib op_cost(0); // Required cost attribute
4229
4230 //----------Instruction Attributes---------------------------------------------
4231 ins_attrib ins_cost(100); // Required cost attribute
4232 ins_attrib ins_size(8); // Required size attribute (in bits)
4233 ins_attrib ins_pc_relative(0); // Required PC Relative flag
4234 ins_attrib ins_short_branch(0); // Required flag: is this instruction
4235 // a non-matching short branch variant
4236 // of some long branch?
4237 ins_attrib ins_alignment(1); // Required alignment attribute (must
4238 // be a power of 2) specifies the
4239 // alignment that some part of the
4240 // instruction (not necessarily the
4241 // start) requires. If > 1, a
4242 // compute_padding() function must be
4243 // provided for the instruction
4244
4245 //----------OPERANDS-----------------------------------------------------------
4246 // Operand definitions must precede instruction definitions for correct parsing
4247 // in the ADLC because operands constitute user defined types which are used in
4248 // instruction definitions.
4249
4250 //----------Simple Operands----------------------------------------------------
4251 // Immediate Operands
4252 // Integer Immediate
4253 operand immI()
4254 %{
4255 match(ConI);
4256
4257 op_cost(10);
4258 format %{ %}
4259 interface(CONST_INTER);
4260 %}
4261
4262 // Constant for test vs zero
4263 operand immI0()
4264 %{
4265 predicate(n->get_int() == 0);
4266 match(ConI);
4267
4268 op_cost(0);
4269 format %{ %}
4270 interface(CONST_INTER);
4271 %}
4272
4273 // Constant for increment
4274 operand immI1()
4275 %{
4276 predicate(n->get_int() == 1);
4277 match(ConI);
4278
4279 op_cost(0);
4280 format %{ %}
4281 interface(CONST_INTER);
4282 %}
4283
4284 // Constant for decrement
4285 operand immI_M1()
4286 %{
4287 predicate(n->get_int() == -1);
4288 match(ConI);
4289
4290 op_cost(0);
4291 format %{ %}
4292 interface(CONST_INTER);
4293 %}
4294
4295 // Valid scale values for addressing modes
4296 operand immI2()
4297 %{
4298 predicate(0 <= n->get_int() && (n->get_int() <= 3));
4299 match(ConI);
4300
4301 format %{ %}
4302 interface(CONST_INTER);
4303 %}
4304
4305 operand immI8()
4306 %{
4307 predicate((-0x80 <= n->get_int()) && (n->get_int() < 0x80));
4308 match(ConI);
4309
4310 op_cost(5);
4311 format %{ %}
4312 interface(CONST_INTER);
4313 %}
4314
4315 operand immI16()
4316 %{
4317 predicate((-32768 <= n->get_int()) && (n->get_int() <= 32767));
4318 match(ConI);
4319
4320 op_cost(10);
4321 format %{ %}
4322 interface(CONST_INTER);
4323 %}
4324
4325 // Constant for long shifts
4326 operand immI_32()
4327 %{
4328 predicate( n->get_int() == 32 );
4329 match(ConI);
4330
4331 op_cost(0);
4332 format %{ %}
4333 interface(CONST_INTER);
4334 %}
4335
4336 // Constant for long shifts
4337 operand immI_64()
4338 %{
4339 predicate( n->get_int() == 64 );
4340 match(ConI);
4341
4342 op_cost(0);
4343 format %{ %}
4344 interface(CONST_INTER);
4345 %}
4346
4347 // Pointer Immediate
4348 operand immP()
4349 %{
4350 match(ConP);
4351
4352 op_cost(10);
4353 format %{ %}
4354 interface(CONST_INTER);
4355 %}
4356
4357 // NULL Pointer Immediate
4358 operand immP0()
4359 %{
4360 predicate(n->get_ptr() == 0);
4361 match(ConP);
4362
4363 op_cost(5);
4364 format %{ %}
4365 interface(CONST_INTER);
4366 %}
4367
4368 // Pointer Immediate
4369 operand immN() %{
4370 match(ConN);
4371
4372 op_cost(10);
4373 format %{ %}
4374 interface(CONST_INTER);
4375 %}
4376
4377 // NULL Pointer Immediate
4378 operand immN0() %{
4379 predicate(n->get_narrowcon() == 0);
4380 match(ConN);
4381
4382 op_cost(5);
4383 format %{ %}
4384 interface(CONST_INTER);
4385 %}
4386
4387 operand immP31()
4388 %{
4389 predicate(!n->as_Type()->type()->isa_oopptr()
4390 && (n->get_ptr() >> 31) == 0);
4391 match(ConP);
4392
4393 op_cost(5);
4394 format %{ %}
4395 interface(CONST_INTER);
4396 %}
4397
4398
4399 // Long Immediate
4400 operand immL()
4401 %{
4402 match(ConL);
4403
4404 op_cost(20);
4405 format %{ %}
4406 interface(CONST_INTER);
4407 %}
4408
4409 // Long Immediate 8-bit
4410 operand immL8()
4411 %{
4412 predicate(-0x80L <= n->get_long() && n->get_long() < 0x80L);
4413 match(ConL);
4414
4415 op_cost(5);
4416 format %{ %}
4417 interface(CONST_INTER);
4418 %}
4419
4420 // Long Immediate 32-bit unsigned
4421 operand immUL32()
4422 %{
4423 predicate(n->get_long() == (unsigned int) (n->get_long()));
4424 match(ConL);
4425
4426 op_cost(10);
4427 format %{ %}
4428 interface(CONST_INTER);
4429 %}
4430
4431 // Long Immediate 32-bit signed
4432 operand immL32()
4433 %{
4434 predicate(n->get_long() == (int) (n->get_long()));
4435 match(ConL);
4436
4437 op_cost(15);
4438 format %{ %}
4439 interface(CONST_INTER);
4440 %}
4441
4442 // Long Immediate zero
4443 operand immL0()
4444 %{
4445 predicate(n->get_long() == 0L);
4446 match(ConL);
4447
4448 op_cost(10);
4449 format %{ %}
4450 interface(CONST_INTER);
4451 %}
4452
4453 // Constant for increment
4454 operand immL1()
4455 %{
4456 predicate(n->get_long() == 1);
4457 match(ConL);
4458
4459 format %{ %}
4460 interface(CONST_INTER);
4461 %}
4462
4463 // Constant for decrement
4464 operand immL_M1()
4465 %{
4466 predicate(n->get_long() == -1);
4467 match(ConL);
4468
4469 format %{ %}
4470 interface(CONST_INTER);
4471 %}
4472
4473 // Long Immediate: the value 10
4474 operand immL10()
4475 %{
4476 predicate(n->get_long() == 10);
4477 match(ConL);
4478
4479 format %{ %}
4480 interface(CONST_INTER);
4481 %}
4482
4483 // Long immediate from 0 to 127.
4484 // Used for a shorter form of long mul by 10.
4485 operand immL_127()
4486 %{
4487 predicate(0 <= n->get_long() && n->get_long() < 0x80);
4488 match(ConL);
4489
4490 op_cost(10);
4491 format %{ %}
4492 interface(CONST_INTER);
4493 %}
4494
4495 // Long Immediate: low 32-bit mask
4496 operand immL_32bits()
4497 %{
4498 predicate(n->get_long() == 0xFFFFFFFFL);
4499 match(ConL);
4500 op_cost(20);
4501
4502 format %{ %}
4503 interface(CONST_INTER);
4504 %}
4505
4506 // Float Immediate zero
4507 operand immF0()
4508 %{
4509 predicate(jint_cast(n->getf()) == 0);
4510 match(ConF);
4511
4512 op_cost(5);
4513 format %{ %}
4514 interface(CONST_INTER);
4515 %}
4516
4517 // Float Immediate
4518 operand immF()
4519 %{
4520 match(ConF);
4521
4522 op_cost(15);
4523 format %{ %}
4524 interface(CONST_INTER);
4525 %}
4526
4527 // Double Immediate zero
4528 operand immD0()
4529 %{
4530 predicate(jlong_cast(n->getd()) == 0);
4531 match(ConD);
4532
4533 op_cost(5);
4534 format %{ %}
4535 interface(CONST_INTER);
4536 %}
4537
4538 // Double Immediate
4539 operand immD()
4540 %{
4541 match(ConD);
4542
4543 op_cost(15);
4544 format %{ %}
4545 interface(CONST_INTER);
4546 %}
4547
4548 // Immediates for special shifts (sign extend)
4549
4550 // Constants for increment
4551 operand immI_16()
4552 %{
4553 predicate(n->get_int() == 16);
4554 match(ConI);
4555
4556 format %{ %}
4557 interface(CONST_INTER);
4558 %}
4559
4560 operand immI_24()
4561 %{
4562 predicate(n->get_int() == 24);
4563 match(ConI);
4564
4565 format %{ %}
4566 interface(CONST_INTER);
4567 %}
4568
4569 // Constant for byte-wide masking
4570 operand immI_255()
4571 %{
4572 predicate(n->get_int() == 255);
4573 match(ConI);
4574
4575 format %{ %}
4576 interface(CONST_INTER);
4577 %}
4578
4579 // Constant for short-wide masking
4580 operand immI_65535()
4581 %{
4582 predicate(n->get_int() == 65535);
4583 match(ConI);
4584
4585 format %{ %}
4586 interface(CONST_INTER);
4587 %}
4588
4589 // Constant for byte-wide masking
4590 operand immL_255()
4591 %{
4592 predicate(n->get_long() == 255);
4593 match(ConL);
4594
4595 format %{ %}
4596 interface(CONST_INTER);
4597 %}
4598
4599 // Constant for short-wide masking
4600 operand immL_65535()
4601 %{
4602 predicate(n->get_long() == 65535);
4603 match(ConL);
4604
4605 format %{ %}
4606 interface(CONST_INTER);
4607 %}
4608
4609 // Register Operands
4610 // Integer Register
4611 operand rRegI()
4612 %{
4613 constraint(ALLOC_IN_RC(int_reg));
4614 match(RegI);
4615
4616 match(rax_RegI);
4617 match(rbx_RegI);
4618 match(rcx_RegI);
4619 match(rdx_RegI);
4620 match(rdi_RegI);
4621
4622 format %{ %}
4623 interface(REG_INTER);
4624 %}
4625
4626 // Special Registers
4627 operand rax_RegI()
4628 %{
4629 constraint(ALLOC_IN_RC(int_rax_reg));
4630 match(RegI);
4631 match(rRegI);
4632
4633 format %{ "RAX" %}
4634 interface(REG_INTER);
4635 %}
4636
4637 // Special Registers
4638 operand rbx_RegI()
4639 %{
4640 constraint(ALLOC_IN_RC(int_rbx_reg));
4641 match(RegI);
4642 match(rRegI);
4643
4644 format %{ "RBX" %}
4645 interface(REG_INTER);
4646 %}
4647
4648 operand rcx_RegI()
4649 %{
4650 constraint(ALLOC_IN_RC(int_rcx_reg));
4651 match(RegI);
4652 match(rRegI);
4653
4654 format %{ "RCX" %}
4655 interface(REG_INTER);
4656 %}
4657
4658 operand rdx_RegI()
4659 %{
4660 constraint(ALLOC_IN_RC(int_rdx_reg));
4661 match(RegI);
4662 match(rRegI);
4663
4664 format %{ "RDX" %}
4665 interface(REG_INTER);
4666 %}
4667
4668 operand rdi_RegI()
4669 %{
4670 constraint(ALLOC_IN_RC(int_rdi_reg));
4671 match(RegI);
4672 match(rRegI);
4673
4674 format %{ "RDI" %}
4675 interface(REG_INTER);
4676 %}
4677
4678 operand no_rcx_RegI()
4679 %{
4680 constraint(ALLOC_IN_RC(int_no_rcx_reg));
4681 match(RegI);
4682 match(rax_RegI);
4683 match(rbx_RegI);
4684 match(rdx_RegI);
4685 match(rdi_RegI);
4686
4687 format %{ %}
4688 interface(REG_INTER);
4689 %}
4690
4691 operand no_rax_rdx_RegI()
4692 %{
4693 constraint(ALLOC_IN_RC(int_no_rax_rdx_reg));
4694 match(RegI);
4695 match(rbx_RegI);
4696 match(rcx_RegI);
4697 match(rdi_RegI);
4698
4699 format %{ %}
4700 interface(REG_INTER);
4701 %}
4702
4703 // Pointer Register
4704 operand any_RegP()
4705 %{
4706 constraint(ALLOC_IN_RC(any_reg));
4707 match(RegP);
4708 match(rax_RegP);
4709 match(rbx_RegP);
4710 match(rdi_RegP);
4711 match(rsi_RegP);
4712 match(rbp_RegP);
4713 match(r15_RegP);
4714 match(rRegP);
4715
4716 format %{ %}
4717 interface(REG_INTER);
4718 %}
4719
4720 operand rRegP()
4721 %{
4722 constraint(ALLOC_IN_RC(ptr_reg));
4723 match(RegP);
4724 match(rax_RegP);
4725 match(rbx_RegP);
4726 match(rdi_RegP);
4727 match(rsi_RegP);
4728 match(rbp_RegP);
4729 match(r15_RegP); // See Q&A below about r15_RegP.
4730
4731 format %{ %}
4732 interface(REG_INTER);
4733 %}
4734
4735 operand rRegN() %{
4736 constraint(ALLOC_IN_RC(int_reg));
4737 match(RegN);
4738
4739 format %{ %}
4740 interface(REG_INTER);
4741 %}
4742
4743 // Question: Why is r15_RegP (the read-only TLS register) a match for rRegP?
4744 // Answer: Operand match rules govern the DFA as it processes instruction inputs.
4745 // It's fine for an instruction input which expects rRegP to match a r15_RegP.
4746 // The output of an instruction is controlled by the allocator, which respects
4747 // register class masks, not match rules. Unless an instruction mentions
4748 // r15_RegP or any_RegP explicitly as its output, r15 will not be considered
4749 // by the allocator as an input.
4750
4751 operand no_rax_RegP()
4752 %{
4753 constraint(ALLOC_IN_RC(ptr_no_rax_reg));
4754 match(RegP);
4755 match(rbx_RegP);
4756 match(rsi_RegP);
4757 match(rdi_RegP);
4758
4759 format %{ %}
4760 interface(REG_INTER);
4761 %}
4762
4763 operand no_rbp_RegP()
4764 %{
4765 constraint(ALLOC_IN_RC(ptr_no_rbp_reg));
4766 match(RegP);
4767 match(rbx_RegP);
4768 match(rsi_RegP);
4769 match(rdi_RegP);
4770
4771 format %{ %}
4772 interface(REG_INTER);
4773 %}
4774
4775 operand no_rax_rbx_RegP()
4776 %{
4777 constraint(ALLOC_IN_RC(ptr_no_rax_rbx_reg));
4778 match(RegP);
4779 match(rsi_RegP);
4780 match(rdi_RegP);
4781
4782 format %{ %}
4783 interface(REG_INTER);
4784 %}
4785
4786 // Special Registers
4787 // Return a pointer value
4788 operand rax_RegP()
4789 %{
4790 constraint(ALLOC_IN_RC(ptr_rax_reg));
4791 match(RegP);
4792 match(rRegP);
4793
4794 format %{ %}
4795 interface(REG_INTER);
4796 %}
4797
4798 // Special Registers
4799 // Return a compressed pointer value
4800 operand rax_RegN()
4801 %{
4802 constraint(ALLOC_IN_RC(int_rax_reg));
4803 match(RegN);
4804 match(rRegN);
4805
4806 format %{ %}
4807 interface(REG_INTER);
4808 %}
4809
4810 // Used in AtomicAdd
4811 operand rbx_RegP()
4812 %{
4813 constraint(ALLOC_IN_RC(ptr_rbx_reg));
4814 match(RegP);
4815 match(rRegP);
4816
4817 format %{ %}
4818 interface(REG_INTER);
4819 %}
4820
4821 operand rsi_RegP()
4822 %{
4823 constraint(ALLOC_IN_RC(ptr_rsi_reg));
4824 match(RegP);
4825 match(rRegP);
4826
4827 format %{ %}
4828 interface(REG_INTER);
4829 %}
4830
4831 // Used in rep stosq
4832 operand rdi_RegP()
4833 %{
4834 constraint(ALLOC_IN_RC(ptr_rdi_reg));
4835 match(RegP);
4836 match(rRegP);
4837
4838 format %{ %}
4839 interface(REG_INTER);
4840 %}
4841
4842 operand rbp_RegP()
4843 %{
4844 constraint(ALLOC_IN_RC(ptr_rbp_reg));
4845 match(RegP);
4846 match(rRegP);
4847
4848 format %{ %}
4849 interface(REG_INTER);
4850 %}
4851
4852 operand r15_RegP()
4853 %{
4854 constraint(ALLOC_IN_RC(ptr_r15_reg));
4855 match(RegP);
4856 match(rRegP);
4857
4858 format %{ %}
4859 interface(REG_INTER);
4860 %}
4861
4862 operand rRegL()
4863 %{
4864 constraint(ALLOC_IN_RC(long_reg));
4865 match(RegL);
4866 match(rax_RegL);
4867 match(rdx_RegL);
4868
4869 format %{ %}
4870 interface(REG_INTER);
4871 %}
4872
4873 // Special Registers
4874 operand no_rax_rdx_RegL()
4875 %{
4876 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
4877 match(RegL);
4878 match(rRegL);
4879
4880 format %{ %}
4881 interface(REG_INTER);
4882 %}
4883
4884 operand no_rax_RegL()
4885 %{
4886 constraint(ALLOC_IN_RC(long_no_rax_rdx_reg));
4887 match(RegL);
4888 match(rRegL);
4889 match(rdx_RegL);
4890
4891 format %{ %}
4892 interface(REG_INTER);
4893 %}
4894
4895 operand no_rcx_RegL()
4896 %{
4897 constraint(ALLOC_IN_RC(long_no_rcx_reg));
4898 match(RegL);
4899 match(rRegL);
4900
4901 format %{ %}
4902 interface(REG_INTER);
4903 %}
4904
4905 operand rax_RegL()
4906 %{
4907 constraint(ALLOC_IN_RC(long_rax_reg));
4908 match(RegL);
4909 match(rRegL);
4910
4911 format %{ "RAX" %}
4912 interface(REG_INTER);
4913 %}
4914
4915 operand rcx_RegL()
4916 %{
4917 constraint(ALLOC_IN_RC(long_rcx_reg));
4918 match(RegL);
4919 match(rRegL);
4920
4921 format %{ %}
4922 interface(REG_INTER);
4923 %}
4924
4925 operand rdx_RegL()
4926 %{
4927 constraint(ALLOC_IN_RC(long_rdx_reg));
4928 match(RegL);
4929 match(rRegL);
4930
4931 format %{ %}
4932 interface(REG_INTER);
4933 %}
4934
4935 // Flags register, used as output of compare instructions
4936 operand rFlagsReg()
4937 %{
4938 constraint(ALLOC_IN_RC(int_flags));
4939 match(RegFlags);
4940
4941 format %{ "RFLAGS" %}
4942 interface(REG_INTER);
4943 %}
4944
4945 // Flags register, used as output of FLOATING POINT compare instructions
4946 operand rFlagsRegU()
4947 %{
4948 constraint(ALLOC_IN_RC(int_flags));
4949 match(RegFlags);
4950
4951 format %{ "RFLAGS_U" %}
4952 interface(REG_INTER);
4953 %}
4954
4955 operand rFlagsRegUCF() %{
4956 constraint(ALLOC_IN_RC(int_flags));
4957 match(RegFlags);
4958 predicate(false);
4959
4960 format %{ "RFLAGS_U_CF" %}
4961 interface(REG_INTER);
4962 %}
4963
4964 // Float register operands
4965 operand regF()
4966 %{
4967 constraint(ALLOC_IN_RC(float_reg));
4968 match(RegF);
4969
4970 format %{ %}
4971 interface(REG_INTER);
4972 %}
4973
4974 // Double register operands
4975 operand regD()
4976 %{
4977 constraint(ALLOC_IN_RC(double_reg));
4978 match(RegD);
4979
4980 format %{ %}
4981 interface(REG_INTER);
4982 %}
4983
4984
4985 //----------Memory Operands----------------------------------------------------
4986 // Direct Memory Operand
4987 // operand direct(immP addr)
4988 // %{
4989 // match(addr);
4990
4991 // format %{ "[$addr]" %}
4992 // interface(MEMORY_INTER) %{
4993 // base(0xFFFFFFFF);
4994 // index(0x4);
4995 // scale(0x0);
4996 // disp($addr);
4997 // %}
4998 // %}
4999
5000 // Indirect Memory Operand
5001 operand indirect(any_RegP reg)
5002 %{
5003 constraint(ALLOC_IN_RC(ptr_reg));
5004 match(reg);
5005
5006 format %{ "[$reg]" %}
5007 interface(MEMORY_INTER) %{
5008 base($reg);
5009 index(0x4);
5010 scale(0x0);
5011 disp(0x0);
5012 %}
5013 %}
5014
5015 // Indirect Memory Plus Short Offset Operand
5016 operand indOffset8(any_RegP reg, immL8 off)
5017 %{
5018 constraint(ALLOC_IN_RC(ptr_reg));
5019 match(AddP reg off);
5020
5021 format %{ "[$reg + $off (8-bit)]" %}
5022 interface(MEMORY_INTER) %{
5023 base($reg);
5024 index(0x4);
5025 scale(0x0);
5026 disp($off);
5027 %}
5028 %}
5029
5030 // Indirect Memory Plus Long Offset Operand
5031 operand indOffset32(any_RegP reg, immL32 off)
5032 %{
5033 constraint(ALLOC_IN_RC(ptr_reg));
5034 match(AddP reg off);
5035
5036 format %{ "[$reg + $off (32-bit)]" %}
5037 interface(MEMORY_INTER) %{
5038 base($reg);
5039 index(0x4);
5040 scale(0x0);
5041 disp($off);
5042 %}
5043 %}
5044
5045 // Indirect Memory Plus Index Register Plus Offset Operand
5046 operand indIndexOffset(any_RegP reg, rRegL lreg, immL32 off)
5047 %{
5048 constraint(ALLOC_IN_RC(ptr_reg));
5049 match(AddP (AddP reg lreg) off);
5050
5051 op_cost(10);
5052 format %{"[$reg + $off + $lreg]" %}
5053 interface(MEMORY_INTER) %{
5054 base($reg);
5055 index($lreg);
5056 scale(0x0);
5057 disp($off);
5058 %}
5059 %}
5060
5061 // Indirect Memory Plus Index Register Plus Offset Operand
5062 operand indIndex(any_RegP reg, rRegL lreg)
5063 %{
5064 constraint(ALLOC_IN_RC(ptr_reg));
5065 match(AddP reg lreg);
5066
5067 op_cost(10);
5068 format %{"[$reg + $lreg]" %}
5069 interface(MEMORY_INTER) %{
5070 base($reg);
5071 index($lreg);
5072 scale(0x0);
5073 disp(0x0);
5074 %}
5075 %}
5076
5077 // Indirect Memory Times Scale Plus Index Register
5078 operand indIndexScale(any_RegP reg, rRegL lreg, immI2 scale)
5079 %{
5080 constraint(ALLOC_IN_RC(ptr_reg));
5081 match(AddP reg (LShiftL lreg scale));
5082
5083 op_cost(10);
5084 format %{"[$reg + $lreg << $scale]" %}
5085 interface(MEMORY_INTER) %{
5086 base($reg);
5087 index($lreg);
5088 scale($scale);
5089 disp(0x0);
5090 %}
5091 %}
5092
5093 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5094 operand indIndexScaleOffset(any_RegP reg, immL32 off, rRegL lreg, immI2 scale)
5095 %{
5096 constraint(ALLOC_IN_RC(ptr_reg));
5097 match(AddP (AddP reg (LShiftL lreg scale)) off);
5098
5099 op_cost(10);
5100 format %{"[$reg + $off + $lreg << $scale]" %}
5101 interface(MEMORY_INTER) %{
5102 base($reg);
5103 index($lreg);
5104 scale($scale);
5105 disp($off);
5106 %}
5107 %}
5108
5109 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
5110 operand indPosIndexScaleOffset(any_RegP reg, immL32 off, rRegI idx, immI2 scale)
5111 %{
5112 constraint(ALLOC_IN_RC(ptr_reg));
5113 predicate(n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
5114 match(AddP (AddP reg (LShiftL (ConvI2L idx) scale)) off);
5115
5116 op_cost(10);
5117 format %{"[$reg + $off + $idx << $scale]" %}
5118 interface(MEMORY_INTER) %{
5119 base($reg);
5120 index($idx);
5121 scale($scale);
5122 disp($off);
5123 %}
5124 %}
5125
5126 // Indirect Narrow Oop Plus Offset Operand
5127 // Note: x86 architecture doesn't support "scale * index + offset" without a base
5128 // we can't free r12 even with Universe::narrow_oop_base() == NULL.
5129 operand indCompressedOopOffset(rRegN reg, immL32 off) %{
5130 predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0));
5131 constraint(ALLOC_IN_RC(ptr_reg));
5132 match(AddP (DecodeN reg) off);
5133
5134 op_cost(10);
5135 format %{"[R12 + $reg << 3 + $off] (compressed oop addressing)" %}
5136 interface(MEMORY_INTER) %{
5137 base(0xc); // R12
5138 index($reg);
5139 scale(0x3);
5140 disp($off);
5141 %}
5142 %}
5143
5144 // Indirect Memory Operand
5145 operand indirectNarrow(rRegN reg)
5146 %{
5147 predicate(Universe::narrow_oop_shift() == 0);
5148 constraint(ALLOC_IN_RC(ptr_reg));
5149 match(DecodeN reg);
5150
5151 format %{ "[$reg]" %}
5152 interface(MEMORY_INTER) %{
5153 base($reg);
5154 index(0x4);
5155 scale(0x0);
5156 disp(0x0);
5157 %}
5158 %}
5159
5160 // Indirect Memory Plus Short Offset Operand
5161 operand indOffset8Narrow(rRegN reg, immL8 off)
5162 %{
5163 predicate(Universe::narrow_oop_shift() == 0);
5164 constraint(ALLOC_IN_RC(ptr_reg));
5165 match(AddP (DecodeN reg) off);
5166
5167 format %{ "[$reg + $off (8-bit)]" %}
5168 interface(MEMORY_INTER) %{
5169 base($reg);
5170 index(0x4);
5171 scale(0x0);
5172 disp($off);
5173 %}
5174 %}
5175
5176 // Indirect Memory Plus Long Offset Operand
5177 operand indOffset32Narrow(rRegN reg, immL32 off)
5178 %{
5179 predicate(Universe::narrow_oop_shift() == 0);
5180 constraint(ALLOC_IN_RC(ptr_reg));
5181 match(AddP (DecodeN reg) off);
5182
5183 format %{ "[$reg + $off (32-bit)]" %}
5184 interface(MEMORY_INTER) %{
5185 base($reg);
5186 index(0x4);
5187 scale(0x0);
5188 disp($off);
5189 %}
5190 %}
5191
5192 // Indirect Memory Plus Index Register Plus Offset Operand
5193 operand indIndexOffsetNarrow(rRegN reg, rRegL lreg, immL32 off)
5194 %{
5195 predicate(Universe::narrow_oop_shift() == 0);
5196 constraint(ALLOC_IN_RC(ptr_reg));
5197 match(AddP (AddP (DecodeN reg) lreg) off);
5198
5199 op_cost(10);
5200 format %{"[$reg + $off + $lreg]" %}
5201 interface(MEMORY_INTER) %{
5202 base($reg);
5203 index($lreg);
5204 scale(0x0);
5205 disp($off);
5206 %}
5207 %}
5208
5209 // Indirect Memory Plus Index Register Plus Offset Operand
5210 operand indIndexNarrow(rRegN reg, rRegL lreg)
5211 %{
5212 predicate(Universe::narrow_oop_shift() == 0);
5213 constraint(ALLOC_IN_RC(ptr_reg));
5214 match(AddP (DecodeN reg) lreg);
5215
5216 op_cost(10);
5217 format %{"[$reg + $lreg]" %}
5218 interface(MEMORY_INTER) %{
5219 base($reg);
5220 index($lreg);
5221 scale(0x0);
5222 disp(0x0);
5223 %}
5224 %}
5225
5226 // Indirect Memory Times Scale Plus Index Register
5227 operand indIndexScaleNarrow(rRegN reg, rRegL lreg, immI2 scale)
5228 %{
5229 predicate(Universe::narrow_oop_shift() == 0);
5230 constraint(ALLOC_IN_RC(ptr_reg));
5231 match(AddP (DecodeN reg) (LShiftL lreg scale));
5232
5233 op_cost(10);
5234 format %{"[$reg + $lreg << $scale]" %}
5235 interface(MEMORY_INTER) %{
5236 base($reg);
5237 index($lreg);
5238 scale($scale);
5239 disp(0x0);
5240 %}
5241 %}
5242
5243 // Indirect Memory Times Scale Plus Index Register Plus Offset Operand
5244 operand indIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegL lreg, immI2 scale)
5245 %{
5246 predicate(Universe::narrow_oop_shift() == 0);
5247 constraint(ALLOC_IN_RC(ptr_reg));
5248 match(AddP (AddP (DecodeN reg) (LShiftL lreg scale)) off);
5249
5250 op_cost(10);
5251 format %{"[$reg + $off + $lreg << $scale]" %}
5252 interface(MEMORY_INTER) %{
5253 base($reg);
5254 index($lreg);
5255 scale($scale);
5256 disp($off);
5257 %}
5258 %}
5259
5260 // Indirect Memory Times Scale Plus Positive Index Register Plus Offset Operand
5261 operand indPosIndexScaleOffsetNarrow(rRegN reg, immL32 off, rRegI idx, immI2 scale)
5262 %{
5263 constraint(ALLOC_IN_RC(ptr_reg));
5264 predicate(Universe::narrow_oop_shift() == 0 && n->in(2)->in(3)->in(1)->as_Type()->type()->is_long()->_lo >= 0);
5265 match(AddP (AddP (DecodeN reg) (LShiftL (ConvI2L idx) scale)) off);
5266
5267 op_cost(10);
5268 format %{"[$reg + $off + $idx << $scale]" %}
5269 interface(MEMORY_INTER) %{
5270 base($reg);
5271 index($idx);
5272 scale($scale);
5273 disp($off);
5274 %}
5275 %}
5276
5277
5278 //----------Special Memory Operands--------------------------------------------
5279 // Stack Slot Operand - This operand is used for loading and storing temporary
5280 // values on the stack where a match requires a value to
5281 // flow through memory.
5282 operand stackSlotP(sRegP reg)
5283 %{
5284 constraint(ALLOC_IN_RC(stack_slots));
5285 // No match rule because this operand is only generated in matching
5286
5287 format %{ "[$reg]" %}
5288 interface(MEMORY_INTER) %{
5289 base(0x4); // RSP
5290 index(0x4); // No Index
5291 scale(0x0); // No Scale
5292 disp($reg); // Stack Offset
5293 %}
5294 %}
5295
5296 operand stackSlotI(sRegI reg)
5297 %{
5298 constraint(ALLOC_IN_RC(stack_slots));
5299 // No match rule because this operand is only generated in matching
5300
5301 format %{ "[$reg]" %}
5302 interface(MEMORY_INTER) %{
5303 base(0x4); // RSP
5304 index(0x4); // No Index
5305 scale(0x0); // No Scale
5306 disp($reg); // Stack Offset
5307 %}
5308 %}
5309
5310 operand stackSlotF(sRegF reg)
5311 %{
5312 constraint(ALLOC_IN_RC(stack_slots));
5313 // No match rule because this operand is only generated in matching
5314
5315 format %{ "[$reg]" %}
5316 interface(MEMORY_INTER) %{
5317 base(0x4); // RSP
5318 index(0x4); // No Index
5319 scale(0x0); // No Scale
5320 disp($reg); // Stack Offset
5321 %}
5322 %}
5323
5324 operand stackSlotD(sRegD reg)
5325 %{
5326 constraint(ALLOC_IN_RC(stack_slots));
5327 // No match rule because this operand is only generated in matching
5328
5329 format %{ "[$reg]" %}
5330 interface(MEMORY_INTER) %{
5331 base(0x4); // RSP
5332 index(0x4); // No Index
5333 scale(0x0); // No Scale
5334 disp($reg); // Stack Offset
5335 %}
5336 %}
5337 operand stackSlotL(sRegL reg)
5338 %{
5339 constraint(ALLOC_IN_RC(stack_slots));
5340 // No match rule because this operand is only generated in matching
5341
5342 format %{ "[$reg]" %}
5343 interface(MEMORY_INTER) %{
5344 base(0x4); // RSP
5345 index(0x4); // No Index
5346 scale(0x0); // No Scale
5347 disp($reg); // Stack Offset
5348 %}
5349 %}
5350
5351 //----------Conditional Branch Operands----------------------------------------
5352 // Comparison Op - This is the operation of the comparison, and is limited to
5353 // the following set of codes:
5354 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
5355 //
5356 // Other attributes of the comparison, such as unsignedness, are specified
5357 // by the comparison instruction that sets a condition code flags register.
5358 // That result is represented by a flags operand whose subtype is appropriate
5359 // to the unsignedness (etc.) of the comparison.
5360 //
5361 // Later, the instruction which matches both the Comparison Op (a Bool) and
5362 // the flags (produced by the Cmp) specifies the coding of the comparison op
5363 // by matching a specific subtype of Bool operand below, such as cmpOpU.
5364
5365 // Comparision Code
5366 operand cmpOp()
5367 %{
5368 match(Bool);
5369
5370 format %{ "" %}
5371 interface(COND_INTER) %{
5372 equal(0x4, "e");
5373 not_equal(0x5, "ne");
5374 less(0xC, "l");
5375 greater_equal(0xD, "ge");
5376 less_equal(0xE, "le");
5377 greater(0xF, "g");
5378 %}
5379 %}
5380
5381 // Comparison Code, unsigned compare. Used by FP also, with
5382 // C2 (unordered) turned into GT or LT already. The other bits
5383 // C0 and C3 are turned into Carry & Zero flags.
5384 operand cmpOpU()
5385 %{
5386 match(Bool);
5387
5388 format %{ "" %}
5389 interface(COND_INTER) %{
5390 equal(0x4, "e");
5391 not_equal(0x5, "ne");
5392 less(0x2, "b");
5393 greater_equal(0x3, "nb");
5394 less_equal(0x6, "be");
5395 greater(0x7, "nbe");
5396 %}
5397 %}
5398
5399
5400 // Floating comparisons that don't require any fixup for the unordered case
5401 operand cmpOpUCF() %{
5402 match(Bool);
5403 predicate(n->as_Bool()->_test._test == BoolTest::lt ||
5404 n->as_Bool()->_test._test == BoolTest::ge ||
5405 n->as_Bool()->_test._test == BoolTest::le ||
5406 n->as_Bool()->_test._test == BoolTest::gt);
5407 format %{ "" %}
5408 interface(COND_INTER) %{
5409 equal(0x4, "e");
5410 not_equal(0x5, "ne");
5411 less(0x2, "b");
5412 greater_equal(0x3, "nb");
5413 less_equal(0x6, "be");
5414 greater(0x7, "nbe");
5415 %}
5416 %}
5417
5418
5419 // Floating comparisons that can be fixed up with extra conditional jumps
5420 operand cmpOpUCF2() %{
5421 match(Bool);
5422 predicate(n->as_Bool()->_test._test == BoolTest::ne ||
5423 n->as_Bool()->_test._test == BoolTest::eq);
5424 format %{ "" %}
5425 interface(COND_INTER) %{
5426 equal(0x4, "e");
5427 not_equal(0x5, "ne");
5428 less(0x2, "b");
5429 greater_equal(0x3, "nb");
5430 less_equal(0x6, "be");
5431 greater(0x7, "nbe");
5432 %}
5433 %}
5434
5435
5436 //----------OPERAND CLASSES----------------------------------------------------
5437 // Operand Classes are groups of operands that are used as to simplify
5438 // instruction definitions by not requiring the AD writer to specify separate
5439 // instructions for every form of operand when the instruction accepts
5440 // multiple operand types with the same basic encoding and format. The classic
5441 // case of this is memory operands.
5442
5443 opclass memory(indirect, indOffset8, indOffset32, indIndexOffset, indIndex,
5444 indIndexScale, indIndexScaleOffset, indPosIndexScaleOffset,
5445 indCompressedOopOffset,
5446 indirectNarrow, indOffset8Narrow, indOffset32Narrow,
5447 indIndexOffsetNarrow, indIndexNarrow, indIndexScaleNarrow,
5448 indIndexScaleOffsetNarrow, indPosIndexScaleOffsetNarrow);
5449
5450 //----------PIPELINE-----------------------------------------------------------
5451 // Rules which define the behavior of the target architectures pipeline.
5452 pipeline %{
5453
5454 //----------ATTRIBUTES---------------------------------------------------------
5455 attributes %{
5456 variable_size_instructions; // Fixed size instructions
5457 max_instructions_per_bundle = 3; // Up to 3 instructions per bundle
5458 instruction_unit_size = 1; // An instruction is 1 bytes long
5459 instruction_fetch_unit_size = 16; // The processor fetches one line
5460 instruction_fetch_units = 1; // of 16 bytes
5461
5462 // List of nop instructions
5463 nops( MachNop );
5464 %}
5465
5466 //----------RESOURCES----------------------------------------------------------
5467 // Resources are the functional units available to the machine
5468
5469 // Generic P2/P3 pipeline
5470 // 3 decoders, only D0 handles big operands; a "bundle" is the limit of
5471 // 3 instructions decoded per cycle.
5472 // 2 load/store ops per cycle, 1 branch, 1 FPU,
5473 // 3 ALU op, only ALU0 handles mul instructions.
5474 resources( D0, D1, D2, DECODE = D0 | D1 | D2,
5475 MS0, MS1, MS2, MEM = MS0 | MS1 | MS2,
5476 BR, FPU,
5477 ALU0, ALU1, ALU2, ALU = ALU0 | ALU1 | ALU2);
5478
5479 //----------PIPELINE DESCRIPTION-----------------------------------------------
5480 // Pipeline Description specifies the stages in the machine's pipeline
5481
5482 // Generic P2/P3 pipeline
5483 pipe_desc(S0, S1, S2, S3, S4, S5);
5484
5485 //----------PIPELINE CLASSES---------------------------------------------------
5486 // Pipeline Classes describe the stages in which input and output are
5487 // referenced by the hardware pipeline.
5488
5489 // Naming convention: ialu or fpu
5490 // Then: _reg
5491 // Then: _reg if there is a 2nd register
5492 // Then: _long if it's a pair of instructions implementing a long
5493 // Then: _fat if it requires the big decoder
5494 // Or: _mem if it requires the big decoder and a memory unit.
5495
5496 // Integer ALU reg operation
5497 pipe_class ialu_reg(rRegI dst)
5498 %{
5499 single_instruction;
5500 dst : S4(write);
5501 dst : S3(read);
5502 DECODE : S0; // any decoder
5503 ALU : S3; // any alu
5504 %}
5505
5506 // Long ALU reg operation
5507 pipe_class ialu_reg_long(rRegL dst)
5508 %{
5509 instruction_count(2);
5510 dst : S4(write);
5511 dst : S3(read);
5512 DECODE : S0(2); // any 2 decoders
5513 ALU : S3(2); // both alus
5514 %}
5515
5516 // Integer ALU reg operation using big decoder
5517 pipe_class ialu_reg_fat(rRegI dst)
5518 %{
5519 single_instruction;
5520 dst : S4(write);
5521 dst : S3(read);
5522 D0 : S0; // big decoder only
5523 ALU : S3; // any alu
5524 %}
5525
5526 // Long ALU reg operation using big decoder
5527 pipe_class ialu_reg_long_fat(rRegL dst)
5528 %{
5529 instruction_count(2);
5530 dst : S4(write);
5531 dst : S3(read);
5532 D0 : S0(2); // big decoder only; twice
5533 ALU : S3(2); // any 2 alus
5534 %}
5535
5536 // Integer ALU reg-reg operation
5537 pipe_class ialu_reg_reg(rRegI dst, rRegI src)
5538 %{
5539 single_instruction;
5540 dst : S4(write);
5541 src : S3(read);
5542 DECODE : S0; // any decoder
5543 ALU : S3; // any alu
5544 %}
5545
5546 // Long ALU reg-reg operation
5547 pipe_class ialu_reg_reg_long(rRegL dst, rRegL src)
5548 %{
5549 instruction_count(2);
5550 dst : S4(write);
5551 src : S3(read);
5552 DECODE : S0(2); // any 2 decoders
5553 ALU : S3(2); // both alus
5554 %}
5555
5556 // Integer ALU reg-reg operation
5557 pipe_class ialu_reg_reg_fat(rRegI dst, memory src)
5558 %{
5559 single_instruction;
5560 dst : S4(write);
5561 src : S3(read);
5562 D0 : S0; // big decoder only
5563 ALU : S3; // any alu
5564 %}
5565
5566 // Long ALU reg-reg operation
5567 pipe_class ialu_reg_reg_long_fat(rRegL dst, rRegL src)
5568 %{
5569 instruction_count(2);
5570 dst : S4(write);
5571 src : S3(read);
5572 D0 : S0(2); // big decoder only; twice
5573 ALU : S3(2); // both alus
5574 %}
5575
5576 // Integer ALU reg-mem operation
5577 pipe_class ialu_reg_mem(rRegI dst, memory mem)
5578 %{
5579 single_instruction;
5580 dst : S5(write);
5581 mem : S3(read);
5582 D0 : S0; // big decoder only
5583 ALU : S4; // any alu
5584 MEM : S3; // any mem
5585 %}
5586
5587 // Integer mem operation (prefetch)
5588 pipe_class ialu_mem(memory mem)
5589 %{
5590 single_instruction;
5591 mem : S3(read);
5592 D0 : S0; // big decoder only
5593 MEM : S3; // any mem
5594 %}
5595
5596 // Integer Store to Memory
5597 pipe_class ialu_mem_reg(memory mem, rRegI src)
5598 %{
5599 single_instruction;
5600 mem : S3(read);
5601 src : S5(read);
5602 D0 : S0; // big decoder only
5603 ALU : S4; // any alu
5604 MEM : S3;
5605 %}
5606
5607 // // Long Store to Memory
5608 // pipe_class ialu_mem_long_reg(memory mem, rRegL src)
5609 // %{
5610 // instruction_count(2);
5611 // mem : S3(read);
5612 // src : S5(read);
5613 // D0 : S0(2); // big decoder only; twice
5614 // ALU : S4(2); // any 2 alus
5615 // MEM : S3(2); // Both mems
5616 // %}
5617
5618 // Integer Store to Memory
5619 pipe_class ialu_mem_imm(memory mem)
5620 %{
5621 single_instruction;
5622 mem : S3(read);
5623 D0 : S0; // big decoder only
5624 ALU : S4; // any alu
5625 MEM : S3;
5626 %}
5627
5628 // Integer ALU0 reg-reg operation
5629 pipe_class ialu_reg_reg_alu0(rRegI dst, rRegI src)
5630 %{
5631 single_instruction;
5632 dst : S4(write);
5633 src : S3(read);
5634 D0 : S0; // Big decoder only
5635 ALU0 : S3; // only alu0
5636 %}
5637
5638 // Integer ALU0 reg-mem operation
5639 pipe_class ialu_reg_mem_alu0(rRegI dst, memory mem)
5640 %{
5641 single_instruction;
5642 dst : S5(write);
5643 mem : S3(read);
5644 D0 : S0; // big decoder only
5645 ALU0 : S4; // ALU0 only
5646 MEM : S3; // any mem
5647 %}
5648
5649 // Integer ALU reg-reg operation
5650 pipe_class ialu_cr_reg_reg(rFlagsReg cr, rRegI src1, rRegI src2)
5651 %{
5652 single_instruction;
5653 cr : S4(write);
5654 src1 : S3(read);
5655 src2 : S3(read);
5656 DECODE : S0; // any decoder
5657 ALU : S3; // any alu
5658 %}
5659
5660 // Integer ALU reg-imm operation
5661 pipe_class ialu_cr_reg_imm(rFlagsReg cr, rRegI src1)
5662 %{
5663 single_instruction;
5664 cr : S4(write);
5665 src1 : S3(read);
5666 DECODE : S0; // any decoder
5667 ALU : S3; // any alu
5668 %}
5669
5670 // Integer ALU reg-mem operation
5671 pipe_class ialu_cr_reg_mem(rFlagsReg cr, rRegI src1, memory src2)
5672 %{
5673 single_instruction;
5674 cr : S4(write);
5675 src1 : S3(read);
5676 src2 : S3(read);
5677 D0 : S0; // big decoder only
5678 ALU : S4; // any alu
5679 MEM : S3;
5680 %}
5681
5682 // Conditional move reg-reg
5683 pipe_class pipe_cmplt( rRegI p, rRegI q, rRegI y)
5684 %{
5685 instruction_count(4);
5686 y : S4(read);
5687 q : S3(read);
5688 p : S3(read);
5689 DECODE : S0(4); // any decoder
5690 %}
5691
5692 // Conditional move reg-reg
5693 pipe_class pipe_cmov_reg( rRegI dst, rRegI src, rFlagsReg cr)
5694 %{
5695 single_instruction;
5696 dst : S4(write);
5697 src : S3(read);
5698 cr : S3(read);
5699 DECODE : S0; // any decoder
5700 %}
5701
5702 // Conditional move reg-mem
5703 pipe_class pipe_cmov_mem( rFlagsReg cr, rRegI dst, memory src)
5704 %{
5705 single_instruction;
5706 dst : S4(write);
5707 src : S3(read);
5708 cr : S3(read);
5709 DECODE : S0; // any decoder
5710 MEM : S3;
5711 %}
5712
5713 // Conditional move reg-reg long
5714 pipe_class pipe_cmov_reg_long( rFlagsReg cr, rRegL dst, rRegL src)
5715 %{
5716 single_instruction;
5717 dst : S4(write);
5718 src : S3(read);
5719 cr : S3(read);
5720 DECODE : S0(2); // any 2 decoders
5721 %}
5722
5723 // XXX
5724 // // Conditional move double reg-reg
5725 // pipe_class pipe_cmovD_reg( rFlagsReg cr, regDPR1 dst, regD src)
5726 // %{
5727 // single_instruction;
5728 // dst : S4(write);
5729 // src : S3(read);
5730 // cr : S3(read);
5731 // DECODE : S0; // any decoder
5732 // %}
5733
5734 // Float reg-reg operation
5735 pipe_class fpu_reg(regD dst)
5736 %{
5737 instruction_count(2);
5738 dst : S3(read);
5739 DECODE : S0(2); // any 2 decoders
5740 FPU : S3;
5741 %}
5742
5743 // Float reg-reg operation
5744 pipe_class fpu_reg_reg(regD dst, regD src)
5745 %{
5746 instruction_count(2);
5747 dst : S4(write);
5748 src : S3(read);
5749 DECODE : S0(2); // any 2 decoders
5750 FPU : S3;
5751 %}
5752
5753 // Float reg-reg operation
5754 pipe_class fpu_reg_reg_reg(regD dst, regD src1, regD src2)
5755 %{
5756 instruction_count(3);
5757 dst : S4(write);
5758 src1 : S3(read);
5759 src2 : S3(read);
5760 DECODE : S0(3); // any 3 decoders
5761 FPU : S3(2);
5762 %}
5763
5764 // Float reg-reg operation
5765 pipe_class fpu_reg_reg_reg_reg(regD dst, regD src1, regD src2, regD src3)
5766 %{
5767 instruction_count(4);
5768 dst : S4(write);
5769 src1 : S3(read);
5770 src2 : S3(read);
5771 src3 : S3(read);
5772 DECODE : S0(4); // any 3 decoders
5773 FPU : S3(2);
5774 %}
5775
5776 // Float reg-reg operation
5777 pipe_class fpu_reg_mem_reg_reg(regD dst, memory src1, regD src2, regD src3)
5778 %{
5779 instruction_count(4);
5780 dst : S4(write);
5781 src1 : S3(read);
5782 src2 : S3(read);
5783 src3 : S3(read);
5784 DECODE : S1(3); // any 3 decoders
5785 D0 : S0; // Big decoder only
5786 FPU : S3(2);
5787 MEM : S3;
5788 %}
5789
5790 // Float reg-mem operation
5791 pipe_class fpu_reg_mem(regD dst, memory mem)
5792 %{
5793 instruction_count(2);
5794 dst : S5(write);
5795 mem : S3(read);
5796 D0 : S0; // big decoder only
5797 DECODE : S1; // any decoder for FPU POP
5798 FPU : S4;
5799 MEM : S3; // any mem
5800 %}
5801
5802 // Float reg-mem operation
5803 pipe_class fpu_reg_reg_mem(regD dst, regD src1, memory mem)
5804 %{
5805 instruction_count(3);
5806 dst : S5(write);
5807 src1 : S3(read);
5808 mem : S3(read);
5809 D0 : S0; // big decoder only
5810 DECODE : S1(2); // any decoder for FPU POP
5811 FPU : S4;
5812 MEM : S3; // any mem
5813 %}
5814
5815 // Float mem-reg operation
5816 pipe_class fpu_mem_reg(memory mem, regD src)
5817 %{
5818 instruction_count(2);
5819 src : S5(read);
5820 mem : S3(read);
5821 DECODE : S0; // any decoder for FPU PUSH
5822 D0 : S1; // big decoder only
5823 FPU : S4;
5824 MEM : S3; // any mem
5825 %}
5826
5827 pipe_class fpu_mem_reg_reg(memory mem, regD src1, regD src2)
5828 %{
5829 instruction_count(3);
5830 src1 : S3(read);
5831 src2 : S3(read);
5832 mem : S3(read);
5833 DECODE : S0(2); // any decoder for FPU PUSH
5834 D0 : S1; // big decoder only
5835 FPU : S4;
5836 MEM : S3; // any mem
5837 %}
5838
5839 pipe_class fpu_mem_reg_mem(memory mem, regD src1, memory src2)
5840 %{
5841 instruction_count(3);
5842 src1 : S3(read);
5843 src2 : S3(read);
5844 mem : S4(read);
5845 DECODE : S0; // any decoder for FPU PUSH
5846 D0 : S0(2); // big decoder only
5847 FPU : S4;
5848 MEM : S3(2); // any mem
5849 %}
5850
5851 pipe_class fpu_mem_mem(memory dst, memory src1)
5852 %{
5853 instruction_count(2);
5854 src1 : S3(read);
5855 dst : S4(read);
5856 D0 : S0(2); // big decoder only
5857 MEM : S3(2); // any mem
5858 %}
5859
5860 pipe_class fpu_mem_mem_mem(memory dst, memory src1, memory src2)
5861 %{
5862 instruction_count(3);
5863 src1 : S3(read);
5864 src2 : S3(read);
5865 dst : S4(read);
5866 D0 : S0(3); // big decoder only
5867 FPU : S4;
5868 MEM : S3(3); // any mem
5869 %}
5870
5871 pipe_class fpu_mem_reg_con(memory mem, regD src1)
5872 %{
5873 instruction_count(3);
5874 src1 : S4(read);
5875 mem : S4(read);
5876 DECODE : S0; // any decoder for FPU PUSH
5877 D0 : S0(2); // big decoder only
5878 FPU : S4;
5879 MEM : S3(2); // any mem
5880 %}
5881
5882 // Float load constant
5883 pipe_class fpu_reg_con(regD dst)
5884 %{
5885 instruction_count(2);
5886 dst : S5(write);
5887 D0 : S0; // big decoder only for the load
5888 DECODE : S1; // any decoder for FPU POP
5889 FPU : S4;
5890 MEM : S3; // any mem
5891 %}
5892
5893 // Float load constant
5894 pipe_class fpu_reg_reg_con(regD dst, regD src)
5895 %{
5896 instruction_count(3);
5897 dst : S5(write);
5898 src : S3(read);
5899 D0 : S0; // big decoder only for the load
5900 DECODE : S1(2); // any decoder for FPU POP
5901 FPU : S4;
5902 MEM : S3; // any mem
5903 %}
5904
5905 // UnConditional branch
5906 pipe_class pipe_jmp(label labl)
5907 %{
5908 single_instruction;
5909 BR : S3;
5910 %}
5911
5912 // Conditional branch
5913 pipe_class pipe_jcc(cmpOp cmp, rFlagsReg cr, label labl)
5914 %{
5915 single_instruction;
5916 cr : S1(read);
5917 BR : S3;
5918 %}
5919
5920 // Allocation idiom
5921 pipe_class pipe_cmpxchg(rRegP dst, rRegP heap_ptr)
5922 %{
5923 instruction_count(1); force_serialization;
5924 fixed_latency(6);
5925 heap_ptr : S3(read);
5926 DECODE : S0(3);
5927 D0 : S2;
5928 MEM : S3;
5929 ALU : S3(2);
5930 dst : S5(write);
5931 BR : S5;
5932 %}
5933
5934 // Generic big/slow expanded idiom
5935 pipe_class pipe_slow()
5936 %{
5937 instruction_count(10); multiple_bundles; force_serialization;
5938 fixed_latency(100);
5939 D0 : S0(2);
5940 MEM : S3(2);
5941 %}
5942
5943 // The real do-nothing guy
5944 pipe_class empty()
5945 %{
5946 instruction_count(0);
5947 %}
5948
5949 // Define the class for the Nop node
5950 define
5951 %{
5952 MachNop = empty;
5953 %}
5954
5955 %}
5956
5957 //----------INSTRUCTIONS-------------------------------------------------------
5958 //
5959 // match -- States which machine-independent subtree may be replaced
5960 // by this instruction.
5961 // ins_cost -- The estimated cost of this instruction is used by instruction
5962 // selection to identify a minimum cost tree of machine
5963 // instructions that matches a tree of machine-independent
5964 // instructions.
5965 // format -- A string providing the disassembly for this instruction.
5966 // The value of an instruction's operand may be inserted
5967 // by referring to it with a '$' prefix.
5968 // opcode -- Three instruction opcodes may be provided. These are referred
5969 // to within an encode class as $primary, $secondary, and $tertiary
5970 // rrspectively. The primary opcode is commonly used to
5971 // indicate the type of machine instruction, while secondary
5972 // and tertiary are often used for prefix options or addressing
5973 // modes.
5974 // ins_encode -- A list of encode classes with parameters. The encode class
5975 // name must have been defined in an 'enc_class' specification
5976 // in the encode section of the architecture description.
5977
5978
5979 //----------Load/Store/Move Instructions---------------------------------------
5980 //----------Load Instructions--------------------------------------------------
5981
5982 // Load Byte (8 bit signed)
5983 instruct loadB(rRegI dst, memory mem)
5984 %{
5985 match(Set dst (LoadB mem));
5986
5987 ins_cost(125);
5988 format %{ "movsbl $dst, $mem\t# byte" %}
5989
5990 ins_encode %{
5991 __ movsbl($dst$$Register, $mem$$Address);
5992 %}
5993
5994 ins_pipe(ialu_reg_mem);
5995 %}
5996
5997 // Load Byte (8 bit signed) into Long Register
5998 instruct loadB2L(rRegL dst, memory mem)
5999 %{
6000 match(Set dst (ConvI2L (LoadB mem)));
6001
6002 ins_cost(125);
6003 format %{ "movsbq $dst, $mem\t# byte -> long" %}
6004
6005 ins_encode %{
6006 __ movsbq($dst$$Register, $mem$$Address);
6007 %}
6008
6009 ins_pipe(ialu_reg_mem);
6010 %}
6011
6012 // Load Unsigned Byte (8 bit UNsigned)
6013 instruct loadUB(rRegI dst, memory mem)
6014 %{
6015 match(Set dst (LoadUB mem));
6016
6017 ins_cost(125);
6018 format %{ "movzbl $dst, $mem\t# ubyte" %}
6019
6020 ins_encode %{
6021 __ movzbl($dst$$Register, $mem$$Address);
6022 %}
6023
6024 ins_pipe(ialu_reg_mem);
6025 %}
6026
6027 // Load Unsigned Byte (8 bit UNsigned) into Long Register
6028 instruct loadUB2L(rRegL dst, memory mem)
6029 %{
6030 match(Set dst (ConvI2L (LoadUB mem)));
6031
6032 ins_cost(125);
6033 format %{ "movzbq $dst, $mem\t# ubyte -> long" %}
6034
6035 ins_encode %{
6036 __ movzbq($dst$$Register, $mem$$Address);
6037 %}
6038
6039 ins_pipe(ialu_reg_mem);
6040 %}
6041
6042 // Load Unsigned Byte (8 bit UNsigned) with a 8-bit mask into Long Register
6043 instruct loadUB2L_immI8(rRegL dst, memory mem, immI8 mask, rFlagsReg cr) %{
6044 match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
6045 effect(KILL cr);
6046
6047 format %{ "movzbq $dst, $mem\t# ubyte & 8-bit mask -> long\n\t"
6048 "andl $dst, $mask" %}
6049 ins_encode %{
6050 Register Rdst = $dst$$Register;
6051 __ movzbq(Rdst, $mem$$Address);
6052 __ andl(Rdst, $mask$$constant);
6053 %}
6054 ins_pipe(ialu_reg_mem);
6055 %}
6056
6057 // Load Short (16 bit signed)
6058 instruct loadS(rRegI dst, memory mem)
6059 %{
6060 match(Set dst (LoadS mem));
6061
6062 ins_cost(125);
6063 format %{ "movswl $dst, $mem\t# short" %}
6064
6065 ins_encode %{
6066 __ movswl($dst$$Register, $mem$$Address);
6067 %}
6068
6069 ins_pipe(ialu_reg_mem);
6070 %}
6071
6072 // Load Short (16 bit signed) to Byte (8 bit signed)
6073 instruct loadS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
6074 match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));
6075
6076 ins_cost(125);
6077 format %{ "movsbl $dst, $mem\t# short -> byte" %}
6078 ins_encode %{
6079 __ movsbl($dst$$Register, $mem$$Address);
6080 %}
6081 ins_pipe(ialu_reg_mem);
6082 %}
6083
6084 // Load Short (16 bit signed) into Long Register
6085 instruct loadS2L(rRegL dst, memory mem)
6086 %{
6087 match(Set dst (ConvI2L (LoadS mem)));
6088
6089 ins_cost(125);
6090 format %{ "movswq $dst, $mem\t# short -> long" %}
6091
6092 ins_encode %{
6093 __ movswq($dst$$Register, $mem$$Address);
6094 %}
6095
6096 ins_pipe(ialu_reg_mem);
6097 %}
6098
6099 // Load Unsigned Short/Char (16 bit UNsigned)
6100 instruct loadUS(rRegI dst, memory mem)
6101 %{
6102 match(Set dst (LoadUS mem));
6103
6104 ins_cost(125);
6105 format %{ "movzwl $dst, $mem\t# ushort/char" %}
6106
6107 ins_encode %{
6108 __ movzwl($dst$$Register, $mem$$Address);
6109 %}
6110
6111 ins_pipe(ialu_reg_mem);
6112 %}
6113
6114 // Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
6115 instruct loadUS2B(rRegI dst, memory mem, immI_24 twentyfour) %{
6116 match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));
6117
6118 ins_cost(125);
6119 format %{ "movsbl $dst, $mem\t# ushort -> byte" %}
6120 ins_encode %{
6121 __ movsbl($dst$$Register, $mem$$Address);
6122 %}
6123 ins_pipe(ialu_reg_mem);
6124 %}
6125
6126 // Load Unsigned Short/Char (16 bit UNsigned) into Long Register
6127 instruct loadUS2L(rRegL dst, memory mem)
6128 %{
6129 match(Set dst (ConvI2L (LoadUS mem)));
6130
6131 ins_cost(125);
6132 format %{ "movzwq $dst, $mem\t# ushort/char -> long" %}
6133
6134 ins_encode %{
6135 __ movzwq($dst$$Register, $mem$$Address);
6136 %}
6137
6138 ins_pipe(ialu_reg_mem);
6139 %}
6140
6141 // Load Unsigned Short/Char (16 bit UNsigned) with mask 0xFF into Long Register
6142 instruct loadUS2L_immI_255(rRegL dst, memory mem, immI_255 mask) %{
6143 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
6144
6145 format %{ "movzbq $dst, $mem\t# ushort/char & 0xFF -> long" %}
6146 ins_encode %{
6147 __ movzbq($dst$$Register, $mem$$Address);
6148 %}
6149 ins_pipe(ialu_reg_mem);
6150 %}
6151
6152 // Load Unsigned Short/Char (16 bit UNsigned) with mask into Long Register
6153 instruct loadUS2L_immI16(rRegL dst, memory mem, immI16 mask, rFlagsReg cr) %{
6154 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
6155 effect(KILL cr);
6156
6157 format %{ "movzwq $dst, $mem\t# ushort/char & 16-bit mask -> long\n\t"
6158 "andl $dst, $mask" %}
6159 ins_encode %{
6160 Register Rdst = $dst$$Register;
6161 __ movzwq(Rdst, $mem$$Address);
6162 __ andl(Rdst, $mask$$constant);
6163 %}
6164 ins_pipe(ialu_reg_mem);
6165 %}
6166
6167 // Load Integer
6168 instruct loadI(rRegI dst, memory mem)
6169 %{
6170 match(Set dst (LoadI mem));
6171
6172 ins_cost(125);
6173 format %{ "movl $dst, $mem\t# int" %}
6174
6175 ins_encode %{
6176 __ movl($dst$$Register, $mem$$Address);
6177 %}
6178
6179 ins_pipe(ialu_reg_mem);
6180 %}
6181
6182 // Load Integer (32 bit signed) to Byte (8 bit signed)
6183 instruct loadI2B(rRegI dst, memory mem, immI_24 twentyfour) %{
6184 match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));
6185
6186 ins_cost(125);
6187 format %{ "movsbl $dst, $mem\t# int -> byte" %}
6188 ins_encode %{
6189 __ movsbl($dst$$Register, $mem$$Address);
6190 %}
6191 ins_pipe(ialu_reg_mem);
6192 %}
6193
6194 // Load Integer (32 bit signed) to Unsigned Byte (8 bit UNsigned)
6195 instruct loadI2UB(rRegI dst, memory mem, immI_255 mask) %{
6196 match(Set dst (AndI (LoadI mem) mask));
6197
6198 ins_cost(125);
6199 format %{ "movzbl $dst, $mem\t# int -> ubyte" %}
6200 ins_encode %{
6201 __ movzbl($dst$$Register, $mem$$Address);
6202 %}
6203 ins_pipe(ialu_reg_mem);
6204 %}
6205
6206 // Load Integer (32 bit signed) to Short (16 bit signed)
6207 instruct loadI2S(rRegI dst, memory mem, immI_16 sixteen) %{
6208 match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));
6209
6210 ins_cost(125);
6211 format %{ "movswl $dst, $mem\t# int -> short" %}
6212 ins_encode %{
6213 __ movswl($dst$$Register, $mem$$Address);
6214 %}
6215 ins_pipe(ialu_reg_mem);
6216 %}
6217
6218 // Load Integer (32 bit signed) to Unsigned Short/Char (16 bit UNsigned)
6219 instruct loadI2US(rRegI dst, memory mem, immI_65535 mask) %{
6220 match(Set dst (AndI (LoadI mem) mask));
6221
6222 ins_cost(125);
6223 format %{ "movzwl $dst, $mem\t# int -> ushort/char" %}
6224 ins_encode %{
6225 __ movzwl($dst$$Register, $mem$$Address);
6226 %}
6227 ins_pipe(ialu_reg_mem);
6228 %}
6229
6230 // Load Integer into Long Register
6231 instruct loadI2L(rRegL dst, memory mem)
6232 %{
6233 match(Set dst (ConvI2L (LoadI mem)));
6234
6235 ins_cost(125);
6236 format %{ "movslq $dst, $mem\t# int -> long" %}
6237
6238 ins_encode %{
6239 __ movslq($dst$$Register, $mem$$Address);
6240 %}
6241
6242 ins_pipe(ialu_reg_mem);
6243 %}
6244
6245 // Load Integer with mask 0xFF into Long Register
6246 instruct loadI2L_immI_255(rRegL dst, memory mem, immI_255 mask) %{
6247 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6248
6249 format %{ "movzbq $dst, $mem\t# int & 0xFF -> long" %}
6250 ins_encode %{
6251 __ movzbq($dst$$Register, $mem$$Address);
6252 %}
6253 ins_pipe(ialu_reg_mem);
6254 %}
6255
6256 // Load Integer with mask 0xFFFF into Long Register
6257 instruct loadI2L_immI_65535(rRegL dst, memory mem, immI_65535 mask) %{
6258 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6259
6260 format %{ "movzwq $dst, $mem\t# int & 0xFFFF -> long" %}
6261 ins_encode %{
6262 __ movzwq($dst$$Register, $mem$$Address);
6263 %}
6264 ins_pipe(ialu_reg_mem);
6265 %}
6266
6267 // Load Integer with a 32-bit mask into Long Register
6268 instruct loadI2L_immI(rRegL dst, memory mem, immI mask, rFlagsReg cr) %{
6269 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
6270 effect(KILL cr);
6271
6272 format %{ "movl $dst, $mem\t# int & 32-bit mask -> long\n\t"
6273 "andl $dst, $mask" %}
6274 ins_encode %{
6275 Register Rdst = $dst$$Register;
6276 __ movl(Rdst, $mem$$Address);
6277 __ andl(Rdst, $mask$$constant);
6278 %}
6279 ins_pipe(ialu_reg_mem);
6280 %}
6281
6282 // Load Unsigned Integer into Long Register
6283 instruct loadUI2L(rRegL dst, memory mem)
6284 %{
6285 match(Set dst (LoadUI2L mem));
6286
6287 ins_cost(125);
6288 format %{ "movl $dst, $mem\t# uint -> long" %}
6289
6290 ins_encode %{
6291 __ movl($dst$$Register, $mem$$Address);
6292 %}
6293
6294 ins_pipe(ialu_reg_mem);
6295 %}
6296
6297 // Load Long
6298 instruct loadL(rRegL dst, memory mem)
6299 %{
6300 match(Set dst (LoadL mem));
6301
6302 ins_cost(125);
6303 format %{ "movq $dst, $mem\t# long" %}
6304
6305 ins_encode %{
6306 __ movq($dst$$Register, $mem$$Address);
6307 %}
6308
6309 ins_pipe(ialu_reg_mem); // XXX
6310 %}
6311
6312 // Load Range
6313 instruct loadRange(rRegI dst, memory mem)
6314 %{
6315 match(Set dst (LoadRange mem));
6316
6317 ins_cost(125); // XXX
6318 format %{ "movl $dst, $mem\t# range" %}
6319 opcode(0x8B);
6320 ins_encode(REX_reg_mem(dst, mem), OpcP, reg_mem(dst, mem));
6321 ins_pipe(ialu_reg_mem);
6322 %}
6323
6324 // Load Pointer
6325 instruct loadP(rRegP dst, memory mem)
6326 %{
6327 match(Set dst (LoadP mem));
6328
6329 ins_cost(125); // XXX
6330 format %{ "movq $dst, $mem\t# ptr" %}
6331 opcode(0x8B);
6332 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6333 ins_pipe(ialu_reg_mem); // XXX
6334 %}
6335
6336 // Load Compressed Pointer
6337 instruct loadN(rRegN dst, memory mem)
6338 %{
6339 match(Set dst (LoadN mem));
6340
6341 ins_cost(125); // XXX
6342 format %{ "movl $dst, $mem\t# compressed ptr" %}
6343 ins_encode %{
6344 __ movl($dst$$Register, $mem$$Address);
6345 %}
6346 ins_pipe(ialu_reg_mem); // XXX
6347 %}
6348
6349
6350 // Load Klass Pointer
6351 instruct loadKlass(rRegP dst, memory mem)
6352 %{
6353 match(Set dst (LoadKlass mem));
6354
6355 ins_cost(125); // XXX
6356 format %{ "movq $dst, $mem\t# class" %}
6357 opcode(0x8B);
6358 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6359 ins_pipe(ialu_reg_mem); // XXX
6360 %}
6361
6362 // Load narrow Klass Pointer
6363 instruct loadNKlass(rRegN dst, memory mem)
6364 %{
6365 match(Set dst (LoadNKlass mem));
6366
6367 ins_cost(125); // XXX
6368 format %{ "movl $dst, $mem\t# compressed klass ptr" %}
6369 ins_encode %{
6370 __ movl($dst$$Register, $mem$$Address);
6371 %}
6372 ins_pipe(ialu_reg_mem); // XXX
6373 %}
6374
6375 // Load Float
6376 instruct loadF(regF dst, memory mem)
6377 %{
6378 match(Set dst (LoadF mem));
6379
6380 ins_cost(145); // XXX
6381 format %{ "movss $dst, $mem\t# float" %}
6382 opcode(0xF3, 0x0F, 0x10);
6383 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6384 ins_pipe(pipe_slow); // XXX
6385 %}
6386
6387 // Load Double
6388 instruct loadD_partial(regD dst, memory mem)
6389 %{
6390 predicate(!UseXmmLoadAndClearUpper);
6391 match(Set dst (LoadD mem));
6392
6393 ins_cost(145); // XXX
6394 format %{ "movlpd $dst, $mem\t# double" %}
6395 opcode(0x66, 0x0F, 0x12);
6396 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6397 ins_pipe(pipe_slow); // XXX
6398 %}
6399
6400 instruct loadD(regD dst, memory mem)
6401 %{
6402 predicate(UseXmmLoadAndClearUpper);
6403 match(Set dst (LoadD mem));
6404
6405 ins_cost(145); // XXX
6406 format %{ "movsd $dst, $mem\t# double" %}
6407 opcode(0xF2, 0x0F, 0x10);
6408 ins_encode(OpcP, REX_reg_mem(dst, mem), OpcS, OpcT, reg_mem(dst, mem));
6409 ins_pipe(pipe_slow); // XXX
6410 %}
6411
6412 // Load Aligned Packed Byte to XMM register
6413 instruct loadA8B(regD dst, memory mem) %{
6414 match(Set dst (Load8B mem));
6415 ins_cost(125);
6416 format %{ "MOVQ $dst,$mem\t! packed8B" %}
6417 ins_encode( movq_ld(dst, mem));
6418 ins_pipe( pipe_slow );
6419 %}
6420
6421 // Load Aligned Packed Short to XMM register
6422 instruct loadA4S(regD dst, memory mem) %{
6423 match(Set dst (Load4S mem));
6424 ins_cost(125);
6425 format %{ "MOVQ $dst,$mem\t! packed4S" %}
6426 ins_encode( movq_ld(dst, mem));
6427 ins_pipe( pipe_slow );
6428 %}
6429
6430 // Load Aligned Packed Char to XMM register
6431 instruct loadA4C(regD dst, memory mem) %{
6432 match(Set dst (Load4C mem));
6433 ins_cost(125);
6434 format %{ "MOVQ $dst,$mem\t! packed4C" %}
6435 ins_encode( movq_ld(dst, mem));
6436 ins_pipe( pipe_slow );
6437 %}
6438
6439 // Load Aligned Packed Integer to XMM register
6440 instruct load2IU(regD dst, memory mem) %{
6441 match(Set dst (Load2I mem));
6442 ins_cost(125);
6443 format %{ "MOVQ $dst,$mem\t! packed2I" %}
6444 ins_encode( movq_ld(dst, mem));
6445 ins_pipe( pipe_slow );
6446 %}
6447
6448 // Load Aligned Packed Single to XMM
6449 instruct loadA2F(regD dst, memory mem) %{
6450 match(Set dst (Load2F mem));
6451 ins_cost(145);
6452 format %{ "MOVQ $dst,$mem\t! packed2F" %}
6453 ins_encode( movq_ld(dst, mem));
6454 ins_pipe( pipe_slow );
6455 %}
6456
6457 // Load Effective Address
6458 instruct leaP8(rRegP dst, indOffset8 mem)
6459 %{
6460 match(Set dst mem);
6461
6462 ins_cost(110); // XXX
6463 format %{ "leaq $dst, $mem\t# ptr 8" %}
6464 opcode(0x8D);
6465 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6466 ins_pipe(ialu_reg_reg_fat);
6467 %}
6468
6469 instruct leaP32(rRegP dst, indOffset32 mem)
6470 %{
6471 match(Set dst mem);
6472
6473 ins_cost(110);
6474 format %{ "leaq $dst, $mem\t# ptr 32" %}
6475 opcode(0x8D);
6476 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6477 ins_pipe(ialu_reg_reg_fat);
6478 %}
6479
6480 // instruct leaPIdx(rRegP dst, indIndex mem)
6481 // %{
6482 // match(Set dst mem);
6483
6484 // ins_cost(110);
6485 // format %{ "leaq $dst, $mem\t# ptr idx" %}
6486 // opcode(0x8D);
6487 // ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6488 // ins_pipe(ialu_reg_reg_fat);
6489 // %}
6490
6491 instruct leaPIdxOff(rRegP dst, indIndexOffset mem)
6492 %{
6493 match(Set dst mem);
6494
6495 ins_cost(110);
6496 format %{ "leaq $dst, $mem\t# ptr idxoff" %}
6497 opcode(0x8D);
6498 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6499 ins_pipe(ialu_reg_reg_fat);
6500 %}
6501
6502 instruct leaPIdxScale(rRegP dst, indIndexScale mem)
6503 %{
6504 match(Set dst mem);
6505
6506 ins_cost(110);
6507 format %{ "leaq $dst, $mem\t# ptr idxscale" %}
6508 opcode(0x8D);
6509 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6510 ins_pipe(ialu_reg_reg_fat);
6511 %}
6512
6513 instruct leaPIdxScaleOff(rRegP dst, indIndexScaleOffset mem)
6514 %{
6515 match(Set dst mem);
6516
6517 ins_cost(110);
6518 format %{ "leaq $dst, $mem\t# ptr idxscaleoff" %}
6519 opcode(0x8D);
6520 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6521 ins_pipe(ialu_reg_reg_fat);
6522 %}
6523
6524 instruct leaPPosIdxScaleOff(rRegP dst, indPosIndexScaleOffset mem)
6525 %{
6526 match(Set dst mem);
6527
6528 ins_cost(110);
6529 format %{ "leaq $dst, $mem\t# ptr posidxscaleoff" %}
6530 opcode(0x8D);
6531 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6532 ins_pipe(ialu_reg_reg_fat);
6533 %}
6534
6535 // Load Effective Address which uses Narrow (32-bits) oop
6536 instruct leaPCompressedOopOffset(rRegP dst, indCompressedOopOffset mem)
6537 %{
6538 predicate(UseCompressedOops && (Universe::narrow_oop_shift() != 0));
6539 match(Set dst mem);
6540
6541 ins_cost(110);
6542 format %{ "leaq $dst, $mem\t# ptr compressedoopoff32" %}
6543 opcode(0x8D);
6544 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6545 ins_pipe(ialu_reg_reg_fat);
6546 %}
6547
6548 instruct leaP8Narrow(rRegP dst, indOffset8Narrow mem)
6549 %{
6550 predicate(Universe::narrow_oop_shift() == 0);
6551 match(Set dst mem);
6552
6553 ins_cost(110); // XXX
6554 format %{ "leaq $dst, $mem\t# ptr off8narrow" %}
6555 opcode(0x8D);
6556 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6557 ins_pipe(ialu_reg_reg_fat);
6558 %}
6559
6560 instruct leaP32Narrow(rRegP dst, indOffset32Narrow mem)
6561 %{
6562 predicate(Universe::narrow_oop_shift() == 0);
6563 match(Set dst mem);
6564
6565 ins_cost(110);
6566 format %{ "leaq $dst, $mem\t# ptr off32narrow" %}
6567 opcode(0x8D);
6568 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6569 ins_pipe(ialu_reg_reg_fat);
6570 %}
6571
6572 instruct leaPIdxOffNarrow(rRegP dst, indIndexOffsetNarrow mem)
6573 %{
6574 predicate(Universe::narrow_oop_shift() == 0);
6575 match(Set dst mem);
6576
6577 ins_cost(110);
6578 format %{ "leaq $dst, $mem\t# ptr idxoffnarrow" %}
6579 opcode(0x8D);
6580 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6581 ins_pipe(ialu_reg_reg_fat);
6582 %}
6583
6584 instruct leaPIdxScaleNarrow(rRegP dst, indIndexScaleNarrow mem)
6585 %{
6586 predicate(Universe::narrow_oop_shift() == 0);
6587 match(Set dst mem);
6588
6589 ins_cost(110);
6590 format %{ "leaq $dst, $mem\t# ptr idxscalenarrow" %}
6591 opcode(0x8D);
6592 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6593 ins_pipe(ialu_reg_reg_fat);
6594 %}
6595
6596 instruct leaPIdxScaleOffNarrow(rRegP dst, indIndexScaleOffsetNarrow mem)
6597 %{
6598 predicate(Universe::narrow_oop_shift() == 0);
6599 match(Set dst mem);
6600
6601 ins_cost(110);
6602 format %{ "leaq $dst, $mem\t# ptr idxscaleoffnarrow" %}
6603 opcode(0x8D);
6604 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6605 ins_pipe(ialu_reg_reg_fat);
6606 %}
6607
6608 instruct leaPPosIdxScaleOffNarrow(rRegP dst, indPosIndexScaleOffsetNarrow mem)
6609 %{
6610 predicate(Universe::narrow_oop_shift() == 0);
6611 match(Set dst mem);
6612
6613 ins_cost(110);
6614 format %{ "leaq $dst, $mem\t# ptr posidxscaleoffnarrow" %}
6615 opcode(0x8D);
6616 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
6617 ins_pipe(ialu_reg_reg_fat);
6618 %}
6619
6620 instruct loadConI(rRegI dst, immI src)
6621 %{
6622 match(Set dst src);
6623
6624 format %{ "movl $dst, $src\t# int" %}
6625 ins_encode(load_immI(dst, src));
6626 ins_pipe(ialu_reg_fat); // XXX
6627 %}
6628
6629 instruct loadConI0(rRegI dst, immI0 src, rFlagsReg cr)
6630 %{
6631 match(Set dst src);
6632 effect(KILL cr);
6633
6634 ins_cost(50);
6635 format %{ "xorl $dst, $dst\t# int" %}
6636 opcode(0x33); /* + rd */
6637 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6638 ins_pipe(ialu_reg);
6639 %}
6640
6641 instruct loadConL(rRegL dst, immL src)
6642 %{
6643 match(Set dst src);
6644
6645 ins_cost(150);
6646 format %{ "movq $dst, $src\t# long" %}
6647 ins_encode(load_immL(dst, src));
6648 ins_pipe(ialu_reg);
6649 %}
6650
6651 instruct loadConL0(rRegL dst, immL0 src, rFlagsReg cr)
6652 %{
6653 match(Set dst src);
6654 effect(KILL cr);
6655
6656 ins_cost(50);
6657 format %{ "xorl $dst, $dst\t# long" %}
6658 opcode(0x33); /* + rd */
6659 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6660 ins_pipe(ialu_reg); // XXX
6661 %}
6662
6663 instruct loadConUL32(rRegL dst, immUL32 src)
6664 %{
6665 match(Set dst src);
6666
6667 ins_cost(60);
6668 format %{ "movl $dst, $src\t# long (unsigned 32-bit)" %}
6669 ins_encode(load_immUL32(dst, src));
6670 ins_pipe(ialu_reg);
6671 %}
6672
6673 instruct loadConL32(rRegL dst, immL32 src)
6674 %{
6675 match(Set dst src);
6676
6677 ins_cost(70);
6678 format %{ "movq $dst, $src\t# long (32-bit)" %}
6679 ins_encode(load_immL32(dst, src));
6680 ins_pipe(ialu_reg);
6681 %}
6682
6683 instruct loadConP(rRegP dst, immP src)
6684 %{
6685 match(Set dst src);
6686
6687 format %{ "movq $dst, $src\t# ptr" %}
6688 ins_encode(load_immP(dst, src));
6689 ins_pipe(ialu_reg_fat); // XXX
6690 %}
6691
6692 instruct loadConP0(rRegP dst, immP0 src, rFlagsReg cr)
6693 %{
6694 match(Set dst src);
6695 effect(KILL cr);
6696
6697 ins_cost(50);
6698 format %{ "xorl $dst, $dst\t# ptr" %}
6699 opcode(0x33); /* + rd */
6700 ins_encode(REX_reg_reg(dst, dst), OpcP, reg_reg(dst, dst));
6701 ins_pipe(ialu_reg);
6702 %}
6703
6704 instruct loadConP31(rRegP dst, immP31 src, rFlagsReg cr)
6705 %{
6706 match(Set dst src);
6707 effect(KILL cr);
6708
6709 ins_cost(60);
6710 format %{ "movl $dst, $src\t# ptr (positive 32-bit)" %}
6711 ins_encode(load_immP31(dst, src));
6712 ins_pipe(ialu_reg);
6713 %}
6714
6715 instruct loadConF(regF dst, immF src)
6716 %{
6717 match(Set dst src);
6718 ins_cost(125);
6719
6720 format %{ "movss $dst, [$src]" %}
6721 ins_encode(load_conF(dst, src));
6722 ins_pipe(pipe_slow);
6723 %}
6724
6725 instruct loadConN0(rRegN dst, immN0 src, rFlagsReg cr) %{
6726 match(Set dst src);
6727 effect(KILL cr);
6728 format %{ "xorq $dst, $src\t# compressed NULL ptr" %}
6729 ins_encode %{
6730 __ xorq($dst$$Register, $dst$$Register);
6731 %}
6732 ins_pipe(ialu_reg);
6733 %}
6734
6735 instruct loadConN(rRegN dst, immN src) %{
6736 match(Set dst src);
6737
6738 ins_cost(125);
6739 format %{ "movl $dst, $src\t# compressed ptr" %}
6740 ins_encode %{
6741 address con = (address)$src$$constant;
6742 if (con == NULL) {
6743 ShouldNotReachHere();
6744 } else {
6745 __ set_narrow_oop($dst$$Register, (jobject)$src$$constant);
6746 }
6747 %}
6748 ins_pipe(ialu_reg_fat); // XXX
6749 %}
6750
6751 instruct loadConF0(regF dst, immF0 src)
6752 %{
6753 match(Set dst src);
6754 ins_cost(100);
6755
6756 format %{ "xorps $dst, $dst\t# float 0.0" %}
6757 opcode(0x0F, 0x57);
6758 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
6759 ins_pipe(pipe_slow);
6760 %}
6761
6762 // Use the same format since predicate() can not be used here.
6763 instruct loadConD(regD dst, immD src)
6764 %{
6765 match(Set dst src);
6766 ins_cost(125);
6767
6768 format %{ "movsd $dst, [$src]" %}
6769 ins_encode(load_conD(dst, src));
6770 ins_pipe(pipe_slow);
6771 %}
6772
6773 instruct loadConD0(regD dst, immD0 src)
6774 %{
6775 match(Set dst src);
6776 ins_cost(100);
6777
6778 format %{ "xorpd $dst, $dst\t# double 0.0" %}
6779 opcode(0x66, 0x0F, 0x57);
6780 ins_encode(OpcP, REX_reg_reg(dst, dst), OpcS, OpcT, reg_reg(dst, dst));
6781 ins_pipe(pipe_slow);
6782 %}
6783
6784 instruct loadSSI(rRegI dst, stackSlotI src)
6785 %{
6786 match(Set dst src);
6787
6788 ins_cost(125);
6789 format %{ "movl $dst, $src\t# int stk" %}
6790 opcode(0x8B);
6791 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
6792 ins_pipe(ialu_reg_mem);
6793 %}
6794
6795 instruct loadSSL(rRegL dst, stackSlotL src)
6796 %{
6797 match(Set dst src);
6798
6799 ins_cost(125);
6800 format %{ "movq $dst, $src\t# long stk" %}
6801 opcode(0x8B);
6802 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6803 ins_pipe(ialu_reg_mem);
6804 %}
6805
6806 instruct loadSSP(rRegP dst, stackSlotP src)
6807 %{
6808 match(Set dst src);
6809
6810 ins_cost(125);
6811 format %{ "movq $dst, $src\t# ptr stk" %}
6812 opcode(0x8B);
6813 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
6814 ins_pipe(ialu_reg_mem);
6815 %}
6816
6817 instruct loadSSF(regF dst, stackSlotF src)
6818 %{
6819 match(Set dst src);
6820
6821 ins_cost(125);
6822 format %{ "movss $dst, $src\t# float stk" %}
6823 opcode(0xF3, 0x0F, 0x10);
6824 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
6825 ins_pipe(pipe_slow); // XXX
6826 %}
6827
6828 // Use the same format since predicate() can not be used here.
6829 instruct loadSSD(regD dst, stackSlotD src)
6830 %{
6831 match(Set dst src);
6832
6833 ins_cost(125);
6834 format %{ "movsd $dst, $src\t# double stk" %}
6835 ins_encode %{
6836 __ movdbl($dst$$XMMRegister, Address(rsp, $src$$disp));
6837 %}
6838 ins_pipe(pipe_slow); // XXX
6839 %}
6840
6841 // Prefetch instructions.
6842 // Must be safe to execute with invalid address (cannot fault).
6843
6844 instruct prefetchr( memory mem ) %{
6845 predicate(ReadPrefetchInstr==3);
6846 match(PrefetchRead mem);
6847 ins_cost(125);
6848
6849 format %{ "PREFETCHR $mem\t# Prefetch into level 1 cache" %}
6850 opcode(0x0F, 0x0D); /* Opcode 0F 0D /0 */
6851 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6852 ins_pipe(ialu_mem);
6853 %}
6854
6855 instruct prefetchrNTA( memory mem ) %{
6856 predicate(ReadPrefetchInstr==0);
6857 match(PrefetchRead mem);
6858 ins_cost(125);
6859
6860 format %{ "PREFETCHNTA $mem\t# Prefetch into non-temporal cache for read" %}
6861 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6862 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6863 ins_pipe(ialu_mem);
6864 %}
6865
6866 instruct prefetchrT0( memory mem ) %{
6867 predicate(ReadPrefetchInstr==1);
6868 match(PrefetchRead mem);
6869 ins_cost(125);
6870
6871 format %{ "PREFETCHT0 $mem\t# prefetch into L1 and L2 caches for read" %}
6872 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6873 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6874 ins_pipe(ialu_mem);
6875 %}
6876
6877 instruct prefetchrT2( memory mem ) %{
6878 predicate(ReadPrefetchInstr==2);
6879 match(PrefetchRead mem);
6880 ins_cost(125);
6881
6882 format %{ "PREFETCHT2 $mem\t# prefetch into L2 caches for read" %}
6883 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6884 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6885 ins_pipe(ialu_mem);
6886 %}
6887
6888 instruct prefetchw( memory mem ) %{
6889 predicate(AllocatePrefetchInstr==3);
6890 match(PrefetchWrite mem);
6891 ins_cost(125);
6892
6893 format %{ "PREFETCHW $mem\t# Prefetch into level 1 cache and mark modified" %}
6894 opcode(0x0F, 0x0D); /* Opcode 0F 0D /1 */
6895 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6896 ins_pipe(ialu_mem);
6897 %}
6898
6899 instruct prefetchwNTA( memory mem ) %{
6900 predicate(AllocatePrefetchInstr==0);
6901 match(PrefetchWrite mem);
6902 ins_cost(125);
6903
6904 format %{ "PREFETCHNTA $mem\t# Prefetch to non-temporal cache for write" %}
6905 opcode(0x0F, 0x18); /* Opcode 0F 18 /0 */
6906 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x00, mem));
6907 ins_pipe(ialu_mem);
6908 %}
6909
6910 instruct prefetchwT0( memory mem ) %{
6911 predicate(AllocatePrefetchInstr==1);
6912 match(PrefetchWrite mem);
6913 ins_cost(125);
6914
6915 format %{ "PREFETCHT0 $mem\t# Prefetch to level 1 and 2 caches for write" %}
6916 opcode(0x0F, 0x18); /* Opcode 0F 18 /1 */
6917 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x01, mem));
6918 ins_pipe(ialu_mem);
6919 %}
6920
6921 instruct prefetchwT2( memory mem ) %{
6922 predicate(AllocatePrefetchInstr==2);
6923 match(PrefetchWrite mem);
6924 ins_cost(125);
6925
6926 format %{ "PREFETCHT2 $mem\t# Prefetch to level 2 cache for write" %}
6927 opcode(0x0F, 0x18); /* Opcode 0F 18 /3 */
6928 ins_encode(REX_mem(mem), OpcP, OpcS, RM_opc_mem(0x03, mem));
6929 ins_pipe(ialu_mem);
6930 %}
6931
6932 //----------Store Instructions-------------------------------------------------
6933
6934 // Store Byte
6935 instruct storeB(memory mem, rRegI src)
6936 %{
6937 match(Set mem (StoreB mem src));
6938
6939 ins_cost(125); // XXX
6940 format %{ "movb $mem, $src\t# byte" %}
6941 opcode(0x88);
6942 ins_encode(REX_breg_mem(src, mem), OpcP, reg_mem(src, mem));
6943 ins_pipe(ialu_mem_reg);
6944 %}
6945
6946 // Store Char/Short
6947 instruct storeC(memory mem, rRegI src)
6948 %{
6949 match(Set mem (StoreC mem src));
6950
6951 ins_cost(125); // XXX
6952 format %{ "movw $mem, $src\t# char/short" %}
6953 opcode(0x89);
6954 ins_encode(SizePrefix, REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6955 ins_pipe(ialu_mem_reg);
6956 %}
6957
6958 // Store Integer
6959 instruct storeI(memory mem, rRegI src)
6960 %{
6961 match(Set mem (StoreI mem src));
6962
6963 ins_cost(125); // XXX
6964 format %{ "movl $mem, $src\t# int" %}
6965 opcode(0x89);
6966 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
6967 ins_pipe(ialu_mem_reg);
6968 %}
6969
6970 // Store Long
6971 instruct storeL(memory mem, rRegL src)
6972 %{
6973 match(Set mem (StoreL mem src));
6974
6975 ins_cost(125); // XXX
6976 format %{ "movq $mem, $src\t# long" %}
6977 opcode(0x89);
6978 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6979 ins_pipe(ialu_mem_reg); // XXX
6980 %}
6981
6982 // Store Pointer
6983 instruct storeP(memory mem, any_RegP src)
6984 %{
6985 match(Set mem (StoreP mem src));
6986
6987 ins_cost(125); // XXX
6988 format %{ "movq $mem, $src\t# ptr" %}
6989 opcode(0x89);
6990 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
6991 ins_pipe(ialu_mem_reg);
6992 %}
6993
6994 instruct storeImmP0(memory mem, immP0 zero)
6995 %{
6996 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
6997 match(Set mem (StoreP mem zero));
6998
6999 ins_cost(125); // XXX
7000 format %{ "movq $mem, R12\t# ptr (R12_heapbase==0)" %}
7001 ins_encode %{
7002 __ movq($mem$$Address, r12);
7003 %}
7004 ins_pipe(ialu_mem_reg);
7005 %}
7006
7007 // Store NULL Pointer, mark word, or other simple pointer constant.
7008 instruct storeImmP(memory mem, immP31 src)
7009 %{
7010 match(Set mem (StoreP mem src));
7011
7012 ins_cost(150); // XXX
7013 format %{ "movq $mem, $src\t# ptr" %}
7014 opcode(0xC7); /* C7 /0 */
7015 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
7016 ins_pipe(ialu_mem_imm);
7017 %}
7018
7019 // Store Compressed Pointer
7020 instruct storeN(memory mem, rRegN src)
7021 %{
7022 match(Set mem (StoreN mem src));
7023
7024 ins_cost(125); // XXX
7025 format %{ "movl $mem, $src\t# compressed ptr" %}
7026 ins_encode %{
7027 __ movl($mem$$Address, $src$$Register);
7028 %}
7029 ins_pipe(ialu_mem_reg);
7030 %}
7031
7032 instruct storeImmN0(memory mem, immN0 zero)
7033 %{
7034 predicate(Universe::narrow_oop_base() == NULL);
7035 match(Set mem (StoreN mem zero));
7036
7037 ins_cost(125); // XXX
7038 format %{ "movl $mem, R12\t# compressed ptr (R12_heapbase==0)" %}
7039 ins_encode %{
7040 __ movl($mem$$Address, r12);
7041 %}
7042 ins_pipe(ialu_mem_reg);
7043 %}
7044
7045 instruct storeImmN(memory mem, immN src)
7046 %{
7047 match(Set mem (StoreN mem src));
7048
7049 ins_cost(150); // XXX
7050 format %{ "movl $mem, $src\t# compressed ptr" %}
7051 ins_encode %{
7052 address con = (address)$src$$constant;
7053 if (con == NULL) {
7054 __ movl($mem$$Address, (int32_t)0);
7055 } else {
7056 __ set_narrow_oop($mem$$Address, (jobject)$src$$constant);
7057 }
7058 %}
7059 ins_pipe(ialu_mem_imm);
7060 %}
7061
7062 // Store Integer Immediate
7063 instruct storeImmI0(memory mem, immI0 zero)
7064 %{
7065 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7066 match(Set mem (StoreI mem zero));
7067
7068 ins_cost(125); // XXX
7069 format %{ "movl $mem, R12\t# int (R12_heapbase==0)" %}
7070 ins_encode %{
7071 __ movl($mem$$Address, r12);
7072 %}
7073 ins_pipe(ialu_mem_reg);
7074 %}
7075
7076 instruct storeImmI(memory mem, immI src)
7077 %{
7078 match(Set mem (StoreI mem src));
7079
7080 ins_cost(150);
7081 format %{ "movl $mem, $src\t# int" %}
7082 opcode(0xC7); /* C7 /0 */
7083 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
7084 ins_pipe(ialu_mem_imm);
7085 %}
7086
7087 // Store Long Immediate
7088 instruct storeImmL0(memory mem, immL0 zero)
7089 %{
7090 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7091 match(Set mem (StoreL mem zero));
7092
7093 ins_cost(125); // XXX
7094 format %{ "movq $mem, R12\t# long (R12_heapbase==0)" %}
7095 ins_encode %{
7096 __ movq($mem$$Address, r12);
7097 %}
7098 ins_pipe(ialu_mem_reg);
7099 %}
7100
7101 instruct storeImmL(memory mem, immL32 src)
7102 %{
7103 match(Set mem (StoreL mem src));
7104
7105 ins_cost(150);
7106 format %{ "movq $mem, $src\t# long" %}
7107 opcode(0xC7); /* C7 /0 */
7108 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32(src));
7109 ins_pipe(ialu_mem_imm);
7110 %}
7111
7112 // Store Short/Char Immediate
7113 instruct storeImmC0(memory mem, immI0 zero)
7114 %{
7115 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7116 match(Set mem (StoreC mem zero));
7117
7118 ins_cost(125); // XXX
7119 format %{ "movw $mem, R12\t# short/char (R12_heapbase==0)" %}
7120 ins_encode %{
7121 __ movw($mem$$Address, r12);
7122 %}
7123 ins_pipe(ialu_mem_reg);
7124 %}
7125
7126 instruct storeImmI16(memory mem, immI16 src)
7127 %{
7128 predicate(UseStoreImmI16);
7129 match(Set mem (StoreC mem src));
7130
7131 ins_cost(150);
7132 format %{ "movw $mem, $src\t# short/char" %}
7133 opcode(0xC7); /* C7 /0 Same as 32 store immediate with prefix */
7134 ins_encode(SizePrefix, REX_mem(mem), OpcP, RM_opc_mem(0x00, mem),Con16(src));
7135 ins_pipe(ialu_mem_imm);
7136 %}
7137
7138 // Store Byte Immediate
7139 instruct storeImmB0(memory mem, immI0 zero)
7140 %{
7141 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7142 match(Set mem (StoreB mem zero));
7143
7144 ins_cost(125); // XXX
7145 format %{ "movb $mem, R12\t# short/char (R12_heapbase==0)" %}
7146 ins_encode %{
7147 __ movb($mem$$Address, r12);
7148 %}
7149 ins_pipe(ialu_mem_reg);
7150 %}
7151
7152 instruct storeImmB(memory mem, immI8 src)
7153 %{
7154 match(Set mem (StoreB mem src));
7155
7156 ins_cost(150); // XXX
7157 format %{ "movb $mem, $src\t# byte" %}
7158 opcode(0xC6); /* C6 /0 */
7159 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
7160 ins_pipe(ialu_mem_imm);
7161 %}
7162
7163 // Store Aligned Packed Byte XMM register to memory
7164 instruct storeA8B(memory mem, regD src) %{
7165 match(Set mem (Store8B mem src));
7166 ins_cost(145);
7167 format %{ "MOVQ $mem,$src\t! packed8B" %}
7168 ins_encode( movq_st(mem, src));
7169 ins_pipe( pipe_slow );
7170 %}
7171
7172 // Store Aligned Packed Char/Short XMM register to memory
7173 instruct storeA4C(memory mem, regD src) %{
7174 match(Set mem (Store4C mem src));
7175 ins_cost(145);
7176 format %{ "MOVQ $mem,$src\t! packed4C" %}
7177 ins_encode( movq_st(mem, src));
7178 ins_pipe( pipe_slow );
7179 %}
7180
7181 // Store Aligned Packed Integer XMM register to memory
7182 instruct storeA2I(memory mem, regD src) %{
7183 match(Set mem (Store2I mem src));
7184 ins_cost(145);
7185 format %{ "MOVQ $mem,$src\t! packed2I" %}
7186 ins_encode( movq_st(mem, src));
7187 ins_pipe( pipe_slow );
7188 %}
7189
7190 // Store CMS card-mark Immediate
7191 instruct storeImmCM0_reg(memory mem, immI0 zero)
7192 %{
7193 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7194 match(Set mem (StoreCM mem zero));
7195
7196 ins_cost(125); // XXX
7197 format %{ "movb $mem, R12\t# CMS card-mark byte 0 (R12_heapbase==0)" %}
7198 ins_encode %{
7199 __ movb($mem$$Address, r12);
7200 %}
7201 ins_pipe(ialu_mem_reg);
7202 %}
7203
7204 instruct storeImmCM0(memory mem, immI0 src)
7205 %{
7206 match(Set mem (StoreCM mem src));
7207
7208 ins_cost(150); // XXX
7209 format %{ "movb $mem, $src\t# CMS card-mark byte 0" %}
7210 opcode(0xC6); /* C6 /0 */
7211 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con8or32(src));
7212 ins_pipe(ialu_mem_imm);
7213 %}
7214
7215 // Store Aligned Packed Single Float XMM register to memory
7216 instruct storeA2F(memory mem, regD src) %{
7217 match(Set mem (Store2F mem src));
7218 ins_cost(145);
7219 format %{ "MOVQ $mem,$src\t! packed2F" %}
7220 ins_encode( movq_st(mem, src));
7221 ins_pipe( pipe_slow );
7222 %}
7223
7224 // Store Float
7225 instruct storeF(memory mem, regF src)
7226 %{
7227 match(Set mem (StoreF mem src));
7228
7229 ins_cost(95); // XXX
7230 format %{ "movss $mem, $src\t# float" %}
7231 opcode(0xF3, 0x0F, 0x11);
7232 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7233 ins_pipe(pipe_slow); // XXX
7234 %}
7235
7236 // Store immediate Float value (it is faster than store from XMM register)
7237 instruct storeF0(memory mem, immF0 zero)
7238 %{
7239 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7240 match(Set mem (StoreF mem zero));
7241
7242 ins_cost(25); // XXX
7243 format %{ "movl $mem, R12\t# float 0. (R12_heapbase==0)" %}
7244 ins_encode %{
7245 __ movl($mem$$Address, r12);
7246 %}
7247 ins_pipe(ialu_mem_reg);
7248 %}
7249
7250 instruct storeF_imm(memory mem, immF src)
7251 %{
7252 match(Set mem (StoreF mem src));
7253
7254 ins_cost(50);
7255 format %{ "movl $mem, $src\t# float" %}
7256 opcode(0xC7); /* C7 /0 */
7257 ins_encode(REX_mem(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7258 ins_pipe(ialu_mem_imm);
7259 %}
7260
7261 // Store Double
7262 instruct storeD(memory mem, regD src)
7263 %{
7264 match(Set mem (StoreD mem src));
7265
7266 ins_cost(95); // XXX
7267 format %{ "movsd $mem, $src\t# double" %}
7268 opcode(0xF2, 0x0F, 0x11);
7269 ins_encode(OpcP, REX_reg_mem(src, mem), OpcS, OpcT, reg_mem(src, mem));
7270 ins_pipe(pipe_slow); // XXX
7271 %}
7272
7273 // Store immediate double 0.0 (it is faster than store from XMM register)
7274 instruct storeD0_imm(memory mem, immD0 src)
7275 %{
7276 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
7277 match(Set mem (StoreD mem src));
7278
7279 ins_cost(50);
7280 format %{ "movq $mem, $src\t# double 0." %}
7281 opcode(0xC7); /* C7 /0 */
7282 ins_encode(REX_mem_wide(mem), OpcP, RM_opc_mem(0x00, mem), Con32F_as_bits(src));
7283 ins_pipe(ialu_mem_imm);
7284 %}
7285
7286 instruct storeD0(memory mem, immD0 zero)
7287 %{
7288 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
7289 match(Set mem (StoreD mem zero));
7290
7291 ins_cost(25); // XXX
7292 format %{ "movq $mem, R12\t# double 0. (R12_heapbase==0)" %}
7293 ins_encode %{
7294 __ movq($mem$$Address, r12);
7295 %}
7296 ins_pipe(ialu_mem_reg);
7297 %}
7298
7299 instruct storeSSI(stackSlotI dst, rRegI src)
7300 %{
7301 match(Set dst src);
7302
7303 ins_cost(100);
7304 format %{ "movl $dst, $src\t# int stk" %}
7305 opcode(0x89);
7306 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
7307 ins_pipe( ialu_mem_reg );
7308 %}
7309
7310 instruct storeSSL(stackSlotL dst, rRegL src)
7311 %{
7312 match(Set dst src);
7313
7314 ins_cost(100);
7315 format %{ "movq $dst, $src\t# long stk" %}
7316 opcode(0x89);
7317 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7318 ins_pipe(ialu_mem_reg);
7319 %}
7320
7321 instruct storeSSP(stackSlotP dst, rRegP src)
7322 %{
7323 match(Set dst src);
7324
7325 ins_cost(100);
7326 format %{ "movq $dst, $src\t# ptr stk" %}
7327 opcode(0x89);
7328 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
7329 ins_pipe(ialu_mem_reg);
7330 %}
7331
7332 instruct storeSSF(stackSlotF dst, regF src)
7333 %{
7334 match(Set dst src);
7335
7336 ins_cost(95); // XXX
7337 format %{ "movss $dst, $src\t# float stk" %}
7338 opcode(0xF3, 0x0F, 0x11);
7339 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7340 ins_pipe(pipe_slow); // XXX
7341 %}
7342
7343 instruct storeSSD(stackSlotD dst, regD src)
7344 %{
7345 match(Set dst src);
7346
7347 ins_cost(95); // XXX
7348 format %{ "movsd $dst, $src\t# double stk" %}
7349 opcode(0xF2, 0x0F, 0x11);
7350 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
7351 ins_pipe(pipe_slow); // XXX
7352 %}
7353
7354 //----------BSWAP Instructions-------------------------------------------------
7355 instruct bytes_reverse_int(rRegI dst) %{
7356 match(Set dst (ReverseBytesI dst));
7357
7358 format %{ "bswapl $dst" %}
7359 opcode(0x0F, 0xC8); /*Opcode 0F /C8 */
7360 ins_encode( REX_reg(dst), OpcP, opc2_reg(dst) );
7361 ins_pipe( ialu_reg );
7362 %}
7363
7364 instruct bytes_reverse_long(rRegL dst) %{
7365 match(Set dst (ReverseBytesL dst));
7366
7367 format %{ "bswapq $dst" %}
7368
7369 opcode(0x0F, 0xC8); /* Opcode 0F /C8 */
7370 ins_encode( REX_reg_wide(dst), OpcP, opc2_reg(dst) );
7371 ins_pipe( ialu_reg);
7372 %}
7373
7374 instruct bytes_reverse_char(rRegI dst) %{
7375 match(Set dst (ReverseBytesC dst));
7376
7377 format %{ "bswapl $dst\n\t"
7378 "shrl $dst,16\n\t" %}
7379 ins_encode %{
7380 __ bswapl($dst$$Register);
7381 __ shrl($dst$$Register, 16);
7382 %}
7383 ins_pipe( ialu_reg );
7384 %}
7385
7386 instruct bytes_reverse_short(rRegI dst) %{
7387 match(Set dst (ReverseBytesS dst));
7388
7389 format %{ "bswapl $dst\n\t"
7390 "sar $dst,16\n\t" %}
7391 ins_encode %{
7392 __ bswapl($dst$$Register);
7393 __ sarl($dst$$Register, 16);
7394 %}
7395 ins_pipe( ialu_reg );
7396 %}
7397
7398 instruct loadI_reversed(rRegI dst, memory src) %{
7399 match(Set dst (ReverseBytesI (LoadI src)));
7400
7401 format %{ "bswap_movl $dst, $src" %}
7402 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
7403 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src), REX_reg(dst), OpcS, opc3_reg(dst));
7404 ins_pipe( ialu_reg_mem );
7405 %}
7406
7407 instruct loadL_reversed(rRegL dst, memory src) %{
7408 match(Set dst (ReverseBytesL (LoadL src)));
7409
7410 format %{ "bswap_movq $dst, $src" %}
7411 opcode(0x8B, 0x0F, 0xC8); /* Opcode 8B 0F C8 */
7412 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src), REX_reg_wide(dst), OpcS, opc3_reg(dst));
7413 ins_pipe( ialu_reg_mem );
7414 %}
7415
7416 instruct storeI_reversed(memory dst, rRegI src) %{
7417 match(Set dst (StoreI dst (ReverseBytesI src)));
7418
7419 format %{ "movl_bswap $dst, $src" %}
7420 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
7421 ins_encode( REX_reg(src), OpcP, opc2_reg(src), REX_reg_mem(src, dst), OpcT, reg_mem(src, dst) );
7422 ins_pipe( ialu_mem_reg );
7423 %}
7424
7425 instruct storeL_reversed(memory dst, rRegL src) %{
7426 match(Set dst (StoreL dst (ReverseBytesL src)));
7427
7428 format %{ "movq_bswap $dst, $src" %}
7429 opcode(0x0F, 0xC8, 0x89); /* Opcode 0F C8 89 */
7430 ins_encode( REX_reg_wide(src), OpcP, opc2_reg(src), REX_reg_mem_wide(src, dst), OpcT, reg_mem(src, dst) );
7431 ins_pipe( ialu_mem_reg );
7432 %}
7433
7434
7435 //---------- Zeros Count Instructions ------------------------------------------
7436
7437 instruct countLeadingZerosI(rRegI dst, rRegI src, rFlagsReg cr) %{
7438 predicate(UseCountLeadingZerosInstruction);
7439 match(Set dst (CountLeadingZerosI src));
7440 effect(KILL cr);
7441
7442 format %{ "lzcntl $dst, $src\t# count leading zeros (int)" %}
7443 ins_encode %{
7444 __ lzcntl($dst$$Register, $src$$Register);
7445 %}
7446 ins_pipe(ialu_reg);
7447 %}
7448
7449 instruct countLeadingZerosI_bsr(rRegI dst, rRegI src, rFlagsReg cr) %{
7450 predicate(!UseCountLeadingZerosInstruction);
7451 match(Set dst (CountLeadingZerosI src));
7452 effect(KILL cr);
7453
7454 format %{ "bsrl $dst, $src\t# count leading zeros (int)\n\t"
7455 "jnz skip\n\t"
7456 "movl $dst, -1\n"
7457 "skip:\n\t"
7458 "negl $dst\n\t"
7459 "addl $dst, 31" %}
7460 ins_encode %{
7461 Register Rdst = $dst$$Register;
7462 Register Rsrc = $src$$Register;
7463 Label skip;
7464 __ bsrl(Rdst, Rsrc);
7465 __ jccb(Assembler::notZero, skip);
7466 __ movl(Rdst, -1);
7467 __ bind(skip);
7468 __ negl(Rdst);
7469 __ addl(Rdst, BitsPerInt - 1);
7470 %}
7471 ins_pipe(ialu_reg);
7472 %}
7473
7474 instruct countLeadingZerosL(rRegI dst, rRegL src, rFlagsReg cr) %{
7475 predicate(UseCountLeadingZerosInstruction);
7476 match(Set dst (CountLeadingZerosL src));
7477 effect(KILL cr);
7478
7479 format %{ "lzcntq $dst, $src\t# count leading zeros (long)" %}
7480 ins_encode %{
7481 __ lzcntq($dst$$Register, $src$$Register);
7482 %}
7483 ins_pipe(ialu_reg);
7484 %}
7485
7486 instruct countLeadingZerosL_bsr(rRegI dst, rRegL src, rFlagsReg cr) %{
7487 predicate(!UseCountLeadingZerosInstruction);
7488 match(Set dst (CountLeadingZerosL src));
7489 effect(KILL cr);
7490
7491 format %{ "bsrq $dst, $src\t# count leading zeros (long)\n\t"
7492 "jnz skip\n\t"
7493 "movl $dst, -1\n"
7494 "skip:\n\t"
7495 "negl $dst\n\t"
7496 "addl $dst, 63" %}
7497 ins_encode %{
7498 Register Rdst = $dst$$Register;
7499 Register Rsrc = $src$$Register;
7500 Label skip;
7501 __ bsrq(Rdst, Rsrc);
7502 __ jccb(Assembler::notZero, skip);
7503 __ movl(Rdst, -1);
7504 __ bind(skip);
7505 __ negl(Rdst);
7506 __ addl(Rdst, BitsPerLong - 1);
7507 %}
7508 ins_pipe(ialu_reg);
7509 %}
7510
7511 instruct countTrailingZerosI(rRegI dst, rRegI src, rFlagsReg cr) %{
7512 match(Set dst (CountTrailingZerosI src));
7513 effect(KILL cr);
7514
7515 format %{ "bsfl $dst, $src\t# count trailing zeros (int)\n\t"
7516 "jnz done\n\t"
7517 "movl $dst, 32\n"
7518 "done:" %}
7519 ins_encode %{
7520 Register Rdst = $dst$$Register;
7521 Label done;
7522 __ bsfl(Rdst, $src$$Register);
7523 __ jccb(Assembler::notZero, done);
7524 __ movl(Rdst, BitsPerInt);
7525 __ bind(done);
7526 %}
7527 ins_pipe(ialu_reg);
7528 %}
7529
7530 instruct countTrailingZerosL(rRegI dst, rRegL src, rFlagsReg cr) %{
7531 match(Set dst (CountTrailingZerosL src));
7532 effect(KILL cr);
7533
7534 format %{ "bsfq $dst, $src\t# count trailing zeros (long)\n\t"
7535 "jnz done\n\t"
7536 "movl $dst, 64\n"
7537 "done:" %}
7538 ins_encode %{
7539 Register Rdst = $dst$$Register;
7540 Label done;
7541 __ bsfq(Rdst, $src$$Register);
7542 __ jccb(Assembler::notZero, done);
7543 __ movl(Rdst, BitsPerLong);
7544 __ bind(done);
7545 %}
7546 ins_pipe(ialu_reg);
7547 %}
7548
7549
7550 //---------- Population Count Instructions -------------------------------------
7551
7552 instruct popCountI(rRegI dst, rRegI src) %{
7553 predicate(UsePopCountInstruction);
7554 match(Set dst (PopCountI src));
7555
7556 format %{ "popcnt $dst, $src" %}
7557 ins_encode %{
7558 __ popcntl($dst$$Register, $src$$Register);
7559 %}
7560 ins_pipe(ialu_reg);
7561 %}
7562
7563 instruct popCountI_mem(rRegI dst, memory mem) %{
7564 predicate(UsePopCountInstruction);
7565 match(Set dst (PopCountI (LoadI mem)));
7566
7567 format %{ "popcnt $dst, $mem" %}
7568 ins_encode %{
7569 __ popcntl($dst$$Register, $mem$$Address);
7570 %}
7571 ins_pipe(ialu_reg);
7572 %}
7573
7574 // Note: Long.bitCount(long) returns an int.
7575 instruct popCountL(rRegI dst, rRegL src) %{
7576 predicate(UsePopCountInstruction);
7577 match(Set dst (PopCountL src));
7578
7579 format %{ "popcnt $dst, $src" %}
7580 ins_encode %{
7581 __ popcntq($dst$$Register, $src$$Register);
7582 %}
7583 ins_pipe(ialu_reg);
7584 %}
7585
7586 // Note: Long.bitCount(long) returns an int.
7587 instruct popCountL_mem(rRegI dst, memory mem) %{
7588 predicate(UsePopCountInstruction);
7589 match(Set dst (PopCountL (LoadL mem)));
7590
7591 format %{ "popcnt $dst, $mem" %}
7592 ins_encode %{
7593 __ popcntq($dst$$Register, $mem$$Address);
7594 %}
7595 ins_pipe(ialu_reg);
7596 %}
7597
7598
7599 //----------MemBar Instructions-----------------------------------------------
7600 // Memory barrier flavors
7601
7602 instruct membar_acquire()
7603 %{
7604 match(MemBarAcquire);
7605 ins_cost(0);
7606
7607 size(0);
7608 format %{ "MEMBAR-acquire ! (empty encoding)" %}
7609 ins_encode();
7610 ins_pipe(empty);
7611 %}
7612
7613 instruct membar_acquire_lock()
7614 %{
7615 match(MemBarAcquire);
7616 predicate(Matcher::prior_fast_lock(n));
7617 ins_cost(0);
7618
7619 size(0);
7620 format %{ "MEMBAR-acquire (prior CMPXCHG in FastLock so empty encoding)" %}
7621 ins_encode();
7622 ins_pipe(empty);
7623 %}
7624
7625 instruct membar_release()
7626 %{
7627 match(MemBarRelease);
7628 ins_cost(0);
7629
7630 size(0);
7631 format %{ "MEMBAR-release ! (empty encoding)" %}
7632 ins_encode();
7633 ins_pipe(empty);
7634 %}
7635
7636 instruct membar_release_lock()
7637 %{
7638 match(MemBarRelease);
7639 predicate(Matcher::post_fast_unlock(n));
7640 ins_cost(0);
7641
7642 size(0);
7643 format %{ "MEMBAR-release (a FastUnlock follows so empty encoding)" %}
7644 ins_encode();
7645 ins_pipe(empty);
7646 %}
7647
7648 instruct membar_volatile(rFlagsReg cr) %{
7649 match(MemBarVolatile);
7650 effect(KILL cr);
7651 ins_cost(400);
7652
7653 format %{
7654 $$template
7655 if (os::is_MP()) {
7656 $$emit$$"lock addl [rsp + #0], 0\t! membar_volatile"
7657 } else {
7658 $$emit$$"MEMBAR-volatile ! (empty encoding)"
7659 }
7660 %}
7661 ins_encode %{
7662 __ membar(Assembler::StoreLoad);
7663 %}
7664 ins_pipe(pipe_slow);
7665 %}
7666
7667 instruct unnecessary_membar_volatile()
7668 %{
7669 match(MemBarVolatile);
7670 predicate(Matcher::post_store_load_barrier(n));
7671 ins_cost(0);
7672
7673 size(0);
7674 format %{ "MEMBAR-volatile (unnecessary so empty encoding)" %}
7675 ins_encode();
7676 ins_pipe(empty);
7677 %}
7678
7679 //----------Move Instructions--------------------------------------------------
7680
7681 instruct castX2P(rRegP dst, rRegL src)
7682 %{
7683 match(Set dst (CastX2P src));
7684
7685 format %{ "movq $dst, $src\t# long->ptr" %}
7686 ins_encode(enc_copy_wide(dst, src));
7687 ins_pipe(ialu_reg_reg); // XXX
7688 %}
7689
7690 instruct castP2X(rRegL dst, rRegP src)
7691 %{
7692 match(Set dst (CastP2X src));
7693
7694 format %{ "movq $dst, $src\t# ptr -> long" %}
7695 ins_encode(enc_copy_wide(dst, src));
7696 ins_pipe(ialu_reg_reg); // XXX
7697 %}
7698
7699
7700 // Convert oop pointer into compressed form
7701 instruct encodeHeapOop(rRegN dst, rRegP src, rFlagsReg cr) %{
7702 predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
7703 match(Set dst (EncodeP src));
7704 effect(KILL cr);
7705 format %{ "encode_heap_oop $dst,$src" %}
7706 ins_encode %{
7707 Register s = $src$$Register;
7708 Register d = $dst$$Register;
7709 if (s != d) {
7710 __ movq(d, s);
7711 }
7712 __ encode_heap_oop(d);
7713 %}
7714 ins_pipe(ialu_reg_long);
7715 %}
7716
7717 instruct encodeHeapOop_not_null(rRegN dst, rRegP src, rFlagsReg cr) %{
7718 predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
7719 match(Set dst (EncodeP src));
7720 effect(KILL cr);
7721 format %{ "encode_heap_oop_not_null $dst,$src" %}
7722 ins_encode %{
7723 __ encode_heap_oop_not_null($dst$$Register, $src$$Register);
7724 %}
7725 ins_pipe(ialu_reg_long);
7726 %}
7727
7728 instruct decodeHeapOop(rRegP dst, rRegN src, rFlagsReg cr) %{
7729 predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
7730 n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
7731 match(Set dst (DecodeN src));
7732 effect(KILL cr);
7733 format %{ "decode_heap_oop $dst,$src" %}
7734 ins_encode %{
7735 Register s = $src$$Register;
7736 Register d = $dst$$Register;
7737 if (s != d) {
7738 __ movq(d, s);
7739 }
7740 __ decode_heap_oop(d);
7741 %}
7742 ins_pipe(ialu_reg_long);
7743 %}
7744
7745 instruct decodeHeapOop_not_null(rRegP dst, rRegN src) %{
7746 predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
7747 n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
7748 match(Set dst (DecodeN src));
7749 format %{ "decode_heap_oop_not_null $dst,$src" %}
7750 ins_encode %{
7751 Register s = $src$$Register;
7752 Register d = $dst$$Register;
7753 if (s != d) {
7754 __ decode_heap_oop_not_null(d, s);
7755 } else {
7756 __ decode_heap_oop_not_null(d);
7757 }
7758 %}
7759 ins_pipe(ialu_reg_long);
7760 %}
7761
7762
7763 //----------Conditional Move---------------------------------------------------
7764 // Jump
7765 // dummy instruction for generating temp registers
7766 instruct jumpXtnd_offset(rRegL switch_val, immI2 shift, rRegI dest) %{
7767 match(Jump (LShiftL switch_val shift));
7768 ins_cost(350);
7769 predicate(false);
7770 effect(TEMP dest);
7771
7772 format %{ "leaq $dest, table_base\n\t"
7773 "jmp [$dest + $switch_val << $shift]\n\t" %}
7774 ins_encode(jump_enc_offset(switch_val, shift, dest));
7775 ins_pipe(pipe_jmp);
7776 ins_pc_relative(1);
7777 %}
7778
7779 instruct jumpXtnd_addr(rRegL switch_val, immI2 shift, immL32 offset, rRegI dest) %{
7780 match(Jump (AddL (LShiftL switch_val shift) offset));
7781 ins_cost(350);
7782 effect(TEMP dest);
7783
7784 format %{ "leaq $dest, table_base\n\t"
7785 "jmp [$dest + $switch_val << $shift + $offset]\n\t" %}
7786 ins_encode(jump_enc_addr(switch_val, shift, offset, dest));
7787 ins_pipe(pipe_jmp);
7788 ins_pc_relative(1);
7789 %}
7790
7791 instruct jumpXtnd(rRegL switch_val, rRegI dest) %{
7792 match(Jump switch_val);
7793 ins_cost(350);
7794 effect(TEMP dest);
7795
7796 format %{ "leaq $dest, table_base\n\t"
7797 "jmp [$dest + $switch_val]\n\t" %}
7798 ins_encode(jump_enc(switch_val, dest));
7799 ins_pipe(pipe_jmp);
7800 ins_pc_relative(1);
7801 %}
7802
7803 // Conditional move
7804 instruct cmovI_reg(rRegI dst, rRegI src, rFlagsReg cr, cmpOp cop)
7805 %{
7806 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7807
7808 ins_cost(200); // XXX
7809 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7810 opcode(0x0F, 0x40);
7811 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7812 ins_pipe(pipe_cmov_reg);
7813 %}
7814
7815 instruct cmovI_regU(cmpOpU cop, rFlagsRegU cr, rRegI dst, rRegI src) %{
7816 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7817
7818 ins_cost(200); // XXX
7819 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7820 opcode(0x0F, 0x40);
7821 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7822 ins_pipe(pipe_cmov_reg);
7823 %}
7824
7825 instruct cmovI_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, rRegI src) %{
7826 match(Set dst (CMoveI (Binary cop cr) (Binary dst src)));
7827 ins_cost(200);
7828 expand %{
7829 cmovI_regU(cop, cr, dst, src);
7830 %}
7831 %}
7832
7833 // Conditional move
7834 instruct cmovI_mem(cmpOp cop, rFlagsReg cr, rRegI dst, memory src) %{
7835 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7836
7837 ins_cost(250); // XXX
7838 format %{ "cmovl$cop $dst, $src\t# signed, int" %}
7839 opcode(0x0F, 0x40);
7840 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7841 ins_pipe(pipe_cmov_mem);
7842 %}
7843
7844 // Conditional move
7845 instruct cmovI_memU(cmpOpU cop, rFlagsRegU cr, rRegI dst, memory src)
7846 %{
7847 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7848
7849 ins_cost(250); // XXX
7850 format %{ "cmovl$cop $dst, $src\t# unsigned, int" %}
7851 opcode(0x0F, 0x40);
7852 ins_encode(REX_reg_mem(dst, src), enc_cmov(cop), reg_mem(dst, src));
7853 ins_pipe(pipe_cmov_mem);
7854 %}
7855
7856 instruct cmovI_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegI dst, memory src) %{
7857 match(Set dst (CMoveI (Binary cop cr) (Binary dst (LoadI src))));
7858 ins_cost(250);
7859 expand %{
7860 cmovI_memU(cop, cr, dst, src);
7861 %}
7862 %}
7863
7864 // Conditional move
7865 instruct cmovN_reg(rRegN dst, rRegN src, rFlagsReg cr, cmpOp cop)
7866 %{
7867 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7868
7869 ins_cost(200); // XXX
7870 format %{ "cmovl$cop $dst, $src\t# signed, compressed ptr" %}
7871 opcode(0x0F, 0x40);
7872 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7873 ins_pipe(pipe_cmov_reg);
7874 %}
7875
7876 // Conditional move
7877 instruct cmovN_regU(cmpOpU cop, rFlagsRegU cr, rRegN dst, rRegN src)
7878 %{
7879 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7880
7881 ins_cost(200); // XXX
7882 format %{ "cmovl$cop $dst, $src\t# unsigned, compressed ptr" %}
7883 opcode(0x0F, 0x40);
7884 ins_encode(REX_reg_reg(dst, src), enc_cmov(cop), reg_reg(dst, src));
7885 ins_pipe(pipe_cmov_reg);
7886 %}
7887
7888 instruct cmovN_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegN dst, rRegN src) %{
7889 match(Set dst (CMoveN (Binary cop cr) (Binary dst src)));
7890 ins_cost(200);
7891 expand %{
7892 cmovN_regU(cop, cr, dst, src);
7893 %}
7894 %}
7895
7896 // Conditional move
7897 instruct cmovP_reg(rRegP dst, rRegP src, rFlagsReg cr, cmpOp cop)
7898 %{
7899 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7900
7901 ins_cost(200); // XXX
7902 format %{ "cmovq$cop $dst, $src\t# signed, ptr" %}
7903 opcode(0x0F, 0x40);
7904 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7905 ins_pipe(pipe_cmov_reg); // XXX
7906 %}
7907
7908 // Conditional move
7909 instruct cmovP_regU(cmpOpU cop, rFlagsRegU cr, rRegP dst, rRegP src)
7910 %{
7911 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7912
7913 ins_cost(200); // XXX
7914 format %{ "cmovq$cop $dst, $src\t# unsigned, ptr" %}
7915 opcode(0x0F, 0x40);
7916 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7917 ins_pipe(pipe_cmov_reg); // XXX
7918 %}
7919
7920 instruct cmovP_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegP dst, rRegP src) %{
7921 match(Set dst (CMoveP (Binary cop cr) (Binary dst src)));
7922 ins_cost(200);
7923 expand %{
7924 cmovP_regU(cop, cr, dst, src);
7925 %}
7926 %}
7927
7928 // DISABLED: Requires the ADLC to emit a bottom_type call that
7929 // correctly meets the two pointer arguments; one is an incoming
7930 // register but the other is a memory operand. ALSO appears to
7931 // be buggy with implicit null checks.
7932 //
7933 //// Conditional move
7934 //instruct cmovP_mem(cmpOp cop, rFlagsReg cr, rRegP dst, memory src)
7935 //%{
7936 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7937 // ins_cost(250);
7938 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7939 // opcode(0x0F,0x40);
7940 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7941 // ins_pipe( pipe_cmov_mem );
7942 //%}
7943 //
7944 //// Conditional move
7945 //instruct cmovP_memU(cmpOpU cop, rFlagsRegU cr, rRegP dst, memory src)
7946 //%{
7947 // match(Set dst (CMoveP (Binary cop cr) (Binary dst (LoadP src))));
7948 // ins_cost(250);
7949 // format %{ "CMOV$cop $dst,$src\t# ptr" %}
7950 // opcode(0x0F,0x40);
7951 // ins_encode( enc_cmov(cop), reg_mem( dst, src ) );
7952 // ins_pipe( pipe_cmov_mem );
7953 //%}
7954
7955 instruct cmovL_reg(cmpOp cop, rFlagsReg cr, rRegL dst, rRegL src)
7956 %{
7957 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7958
7959 ins_cost(200); // XXX
7960 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7961 opcode(0x0F, 0x40);
7962 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7963 ins_pipe(pipe_cmov_reg); // XXX
7964 %}
7965
7966 instruct cmovL_mem(cmpOp cop, rFlagsReg cr, rRegL dst, memory src)
7967 %{
7968 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7969
7970 ins_cost(200); // XXX
7971 format %{ "cmovq$cop $dst, $src\t# signed, long" %}
7972 opcode(0x0F, 0x40);
7973 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
7974 ins_pipe(pipe_cmov_mem); // XXX
7975 %}
7976
7977 instruct cmovL_regU(cmpOpU cop, rFlagsRegU cr, rRegL dst, rRegL src)
7978 %{
7979 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7980
7981 ins_cost(200); // XXX
7982 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
7983 opcode(0x0F, 0x40);
7984 ins_encode(REX_reg_reg_wide(dst, src), enc_cmov(cop), reg_reg(dst, src));
7985 ins_pipe(pipe_cmov_reg); // XXX
7986 %}
7987
7988 instruct cmovL_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, rRegL src) %{
7989 match(Set dst (CMoveL (Binary cop cr) (Binary dst src)));
7990 ins_cost(200);
7991 expand %{
7992 cmovL_regU(cop, cr, dst, src);
7993 %}
7994 %}
7995
7996 instruct cmovL_memU(cmpOpU cop, rFlagsRegU cr, rRegL dst, memory src)
7997 %{
7998 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
7999
8000 ins_cost(200); // XXX
8001 format %{ "cmovq$cop $dst, $src\t# unsigned, long" %}
8002 opcode(0x0F, 0x40);
8003 ins_encode(REX_reg_mem_wide(dst, src), enc_cmov(cop), reg_mem(dst, src));
8004 ins_pipe(pipe_cmov_mem); // XXX
8005 %}
8006
8007 instruct cmovL_memUCF(cmpOpUCF cop, rFlagsRegUCF cr, rRegL dst, memory src) %{
8008 match(Set dst (CMoveL (Binary cop cr) (Binary dst (LoadL src))));
8009 ins_cost(200);
8010 expand %{
8011 cmovL_memU(cop, cr, dst, src);
8012 %}
8013 %}
8014
8015 instruct cmovF_reg(cmpOp cop, rFlagsReg cr, regF dst, regF src)
8016 %{
8017 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
8018
8019 ins_cost(200); // XXX
8020 format %{ "jn$cop skip\t# signed cmove float\n\t"
8021 "movss $dst, $src\n"
8022 "skip:" %}
8023 ins_encode(enc_cmovf_branch(cop, dst, src));
8024 ins_pipe(pipe_slow);
8025 %}
8026
8027 // instruct cmovF_mem(cmpOp cop, rFlagsReg cr, regF dst, memory src)
8028 // %{
8029 // match(Set dst (CMoveF (Binary cop cr) (Binary dst (LoadL src))));
8030
8031 // ins_cost(200); // XXX
8032 // format %{ "jn$cop skip\t# signed cmove float\n\t"
8033 // "movss $dst, $src\n"
8034 // "skip:" %}
8035 // ins_encode(enc_cmovf_mem_branch(cop, dst, src));
8036 // ins_pipe(pipe_slow);
8037 // %}
8038
8039 instruct cmovF_regU(cmpOpU cop, rFlagsRegU cr, regF dst, regF src)
8040 %{
8041 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
8042
8043 ins_cost(200); // XXX
8044 format %{ "jn$cop skip\t# unsigned cmove float\n\t"
8045 "movss $dst, $src\n"
8046 "skip:" %}
8047 ins_encode(enc_cmovf_branch(cop, dst, src));
8048 ins_pipe(pipe_slow);
8049 %}
8050
8051 instruct cmovF_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regF dst, regF src) %{
8052 match(Set dst (CMoveF (Binary cop cr) (Binary dst src)));
8053 ins_cost(200);
8054 expand %{
8055 cmovF_regU(cop, cr, dst, src);
8056 %}
8057 %}
8058
8059 instruct cmovD_reg(cmpOp cop, rFlagsReg cr, regD dst, regD src)
8060 %{
8061 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
8062
8063 ins_cost(200); // XXX
8064 format %{ "jn$cop skip\t# signed cmove double\n\t"
8065 "movsd $dst, $src\n"
8066 "skip:" %}
8067 ins_encode(enc_cmovd_branch(cop, dst, src));
8068 ins_pipe(pipe_slow);
8069 %}
8070
8071 instruct cmovD_regU(cmpOpU cop, rFlagsRegU cr, regD dst, regD src)
8072 %{
8073 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
8074
8075 ins_cost(200); // XXX
8076 format %{ "jn$cop skip\t# unsigned cmove double\n\t"
8077 "movsd $dst, $src\n"
8078 "skip:" %}
8079 ins_encode(enc_cmovd_branch(cop, dst, src));
8080 ins_pipe(pipe_slow);
8081 %}
8082
8083 instruct cmovD_regUCF(cmpOpUCF cop, rFlagsRegUCF cr, regD dst, regD src) %{
8084 match(Set dst (CMoveD (Binary cop cr) (Binary dst src)));
8085 ins_cost(200);
8086 expand %{
8087 cmovD_regU(cop, cr, dst, src);
8088 %}
8089 %}
8090
8091 //----------Arithmetic Instructions--------------------------------------------
8092 //----------Addition Instructions----------------------------------------------
8093
8094 instruct addI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8095 %{
8096 match(Set dst (AddI dst src));
8097 effect(KILL cr);
8098
8099 format %{ "addl $dst, $src\t# int" %}
8100 opcode(0x03);
8101 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
8102 ins_pipe(ialu_reg_reg);
8103 %}
8104
8105 instruct addI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
8106 %{
8107 match(Set dst (AddI dst src));
8108 effect(KILL cr);
8109
8110 format %{ "addl $dst, $src\t# int" %}
8111 opcode(0x81, 0x00); /* /0 id */
8112 ins_encode(OpcSErm(dst, src), Con8or32(src));
8113 ins_pipe( ialu_reg );
8114 %}
8115
8116 instruct addI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
8117 %{
8118 match(Set dst (AddI dst (LoadI src)));
8119 effect(KILL cr);
8120
8121 ins_cost(125); // XXX
8122 format %{ "addl $dst, $src\t# int" %}
8123 opcode(0x03);
8124 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
8125 ins_pipe(ialu_reg_mem);
8126 %}
8127
8128 instruct addI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
8129 %{
8130 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
8131 effect(KILL cr);
8132
8133 ins_cost(150); // XXX
8134 format %{ "addl $dst, $src\t# int" %}
8135 opcode(0x01); /* Opcode 01 /r */
8136 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
8137 ins_pipe(ialu_mem_reg);
8138 %}
8139
8140 instruct addI_mem_imm(memory dst, immI src, rFlagsReg cr)
8141 %{
8142 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
8143 effect(KILL cr);
8144
8145 ins_cost(125); // XXX
8146 format %{ "addl $dst, $src\t# int" %}
8147 opcode(0x81); /* Opcode 81 /0 id */
8148 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
8149 ins_pipe(ialu_mem_imm);
8150 %}
8151
8152 instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
8153 %{
8154 predicate(UseIncDec);
8155 match(Set dst (AddI dst src));
8156 effect(KILL cr);
8157
8158 format %{ "incl $dst\t# int" %}
8159 opcode(0xFF, 0x00); // FF /0
8160 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8161 ins_pipe(ialu_reg);
8162 %}
8163
8164 instruct incI_mem(memory dst, immI1 src, rFlagsReg cr)
8165 %{
8166 predicate(UseIncDec);
8167 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
8168 effect(KILL cr);
8169
8170 ins_cost(125); // XXX
8171 format %{ "incl $dst\t# int" %}
8172 opcode(0xFF); /* Opcode FF /0 */
8173 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x00, dst));
8174 ins_pipe(ialu_mem_imm);
8175 %}
8176
8177 // XXX why does that use AddI
8178 instruct decI_rReg(rRegI dst, immI_M1 src, rFlagsReg cr)
8179 %{
8180 predicate(UseIncDec);
8181 match(Set dst (AddI dst src));
8182 effect(KILL cr);
8183
8184 format %{ "decl $dst\t# int" %}
8185 opcode(0xFF, 0x01); // FF /1
8186 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8187 ins_pipe(ialu_reg);
8188 %}
8189
8190 // XXX why does that use AddI
8191 instruct decI_mem(memory dst, immI_M1 src, rFlagsReg cr)
8192 %{
8193 predicate(UseIncDec);
8194 match(Set dst (StoreI dst (AddI (LoadI dst) src)));
8195 effect(KILL cr);
8196
8197 ins_cost(125); // XXX
8198 format %{ "decl $dst\t# int" %}
8199 opcode(0xFF); /* Opcode FF /1 */
8200 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(0x01, dst));
8201 ins_pipe(ialu_mem_imm);
8202 %}
8203
8204 instruct leaI_rReg_immI(rRegI dst, rRegI src0, immI src1)
8205 %{
8206 match(Set dst (AddI src0 src1));
8207
8208 ins_cost(110);
8209 format %{ "addr32 leal $dst, [$src0 + $src1]\t# int" %}
8210 opcode(0x8D); /* 0x8D /r */
8211 ins_encode(Opcode(0x67), REX_reg_reg(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
8212 ins_pipe(ialu_reg_reg);
8213 %}
8214
8215 instruct addL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8216 %{
8217 match(Set dst (AddL dst src));
8218 effect(KILL cr);
8219
8220 format %{ "addq $dst, $src\t# long" %}
8221 opcode(0x03);
8222 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8223 ins_pipe(ialu_reg_reg);
8224 %}
8225
8226 instruct addL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
8227 %{
8228 match(Set dst (AddL dst src));
8229 effect(KILL cr);
8230
8231 format %{ "addq $dst, $src\t# long" %}
8232 opcode(0x81, 0x00); /* /0 id */
8233 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8234 ins_pipe( ialu_reg );
8235 %}
8236
8237 instruct addL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8238 %{
8239 match(Set dst (AddL dst (LoadL src)));
8240 effect(KILL cr);
8241
8242 ins_cost(125); // XXX
8243 format %{ "addq $dst, $src\t# long" %}
8244 opcode(0x03);
8245 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8246 ins_pipe(ialu_reg_mem);
8247 %}
8248
8249 instruct addL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8250 %{
8251 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8252 effect(KILL cr);
8253
8254 ins_cost(150); // XXX
8255 format %{ "addq $dst, $src\t# long" %}
8256 opcode(0x01); /* Opcode 01 /r */
8257 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8258 ins_pipe(ialu_mem_reg);
8259 %}
8260
8261 instruct addL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8262 %{
8263 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8264 effect(KILL cr);
8265
8266 ins_cost(125); // XXX
8267 format %{ "addq $dst, $src\t# long" %}
8268 opcode(0x81); /* Opcode 81 /0 id */
8269 ins_encode(REX_mem_wide(dst),
8270 OpcSE(src), RM_opc_mem(0x00, dst), Con8or32(src));
8271 ins_pipe(ialu_mem_imm);
8272 %}
8273
8274 instruct incL_rReg(rRegI dst, immL1 src, rFlagsReg cr)
8275 %{
8276 predicate(UseIncDec);
8277 match(Set dst (AddL dst src));
8278 effect(KILL cr);
8279
8280 format %{ "incq $dst\t# long" %}
8281 opcode(0xFF, 0x00); // FF /0
8282 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8283 ins_pipe(ialu_reg);
8284 %}
8285
8286 instruct incL_mem(memory dst, immL1 src, rFlagsReg cr)
8287 %{
8288 predicate(UseIncDec);
8289 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8290 effect(KILL cr);
8291
8292 ins_cost(125); // XXX
8293 format %{ "incq $dst\t# long" %}
8294 opcode(0xFF); /* Opcode FF /0 */
8295 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x00, dst));
8296 ins_pipe(ialu_mem_imm);
8297 %}
8298
8299 // XXX why does that use AddL
8300 instruct decL_rReg(rRegL dst, immL_M1 src, rFlagsReg cr)
8301 %{
8302 predicate(UseIncDec);
8303 match(Set dst (AddL dst src));
8304 effect(KILL cr);
8305
8306 format %{ "decq $dst\t# long" %}
8307 opcode(0xFF, 0x01); // FF /1
8308 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8309 ins_pipe(ialu_reg);
8310 %}
8311
8312 // XXX why does that use AddL
8313 instruct decL_mem(memory dst, immL_M1 src, rFlagsReg cr)
8314 %{
8315 predicate(UseIncDec);
8316 match(Set dst (StoreL dst (AddL (LoadL dst) src)));
8317 effect(KILL cr);
8318
8319 ins_cost(125); // XXX
8320 format %{ "decq $dst\t# long" %}
8321 opcode(0xFF); /* Opcode FF /1 */
8322 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(0x01, dst));
8323 ins_pipe(ialu_mem_imm);
8324 %}
8325
8326 instruct leaL_rReg_immL(rRegL dst, rRegL src0, immL32 src1)
8327 %{
8328 match(Set dst (AddL src0 src1));
8329
8330 ins_cost(110);
8331 format %{ "leaq $dst, [$src0 + $src1]\t# long" %}
8332 opcode(0x8D); /* 0x8D /r */
8333 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1)); // XXX
8334 ins_pipe(ialu_reg_reg);
8335 %}
8336
8337 instruct addP_rReg(rRegP dst, rRegL src, rFlagsReg cr)
8338 %{
8339 match(Set dst (AddP dst src));
8340 effect(KILL cr);
8341
8342 format %{ "addq $dst, $src\t# ptr" %}
8343 opcode(0x03);
8344 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8345 ins_pipe(ialu_reg_reg);
8346 %}
8347
8348 instruct addP_rReg_imm(rRegP dst, immL32 src, rFlagsReg cr)
8349 %{
8350 match(Set dst (AddP dst src));
8351 effect(KILL cr);
8352
8353 format %{ "addq $dst, $src\t# ptr" %}
8354 opcode(0x81, 0x00); /* /0 id */
8355 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8356 ins_pipe( ialu_reg );
8357 %}
8358
8359 // XXX addP mem ops ????
8360
8361 instruct leaP_rReg_imm(rRegP dst, rRegP src0, immL32 src1)
8362 %{
8363 match(Set dst (AddP src0 src1));
8364
8365 ins_cost(110);
8366 format %{ "leaq $dst, [$src0 + $src1]\t# ptr" %}
8367 opcode(0x8D); /* 0x8D /r */
8368 ins_encode(REX_reg_reg_wide(dst, src0), OpcP, reg_lea(dst, src0, src1));// XXX
8369 ins_pipe(ialu_reg_reg);
8370 %}
8371
8372 instruct checkCastPP(rRegP dst)
8373 %{
8374 match(Set dst (CheckCastPP dst));
8375
8376 size(0);
8377 format %{ "# checkcastPP of $dst" %}
8378 ins_encode(/* empty encoding */);
8379 ins_pipe(empty);
8380 %}
8381
8382 instruct castPP(rRegP dst)
8383 %{
8384 match(Set dst (CastPP dst));
8385
8386 size(0);
8387 format %{ "# castPP of $dst" %}
8388 ins_encode(/* empty encoding */);
8389 ins_pipe(empty);
8390 %}
8391
8392 instruct castII(rRegI dst)
8393 %{
8394 match(Set dst (CastII dst));
8395
8396 size(0);
8397 format %{ "# castII of $dst" %}
8398 ins_encode(/* empty encoding */);
8399 ins_cost(0);
8400 ins_pipe(empty);
8401 %}
8402
8403 // LoadP-locked same as a regular LoadP when used with compare-swap
8404 instruct loadPLocked(rRegP dst, memory mem)
8405 %{
8406 match(Set dst (LoadPLocked mem));
8407
8408 ins_cost(125); // XXX
8409 format %{ "movq $dst, $mem\t# ptr locked" %}
8410 opcode(0x8B);
8411 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8412 ins_pipe(ialu_reg_mem); // XXX
8413 %}
8414
8415 // LoadL-locked - same as a regular LoadL when used with compare-swap
8416 instruct loadLLocked(rRegL dst, memory mem)
8417 %{
8418 match(Set dst (LoadLLocked mem));
8419
8420 ins_cost(125); // XXX
8421 format %{ "movq $dst, $mem\t# long locked" %}
8422 opcode(0x8B);
8423 ins_encode(REX_reg_mem_wide(dst, mem), OpcP, reg_mem(dst, mem));
8424 ins_pipe(ialu_reg_mem); // XXX
8425 %}
8426
8427 // Conditional-store of the updated heap-top.
8428 // Used during allocation of the shared heap.
8429 // Sets flags (EQ) on success. Implemented with a CMPXCHG on Intel.
8430
8431 instruct storePConditional(memory heap_top_ptr,
8432 rax_RegP oldval, rRegP newval,
8433 rFlagsReg cr)
8434 %{
8435 match(Set cr (StorePConditional heap_top_ptr (Binary oldval newval)));
8436
8437 format %{ "cmpxchgq $heap_top_ptr, $newval\t# (ptr) "
8438 "If rax == $heap_top_ptr then store $newval into $heap_top_ptr" %}
8439 opcode(0x0F, 0xB1);
8440 ins_encode(lock_prefix,
8441 REX_reg_mem_wide(newval, heap_top_ptr),
8442 OpcP, OpcS,
8443 reg_mem(newval, heap_top_ptr));
8444 ins_pipe(pipe_cmpxchg);
8445 %}
8446
8447 // Conditional-store of an int value.
8448 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG.
8449 instruct storeIConditional(memory mem, rax_RegI oldval, rRegI newval, rFlagsReg cr)
8450 %{
8451 match(Set cr (StoreIConditional mem (Binary oldval newval)));
8452 effect(KILL oldval);
8453
8454 format %{ "cmpxchgl $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8455 opcode(0x0F, 0xB1);
8456 ins_encode(lock_prefix,
8457 REX_reg_mem(newval, mem),
8458 OpcP, OpcS,
8459 reg_mem(newval, mem));
8460 ins_pipe(pipe_cmpxchg);
8461 %}
8462
8463 // Conditional-store of a long value.
8464 // ZF flag is set on success, reset otherwise. Implemented with a CMPXCHG.
8465 instruct storeLConditional(memory mem, rax_RegL oldval, rRegL newval, rFlagsReg cr)
8466 %{
8467 match(Set cr (StoreLConditional mem (Binary oldval newval)));
8468 effect(KILL oldval);
8469
8470 format %{ "cmpxchgq $mem, $newval\t# If rax == $mem then store $newval into $mem" %}
8471 opcode(0x0F, 0xB1);
8472 ins_encode(lock_prefix,
8473 REX_reg_mem_wide(newval, mem),
8474 OpcP, OpcS,
8475 reg_mem(newval, mem));
8476 ins_pipe(pipe_cmpxchg);
8477 %}
8478
8479
8480 // XXX No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
8481 instruct compareAndSwapP(rRegI res,
8482 memory mem_ptr,
8483 rax_RegP oldval, rRegP newval,
8484 rFlagsReg cr)
8485 %{
8486 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
8487 effect(KILL cr, KILL oldval);
8488
8489 format %{ "cmpxchgq $mem_ptr,$newval\t# "
8490 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8491 "sete $res\n\t"
8492 "movzbl $res, $res" %}
8493 opcode(0x0F, 0xB1);
8494 ins_encode(lock_prefix,
8495 REX_reg_mem_wide(newval, mem_ptr),
8496 OpcP, OpcS,
8497 reg_mem(newval, mem_ptr),
8498 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8499 REX_reg_breg(res, res), // movzbl
8500 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8501 ins_pipe( pipe_cmpxchg );
8502 %}
8503
8504 instruct compareAndSwapL(rRegI res,
8505 memory mem_ptr,
8506 rax_RegL oldval, rRegL newval,
8507 rFlagsReg cr)
8508 %{
8509 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
8510 effect(KILL cr, KILL oldval);
8511
8512 format %{ "cmpxchgq $mem_ptr,$newval\t# "
8513 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8514 "sete $res\n\t"
8515 "movzbl $res, $res" %}
8516 opcode(0x0F, 0xB1);
8517 ins_encode(lock_prefix,
8518 REX_reg_mem_wide(newval, mem_ptr),
8519 OpcP, OpcS,
8520 reg_mem(newval, mem_ptr),
8521 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8522 REX_reg_breg(res, res), // movzbl
8523 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8524 ins_pipe( pipe_cmpxchg );
8525 %}
8526
8527 instruct compareAndSwapI(rRegI res,
8528 memory mem_ptr,
8529 rax_RegI oldval, rRegI newval,
8530 rFlagsReg cr)
8531 %{
8532 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
8533 effect(KILL cr, KILL oldval);
8534
8535 format %{ "cmpxchgl $mem_ptr,$newval\t# "
8536 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8537 "sete $res\n\t"
8538 "movzbl $res, $res" %}
8539 opcode(0x0F, 0xB1);
8540 ins_encode(lock_prefix,
8541 REX_reg_mem(newval, mem_ptr),
8542 OpcP, OpcS,
8543 reg_mem(newval, mem_ptr),
8544 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8545 REX_reg_breg(res, res), // movzbl
8546 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8547 ins_pipe( pipe_cmpxchg );
8548 %}
8549
8550
8551 instruct compareAndSwapN(rRegI res,
8552 memory mem_ptr,
8553 rax_RegN oldval, rRegN newval,
8554 rFlagsReg cr) %{
8555 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
8556 effect(KILL cr, KILL oldval);
8557
8558 format %{ "cmpxchgl $mem_ptr,$newval\t# "
8559 "If rax == $mem_ptr then store $newval into $mem_ptr\n\t"
8560 "sete $res\n\t"
8561 "movzbl $res, $res" %}
8562 opcode(0x0F, 0xB1);
8563 ins_encode(lock_prefix,
8564 REX_reg_mem(newval, mem_ptr),
8565 OpcP, OpcS,
8566 reg_mem(newval, mem_ptr),
8567 REX_breg(res), Opcode(0x0F), Opcode(0x94), reg(res), // sete
8568 REX_reg_breg(res, res), // movzbl
8569 Opcode(0xF), Opcode(0xB6), reg_reg(res, res));
8570 ins_pipe( pipe_cmpxchg );
8571 %}
8572
8573 //----------Subtraction Instructions-------------------------------------------
8574
8575 // Integer Subtraction Instructions
8576 instruct subI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8577 %{
8578 match(Set dst (SubI dst src));
8579 effect(KILL cr);
8580
8581 format %{ "subl $dst, $src\t# int" %}
8582 opcode(0x2B);
8583 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
8584 ins_pipe(ialu_reg_reg);
8585 %}
8586
8587 instruct subI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
8588 %{
8589 match(Set dst (SubI dst src));
8590 effect(KILL cr);
8591
8592 format %{ "subl $dst, $src\t# int" %}
8593 opcode(0x81, 0x05); /* Opcode 81 /5 */
8594 ins_encode(OpcSErm(dst, src), Con8or32(src));
8595 ins_pipe(ialu_reg);
8596 %}
8597
8598 instruct subI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
8599 %{
8600 match(Set dst (SubI dst (LoadI src)));
8601 effect(KILL cr);
8602
8603 ins_cost(125);
8604 format %{ "subl $dst, $src\t# int" %}
8605 opcode(0x2B);
8606 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
8607 ins_pipe(ialu_reg_mem);
8608 %}
8609
8610 instruct subI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
8611 %{
8612 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8613 effect(KILL cr);
8614
8615 ins_cost(150);
8616 format %{ "subl $dst, $src\t# int" %}
8617 opcode(0x29); /* Opcode 29 /r */
8618 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
8619 ins_pipe(ialu_mem_reg);
8620 %}
8621
8622 instruct subI_mem_imm(memory dst, immI src, rFlagsReg cr)
8623 %{
8624 match(Set dst (StoreI dst (SubI (LoadI dst) src)));
8625 effect(KILL cr);
8626
8627 ins_cost(125); // XXX
8628 format %{ "subl $dst, $src\t# int" %}
8629 opcode(0x81); /* Opcode 81 /5 id */
8630 ins_encode(REX_mem(dst), OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8631 ins_pipe(ialu_mem_imm);
8632 %}
8633
8634 instruct subL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8635 %{
8636 match(Set dst (SubL dst src));
8637 effect(KILL cr);
8638
8639 format %{ "subq $dst, $src\t# long" %}
8640 opcode(0x2B);
8641 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8642 ins_pipe(ialu_reg_reg);
8643 %}
8644
8645 instruct subL_rReg_imm(rRegI dst, immL32 src, rFlagsReg cr)
8646 %{
8647 match(Set dst (SubL dst src));
8648 effect(KILL cr);
8649
8650 format %{ "subq $dst, $src\t# long" %}
8651 opcode(0x81, 0x05); /* Opcode 81 /5 */
8652 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
8653 ins_pipe(ialu_reg);
8654 %}
8655
8656 instruct subL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
8657 %{
8658 match(Set dst (SubL dst (LoadL src)));
8659 effect(KILL cr);
8660
8661 ins_cost(125);
8662 format %{ "subq $dst, $src\t# long" %}
8663 opcode(0x2B);
8664 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
8665 ins_pipe(ialu_reg_mem);
8666 %}
8667
8668 instruct subL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
8669 %{
8670 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8671 effect(KILL cr);
8672
8673 ins_cost(150);
8674 format %{ "subq $dst, $src\t# long" %}
8675 opcode(0x29); /* Opcode 29 /r */
8676 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
8677 ins_pipe(ialu_mem_reg);
8678 %}
8679
8680 instruct subL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
8681 %{
8682 match(Set dst (StoreL dst (SubL (LoadL dst) src)));
8683 effect(KILL cr);
8684
8685 ins_cost(125); // XXX
8686 format %{ "subq $dst, $src\t# long" %}
8687 opcode(0x81); /* Opcode 81 /5 id */
8688 ins_encode(REX_mem_wide(dst),
8689 OpcSE(src), RM_opc_mem(0x05, dst), Con8or32(src));
8690 ins_pipe(ialu_mem_imm);
8691 %}
8692
8693 // Subtract from a pointer
8694 // XXX hmpf???
8695 instruct subP_rReg(rRegP dst, rRegI src, immI0 zero, rFlagsReg cr)
8696 %{
8697 match(Set dst (AddP dst (SubI zero src)));
8698 effect(KILL cr);
8699
8700 format %{ "subq $dst, $src\t# ptr - int" %}
8701 opcode(0x2B);
8702 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
8703 ins_pipe(ialu_reg_reg);
8704 %}
8705
8706 instruct negI_rReg(rRegI dst, immI0 zero, rFlagsReg cr)
8707 %{
8708 match(Set dst (SubI zero dst));
8709 effect(KILL cr);
8710
8711 format %{ "negl $dst\t# int" %}
8712 opcode(0xF7, 0x03); // Opcode F7 /3
8713 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
8714 ins_pipe(ialu_reg);
8715 %}
8716
8717 instruct negI_mem(memory dst, immI0 zero, rFlagsReg cr)
8718 %{
8719 match(Set dst (StoreI dst (SubI zero (LoadI dst))));
8720 effect(KILL cr);
8721
8722 format %{ "negl $dst\t# int" %}
8723 opcode(0xF7, 0x03); // Opcode F7 /3
8724 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
8725 ins_pipe(ialu_reg);
8726 %}
8727
8728 instruct negL_rReg(rRegL dst, immL0 zero, rFlagsReg cr)
8729 %{
8730 match(Set dst (SubL zero dst));
8731 effect(KILL cr);
8732
8733 format %{ "negq $dst\t# long" %}
8734 opcode(0xF7, 0x03); // Opcode F7 /3
8735 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
8736 ins_pipe(ialu_reg);
8737 %}
8738
8739 instruct negL_mem(memory dst, immL0 zero, rFlagsReg cr)
8740 %{
8741 match(Set dst (StoreL dst (SubL zero (LoadL dst))));
8742 effect(KILL cr);
8743
8744 format %{ "negq $dst\t# long" %}
8745 opcode(0xF7, 0x03); // Opcode F7 /3
8746 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
8747 ins_pipe(ialu_reg);
8748 %}
8749
8750
8751 //----------Multiplication/Division Instructions-------------------------------
8752 // Integer Multiplication Instructions
8753 // Multiply Register
8754
8755 instruct mulI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
8756 %{
8757 match(Set dst (MulI dst src));
8758 effect(KILL cr);
8759
8760 ins_cost(300);
8761 format %{ "imull $dst, $src\t# int" %}
8762 opcode(0x0F, 0xAF);
8763 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
8764 ins_pipe(ialu_reg_reg_alu0);
8765 %}
8766
8767 instruct mulI_rReg_imm(rRegI dst, rRegI src, immI imm, rFlagsReg cr)
8768 %{
8769 match(Set dst (MulI src imm));
8770 effect(KILL cr);
8771
8772 ins_cost(300);
8773 format %{ "imull $dst, $src, $imm\t# int" %}
8774 opcode(0x69); /* 69 /r id */
8775 ins_encode(REX_reg_reg(dst, src),
8776 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8777 ins_pipe(ialu_reg_reg_alu0);
8778 %}
8779
8780 instruct mulI_mem(rRegI dst, memory src, rFlagsReg cr)
8781 %{
8782 match(Set dst (MulI dst (LoadI src)));
8783 effect(KILL cr);
8784
8785 ins_cost(350);
8786 format %{ "imull $dst, $src\t# int" %}
8787 opcode(0x0F, 0xAF);
8788 ins_encode(REX_reg_mem(dst, src), OpcP, OpcS, reg_mem(dst, src));
8789 ins_pipe(ialu_reg_mem_alu0);
8790 %}
8791
8792 instruct mulI_mem_imm(rRegI dst, memory src, immI imm, rFlagsReg cr)
8793 %{
8794 match(Set dst (MulI (LoadI src) imm));
8795 effect(KILL cr);
8796
8797 ins_cost(300);
8798 format %{ "imull $dst, $src, $imm\t# int" %}
8799 opcode(0x69); /* 69 /r id */
8800 ins_encode(REX_reg_mem(dst, src),
8801 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8802 ins_pipe(ialu_reg_mem_alu0);
8803 %}
8804
8805 instruct mulL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
8806 %{
8807 match(Set dst (MulL dst src));
8808 effect(KILL cr);
8809
8810 ins_cost(300);
8811 format %{ "imulq $dst, $src\t# long" %}
8812 opcode(0x0F, 0xAF);
8813 ins_encode(REX_reg_reg_wide(dst, src), OpcP, OpcS, reg_reg(dst, src));
8814 ins_pipe(ialu_reg_reg_alu0);
8815 %}
8816
8817 instruct mulL_rReg_imm(rRegL dst, rRegL src, immL32 imm, rFlagsReg cr)
8818 %{
8819 match(Set dst (MulL src imm));
8820 effect(KILL cr);
8821
8822 ins_cost(300);
8823 format %{ "imulq $dst, $src, $imm\t# long" %}
8824 opcode(0x69); /* 69 /r id */
8825 ins_encode(REX_reg_reg_wide(dst, src),
8826 OpcSE(imm), reg_reg(dst, src), Con8or32(imm));
8827 ins_pipe(ialu_reg_reg_alu0);
8828 %}
8829
8830 instruct mulL_mem(rRegL dst, memory src, rFlagsReg cr)
8831 %{
8832 match(Set dst (MulL dst (LoadL src)));
8833 effect(KILL cr);
8834
8835 ins_cost(350);
8836 format %{ "imulq $dst, $src\t# long" %}
8837 opcode(0x0F, 0xAF);
8838 ins_encode(REX_reg_mem_wide(dst, src), OpcP, OpcS, reg_mem(dst, src));
8839 ins_pipe(ialu_reg_mem_alu0);
8840 %}
8841
8842 instruct mulL_mem_imm(rRegL dst, memory src, immL32 imm, rFlagsReg cr)
8843 %{
8844 match(Set dst (MulL (LoadL src) imm));
8845 effect(KILL cr);
8846
8847 ins_cost(300);
8848 format %{ "imulq $dst, $src, $imm\t# long" %}
8849 opcode(0x69); /* 69 /r id */
8850 ins_encode(REX_reg_mem_wide(dst, src),
8851 OpcSE(imm), reg_mem(dst, src), Con8or32(imm));
8852 ins_pipe(ialu_reg_mem_alu0);
8853 %}
8854
8855 instruct mulHiL_rReg(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8856 %{
8857 match(Set dst (MulHiL src rax));
8858 effect(USE_KILL rax, KILL cr);
8859
8860 ins_cost(300);
8861 format %{ "imulq RDX:RAX, RAX, $src\t# mulhi" %}
8862 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8863 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8864 ins_pipe(ialu_reg_reg_alu0);
8865 %}
8866
8867 instruct divI_rReg(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8868 rFlagsReg cr)
8869 %{
8870 match(Set rax (DivI rax div));
8871 effect(KILL rdx, KILL cr);
8872
8873 ins_cost(30*100+10*100); // XXX
8874 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8875 "jne,s normal\n\t"
8876 "xorl rdx, rdx\n\t"
8877 "cmpl $div, -1\n\t"
8878 "je,s done\n"
8879 "normal: cdql\n\t"
8880 "idivl $div\n"
8881 "done:" %}
8882 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8883 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8884 ins_pipe(ialu_reg_reg_alu0);
8885 %}
8886
8887 instruct divL_rReg(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8888 rFlagsReg cr)
8889 %{
8890 match(Set rax (DivL rax div));
8891 effect(KILL rdx, KILL cr);
8892
8893 ins_cost(30*100+10*100); // XXX
8894 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8895 "cmpq rax, rdx\n\t"
8896 "jne,s normal\n\t"
8897 "xorl rdx, rdx\n\t"
8898 "cmpq $div, -1\n\t"
8899 "je,s done\n"
8900 "normal: cdqq\n\t"
8901 "idivq $div\n"
8902 "done:" %}
8903 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8904 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8905 ins_pipe(ialu_reg_reg_alu0);
8906 %}
8907
8908 // Integer DIVMOD with Register, both quotient and mod results
8909 instruct divModI_rReg_divmod(rax_RegI rax, rdx_RegI rdx, no_rax_rdx_RegI div,
8910 rFlagsReg cr)
8911 %{
8912 match(DivModI rax div);
8913 effect(KILL cr);
8914
8915 ins_cost(30*100+10*100); // XXX
8916 format %{ "cmpl rax, 0x80000000\t# idiv\n\t"
8917 "jne,s normal\n\t"
8918 "xorl rdx, rdx\n\t"
8919 "cmpl $div, -1\n\t"
8920 "je,s done\n"
8921 "normal: cdql\n\t"
8922 "idivl $div\n"
8923 "done:" %}
8924 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8925 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
8926 ins_pipe(pipe_slow);
8927 %}
8928
8929 // Long DIVMOD with Register, both quotient and mod results
8930 instruct divModL_rReg_divmod(rax_RegL rax, rdx_RegL rdx, no_rax_rdx_RegL div,
8931 rFlagsReg cr)
8932 %{
8933 match(DivModL rax div);
8934 effect(KILL cr);
8935
8936 ins_cost(30*100+10*100); // XXX
8937 format %{ "movq rdx, 0x8000000000000000\t# ldiv\n\t"
8938 "cmpq rax, rdx\n\t"
8939 "jne,s normal\n\t"
8940 "xorl rdx, rdx\n\t"
8941 "cmpq $div, -1\n\t"
8942 "je,s done\n"
8943 "normal: cdqq\n\t"
8944 "idivq $div\n"
8945 "done:" %}
8946 opcode(0xF7, 0x7); /* Opcode F7 /7 */
8947 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
8948 ins_pipe(pipe_slow);
8949 %}
8950
8951 //----------- DivL-By-Constant-Expansions--------------------------------------
8952 // DivI cases are handled by the compiler
8953
8954 // Magic constant, reciprocal of 10
8955 instruct loadConL_0x6666666666666667(rRegL dst)
8956 %{
8957 effect(DEF dst);
8958
8959 format %{ "movq $dst, #0x666666666666667\t# Used in div-by-10" %}
8960 ins_encode(load_immL(dst, 0x6666666666666667));
8961 ins_pipe(ialu_reg);
8962 %}
8963
8964 instruct mul_hi(rdx_RegL dst, no_rax_RegL src, rax_RegL rax, rFlagsReg cr)
8965 %{
8966 effect(DEF dst, USE src, USE_KILL rax, KILL cr);
8967
8968 format %{ "imulq rdx:rax, rax, $src\t# Used in div-by-10" %}
8969 opcode(0xF7, 0x5); /* Opcode F7 /5 */
8970 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src));
8971 ins_pipe(ialu_reg_reg_alu0);
8972 %}
8973
8974 instruct sarL_rReg_63(rRegL dst, rFlagsReg cr)
8975 %{
8976 effect(USE_DEF dst, KILL cr);
8977
8978 format %{ "sarq $dst, #63\t# Used in div-by-10" %}
8979 opcode(0xC1, 0x7); /* C1 /7 ib */
8980 ins_encode(reg_opc_imm_wide(dst, 0x3F));
8981 ins_pipe(ialu_reg);
8982 %}
8983
8984 instruct sarL_rReg_2(rRegL dst, rFlagsReg cr)
8985 %{
8986 effect(USE_DEF dst, KILL cr);
8987
8988 format %{ "sarq $dst, #2\t# Used in div-by-10" %}
8989 opcode(0xC1, 0x7); /* C1 /7 ib */
8990 ins_encode(reg_opc_imm_wide(dst, 0x2));
8991 ins_pipe(ialu_reg);
8992 %}
8993
8994 instruct divL_10(rdx_RegL dst, no_rax_RegL src, immL10 div)
8995 %{
8996 match(Set dst (DivL src div));
8997
8998 ins_cost((5+8)*100);
8999 expand %{
9000 rax_RegL rax; // Killed temp
9001 rFlagsReg cr; // Killed
9002 loadConL_0x6666666666666667(rax); // movq rax, 0x6666666666666667
9003 mul_hi(dst, src, rax, cr); // mulq rdx:rax <= rax * $src
9004 sarL_rReg_63(src, cr); // sarq src, 63
9005 sarL_rReg_2(dst, cr); // sarq rdx, 2
9006 subL_rReg(dst, src, cr); // subl rdx, src
9007 %}
9008 %}
9009
9010 //-----------------------------------------------------------------------------
9011
9012 instruct modI_rReg(rdx_RegI rdx, rax_RegI rax, no_rax_rdx_RegI div,
9013 rFlagsReg cr)
9014 %{
9015 match(Set rdx (ModI rax div));
9016 effect(KILL rax, KILL cr);
9017
9018 ins_cost(300); // XXX
9019 format %{ "cmpl rax, 0x80000000\t# irem\n\t"
9020 "jne,s normal\n\t"
9021 "xorl rdx, rdx\n\t"
9022 "cmpl $div, -1\n\t"
9023 "je,s done\n"
9024 "normal: cdql\n\t"
9025 "idivl $div\n"
9026 "done:" %}
9027 opcode(0xF7, 0x7); /* Opcode F7 /7 */
9028 ins_encode(cdql_enc(div), REX_reg(div), OpcP, reg_opc(div));
9029 ins_pipe(ialu_reg_reg_alu0);
9030 %}
9031
9032 instruct modL_rReg(rdx_RegL rdx, rax_RegL rax, no_rax_rdx_RegL div,
9033 rFlagsReg cr)
9034 %{
9035 match(Set rdx (ModL rax div));
9036 effect(KILL rax, KILL cr);
9037
9038 ins_cost(300); // XXX
9039 format %{ "movq rdx, 0x8000000000000000\t# lrem\n\t"
9040 "cmpq rax, rdx\n\t"
9041 "jne,s normal\n\t"
9042 "xorl rdx, rdx\n\t"
9043 "cmpq $div, -1\n\t"
9044 "je,s done\n"
9045 "normal: cdqq\n\t"
9046 "idivq $div\n"
9047 "done:" %}
9048 opcode(0xF7, 0x7); /* Opcode F7 /7 */
9049 ins_encode(cdqq_enc(div), REX_reg_wide(div), OpcP, reg_opc(div));
9050 ins_pipe(ialu_reg_reg_alu0);
9051 %}
9052
9053 // Integer Shift Instructions
9054 // Shift Left by one
9055 instruct salI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
9056 %{
9057 match(Set dst (LShiftI dst shift));
9058 effect(KILL cr);
9059
9060 format %{ "sall $dst, $shift" %}
9061 opcode(0xD1, 0x4); /* D1 /4 */
9062 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9063 ins_pipe(ialu_reg);
9064 %}
9065
9066 // Shift Left by one
9067 instruct salI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9068 %{
9069 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
9070 effect(KILL cr);
9071
9072 format %{ "sall $dst, $shift\t" %}
9073 opcode(0xD1, 0x4); /* D1 /4 */
9074 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9075 ins_pipe(ialu_mem_imm);
9076 %}
9077
9078 // Shift Left by 8-bit immediate
9079 instruct salI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
9080 %{
9081 match(Set dst (LShiftI dst shift));
9082 effect(KILL cr);
9083
9084 format %{ "sall $dst, $shift" %}
9085 opcode(0xC1, 0x4); /* C1 /4 ib */
9086 ins_encode(reg_opc_imm(dst, shift));
9087 ins_pipe(ialu_reg);
9088 %}
9089
9090 // Shift Left by 8-bit immediate
9091 instruct salI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9092 %{
9093 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
9094 effect(KILL cr);
9095
9096 format %{ "sall $dst, $shift" %}
9097 opcode(0xC1, 0x4); /* C1 /4 ib */
9098 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
9099 ins_pipe(ialu_mem_imm);
9100 %}
9101
9102 // Shift Left by variable
9103 instruct salI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
9104 %{
9105 match(Set dst (LShiftI dst shift));
9106 effect(KILL cr);
9107
9108 format %{ "sall $dst, $shift" %}
9109 opcode(0xD3, 0x4); /* D3 /4 */
9110 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9111 ins_pipe(ialu_reg_reg);
9112 %}
9113
9114 // Shift Left by variable
9115 instruct salI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9116 %{
9117 match(Set dst (StoreI dst (LShiftI (LoadI dst) shift)));
9118 effect(KILL cr);
9119
9120 format %{ "sall $dst, $shift" %}
9121 opcode(0xD3, 0x4); /* D3 /4 */
9122 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9123 ins_pipe(ialu_mem_reg);
9124 %}
9125
9126 // Arithmetic shift right by one
9127 instruct sarI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
9128 %{
9129 match(Set dst (RShiftI dst shift));
9130 effect(KILL cr);
9131
9132 format %{ "sarl $dst, $shift" %}
9133 opcode(0xD1, 0x7); /* D1 /7 */
9134 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9135 ins_pipe(ialu_reg);
9136 %}
9137
9138 // Arithmetic shift right by one
9139 instruct sarI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9140 %{
9141 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
9142 effect(KILL cr);
9143
9144 format %{ "sarl $dst, $shift" %}
9145 opcode(0xD1, 0x7); /* D1 /7 */
9146 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9147 ins_pipe(ialu_mem_imm);
9148 %}
9149
9150 // Arithmetic Shift Right by 8-bit immediate
9151 instruct sarI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
9152 %{
9153 match(Set dst (RShiftI dst shift));
9154 effect(KILL cr);
9155
9156 format %{ "sarl $dst, $shift" %}
9157 opcode(0xC1, 0x7); /* C1 /7 ib */
9158 ins_encode(reg_opc_imm(dst, shift));
9159 ins_pipe(ialu_mem_imm);
9160 %}
9161
9162 // Arithmetic Shift Right by 8-bit immediate
9163 instruct sarI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9164 %{
9165 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
9166 effect(KILL cr);
9167
9168 format %{ "sarl $dst, $shift" %}
9169 opcode(0xC1, 0x7); /* C1 /7 ib */
9170 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
9171 ins_pipe(ialu_mem_imm);
9172 %}
9173
9174 // Arithmetic Shift Right by variable
9175 instruct sarI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
9176 %{
9177 match(Set dst (RShiftI dst shift));
9178 effect(KILL cr);
9179
9180 format %{ "sarl $dst, $shift" %}
9181 opcode(0xD3, 0x7); /* D3 /7 */
9182 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9183 ins_pipe(ialu_reg_reg);
9184 %}
9185
9186 // Arithmetic Shift Right by variable
9187 instruct sarI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9188 %{
9189 match(Set dst (StoreI dst (RShiftI (LoadI dst) shift)));
9190 effect(KILL cr);
9191
9192 format %{ "sarl $dst, $shift" %}
9193 opcode(0xD3, 0x7); /* D3 /7 */
9194 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9195 ins_pipe(ialu_mem_reg);
9196 %}
9197
9198 // Logical shift right by one
9199 instruct shrI_rReg_1(rRegI dst, immI1 shift, rFlagsReg cr)
9200 %{
9201 match(Set dst (URShiftI dst shift));
9202 effect(KILL cr);
9203
9204 format %{ "shrl $dst, $shift" %}
9205 opcode(0xD1, 0x5); /* D1 /5 */
9206 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9207 ins_pipe(ialu_reg);
9208 %}
9209
9210 // Logical shift right by one
9211 instruct shrI_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9212 %{
9213 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9214 effect(KILL cr);
9215
9216 format %{ "shrl $dst, $shift" %}
9217 opcode(0xD1, 0x5); /* D1 /5 */
9218 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9219 ins_pipe(ialu_mem_imm);
9220 %}
9221
9222 // Logical Shift Right by 8-bit immediate
9223 instruct shrI_rReg_imm(rRegI dst, immI8 shift, rFlagsReg cr)
9224 %{
9225 match(Set dst (URShiftI dst shift));
9226 effect(KILL cr);
9227
9228 format %{ "shrl $dst, $shift" %}
9229 opcode(0xC1, 0x5); /* C1 /5 ib */
9230 ins_encode(reg_opc_imm(dst, shift));
9231 ins_pipe(ialu_reg);
9232 %}
9233
9234 // Logical Shift Right by 8-bit immediate
9235 instruct shrI_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9236 %{
9237 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9238 effect(KILL cr);
9239
9240 format %{ "shrl $dst, $shift" %}
9241 opcode(0xC1, 0x5); /* C1 /5 ib */
9242 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst), Con8or32(shift));
9243 ins_pipe(ialu_mem_imm);
9244 %}
9245
9246 // Logical Shift Right by variable
9247 instruct shrI_rReg_CL(rRegI dst, rcx_RegI shift, rFlagsReg cr)
9248 %{
9249 match(Set dst (URShiftI dst shift));
9250 effect(KILL cr);
9251
9252 format %{ "shrl $dst, $shift" %}
9253 opcode(0xD3, 0x5); /* D3 /5 */
9254 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9255 ins_pipe(ialu_reg_reg);
9256 %}
9257
9258 // Logical Shift Right by variable
9259 instruct shrI_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9260 %{
9261 match(Set dst (StoreI dst (URShiftI (LoadI dst) shift)));
9262 effect(KILL cr);
9263
9264 format %{ "shrl $dst, $shift" %}
9265 opcode(0xD3, 0x5); /* D3 /5 */
9266 ins_encode(REX_mem(dst), OpcP, RM_opc_mem(secondary, dst));
9267 ins_pipe(ialu_mem_reg);
9268 %}
9269
9270 // Long Shift Instructions
9271 // Shift Left by one
9272 instruct salL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9273 %{
9274 match(Set dst (LShiftL dst shift));
9275 effect(KILL cr);
9276
9277 format %{ "salq $dst, $shift" %}
9278 opcode(0xD1, 0x4); /* D1 /4 */
9279 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9280 ins_pipe(ialu_reg);
9281 %}
9282
9283 // Shift Left by one
9284 instruct salL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9285 %{
9286 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9287 effect(KILL cr);
9288
9289 format %{ "salq $dst, $shift" %}
9290 opcode(0xD1, 0x4); /* D1 /4 */
9291 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9292 ins_pipe(ialu_mem_imm);
9293 %}
9294
9295 // Shift Left by 8-bit immediate
9296 instruct salL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9297 %{
9298 match(Set dst (LShiftL dst shift));
9299 effect(KILL cr);
9300
9301 format %{ "salq $dst, $shift" %}
9302 opcode(0xC1, 0x4); /* C1 /4 ib */
9303 ins_encode(reg_opc_imm_wide(dst, shift));
9304 ins_pipe(ialu_reg);
9305 %}
9306
9307 // Shift Left by 8-bit immediate
9308 instruct salL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9309 %{
9310 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9311 effect(KILL cr);
9312
9313 format %{ "salq $dst, $shift" %}
9314 opcode(0xC1, 0x4); /* C1 /4 ib */
9315 ins_encode(REX_mem_wide(dst), OpcP,
9316 RM_opc_mem(secondary, dst), Con8or32(shift));
9317 ins_pipe(ialu_mem_imm);
9318 %}
9319
9320 // Shift Left by variable
9321 instruct salL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9322 %{
9323 match(Set dst (LShiftL dst shift));
9324 effect(KILL cr);
9325
9326 format %{ "salq $dst, $shift" %}
9327 opcode(0xD3, 0x4); /* D3 /4 */
9328 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9329 ins_pipe(ialu_reg_reg);
9330 %}
9331
9332 // Shift Left by variable
9333 instruct salL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9334 %{
9335 match(Set dst (StoreL dst (LShiftL (LoadL dst) shift)));
9336 effect(KILL cr);
9337
9338 format %{ "salq $dst, $shift" %}
9339 opcode(0xD3, 0x4); /* D3 /4 */
9340 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9341 ins_pipe(ialu_mem_reg);
9342 %}
9343
9344 // Arithmetic shift right by one
9345 instruct sarL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9346 %{
9347 match(Set dst (RShiftL dst shift));
9348 effect(KILL cr);
9349
9350 format %{ "sarq $dst, $shift" %}
9351 opcode(0xD1, 0x7); /* D1 /7 */
9352 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9353 ins_pipe(ialu_reg);
9354 %}
9355
9356 // Arithmetic shift right by one
9357 instruct sarL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9358 %{
9359 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9360 effect(KILL cr);
9361
9362 format %{ "sarq $dst, $shift" %}
9363 opcode(0xD1, 0x7); /* D1 /7 */
9364 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9365 ins_pipe(ialu_mem_imm);
9366 %}
9367
9368 // Arithmetic Shift Right by 8-bit immediate
9369 instruct sarL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9370 %{
9371 match(Set dst (RShiftL dst shift));
9372 effect(KILL cr);
9373
9374 format %{ "sarq $dst, $shift" %}
9375 opcode(0xC1, 0x7); /* C1 /7 ib */
9376 ins_encode(reg_opc_imm_wide(dst, shift));
9377 ins_pipe(ialu_mem_imm);
9378 %}
9379
9380 // Arithmetic Shift Right by 8-bit immediate
9381 instruct sarL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9382 %{
9383 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9384 effect(KILL cr);
9385
9386 format %{ "sarq $dst, $shift" %}
9387 opcode(0xC1, 0x7); /* C1 /7 ib */
9388 ins_encode(REX_mem_wide(dst), OpcP,
9389 RM_opc_mem(secondary, dst), Con8or32(shift));
9390 ins_pipe(ialu_mem_imm);
9391 %}
9392
9393 // Arithmetic Shift Right by variable
9394 instruct sarL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9395 %{
9396 match(Set dst (RShiftL dst shift));
9397 effect(KILL cr);
9398
9399 format %{ "sarq $dst, $shift" %}
9400 opcode(0xD3, 0x7); /* D3 /7 */
9401 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9402 ins_pipe(ialu_reg_reg);
9403 %}
9404
9405 // Arithmetic Shift Right by variable
9406 instruct sarL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9407 %{
9408 match(Set dst (StoreL dst (RShiftL (LoadL dst) shift)));
9409 effect(KILL cr);
9410
9411 format %{ "sarq $dst, $shift" %}
9412 opcode(0xD3, 0x7); /* D3 /7 */
9413 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9414 ins_pipe(ialu_mem_reg);
9415 %}
9416
9417 // Logical shift right by one
9418 instruct shrL_rReg_1(rRegL dst, immI1 shift, rFlagsReg cr)
9419 %{
9420 match(Set dst (URShiftL dst shift));
9421 effect(KILL cr);
9422
9423 format %{ "shrq $dst, $shift" %}
9424 opcode(0xD1, 0x5); /* D1 /5 */
9425 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst ));
9426 ins_pipe(ialu_reg);
9427 %}
9428
9429 // Logical shift right by one
9430 instruct shrL_mem_1(memory dst, immI1 shift, rFlagsReg cr)
9431 %{
9432 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9433 effect(KILL cr);
9434
9435 format %{ "shrq $dst, $shift" %}
9436 opcode(0xD1, 0x5); /* D1 /5 */
9437 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9438 ins_pipe(ialu_mem_imm);
9439 %}
9440
9441 // Logical Shift Right by 8-bit immediate
9442 instruct shrL_rReg_imm(rRegL dst, immI8 shift, rFlagsReg cr)
9443 %{
9444 match(Set dst (URShiftL dst shift));
9445 effect(KILL cr);
9446
9447 format %{ "shrq $dst, $shift" %}
9448 opcode(0xC1, 0x5); /* C1 /5 ib */
9449 ins_encode(reg_opc_imm_wide(dst, shift));
9450 ins_pipe(ialu_reg);
9451 %}
9452
9453
9454 // Logical Shift Right by 8-bit immediate
9455 instruct shrL_mem_imm(memory dst, immI8 shift, rFlagsReg cr)
9456 %{
9457 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9458 effect(KILL cr);
9459
9460 format %{ "shrq $dst, $shift" %}
9461 opcode(0xC1, 0x5); /* C1 /5 ib */
9462 ins_encode(REX_mem_wide(dst), OpcP,
9463 RM_opc_mem(secondary, dst), Con8or32(shift));
9464 ins_pipe(ialu_mem_imm);
9465 %}
9466
9467 // Logical Shift Right by variable
9468 instruct shrL_rReg_CL(rRegL dst, rcx_RegI shift, rFlagsReg cr)
9469 %{
9470 match(Set dst (URShiftL dst shift));
9471 effect(KILL cr);
9472
9473 format %{ "shrq $dst, $shift" %}
9474 opcode(0xD3, 0x5); /* D3 /5 */
9475 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9476 ins_pipe(ialu_reg_reg);
9477 %}
9478
9479 // Logical Shift Right by variable
9480 instruct shrL_mem_CL(memory dst, rcx_RegI shift, rFlagsReg cr)
9481 %{
9482 match(Set dst (StoreL dst (URShiftL (LoadL dst) shift)));
9483 effect(KILL cr);
9484
9485 format %{ "shrq $dst, $shift" %}
9486 opcode(0xD3, 0x5); /* D3 /5 */
9487 ins_encode(REX_mem_wide(dst), OpcP, RM_opc_mem(secondary, dst));
9488 ins_pipe(ialu_mem_reg);
9489 %}
9490
9491 // Logical Shift Right by 24, followed by Arithmetic Shift Left by 24.
9492 // This idiom is used by the compiler for the i2b bytecode.
9493 instruct i2b(rRegI dst, rRegI src, immI_24 twentyfour)
9494 %{
9495 match(Set dst (RShiftI (LShiftI src twentyfour) twentyfour));
9496
9497 format %{ "movsbl $dst, $src\t# i2b" %}
9498 opcode(0x0F, 0xBE);
9499 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9500 ins_pipe(ialu_reg_reg);
9501 %}
9502
9503 // Logical Shift Right by 16, followed by Arithmetic Shift Left by 16.
9504 // This idiom is used by the compiler the i2s bytecode.
9505 instruct i2s(rRegI dst, rRegI src, immI_16 sixteen)
9506 %{
9507 match(Set dst (RShiftI (LShiftI src sixteen) sixteen));
9508
9509 format %{ "movswl $dst, $src\t# i2s" %}
9510 opcode(0x0F, 0xBF);
9511 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9512 ins_pipe(ialu_reg_reg);
9513 %}
9514
9515 // ROL/ROR instructions
9516
9517 // ROL expand
9518 instruct rolI_rReg_imm1(rRegI dst, rFlagsReg cr) %{
9519 effect(KILL cr, USE_DEF dst);
9520
9521 format %{ "roll $dst" %}
9522 opcode(0xD1, 0x0); /* Opcode D1 /0 */
9523 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9524 ins_pipe(ialu_reg);
9525 %}
9526
9527 instruct rolI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr) %{
9528 effect(USE_DEF dst, USE shift, KILL cr);
9529
9530 format %{ "roll $dst, $shift" %}
9531 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9532 ins_encode( reg_opc_imm(dst, shift) );
9533 ins_pipe(ialu_reg);
9534 %}
9535
9536 instruct rolI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9537 %{
9538 effect(USE_DEF dst, USE shift, KILL cr);
9539
9540 format %{ "roll $dst, $shift" %}
9541 opcode(0xD3, 0x0); /* Opcode D3 /0 */
9542 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9543 ins_pipe(ialu_reg_reg);
9544 %}
9545 // end of ROL expand
9546
9547 // Rotate Left by one
9548 instruct rolI_rReg_i1(rRegI dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9549 %{
9550 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9551
9552 expand %{
9553 rolI_rReg_imm1(dst, cr);
9554 %}
9555 %}
9556
9557 // Rotate Left by 8-bit immediate
9558 instruct rolI_rReg_i8(rRegI dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9559 %{
9560 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9561 match(Set dst (OrI (LShiftI dst lshift) (URShiftI dst rshift)));
9562
9563 expand %{
9564 rolI_rReg_imm8(dst, lshift, cr);
9565 %}
9566 %}
9567
9568 // Rotate Left by variable
9569 instruct rolI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9570 %{
9571 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI zero shift))));
9572
9573 expand %{
9574 rolI_rReg_CL(dst, shift, cr);
9575 %}
9576 %}
9577
9578 // Rotate Left by variable
9579 instruct rolI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9580 %{
9581 match(Set dst (OrI (LShiftI dst shift) (URShiftI dst (SubI c32 shift))));
9582
9583 expand %{
9584 rolI_rReg_CL(dst, shift, cr);
9585 %}
9586 %}
9587
9588 // ROR expand
9589 instruct rorI_rReg_imm1(rRegI dst, rFlagsReg cr)
9590 %{
9591 effect(USE_DEF dst, KILL cr);
9592
9593 format %{ "rorl $dst" %}
9594 opcode(0xD1, 0x1); /* D1 /1 */
9595 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9596 ins_pipe(ialu_reg);
9597 %}
9598
9599 instruct rorI_rReg_imm8(rRegI dst, immI8 shift, rFlagsReg cr)
9600 %{
9601 effect(USE_DEF dst, USE shift, KILL cr);
9602
9603 format %{ "rorl $dst, $shift" %}
9604 opcode(0xC1, 0x1); /* C1 /1 ib */
9605 ins_encode(reg_opc_imm(dst, shift));
9606 ins_pipe(ialu_reg);
9607 %}
9608
9609 instruct rorI_rReg_CL(no_rcx_RegI dst, rcx_RegI shift, rFlagsReg cr)
9610 %{
9611 effect(USE_DEF dst, USE shift, KILL cr);
9612
9613 format %{ "rorl $dst, $shift" %}
9614 opcode(0xD3, 0x1); /* D3 /1 */
9615 ins_encode(REX_reg(dst), OpcP, reg_opc(dst));
9616 ins_pipe(ialu_reg_reg);
9617 %}
9618 // end of ROR expand
9619
9620 // Rotate Right by one
9621 instruct rorI_rReg_i1(rRegI dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9622 %{
9623 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9624
9625 expand %{
9626 rorI_rReg_imm1(dst, cr);
9627 %}
9628 %}
9629
9630 // Rotate Right by 8-bit immediate
9631 instruct rorI_rReg_i8(rRegI dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9632 %{
9633 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x1f));
9634 match(Set dst (OrI (URShiftI dst rshift) (LShiftI dst lshift)));
9635
9636 expand %{
9637 rorI_rReg_imm8(dst, rshift, cr);
9638 %}
9639 %}
9640
9641 // Rotate Right by variable
9642 instruct rorI_rReg_Var_C0(no_rcx_RegI dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9643 %{
9644 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI zero shift))));
9645
9646 expand %{
9647 rorI_rReg_CL(dst, shift, cr);
9648 %}
9649 %}
9650
9651 // Rotate Right by variable
9652 instruct rorI_rReg_Var_C32(no_rcx_RegI dst, rcx_RegI shift, immI_32 c32, rFlagsReg cr)
9653 %{
9654 match(Set dst (OrI (URShiftI dst shift) (LShiftI dst (SubI c32 shift))));
9655
9656 expand %{
9657 rorI_rReg_CL(dst, shift, cr);
9658 %}
9659 %}
9660
9661 // for long rotate
9662 // ROL expand
9663 instruct rolL_rReg_imm1(rRegL dst, rFlagsReg cr) %{
9664 effect(USE_DEF dst, KILL cr);
9665
9666 format %{ "rolq $dst" %}
9667 opcode(0xD1, 0x0); /* Opcode D1 /0 */
9668 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9669 ins_pipe(ialu_reg);
9670 %}
9671
9672 instruct rolL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr) %{
9673 effect(USE_DEF dst, USE shift, KILL cr);
9674
9675 format %{ "rolq $dst, $shift" %}
9676 opcode(0xC1, 0x0); /* Opcode C1 /0 ib */
9677 ins_encode( reg_opc_imm_wide(dst, shift) );
9678 ins_pipe(ialu_reg);
9679 %}
9680
9681 instruct rolL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9682 %{
9683 effect(USE_DEF dst, USE shift, KILL cr);
9684
9685 format %{ "rolq $dst, $shift" %}
9686 opcode(0xD3, 0x0); /* Opcode D3 /0 */
9687 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9688 ins_pipe(ialu_reg_reg);
9689 %}
9690 // end of ROL expand
9691
9692 // Rotate Left by one
9693 instruct rolL_rReg_i1(rRegL dst, immI1 lshift, immI_M1 rshift, rFlagsReg cr)
9694 %{
9695 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9696
9697 expand %{
9698 rolL_rReg_imm1(dst, cr);
9699 %}
9700 %}
9701
9702 // Rotate Left by 8-bit immediate
9703 instruct rolL_rReg_i8(rRegL dst, immI8 lshift, immI8 rshift, rFlagsReg cr)
9704 %{
9705 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9706 match(Set dst (OrL (LShiftL dst lshift) (URShiftL dst rshift)));
9707
9708 expand %{
9709 rolL_rReg_imm8(dst, lshift, cr);
9710 %}
9711 %}
9712
9713 // Rotate Left by variable
9714 instruct rolL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9715 %{
9716 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI zero shift))));
9717
9718 expand %{
9719 rolL_rReg_CL(dst, shift, cr);
9720 %}
9721 %}
9722
9723 // Rotate Left by variable
9724 instruct rolL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9725 %{
9726 match(Set dst (OrL (LShiftL dst shift) (URShiftL dst (SubI c64 shift))));
9727
9728 expand %{
9729 rolL_rReg_CL(dst, shift, cr);
9730 %}
9731 %}
9732
9733 // ROR expand
9734 instruct rorL_rReg_imm1(rRegL dst, rFlagsReg cr)
9735 %{
9736 effect(USE_DEF dst, KILL cr);
9737
9738 format %{ "rorq $dst" %}
9739 opcode(0xD1, 0x1); /* D1 /1 */
9740 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9741 ins_pipe(ialu_reg);
9742 %}
9743
9744 instruct rorL_rReg_imm8(rRegL dst, immI8 shift, rFlagsReg cr)
9745 %{
9746 effect(USE_DEF dst, USE shift, KILL cr);
9747
9748 format %{ "rorq $dst, $shift" %}
9749 opcode(0xC1, 0x1); /* C1 /1 ib */
9750 ins_encode(reg_opc_imm_wide(dst, shift));
9751 ins_pipe(ialu_reg);
9752 %}
9753
9754 instruct rorL_rReg_CL(no_rcx_RegL dst, rcx_RegI shift, rFlagsReg cr)
9755 %{
9756 effect(USE_DEF dst, USE shift, KILL cr);
9757
9758 format %{ "rorq $dst, $shift" %}
9759 opcode(0xD3, 0x1); /* D3 /1 */
9760 ins_encode(REX_reg_wide(dst), OpcP, reg_opc(dst));
9761 ins_pipe(ialu_reg_reg);
9762 %}
9763 // end of ROR expand
9764
9765 // Rotate Right by one
9766 instruct rorL_rReg_i1(rRegL dst, immI1 rshift, immI_M1 lshift, rFlagsReg cr)
9767 %{
9768 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9769
9770 expand %{
9771 rorL_rReg_imm1(dst, cr);
9772 %}
9773 %}
9774
9775 // Rotate Right by 8-bit immediate
9776 instruct rorL_rReg_i8(rRegL dst, immI8 rshift, immI8 lshift, rFlagsReg cr)
9777 %{
9778 predicate(0 == ((n->in(1)->in(2)->get_int() + n->in(2)->in(2)->get_int()) & 0x3f));
9779 match(Set dst (OrL (URShiftL dst rshift) (LShiftL dst lshift)));
9780
9781 expand %{
9782 rorL_rReg_imm8(dst, rshift, cr);
9783 %}
9784 %}
9785
9786 // Rotate Right by variable
9787 instruct rorL_rReg_Var_C0(no_rcx_RegL dst, rcx_RegI shift, immI0 zero, rFlagsReg cr)
9788 %{
9789 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI zero shift))));
9790
9791 expand %{
9792 rorL_rReg_CL(dst, shift, cr);
9793 %}
9794 %}
9795
9796 // Rotate Right by variable
9797 instruct rorL_rReg_Var_C64(no_rcx_RegL dst, rcx_RegI shift, immI_64 c64, rFlagsReg cr)
9798 %{
9799 match(Set dst (OrL (URShiftL dst shift) (LShiftL dst (SubI c64 shift))));
9800
9801 expand %{
9802 rorL_rReg_CL(dst, shift, cr);
9803 %}
9804 %}
9805
9806 // Logical Instructions
9807
9808 // Integer Logical Instructions
9809
9810 // And Instructions
9811 // And Register with Register
9812 instruct andI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9813 %{
9814 match(Set dst (AndI dst src));
9815 effect(KILL cr);
9816
9817 format %{ "andl $dst, $src\t# int" %}
9818 opcode(0x23);
9819 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9820 ins_pipe(ialu_reg_reg);
9821 %}
9822
9823 // And Register with Immediate 255
9824 instruct andI_rReg_imm255(rRegI dst, immI_255 src)
9825 %{
9826 match(Set dst (AndI dst src));
9827
9828 format %{ "movzbl $dst, $dst\t# int & 0xFF" %}
9829 opcode(0x0F, 0xB6);
9830 ins_encode(REX_reg_breg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9831 ins_pipe(ialu_reg);
9832 %}
9833
9834 // And Register with Immediate 255 and promote to long
9835 instruct andI2L_rReg_imm255(rRegL dst, rRegI src, immI_255 mask)
9836 %{
9837 match(Set dst (ConvI2L (AndI src mask)));
9838
9839 format %{ "movzbl $dst, $src\t# int & 0xFF -> long" %}
9840 opcode(0x0F, 0xB6);
9841 ins_encode(REX_reg_breg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9842 ins_pipe(ialu_reg);
9843 %}
9844
9845 // And Register with Immediate 65535
9846 instruct andI_rReg_imm65535(rRegI dst, immI_65535 src)
9847 %{
9848 match(Set dst (AndI dst src));
9849
9850 format %{ "movzwl $dst, $dst\t# int & 0xFFFF" %}
9851 opcode(0x0F, 0xB7);
9852 ins_encode(REX_reg_reg(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
9853 ins_pipe(ialu_reg);
9854 %}
9855
9856 // And Register with Immediate 65535 and promote to long
9857 instruct andI2L_rReg_imm65535(rRegL dst, rRegI src, immI_65535 mask)
9858 %{
9859 match(Set dst (ConvI2L (AndI src mask)));
9860
9861 format %{ "movzwl $dst, $src\t# int & 0xFFFF -> long" %}
9862 opcode(0x0F, 0xB7);
9863 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
9864 ins_pipe(ialu_reg);
9865 %}
9866
9867 // And Register with Immediate
9868 instruct andI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9869 %{
9870 match(Set dst (AndI dst src));
9871 effect(KILL cr);
9872
9873 format %{ "andl $dst, $src\t# int" %}
9874 opcode(0x81, 0x04); /* Opcode 81 /4 */
9875 ins_encode(OpcSErm(dst, src), Con8or32(src));
9876 ins_pipe(ialu_reg);
9877 %}
9878
9879 // And Register with Memory
9880 instruct andI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9881 %{
9882 match(Set dst (AndI dst (LoadI src)));
9883 effect(KILL cr);
9884
9885 ins_cost(125);
9886 format %{ "andl $dst, $src\t# int" %}
9887 opcode(0x23);
9888 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9889 ins_pipe(ialu_reg_mem);
9890 %}
9891
9892 // And Memory with Register
9893 instruct andI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9894 %{
9895 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9896 effect(KILL cr);
9897
9898 ins_cost(150);
9899 format %{ "andl $dst, $src\t# int" %}
9900 opcode(0x21); /* Opcode 21 /r */
9901 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9902 ins_pipe(ialu_mem_reg);
9903 %}
9904
9905 // And Memory with Immediate
9906 instruct andI_mem_imm(memory dst, immI src, rFlagsReg cr)
9907 %{
9908 match(Set dst (StoreI dst (AndI (LoadI dst) src)));
9909 effect(KILL cr);
9910
9911 ins_cost(125);
9912 format %{ "andl $dst, $src\t# int" %}
9913 opcode(0x81, 0x4); /* Opcode 81 /4 id */
9914 ins_encode(REX_mem(dst), OpcSE(src),
9915 RM_opc_mem(secondary, dst), Con8or32(src));
9916 ins_pipe(ialu_mem_imm);
9917 %}
9918
9919 // Or Instructions
9920 // Or Register with Register
9921 instruct orI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9922 %{
9923 match(Set dst (OrI dst src));
9924 effect(KILL cr);
9925
9926 format %{ "orl $dst, $src\t# int" %}
9927 opcode(0x0B);
9928 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9929 ins_pipe(ialu_reg_reg);
9930 %}
9931
9932 // Or Register with Immediate
9933 instruct orI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
9934 %{
9935 match(Set dst (OrI dst src));
9936 effect(KILL cr);
9937
9938 format %{ "orl $dst, $src\t# int" %}
9939 opcode(0x81, 0x01); /* Opcode 81 /1 id */
9940 ins_encode(OpcSErm(dst, src), Con8or32(src));
9941 ins_pipe(ialu_reg);
9942 %}
9943
9944 // Or Register with Memory
9945 instruct orI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
9946 %{
9947 match(Set dst (OrI dst (LoadI src)));
9948 effect(KILL cr);
9949
9950 ins_cost(125);
9951 format %{ "orl $dst, $src\t# int" %}
9952 opcode(0x0B);
9953 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
9954 ins_pipe(ialu_reg_mem);
9955 %}
9956
9957 // Or Memory with Register
9958 instruct orI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
9959 %{
9960 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9961 effect(KILL cr);
9962
9963 ins_cost(150);
9964 format %{ "orl $dst, $src\t# int" %}
9965 opcode(0x09); /* Opcode 09 /r */
9966 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
9967 ins_pipe(ialu_mem_reg);
9968 %}
9969
9970 // Or Memory with Immediate
9971 instruct orI_mem_imm(memory dst, immI src, rFlagsReg cr)
9972 %{
9973 match(Set dst (StoreI dst (OrI (LoadI dst) src)));
9974 effect(KILL cr);
9975
9976 ins_cost(125);
9977 format %{ "orl $dst, $src\t# int" %}
9978 opcode(0x81, 0x1); /* Opcode 81 /1 id */
9979 ins_encode(REX_mem(dst), OpcSE(src),
9980 RM_opc_mem(secondary, dst), Con8or32(src));
9981 ins_pipe(ialu_mem_imm);
9982 %}
9983
9984 // Xor Instructions
9985 // Xor Register with Register
9986 instruct xorI_rReg(rRegI dst, rRegI src, rFlagsReg cr)
9987 %{
9988 match(Set dst (XorI dst src));
9989 effect(KILL cr);
9990
9991 format %{ "xorl $dst, $src\t# int" %}
9992 opcode(0x33);
9993 ins_encode(REX_reg_reg(dst, src), OpcP, reg_reg(dst, src));
9994 ins_pipe(ialu_reg_reg);
9995 %}
9996
9997 // Xor Register with Immediate -1
9998 instruct xorI_rReg_im1(rRegI dst, immI_M1 imm) %{
9999 match(Set dst (XorI dst imm));
10000
10001 format %{ "not $dst" %}
10002 ins_encode %{
10003 __ notl($dst$$Register);
10004 %}
10005 ins_pipe(ialu_reg);
10006 %}
10007
10008 // Xor Register with Immediate
10009 instruct xorI_rReg_imm(rRegI dst, immI src, rFlagsReg cr)
10010 %{
10011 match(Set dst (XorI dst src));
10012 effect(KILL cr);
10013
10014 format %{ "xorl $dst, $src\t# int" %}
10015 opcode(0x81, 0x06); /* Opcode 81 /6 id */
10016 ins_encode(OpcSErm(dst, src), Con8or32(src));
10017 ins_pipe(ialu_reg);
10018 %}
10019
10020 // Xor Register with Memory
10021 instruct xorI_rReg_mem(rRegI dst, memory src, rFlagsReg cr)
10022 %{
10023 match(Set dst (XorI dst (LoadI src)));
10024 effect(KILL cr);
10025
10026 ins_cost(125);
10027 format %{ "xorl $dst, $src\t# int" %}
10028 opcode(0x33);
10029 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
10030 ins_pipe(ialu_reg_mem);
10031 %}
10032
10033 // Xor Memory with Register
10034 instruct xorI_mem_rReg(memory dst, rRegI src, rFlagsReg cr)
10035 %{
10036 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
10037 effect(KILL cr);
10038
10039 ins_cost(150);
10040 format %{ "xorl $dst, $src\t# int" %}
10041 opcode(0x31); /* Opcode 31 /r */
10042 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
10043 ins_pipe(ialu_mem_reg);
10044 %}
10045
10046 // Xor Memory with Immediate
10047 instruct xorI_mem_imm(memory dst, immI src, rFlagsReg cr)
10048 %{
10049 match(Set dst (StoreI dst (XorI (LoadI dst) src)));
10050 effect(KILL cr);
10051
10052 ins_cost(125);
10053 format %{ "xorl $dst, $src\t# int" %}
10054 opcode(0x81, 0x6); /* Opcode 81 /6 id */
10055 ins_encode(REX_mem(dst), OpcSE(src),
10056 RM_opc_mem(secondary, dst), Con8or32(src));
10057 ins_pipe(ialu_mem_imm);
10058 %}
10059
10060
10061 // Long Logical Instructions
10062
10063 // And Instructions
10064 // And Register with Register
10065 instruct andL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
10066 %{
10067 match(Set dst (AndL dst src));
10068 effect(KILL cr);
10069
10070 format %{ "andq $dst, $src\t# long" %}
10071 opcode(0x23);
10072 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10073 ins_pipe(ialu_reg_reg);
10074 %}
10075
10076 // And Register with Immediate 255
10077 instruct andL_rReg_imm255(rRegL dst, immL_255 src)
10078 %{
10079 match(Set dst (AndL dst src));
10080
10081 format %{ "movzbq $dst, $dst\t# long & 0xFF" %}
10082 opcode(0x0F, 0xB6);
10083 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
10084 ins_pipe(ialu_reg);
10085 %}
10086
10087 // And Register with Immediate 65535
10088 instruct andL_rReg_imm65535(rRegL dst, immL_65535 src)
10089 %{
10090 match(Set dst (AndL dst src));
10091
10092 format %{ "movzwq $dst, $dst\t# long & 0xFFFF" %}
10093 opcode(0x0F, 0xB7);
10094 ins_encode(REX_reg_reg_wide(dst, dst), OpcP, OpcS, reg_reg(dst, dst));
10095 ins_pipe(ialu_reg);
10096 %}
10097
10098 // And Register with Immediate
10099 instruct andL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
10100 %{
10101 match(Set dst (AndL dst src));
10102 effect(KILL cr);
10103
10104 format %{ "andq $dst, $src\t# long" %}
10105 opcode(0x81, 0x04); /* Opcode 81 /4 */
10106 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
10107 ins_pipe(ialu_reg);
10108 %}
10109
10110 // And Register with Memory
10111 instruct andL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
10112 %{
10113 match(Set dst (AndL dst (LoadL src)));
10114 effect(KILL cr);
10115
10116 ins_cost(125);
10117 format %{ "andq $dst, $src\t# long" %}
10118 opcode(0x23);
10119 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10120 ins_pipe(ialu_reg_mem);
10121 %}
10122
10123 // And Memory with Register
10124 instruct andL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
10125 %{
10126 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
10127 effect(KILL cr);
10128
10129 ins_cost(150);
10130 format %{ "andq $dst, $src\t# long" %}
10131 opcode(0x21); /* Opcode 21 /r */
10132 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10133 ins_pipe(ialu_mem_reg);
10134 %}
10135
10136 // And Memory with Immediate
10137 instruct andL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
10138 %{
10139 match(Set dst (StoreL dst (AndL (LoadL dst) src)));
10140 effect(KILL cr);
10141
10142 ins_cost(125);
10143 format %{ "andq $dst, $src\t# long" %}
10144 opcode(0x81, 0x4); /* Opcode 81 /4 id */
10145 ins_encode(REX_mem_wide(dst), OpcSE(src),
10146 RM_opc_mem(secondary, dst), Con8or32(src));
10147 ins_pipe(ialu_mem_imm);
10148 %}
10149
10150 // Or Instructions
10151 // Or Register with Register
10152 instruct orL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
10153 %{
10154 match(Set dst (OrL dst src));
10155 effect(KILL cr);
10156
10157 format %{ "orq $dst, $src\t# long" %}
10158 opcode(0x0B);
10159 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10160 ins_pipe(ialu_reg_reg);
10161 %}
10162
10163 // Use any_RegP to match R15 (TLS register) without spilling.
10164 instruct orL_rReg_castP2X(rRegL dst, any_RegP src, rFlagsReg cr) %{
10165 match(Set dst (OrL dst (CastP2X src)));
10166 effect(KILL cr);
10167
10168 format %{ "orq $dst, $src\t# long" %}
10169 opcode(0x0B);
10170 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10171 ins_pipe(ialu_reg_reg);
10172 %}
10173
10174
10175 // Or Register with Immediate
10176 instruct orL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
10177 %{
10178 match(Set dst (OrL dst src));
10179 effect(KILL cr);
10180
10181 format %{ "orq $dst, $src\t# long" %}
10182 opcode(0x81, 0x01); /* Opcode 81 /1 id */
10183 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
10184 ins_pipe(ialu_reg);
10185 %}
10186
10187 // Or Register with Memory
10188 instruct orL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
10189 %{
10190 match(Set dst (OrL dst (LoadL src)));
10191 effect(KILL cr);
10192
10193 ins_cost(125);
10194 format %{ "orq $dst, $src\t# long" %}
10195 opcode(0x0B);
10196 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10197 ins_pipe(ialu_reg_mem);
10198 %}
10199
10200 // Or Memory with Register
10201 instruct orL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
10202 %{
10203 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
10204 effect(KILL cr);
10205
10206 ins_cost(150);
10207 format %{ "orq $dst, $src\t# long" %}
10208 opcode(0x09); /* Opcode 09 /r */
10209 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10210 ins_pipe(ialu_mem_reg);
10211 %}
10212
10213 // Or Memory with Immediate
10214 instruct orL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
10215 %{
10216 match(Set dst (StoreL dst (OrL (LoadL dst) src)));
10217 effect(KILL cr);
10218
10219 ins_cost(125);
10220 format %{ "orq $dst, $src\t# long" %}
10221 opcode(0x81, 0x1); /* Opcode 81 /1 id */
10222 ins_encode(REX_mem_wide(dst), OpcSE(src),
10223 RM_opc_mem(secondary, dst), Con8or32(src));
10224 ins_pipe(ialu_mem_imm);
10225 %}
10226
10227 // Xor Instructions
10228 // Xor Register with Register
10229 instruct xorL_rReg(rRegL dst, rRegL src, rFlagsReg cr)
10230 %{
10231 match(Set dst (XorL dst src));
10232 effect(KILL cr);
10233
10234 format %{ "xorq $dst, $src\t# long" %}
10235 opcode(0x33);
10236 ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst, src));
10237 ins_pipe(ialu_reg_reg);
10238 %}
10239
10240 // Xor Register with Immediate -1
10241 instruct xorL_rReg_im1(rRegL dst, immL_M1 imm) %{
10242 match(Set dst (XorL dst imm));
10243
10244 format %{ "notq $dst" %}
10245 ins_encode %{
10246 __ notq($dst$$Register);
10247 %}
10248 ins_pipe(ialu_reg);
10249 %}
10250
10251 // Xor Register with Immediate
10252 instruct xorL_rReg_imm(rRegL dst, immL32 src, rFlagsReg cr)
10253 %{
10254 match(Set dst (XorL dst src));
10255 effect(KILL cr);
10256
10257 format %{ "xorq $dst, $src\t# long" %}
10258 opcode(0x81, 0x06); /* Opcode 81 /6 id */
10259 ins_encode(OpcSErm_wide(dst, src), Con8or32(src));
10260 ins_pipe(ialu_reg);
10261 %}
10262
10263 // Xor Register with Memory
10264 instruct xorL_rReg_mem(rRegL dst, memory src, rFlagsReg cr)
10265 %{
10266 match(Set dst (XorL dst (LoadL src)));
10267 effect(KILL cr);
10268
10269 ins_cost(125);
10270 format %{ "xorq $dst, $src\t# long" %}
10271 opcode(0x33);
10272 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
10273 ins_pipe(ialu_reg_mem);
10274 %}
10275
10276 // Xor Memory with Register
10277 instruct xorL_mem_rReg(memory dst, rRegL src, rFlagsReg cr)
10278 %{
10279 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
10280 effect(KILL cr);
10281
10282 ins_cost(150);
10283 format %{ "xorq $dst, $src\t# long" %}
10284 opcode(0x31); /* Opcode 31 /r */
10285 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
10286 ins_pipe(ialu_mem_reg);
10287 %}
10288
10289 // Xor Memory with Immediate
10290 instruct xorL_mem_imm(memory dst, immL32 src, rFlagsReg cr)
10291 %{
10292 match(Set dst (StoreL dst (XorL (LoadL dst) src)));
10293 effect(KILL cr);
10294
10295 ins_cost(125);
10296 format %{ "xorq $dst, $src\t# long" %}
10297 opcode(0x81, 0x6); /* Opcode 81 /6 id */
10298 ins_encode(REX_mem_wide(dst), OpcSE(src),
10299 RM_opc_mem(secondary, dst), Con8or32(src));
10300 ins_pipe(ialu_mem_imm);
10301 %}
10302
10303 // Convert Int to Boolean
10304 instruct convI2B(rRegI dst, rRegI src, rFlagsReg cr)
10305 %{
10306 match(Set dst (Conv2B src));
10307 effect(KILL cr);
10308
10309 format %{ "testl $src, $src\t# ci2b\n\t"
10310 "setnz $dst\n\t"
10311 "movzbl $dst, $dst" %}
10312 ins_encode(REX_reg_reg(src, src), opc_reg_reg(0x85, src, src), // testl
10313 setNZ_reg(dst),
10314 REX_reg_breg(dst, dst), // movzbl
10315 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10316 ins_pipe(pipe_slow); // XXX
10317 %}
10318
10319 // Convert Pointer to Boolean
10320 instruct convP2B(rRegI dst, rRegP src, rFlagsReg cr)
10321 %{
10322 match(Set dst (Conv2B src));
10323 effect(KILL cr);
10324
10325 format %{ "testq $src, $src\t# cp2b\n\t"
10326 "setnz $dst\n\t"
10327 "movzbl $dst, $dst" %}
10328 ins_encode(REX_reg_reg_wide(src, src), opc_reg_reg(0x85, src, src), // testq
10329 setNZ_reg(dst),
10330 REX_reg_breg(dst, dst), // movzbl
10331 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst));
10332 ins_pipe(pipe_slow); // XXX
10333 %}
10334
10335 instruct cmpLTMask(rRegI dst, rRegI p, rRegI q, rFlagsReg cr)
10336 %{
10337 match(Set dst (CmpLTMask p q));
10338 effect(KILL cr);
10339
10340 ins_cost(400); // XXX
10341 format %{ "cmpl $p, $q\t# cmpLTMask\n\t"
10342 "setlt $dst\n\t"
10343 "movzbl $dst, $dst\n\t"
10344 "negl $dst" %}
10345 ins_encode(REX_reg_reg(p, q), opc_reg_reg(0x3B, p, q), // cmpl
10346 setLT_reg(dst),
10347 REX_reg_breg(dst, dst), // movzbl
10348 Opcode(0x0F), Opcode(0xB6), reg_reg(dst, dst),
10349 neg_reg(dst));
10350 ins_pipe(pipe_slow);
10351 %}
10352
10353 instruct cmpLTMask0(rRegI dst, immI0 zero, rFlagsReg cr)
10354 %{
10355 match(Set dst (CmpLTMask dst zero));
10356 effect(KILL cr);
10357
10358 ins_cost(100); // XXX
10359 format %{ "sarl $dst, #31\t# cmpLTMask0" %}
10360 opcode(0xC1, 0x7); /* C1 /7 ib */
10361 ins_encode(reg_opc_imm(dst, 0x1F));
10362 ins_pipe(ialu_reg);
10363 %}
10364
10365
10366 instruct cadd_cmpLTMask(rRegI p, rRegI q, rRegI y,
10367 rRegI tmp,
10368 rFlagsReg cr)
10369 %{
10370 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
10371 effect(TEMP tmp, KILL cr);
10372
10373 ins_cost(400); // XXX
10374 format %{ "subl $p, $q\t# cadd_cmpLTMask1\n\t"
10375 "sbbl $tmp, $tmp\n\t"
10376 "andl $tmp, $y\n\t"
10377 "addl $p, $tmp" %}
10378 ins_encode(enc_cmpLTP(p, q, y, tmp));
10379 ins_pipe(pipe_cmplt);
10380 %}
10381
10382 /* If I enable this, I encourage spilling in the inner loop of compress.
10383 instruct cadd_cmpLTMask_mem( rRegI p, rRegI q, memory y, rRegI tmp, rFlagsReg cr )
10384 %{
10385 match(Set p (AddI (AndI (CmpLTMask p q) (LoadI y)) (SubI p q)));
10386 effect( TEMP tmp, KILL cr );
10387 ins_cost(400);
10388
10389 format %{ "SUB $p,$q\n\t"
10390 "SBB RCX,RCX\n\t"
10391 "AND RCX,$y\n\t"
10392 "ADD $p,RCX" %}
10393 ins_encode( enc_cmpLTP_mem(p,q,y,tmp) );
10394 %}
10395 */
10396
10397 //---------- FP Instructions------------------------------------------------
10398
10399 instruct cmpF_cc_reg(rFlagsRegU cr, regF src1, regF src2)
10400 %{
10401 match(Set cr (CmpF src1 src2));
10402
10403 ins_cost(145);
10404 format %{ "ucomiss $src1, $src2\n\t"
10405 "jnp,s exit\n\t"
10406 "pushfq\t# saw NaN, set CF\n\t"
10407 "andq [rsp], #0xffffff2b\n\t"
10408 "popfq\n"
10409 "exit: nop\t# avoid branch to branch" %}
10410 opcode(0x0F, 0x2E);
10411 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10412 cmpfp_fixup);
10413 ins_pipe(pipe_slow);
10414 %}
10415
10416 instruct cmpF_cc_reg_CF(rFlagsRegUCF cr, regF src1, regF src2) %{
10417 match(Set cr (CmpF src1 src2));
10418
10419 ins_cost(145);
10420 format %{ "ucomiss $src1, $src2" %}
10421 ins_encode %{
10422 __ ucomiss($src1$$XMMRegister, $src2$$XMMRegister);
10423 %}
10424 ins_pipe(pipe_slow);
10425 %}
10426
10427 instruct cmpF_cc_mem(rFlagsRegU cr, regF src1, memory src2)
10428 %{
10429 match(Set cr (CmpF src1 (LoadF src2)));
10430
10431 ins_cost(145);
10432 format %{ "ucomiss $src1, $src2\n\t"
10433 "jnp,s exit\n\t"
10434 "pushfq\t# saw NaN, set CF\n\t"
10435 "andq [rsp], #0xffffff2b\n\t"
10436 "popfq\n"
10437 "exit: nop\t# avoid branch to branch" %}
10438 opcode(0x0F, 0x2E);
10439 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10440 cmpfp_fixup);
10441 ins_pipe(pipe_slow);
10442 %}
10443
10444 instruct cmpF_cc_memCF(rFlagsRegUCF cr, regF src1, memory src2) %{
10445 match(Set cr (CmpF src1 (LoadF src2)));
10446
10447 ins_cost(100);
10448 format %{ "ucomiss $src1, $src2" %}
10449 opcode(0x0F, 0x2E);
10450 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2));
10451 ins_pipe(pipe_slow);
10452 %}
10453
10454 instruct cmpF_cc_imm(rFlagsRegU cr, regF src1, immF src2)
10455 %{
10456 match(Set cr (CmpF src1 src2));
10457
10458 ins_cost(145);
10459 format %{ "ucomiss $src1, $src2\n\t"
10460 "jnp,s exit\n\t"
10461 "pushfq\t# saw NaN, set CF\n\t"
10462 "andq [rsp], #0xffffff2b\n\t"
10463 "popfq\n"
10464 "exit: nop\t# avoid branch to branch" %}
10465 opcode(0x0F, 0x2E);
10466 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
10467 cmpfp_fixup);
10468 ins_pipe(pipe_slow);
10469 %}
10470
10471 instruct cmpF_cc_immCF(rFlagsRegUCF cr, regF src1, immF src2) %{
10472 match(Set cr (CmpF src1 src2));
10473
10474 ins_cost(100);
10475 format %{ "ucomiss $src1, $src2" %}
10476 opcode(0x0F, 0x2E);
10477 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2));
10478 ins_pipe(pipe_slow);
10479 %}
10480
10481 instruct cmpD_cc_reg(rFlagsRegU cr, regD src1, regD src2)
10482 %{
10483 match(Set cr (CmpD src1 src2));
10484
10485 ins_cost(145);
10486 format %{ "ucomisd $src1, $src2\n\t"
10487 "jnp,s exit\n\t"
10488 "pushfq\t# saw NaN, set CF\n\t"
10489 "andq [rsp], #0xffffff2b\n\t"
10490 "popfq\n"
10491 "exit: nop\t# avoid branch to branch" %}
10492 opcode(0x66, 0x0F, 0x2E);
10493 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10494 cmpfp_fixup);
10495 ins_pipe(pipe_slow);
10496 %}
10497
10498 instruct cmpD_cc_reg_CF(rFlagsRegUCF cr, regD src1, regD src2) %{
10499 match(Set cr (CmpD src1 src2));
10500
10501 ins_cost(100);
10502 format %{ "ucomisd $src1, $src2 test" %}
10503 ins_encode %{
10504 __ ucomisd($src1$$XMMRegister, $src2$$XMMRegister);
10505 %}
10506 ins_pipe(pipe_slow);
10507 %}
10508
10509 instruct cmpD_cc_mem(rFlagsRegU cr, regD src1, memory src2)
10510 %{
10511 match(Set cr (CmpD src1 (LoadD src2)));
10512
10513 ins_cost(145);
10514 format %{ "ucomisd $src1, $src2\n\t"
10515 "jnp,s exit\n\t"
10516 "pushfq\t# saw NaN, set CF\n\t"
10517 "andq [rsp], #0xffffff2b\n\t"
10518 "popfq\n"
10519 "exit: nop\t# avoid branch to branch" %}
10520 opcode(0x66, 0x0F, 0x2E);
10521 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10522 cmpfp_fixup);
10523 ins_pipe(pipe_slow);
10524 %}
10525
10526 instruct cmpD_cc_memCF(rFlagsRegUCF cr, regD src1, memory src2) %{
10527 match(Set cr (CmpD src1 (LoadD src2)));
10528
10529 ins_cost(100);
10530 format %{ "ucomisd $src1, $src2" %}
10531 opcode(0x66, 0x0F, 0x2E);
10532 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2));
10533 ins_pipe(pipe_slow);
10534 %}
10535
10536 instruct cmpD_cc_imm(rFlagsRegU cr, regD src1, immD src2)
10537 %{
10538 match(Set cr (CmpD src1 src2));
10539
10540 ins_cost(145);
10541 format %{ "ucomisd $src1, [$src2]\n\t"
10542 "jnp,s exit\n\t"
10543 "pushfq\t# saw NaN, set CF\n\t"
10544 "andq [rsp], #0xffffff2b\n\t"
10545 "popfq\n"
10546 "exit: nop\t# avoid branch to branch" %}
10547 opcode(0x66, 0x0F, 0x2E);
10548 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
10549 cmpfp_fixup);
10550 ins_pipe(pipe_slow);
10551 %}
10552
10553 instruct cmpD_cc_immCF(rFlagsRegUCF cr, regD src1, immD src2) %{
10554 match(Set cr (CmpD src1 src2));
10555
10556 ins_cost(100);
10557 format %{ "ucomisd $src1, [$src2]" %}
10558 opcode(0x66, 0x0F, 0x2E);
10559 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2));
10560 ins_pipe(pipe_slow);
10561 %}
10562
10563 // Compare into -1,0,1
10564 instruct cmpF_reg(rRegI dst, regF src1, regF src2, rFlagsReg cr)
10565 %{
10566 match(Set dst (CmpF3 src1 src2));
10567 effect(KILL cr);
10568
10569 ins_cost(275);
10570 format %{ "ucomiss $src1, $src2\n\t"
10571 "movl $dst, #-1\n\t"
10572 "jp,s done\n\t"
10573 "jb,s done\n\t"
10574 "setne $dst\n\t"
10575 "movzbl $dst, $dst\n"
10576 "done:" %}
10577
10578 opcode(0x0F, 0x2E);
10579 ins_encode(REX_reg_reg(src1, src2), OpcP, OpcS, reg_reg(src1, src2),
10580 cmpfp3(dst));
10581 ins_pipe(pipe_slow);
10582 %}
10583
10584 // Compare into -1,0,1
10585 instruct cmpF_mem(rRegI dst, regF src1, memory src2, rFlagsReg cr)
10586 %{
10587 match(Set dst (CmpF3 src1 (LoadF src2)));
10588 effect(KILL cr);
10589
10590 ins_cost(275);
10591 format %{ "ucomiss $src1, $src2\n\t"
10592 "movl $dst, #-1\n\t"
10593 "jp,s done\n\t"
10594 "jb,s done\n\t"
10595 "setne $dst\n\t"
10596 "movzbl $dst, $dst\n"
10597 "done:" %}
10598
10599 opcode(0x0F, 0x2E);
10600 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, reg_mem(src1, src2),
10601 cmpfp3(dst));
10602 ins_pipe(pipe_slow);
10603 %}
10604
10605 // Compare into -1,0,1
10606 instruct cmpF_imm(rRegI dst, regF src1, immF src2, rFlagsReg cr)
10607 %{
10608 match(Set dst (CmpF3 src1 src2));
10609 effect(KILL cr);
10610
10611 ins_cost(275);
10612 format %{ "ucomiss $src1, [$src2]\n\t"
10613 "movl $dst, #-1\n\t"
10614 "jp,s done\n\t"
10615 "jb,s done\n\t"
10616 "setne $dst\n\t"
10617 "movzbl $dst, $dst\n"
10618 "done:" %}
10619
10620 opcode(0x0F, 0x2E);
10621 ins_encode(REX_reg_mem(src1, src2), OpcP, OpcS, load_immF(src1, src2),
10622 cmpfp3(dst));
10623 ins_pipe(pipe_slow);
10624 %}
10625
10626 // Compare into -1,0,1
10627 instruct cmpD_reg(rRegI dst, regD src1, regD src2, rFlagsReg cr)
10628 %{
10629 match(Set dst (CmpD3 src1 src2));
10630 effect(KILL cr);
10631
10632 ins_cost(275);
10633 format %{ "ucomisd $src1, $src2\n\t"
10634 "movl $dst, #-1\n\t"
10635 "jp,s done\n\t"
10636 "jb,s done\n\t"
10637 "setne $dst\n\t"
10638 "movzbl $dst, $dst\n"
10639 "done:" %}
10640
10641 opcode(0x66, 0x0F, 0x2E);
10642 ins_encode(OpcP, REX_reg_reg(src1, src2), OpcS, OpcT, reg_reg(src1, src2),
10643 cmpfp3(dst));
10644 ins_pipe(pipe_slow);
10645 %}
10646
10647 // Compare into -1,0,1
10648 instruct cmpD_mem(rRegI dst, regD src1, memory src2, rFlagsReg cr)
10649 %{
10650 match(Set dst (CmpD3 src1 (LoadD src2)));
10651 effect(KILL cr);
10652
10653 ins_cost(275);
10654 format %{ "ucomisd $src1, $src2\n\t"
10655 "movl $dst, #-1\n\t"
10656 "jp,s done\n\t"
10657 "jb,s done\n\t"
10658 "setne $dst\n\t"
10659 "movzbl $dst, $dst\n"
10660 "done:" %}
10661
10662 opcode(0x66, 0x0F, 0x2E);
10663 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, reg_mem(src1, src2),
10664 cmpfp3(dst));
10665 ins_pipe(pipe_slow);
10666 %}
10667
10668 // Compare into -1,0,1
10669 instruct cmpD_imm(rRegI dst, regD src1, immD src2, rFlagsReg cr)
10670 %{
10671 match(Set dst (CmpD3 src1 src2));
10672 effect(KILL cr);
10673
10674 ins_cost(275);
10675 format %{ "ucomisd $src1, [$src2]\n\t"
10676 "movl $dst, #-1\n\t"
10677 "jp,s done\n\t"
10678 "jb,s done\n\t"
10679 "setne $dst\n\t"
10680 "movzbl $dst, $dst\n"
10681 "done:" %}
10682
10683 opcode(0x66, 0x0F, 0x2E);
10684 ins_encode(OpcP, REX_reg_mem(src1, src2), OpcS, OpcT, load_immD(src1, src2),
10685 cmpfp3(dst));
10686 ins_pipe(pipe_slow);
10687 %}
10688
10689 instruct addF_reg(regF dst, regF src)
10690 %{
10691 match(Set dst (AddF dst src));
10692
10693 format %{ "addss $dst, $src" %}
10694 ins_cost(150); // XXX
10695 opcode(0xF3, 0x0F, 0x58);
10696 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10697 ins_pipe(pipe_slow);
10698 %}
10699
10700 instruct addF_mem(regF dst, memory src)
10701 %{
10702 match(Set dst (AddF dst (LoadF src)));
10703
10704 format %{ "addss $dst, $src" %}
10705 ins_cost(150); // XXX
10706 opcode(0xF3, 0x0F, 0x58);
10707 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10708 ins_pipe(pipe_slow);
10709 %}
10710
10711 instruct addF_imm(regF dst, immF src)
10712 %{
10713 match(Set dst (AddF dst src));
10714
10715 format %{ "addss $dst, [$src]" %}
10716 ins_cost(150); // XXX
10717 opcode(0xF3, 0x0F, 0x58);
10718 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10719 ins_pipe(pipe_slow);
10720 %}
10721
10722 instruct addD_reg(regD dst, regD src)
10723 %{
10724 match(Set dst (AddD dst src));
10725
10726 format %{ "addsd $dst, $src" %}
10727 ins_cost(150); // XXX
10728 opcode(0xF2, 0x0F, 0x58);
10729 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10730 ins_pipe(pipe_slow);
10731 %}
10732
10733 instruct addD_mem(regD dst, memory src)
10734 %{
10735 match(Set dst (AddD dst (LoadD src)));
10736
10737 format %{ "addsd $dst, $src" %}
10738 ins_cost(150); // XXX
10739 opcode(0xF2, 0x0F, 0x58);
10740 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10741 ins_pipe(pipe_slow);
10742 %}
10743
10744 instruct addD_imm(regD dst, immD src)
10745 %{
10746 match(Set dst (AddD dst src));
10747
10748 format %{ "addsd $dst, [$src]" %}
10749 ins_cost(150); // XXX
10750 opcode(0xF2, 0x0F, 0x58);
10751 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10752 ins_pipe(pipe_slow);
10753 %}
10754
10755 instruct subF_reg(regF dst, regF src)
10756 %{
10757 match(Set dst (SubF dst src));
10758
10759 format %{ "subss $dst, $src" %}
10760 ins_cost(150); // XXX
10761 opcode(0xF3, 0x0F, 0x5C);
10762 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10763 ins_pipe(pipe_slow);
10764 %}
10765
10766 instruct subF_mem(regF dst, memory src)
10767 %{
10768 match(Set dst (SubF dst (LoadF src)));
10769
10770 format %{ "subss $dst, $src" %}
10771 ins_cost(150); // XXX
10772 opcode(0xF3, 0x0F, 0x5C);
10773 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10774 ins_pipe(pipe_slow);
10775 %}
10776
10777 instruct subF_imm(regF dst, immF src)
10778 %{
10779 match(Set dst (SubF dst src));
10780
10781 format %{ "subss $dst, [$src]" %}
10782 ins_cost(150); // XXX
10783 opcode(0xF3, 0x0F, 0x5C);
10784 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10785 ins_pipe(pipe_slow);
10786 %}
10787
10788 instruct subD_reg(regD dst, regD src)
10789 %{
10790 match(Set dst (SubD dst src));
10791
10792 format %{ "subsd $dst, $src" %}
10793 ins_cost(150); // XXX
10794 opcode(0xF2, 0x0F, 0x5C);
10795 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10796 ins_pipe(pipe_slow);
10797 %}
10798
10799 instruct subD_mem(regD dst, memory src)
10800 %{
10801 match(Set dst (SubD dst (LoadD src)));
10802
10803 format %{ "subsd $dst, $src" %}
10804 ins_cost(150); // XXX
10805 opcode(0xF2, 0x0F, 0x5C);
10806 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10807 ins_pipe(pipe_slow);
10808 %}
10809
10810 instruct subD_imm(regD dst, immD src)
10811 %{
10812 match(Set dst (SubD dst src));
10813
10814 format %{ "subsd $dst, [$src]" %}
10815 ins_cost(150); // XXX
10816 opcode(0xF2, 0x0F, 0x5C);
10817 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10818 ins_pipe(pipe_slow);
10819 %}
10820
10821 instruct mulF_reg(regF dst, regF src)
10822 %{
10823 match(Set dst (MulF dst src));
10824
10825 format %{ "mulss $dst, $src" %}
10826 ins_cost(150); // XXX
10827 opcode(0xF3, 0x0F, 0x59);
10828 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10829 ins_pipe(pipe_slow);
10830 %}
10831
10832 instruct mulF_mem(regF dst, memory src)
10833 %{
10834 match(Set dst (MulF dst (LoadF src)));
10835
10836 format %{ "mulss $dst, $src" %}
10837 ins_cost(150); // XXX
10838 opcode(0xF3, 0x0F, 0x59);
10839 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10840 ins_pipe(pipe_slow);
10841 %}
10842
10843 instruct mulF_imm(regF dst, immF src)
10844 %{
10845 match(Set dst (MulF dst src));
10846
10847 format %{ "mulss $dst, [$src]" %}
10848 ins_cost(150); // XXX
10849 opcode(0xF3, 0x0F, 0x59);
10850 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10851 ins_pipe(pipe_slow);
10852 %}
10853
10854 instruct mulD_reg(regD dst, regD src)
10855 %{
10856 match(Set dst (MulD dst src));
10857
10858 format %{ "mulsd $dst, $src" %}
10859 ins_cost(150); // XXX
10860 opcode(0xF2, 0x0F, 0x59);
10861 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10862 ins_pipe(pipe_slow);
10863 %}
10864
10865 instruct mulD_mem(regD dst, memory src)
10866 %{
10867 match(Set dst (MulD dst (LoadD src)));
10868
10869 format %{ "mulsd $dst, $src" %}
10870 ins_cost(150); // XXX
10871 opcode(0xF2, 0x0F, 0x59);
10872 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10873 ins_pipe(pipe_slow);
10874 %}
10875
10876 instruct mulD_imm(regD dst, immD src)
10877 %{
10878 match(Set dst (MulD dst src));
10879
10880 format %{ "mulsd $dst, [$src]" %}
10881 ins_cost(150); // XXX
10882 opcode(0xF2, 0x0F, 0x59);
10883 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10884 ins_pipe(pipe_slow);
10885 %}
10886
10887 instruct divF_reg(regF dst, regF src)
10888 %{
10889 match(Set dst (DivF dst src));
10890
10891 format %{ "divss $dst, $src" %}
10892 ins_cost(150); // XXX
10893 opcode(0xF3, 0x0F, 0x5E);
10894 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10895 ins_pipe(pipe_slow);
10896 %}
10897
10898 instruct divF_mem(regF dst, memory src)
10899 %{
10900 match(Set dst (DivF dst (LoadF src)));
10901
10902 format %{ "divss $dst, $src" %}
10903 ins_cost(150); // XXX
10904 opcode(0xF3, 0x0F, 0x5E);
10905 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10906 ins_pipe(pipe_slow);
10907 %}
10908
10909 instruct divF_imm(regF dst, immF src)
10910 %{
10911 match(Set dst (DivF dst src));
10912
10913 format %{ "divss $dst, [$src]" %}
10914 ins_cost(150); // XXX
10915 opcode(0xF3, 0x0F, 0x5E);
10916 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10917 ins_pipe(pipe_slow);
10918 %}
10919
10920 instruct divD_reg(regD dst, regD src)
10921 %{
10922 match(Set dst (DivD dst src));
10923
10924 format %{ "divsd $dst, $src" %}
10925 ins_cost(150); // XXX
10926 opcode(0xF2, 0x0F, 0x5E);
10927 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10928 ins_pipe(pipe_slow);
10929 %}
10930
10931 instruct divD_mem(regD dst, memory src)
10932 %{
10933 match(Set dst (DivD dst (LoadD src)));
10934
10935 format %{ "divsd $dst, $src" %}
10936 ins_cost(150); // XXX
10937 opcode(0xF2, 0x0F, 0x5E);
10938 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10939 ins_pipe(pipe_slow);
10940 %}
10941
10942 instruct divD_imm(regD dst, immD src)
10943 %{
10944 match(Set dst (DivD dst src));
10945
10946 format %{ "divsd $dst, [$src]" %}
10947 ins_cost(150); // XXX
10948 opcode(0xF2, 0x0F, 0x5E);
10949 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
10950 ins_pipe(pipe_slow);
10951 %}
10952
10953 instruct sqrtF_reg(regF dst, regF src)
10954 %{
10955 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10956
10957 format %{ "sqrtss $dst, $src" %}
10958 ins_cost(150); // XXX
10959 opcode(0xF3, 0x0F, 0x51);
10960 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10961 ins_pipe(pipe_slow);
10962 %}
10963
10964 instruct sqrtF_mem(regF dst, memory src)
10965 %{
10966 match(Set dst (ConvD2F (SqrtD (ConvF2D (LoadF src)))));
10967
10968 format %{ "sqrtss $dst, $src" %}
10969 ins_cost(150); // XXX
10970 opcode(0xF3, 0x0F, 0x51);
10971 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
10972 ins_pipe(pipe_slow);
10973 %}
10974
10975 instruct sqrtF_imm(regF dst, immF src)
10976 %{
10977 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
10978
10979 format %{ "sqrtss $dst, [$src]" %}
10980 ins_cost(150); // XXX
10981 opcode(0xF3, 0x0F, 0x51);
10982 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immF(dst, src));
10983 ins_pipe(pipe_slow);
10984 %}
10985
10986 instruct sqrtD_reg(regD dst, regD src)
10987 %{
10988 match(Set dst (SqrtD src));
10989
10990 format %{ "sqrtsd $dst, $src" %}
10991 ins_cost(150); // XXX
10992 opcode(0xF2, 0x0F, 0x51);
10993 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
10994 ins_pipe(pipe_slow);
10995 %}
10996
10997 instruct sqrtD_mem(regD dst, memory src)
10998 %{
10999 match(Set dst (SqrtD (LoadD src)));
11000
11001 format %{ "sqrtsd $dst, $src" %}
11002 ins_cost(150); // XXX
11003 opcode(0xF2, 0x0F, 0x51);
11004 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11005 ins_pipe(pipe_slow);
11006 %}
11007
11008 instruct sqrtD_imm(regD dst, immD src)
11009 %{
11010 match(Set dst (SqrtD src));
11011
11012 format %{ "sqrtsd $dst, [$src]" %}
11013 ins_cost(150); // XXX
11014 opcode(0xF2, 0x0F, 0x51);
11015 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, load_immD(dst, src));
11016 ins_pipe(pipe_slow);
11017 %}
11018
11019 instruct absF_reg(regF dst)
11020 %{
11021 match(Set dst (AbsF dst));
11022
11023 format %{ "andps $dst, [0x7fffffff]\t# abs float by sign masking" %}
11024 ins_encode(absF_encoding(dst));
11025 ins_pipe(pipe_slow);
11026 %}
11027
11028 instruct absD_reg(regD dst)
11029 %{
11030 match(Set dst (AbsD dst));
11031
11032 format %{ "andpd $dst, [0x7fffffffffffffff]\t"
11033 "# abs double by sign masking" %}
11034 ins_encode(absD_encoding(dst));
11035 ins_pipe(pipe_slow);
11036 %}
11037
11038 instruct negF_reg(regF dst)
11039 %{
11040 match(Set dst (NegF dst));
11041
11042 format %{ "xorps $dst, [0x80000000]\t# neg float by sign flipping" %}
11043 ins_encode(negF_encoding(dst));
11044 ins_pipe(pipe_slow);
11045 %}
11046
11047 instruct negD_reg(regD dst)
11048 %{
11049 match(Set dst (NegD dst));
11050
11051 format %{ "xorpd $dst, [0x8000000000000000]\t"
11052 "# neg double by sign flipping" %}
11053 ins_encode(negD_encoding(dst));
11054 ins_pipe(pipe_slow);
11055 %}
11056
11057 // -----------Trig and Trancendental Instructions------------------------------
11058 instruct cosD_reg(regD dst) %{
11059 match(Set dst (CosD dst));
11060
11061 format %{ "dcos $dst\n\t" %}
11062 opcode(0xD9, 0xFF);
11063 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
11064 ins_pipe( pipe_slow );
11065 %}
11066
11067 instruct sinD_reg(regD dst) %{
11068 match(Set dst (SinD dst));
11069
11070 format %{ "dsin $dst\n\t" %}
11071 opcode(0xD9, 0xFE);
11072 ins_encode( Push_SrcXD(dst), OpcP, OpcS, Push_ResultXD(dst) );
11073 ins_pipe( pipe_slow );
11074 %}
11075
11076 instruct tanD_reg(regD dst) %{
11077 match(Set dst (TanD dst));
11078
11079 format %{ "dtan $dst\n\t" %}
11080 ins_encode( Push_SrcXD(dst),
11081 Opcode(0xD9), Opcode(0xF2), //fptan
11082 Opcode(0xDD), Opcode(0xD8), //fstp st
11083 Push_ResultXD(dst) );
11084 ins_pipe( pipe_slow );
11085 %}
11086
11087 instruct log10D_reg(regD dst) %{
11088 // The source and result Double operands in XMM registers
11089 match(Set dst (Log10D dst));
11090 // fldlg2 ; push log_10(2) on the FPU stack; full 80-bit number
11091 // fyl2x ; compute log_10(2) * log_2(x)
11092 format %{ "fldlg2\t\t\t#Log10\n\t"
11093 "fyl2x\t\t\t# Q=Log10*Log_2(x)\n\t"
11094 %}
11095 ins_encode(Opcode(0xD9), Opcode(0xEC), // fldlg2
11096 Push_SrcXD(dst),
11097 Opcode(0xD9), Opcode(0xF1), // fyl2x
11098 Push_ResultXD(dst));
11099
11100 ins_pipe( pipe_slow );
11101 %}
11102
11103 instruct logD_reg(regD dst) %{
11104 // The source and result Double operands in XMM registers
11105 match(Set dst (LogD dst));
11106 // fldln2 ; push log_e(2) on the FPU stack; full 80-bit number
11107 // fyl2x ; compute log_e(2) * log_2(x)
11108 format %{ "fldln2\t\t\t#Log_e\n\t"
11109 "fyl2x\t\t\t# Q=Log_e*Log_2(x)\n\t"
11110 %}
11111 ins_encode( Opcode(0xD9), Opcode(0xED), // fldln2
11112 Push_SrcXD(dst),
11113 Opcode(0xD9), Opcode(0xF1), // fyl2x
11114 Push_ResultXD(dst));
11115 ins_pipe( pipe_slow );
11116 %}
11117
11118
11119
11120 //----------Arithmetic Conversion Instructions---------------------------------
11121
11122 instruct roundFloat_nop(regF dst)
11123 %{
11124 match(Set dst (RoundFloat dst));
11125
11126 ins_cost(0);
11127 ins_encode();
11128 ins_pipe(empty);
11129 %}
11130
11131 instruct roundDouble_nop(regD dst)
11132 %{
11133 match(Set dst (RoundDouble dst));
11134
11135 ins_cost(0);
11136 ins_encode();
11137 ins_pipe(empty);
11138 %}
11139
11140 instruct convF2D_reg_reg(regD dst, regF src)
11141 %{
11142 match(Set dst (ConvF2D src));
11143
11144 format %{ "cvtss2sd $dst, $src" %}
11145 opcode(0xF3, 0x0F, 0x5A);
11146 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11147 ins_pipe(pipe_slow); // XXX
11148 %}
11149
11150 instruct convF2D_reg_mem(regD dst, memory src)
11151 %{
11152 match(Set dst (ConvF2D (LoadF src)));
11153
11154 format %{ "cvtss2sd $dst, $src" %}
11155 opcode(0xF3, 0x0F, 0x5A);
11156 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11157 ins_pipe(pipe_slow); // XXX
11158 %}
11159
11160 instruct convD2F_reg_reg(regF dst, regD src)
11161 %{
11162 match(Set dst (ConvD2F src));
11163
11164 format %{ "cvtsd2ss $dst, $src" %}
11165 opcode(0xF2, 0x0F, 0x5A);
11166 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11167 ins_pipe(pipe_slow); // XXX
11168 %}
11169
11170 instruct convD2F_reg_mem(regF dst, memory src)
11171 %{
11172 match(Set dst (ConvD2F (LoadD src)));
11173
11174 format %{ "cvtsd2ss $dst, $src" %}
11175 opcode(0xF2, 0x0F, 0x5A);
11176 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11177 ins_pipe(pipe_slow); // XXX
11178 %}
11179
11180 // XXX do mem variants
11181 instruct convF2I_reg_reg(rRegI dst, regF src, rFlagsReg cr)
11182 %{
11183 match(Set dst (ConvF2I src));
11184 effect(KILL cr);
11185
11186 format %{ "cvttss2sil $dst, $src\t# f2i\n\t"
11187 "cmpl $dst, #0x80000000\n\t"
11188 "jne,s done\n\t"
11189 "subq rsp, #8\n\t"
11190 "movss [rsp], $src\n\t"
11191 "call f2i_fixup\n\t"
11192 "popq $dst\n"
11193 "done: "%}
11194 opcode(0xF3, 0x0F, 0x2C);
11195 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
11196 f2i_fixup(dst, src));
11197 ins_pipe(pipe_slow);
11198 %}
11199
11200 instruct convF2L_reg_reg(rRegL dst, regF src, rFlagsReg cr)
11201 %{
11202 match(Set dst (ConvF2L src));
11203 effect(KILL cr);
11204
11205 format %{ "cvttss2siq $dst, $src\t# f2l\n\t"
11206 "cmpq $dst, [0x8000000000000000]\n\t"
11207 "jne,s done\n\t"
11208 "subq rsp, #8\n\t"
11209 "movss [rsp], $src\n\t"
11210 "call f2l_fixup\n\t"
11211 "popq $dst\n"
11212 "done: "%}
11213 opcode(0xF3, 0x0F, 0x2C);
11214 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
11215 f2l_fixup(dst, src));
11216 ins_pipe(pipe_slow);
11217 %}
11218
11219 instruct convD2I_reg_reg(rRegI dst, regD src, rFlagsReg cr)
11220 %{
11221 match(Set dst (ConvD2I src));
11222 effect(KILL cr);
11223
11224 format %{ "cvttsd2sil $dst, $src\t# d2i\n\t"
11225 "cmpl $dst, #0x80000000\n\t"
11226 "jne,s done\n\t"
11227 "subq rsp, #8\n\t"
11228 "movsd [rsp], $src\n\t"
11229 "call d2i_fixup\n\t"
11230 "popq $dst\n"
11231 "done: "%}
11232 opcode(0xF2, 0x0F, 0x2C);
11233 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src),
11234 d2i_fixup(dst, src));
11235 ins_pipe(pipe_slow);
11236 %}
11237
11238 instruct convD2L_reg_reg(rRegL dst, regD src, rFlagsReg cr)
11239 %{
11240 match(Set dst (ConvD2L src));
11241 effect(KILL cr);
11242
11243 format %{ "cvttsd2siq $dst, $src\t# d2l\n\t"
11244 "cmpq $dst, [0x8000000000000000]\n\t"
11245 "jne,s done\n\t"
11246 "subq rsp, #8\n\t"
11247 "movsd [rsp], $src\n\t"
11248 "call d2l_fixup\n\t"
11249 "popq $dst\n"
11250 "done: "%}
11251 opcode(0xF2, 0x0F, 0x2C);
11252 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src),
11253 d2l_fixup(dst, src));
11254 ins_pipe(pipe_slow);
11255 %}
11256
11257 instruct convI2F_reg_reg(regF dst, rRegI src)
11258 %{
11259 predicate(!UseXmmI2F);
11260 match(Set dst (ConvI2F src));
11261
11262 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
11263 opcode(0xF3, 0x0F, 0x2A);
11264 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11265 ins_pipe(pipe_slow); // XXX
11266 %}
11267
11268 instruct convI2F_reg_mem(regF dst, memory src)
11269 %{
11270 match(Set dst (ConvI2F (LoadI src)));
11271
11272 format %{ "cvtsi2ssl $dst, $src\t# i2f" %}
11273 opcode(0xF3, 0x0F, 0x2A);
11274 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11275 ins_pipe(pipe_slow); // XXX
11276 %}
11277
11278 instruct convI2D_reg_reg(regD dst, rRegI src)
11279 %{
11280 predicate(!UseXmmI2D);
11281 match(Set dst (ConvI2D src));
11282
11283 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
11284 opcode(0xF2, 0x0F, 0x2A);
11285 ins_encode(OpcP, REX_reg_reg(dst, src), OpcS, OpcT, reg_reg(dst, src));
11286 ins_pipe(pipe_slow); // XXX
11287 %}
11288
11289 instruct convI2D_reg_mem(regD dst, memory src)
11290 %{
11291 match(Set dst (ConvI2D (LoadI src)));
11292
11293 format %{ "cvtsi2sdl $dst, $src\t# i2d" %}
11294 opcode(0xF2, 0x0F, 0x2A);
11295 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11296 ins_pipe(pipe_slow); // XXX
11297 %}
11298
11299 instruct convXI2F_reg(regF dst, rRegI src)
11300 %{
11301 predicate(UseXmmI2F);
11302 match(Set dst (ConvI2F src));
11303
11304 format %{ "movdl $dst, $src\n\t"
11305 "cvtdq2psl $dst, $dst\t# i2f" %}
11306 ins_encode %{
11307 __ movdl($dst$$XMMRegister, $src$$Register);
11308 __ cvtdq2ps($dst$$XMMRegister, $dst$$XMMRegister);
11309 %}
11310 ins_pipe(pipe_slow); // XXX
11311 %}
11312
11313 instruct convXI2D_reg(regD dst, rRegI src)
11314 %{
11315 predicate(UseXmmI2D);
11316 match(Set dst (ConvI2D src));
11317
11318 format %{ "movdl $dst, $src\n\t"
11319 "cvtdq2pdl $dst, $dst\t# i2d" %}
11320 ins_encode %{
11321 __ movdl($dst$$XMMRegister, $src$$Register);
11322 __ cvtdq2pd($dst$$XMMRegister, $dst$$XMMRegister);
11323 %}
11324 ins_pipe(pipe_slow); // XXX
11325 %}
11326
11327 instruct convL2F_reg_reg(regF dst, rRegL src)
11328 %{
11329 match(Set dst (ConvL2F src));
11330
11331 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11332 opcode(0xF3, 0x0F, 0x2A);
11333 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11334 ins_pipe(pipe_slow); // XXX
11335 %}
11336
11337 instruct convL2F_reg_mem(regF dst, memory src)
11338 %{
11339 match(Set dst (ConvL2F (LoadL src)));
11340
11341 format %{ "cvtsi2ssq $dst, $src\t# l2f" %}
11342 opcode(0xF3, 0x0F, 0x2A);
11343 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11344 ins_pipe(pipe_slow); // XXX
11345 %}
11346
11347 instruct convL2D_reg_reg(regD dst, rRegL src)
11348 %{
11349 match(Set dst (ConvL2D src));
11350
11351 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11352 opcode(0xF2, 0x0F, 0x2A);
11353 ins_encode(OpcP, REX_reg_reg_wide(dst, src), OpcS, OpcT, reg_reg(dst, src));
11354 ins_pipe(pipe_slow); // XXX
11355 %}
11356
11357 instruct convL2D_reg_mem(regD dst, memory src)
11358 %{
11359 match(Set dst (ConvL2D (LoadL src)));
11360
11361 format %{ "cvtsi2sdq $dst, $src\t# l2d" %}
11362 opcode(0xF2, 0x0F, 0x2A);
11363 ins_encode(OpcP, REX_reg_mem_wide(dst, src), OpcS, OpcT, reg_mem(dst, src));
11364 ins_pipe(pipe_slow); // XXX
11365 %}
11366
11367 instruct convI2L_reg_reg(rRegL dst, rRegI src)
11368 %{
11369 match(Set dst (ConvI2L src));
11370
11371 ins_cost(125);
11372 format %{ "movslq $dst, $src\t# i2l" %}
11373 ins_encode %{
11374 __ movslq($dst$$Register, $src$$Register);
11375 %}
11376 ins_pipe(ialu_reg_reg);
11377 %}
11378
11379 // instruct convI2L_reg_reg_foo(rRegL dst, rRegI src)
11380 // %{
11381 // match(Set dst (ConvI2L src));
11382 // // predicate(_kids[0]->_leaf->as_Type()->type()->is_int()->_lo >= 0 &&
11383 // // _kids[0]->_leaf->as_Type()->type()->is_int()->_hi >= 0);
11384 // predicate(((const TypeNode*) n)->type()->is_long()->_hi ==
11385 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_hi &&
11386 // ((const TypeNode*) n)->type()->is_long()->_lo ==
11387 // (unsigned int) ((const TypeNode*) n)->type()->is_long()->_lo);
11388
11389 // format %{ "movl $dst, $src\t# unsigned i2l" %}
11390 // ins_encode(enc_copy(dst, src));
11391 // // opcode(0x63); // needs REX.W
11392 // // ins_encode(REX_reg_reg_wide(dst, src), OpcP, reg_reg(dst,src));
11393 // ins_pipe(ialu_reg_reg);
11394 // %}
11395
11396 // Zero-extend convert int to long
11397 instruct convI2L_reg_reg_zex(rRegL dst, rRegI src, immL_32bits mask)
11398 %{
11399 match(Set dst (AndL (ConvI2L src) mask));
11400
11401 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
11402 ins_encode(enc_copy(dst, src));
11403 ins_pipe(ialu_reg_reg);
11404 %}
11405
11406 // Zero-extend convert int to long
11407 instruct convI2L_reg_mem_zex(rRegL dst, memory src, immL_32bits mask)
11408 %{
11409 match(Set dst (AndL (ConvI2L (LoadI src)) mask));
11410
11411 format %{ "movl $dst, $src\t# i2l zero-extend\n\t" %}
11412 opcode(0x8B);
11413 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11414 ins_pipe(ialu_reg_mem);
11415 %}
11416
11417 instruct zerox_long_reg_reg(rRegL dst, rRegL src, immL_32bits mask)
11418 %{
11419 match(Set dst (AndL src mask));
11420
11421 format %{ "movl $dst, $src\t# zero-extend long" %}
11422 ins_encode(enc_copy_always(dst, src));
11423 ins_pipe(ialu_reg_reg);
11424 %}
11425
11426 instruct convL2I_reg_reg(rRegI dst, rRegL src)
11427 %{
11428 match(Set dst (ConvL2I src));
11429
11430 format %{ "movl $dst, $src\t# l2i" %}
11431 ins_encode(enc_copy_always(dst, src));
11432 ins_pipe(ialu_reg_reg);
11433 %}
11434
11435
11436 instruct MoveF2I_stack_reg(rRegI dst, stackSlotF src) %{
11437 match(Set dst (MoveF2I src));
11438 effect(DEF dst, USE src);
11439
11440 ins_cost(125);
11441 format %{ "movl $dst, $src\t# MoveF2I_stack_reg" %}
11442 opcode(0x8B);
11443 ins_encode(REX_reg_mem(dst, src), OpcP, reg_mem(dst, src));
11444 ins_pipe(ialu_reg_mem);
11445 %}
11446
11447 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
11448 match(Set dst (MoveI2F src));
11449 effect(DEF dst, USE src);
11450
11451 ins_cost(125);
11452 format %{ "movss $dst, $src\t# MoveI2F_stack_reg" %}
11453 opcode(0xF3, 0x0F, 0x10);
11454 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11455 ins_pipe(pipe_slow);
11456 %}
11457
11458 instruct MoveD2L_stack_reg(rRegL dst, stackSlotD src) %{
11459 match(Set dst (MoveD2L src));
11460 effect(DEF dst, USE src);
11461
11462 ins_cost(125);
11463 format %{ "movq $dst, $src\t# MoveD2L_stack_reg" %}
11464 opcode(0x8B);
11465 ins_encode(REX_reg_mem_wide(dst, src), OpcP, reg_mem(dst, src));
11466 ins_pipe(ialu_reg_mem);
11467 %}
11468
11469 instruct MoveL2D_stack_reg_partial(regD dst, stackSlotL src) %{
11470 predicate(!UseXmmLoadAndClearUpper);
11471 match(Set dst (MoveL2D src));
11472 effect(DEF dst, USE src);
11473
11474 ins_cost(125);
11475 format %{ "movlpd $dst, $src\t# MoveL2D_stack_reg" %}
11476 opcode(0x66, 0x0F, 0x12);
11477 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11478 ins_pipe(pipe_slow);
11479 %}
11480
11481 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
11482 predicate(UseXmmLoadAndClearUpper);
11483 match(Set dst (MoveL2D src));
11484 effect(DEF dst, USE src);
11485
11486 ins_cost(125);
11487 format %{ "movsd $dst, $src\t# MoveL2D_stack_reg" %}
11488 opcode(0xF2, 0x0F, 0x10);
11489 ins_encode(OpcP, REX_reg_mem(dst, src), OpcS, OpcT, reg_mem(dst, src));
11490 ins_pipe(pipe_slow);
11491 %}
11492
11493
11494 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
11495 match(Set dst (MoveF2I src));
11496 effect(DEF dst, USE src);
11497
11498 ins_cost(95); // XXX
11499 format %{ "movss $dst, $src\t# MoveF2I_reg_stack" %}
11500 opcode(0xF3, 0x0F, 0x11);
11501 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11502 ins_pipe(pipe_slow);
11503 %}
11504
11505 instruct MoveI2F_reg_stack(stackSlotF dst, rRegI src) %{
11506 match(Set dst (MoveI2F src));
11507 effect(DEF dst, USE src);
11508
11509 ins_cost(100);
11510 format %{ "movl $dst, $src\t# MoveI2F_reg_stack" %}
11511 opcode(0x89);
11512 ins_encode(REX_reg_mem(src, dst), OpcP, reg_mem(src, dst));
11513 ins_pipe( ialu_mem_reg );
11514 %}
11515
11516 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
11517 match(Set dst (MoveD2L src));
11518 effect(DEF dst, USE src);
11519
11520 ins_cost(95); // XXX
11521 format %{ "movsd $dst, $src\t# MoveL2D_reg_stack" %}
11522 opcode(0xF2, 0x0F, 0x11);
11523 ins_encode(OpcP, REX_reg_mem(src, dst), OpcS, OpcT, reg_mem(src, dst));
11524 ins_pipe(pipe_slow);
11525 %}
11526
11527 instruct MoveL2D_reg_stack(stackSlotD dst, rRegL src) %{
11528 match(Set dst (MoveL2D src));
11529 effect(DEF dst, USE src);
11530
11531 ins_cost(100);
11532 format %{ "movq $dst, $src\t# MoveL2D_reg_stack" %}
11533 opcode(0x89);
11534 ins_encode(REX_reg_mem_wide(src, dst), OpcP, reg_mem(src, dst));
11535 ins_pipe(ialu_mem_reg);
11536 %}
11537
11538 instruct MoveF2I_reg_reg(rRegI dst, regF src) %{
11539 match(Set dst (MoveF2I src));
11540 effect(DEF dst, USE src);
11541 ins_cost(85);
11542 format %{ "movd $dst,$src\t# MoveF2I" %}
11543 ins_encode %{ __ movdl($dst$$Register, $src$$XMMRegister); %}
11544 ins_pipe( pipe_slow );
11545 %}
11546
11547 instruct MoveD2L_reg_reg(rRegL dst, regD src) %{
11548 match(Set dst (MoveD2L src));
11549 effect(DEF dst, USE src);
11550 ins_cost(85);
11551 format %{ "movd $dst,$src\t# MoveD2L" %}
11552 ins_encode %{ __ movdq($dst$$Register, $src$$XMMRegister); %}
11553 ins_pipe( pipe_slow );
11554 %}
11555
11556 // The next instructions have long latency and use Int unit. Set high cost.
11557 instruct MoveI2F_reg_reg(regF dst, rRegI src) %{
11558 match(Set dst (MoveI2F src));
11559 effect(DEF dst, USE src);
11560 ins_cost(300);
11561 format %{ "movd $dst,$src\t# MoveI2F" %}
11562 ins_encode %{ __ movdl($dst$$XMMRegister, $src$$Register); %}
11563 ins_pipe( pipe_slow );
11564 %}
11565
11566 instruct MoveL2D_reg_reg(regD dst, rRegL src) %{
11567 match(Set dst (MoveL2D src));
11568 effect(DEF dst, USE src);
11569 ins_cost(300);
11570 format %{ "movd $dst,$src\t# MoveL2D" %}
11571 ins_encode %{ __ movdq($dst$$XMMRegister, $src$$Register); %}
11572 ins_pipe( pipe_slow );
11573 %}
11574
11575 // Replicate scalar to packed byte (1 byte) values in xmm
11576 instruct Repl8B_reg(regD dst, regD src) %{
11577 match(Set dst (Replicate8B src));
11578 format %{ "MOVDQA $dst,$src\n\t"
11579 "PUNPCKLBW $dst,$dst\n\t"
11580 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11581 ins_encode( pshufd_8x8(dst, src));
11582 ins_pipe( pipe_slow );
11583 %}
11584
11585 // Replicate scalar to packed byte (1 byte) values in xmm
11586 instruct Repl8B_rRegI(regD dst, rRegI src) %{
11587 match(Set dst (Replicate8B src));
11588 format %{ "MOVD $dst,$src\n\t"
11589 "PUNPCKLBW $dst,$dst\n\t"
11590 "PSHUFLW $dst,$dst,0x00\t! replicate8B" %}
11591 ins_encode( mov_i2x(dst, src), pshufd_8x8(dst, dst));
11592 ins_pipe( pipe_slow );
11593 %}
11594
11595 // Replicate scalar zero to packed byte (1 byte) values in xmm
11596 instruct Repl8B_immI0(regD dst, immI0 zero) %{
11597 match(Set dst (Replicate8B zero));
11598 format %{ "PXOR $dst,$dst\t! replicate8B" %}
11599 ins_encode( pxor(dst, dst));
11600 ins_pipe( fpu_reg_reg );
11601 %}
11602
11603 // Replicate scalar to packed shore (2 byte) values in xmm
11604 instruct Repl4S_reg(regD dst, regD src) %{
11605 match(Set dst (Replicate4S src));
11606 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4S" %}
11607 ins_encode( pshufd_4x16(dst, src));
11608 ins_pipe( fpu_reg_reg );
11609 %}
11610
11611 // Replicate scalar to packed shore (2 byte) values in xmm
11612 instruct Repl4S_rRegI(regD dst, rRegI src) %{
11613 match(Set dst (Replicate4S src));
11614 format %{ "MOVD $dst,$src\n\t"
11615 "PSHUFLW $dst,$dst,0x00\t! replicate4S" %}
11616 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11617 ins_pipe( fpu_reg_reg );
11618 %}
11619
11620 // Replicate scalar zero to packed short (2 byte) values in xmm
11621 instruct Repl4S_immI0(regD dst, immI0 zero) %{
11622 match(Set dst (Replicate4S zero));
11623 format %{ "PXOR $dst,$dst\t! replicate4S" %}
11624 ins_encode( pxor(dst, dst));
11625 ins_pipe( fpu_reg_reg );
11626 %}
11627
11628 // Replicate scalar to packed char (2 byte) values in xmm
11629 instruct Repl4C_reg(regD dst, regD src) %{
11630 match(Set dst (Replicate4C src));
11631 format %{ "PSHUFLW $dst,$src,0x00\t! replicate4C" %}
11632 ins_encode( pshufd_4x16(dst, src));
11633 ins_pipe( fpu_reg_reg );
11634 %}
11635
11636 // Replicate scalar to packed char (2 byte) values in xmm
11637 instruct Repl4C_rRegI(regD dst, rRegI src) %{
11638 match(Set dst (Replicate4C src));
11639 format %{ "MOVD $dst,$src\n\t"
11640 "PSHUFLW $dst,$dst,0x00\t! replicate4C" %}
11641 ins_encode( mov_i2x(dst, src), pshufd_4x16(dst, dst));
11642 ins_pipe( fpu_reg_reg );
11643 %}
11644
11645 // Replicate scalar zero to packed char (2 byte) values in xmm
11646 instruct Repl4C_immI0(regD dst, immI0 zero) %{
11647 match(Set dst (Replicate4C zero));
11648 format %{ "PXOR $dst,$dst\t! replicate4C" %}
11649 ins_encode( pxor(dst, dst));
11650 ins_pipe( fpu_reg_reg );
11651 %}
11652
11653 // Replicate scalar to packed integer (4 byte) values in xmm
11654 instruct Repl2I_reg(regD dst, regD src) %{
11655 match(Set dst (Replicate2I src));
11656 format %{ "PSHUFD $dst,$src,0x00\t! replicate2I" %}
11657 ins_encode( pshufd(dst, src, 0x00));
11658 ins_pipe( fpu_reg_reg );
11659 %}
11660
11661 // Replicate scalar to packed integer (4 byte) values in xmm
11662 instruct Repl2I_rRegI(regD dst, rRegI src) %{
11663 match(Set dst (Replicate2I src));
11664 format %{ "MOVD $dst,$src\n\t"
11665 "PSHUFD $dst,$dst,0x00\t! replicate2I" %}
11666 ins_encode( mov_i2x(dst, src), pshufd(dst, dst, 0x00));
11667 ins_pipe( fpu_reg_reg );
11668 %}
11669
11670 // Replicate scalar zero to packed integer (2 byte) values in xmm
11671 instruct Repl2I_immI0(regD dst, immI0 zero) %{
11672 match(Set dst (Replicate2I zero));
11673 format %{ "PXOR $dst,$dst\t! replicate2I" %}
11674 ins_encode( pxor(dst, dst));
11675 ins_pipe( fpu_reg_reg );
11676 %}
11677
11678 // Replicate scalar to packed single precision floating point values in xmm
11679 instruct Repl2F_reg(regD dst, regD src) %{
11680 match(Set dst (Replicate2F src));
11681 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11682 ins_encode( pshufd(dst, src, 0xe0));
11683 ins_pipe( fpu_reg_reg );
11684 %}
11685
11686 // Replicate scalar to packed single precision floating point values in xmm
11687 instruct Repl2F_regF(regD dst, regF src) %{
11688 match(Set dst (Replicate2F src));
11689 format %{ "PSHUFD $dst,$src,0xe0\t! replicate2F" %}
11690 ins_encode( pshufd(dst, src, 0xe0));
11691 ins_pipe( fpu_reg_reg );
11692 %}
11693
11694 // Replicate scalar to packed single precision floating point values in xmm
11695 instruct Repl2F_immF0(regD dst, immF0 zero) %{
11696 match(Set dst (Replicate2F zero));
11697 format %{ "PXOR $dst,$dst\t! replicate2F" %}
11698 ins_encode( pxor(dst, dst));
11699 ins_pipe( fpu_reg_reg );
11700 %}
11701
11702
11703 // =======================================================================
11704 // fast clearing of an array
11705 instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy,
11706 rFlagsReg cr)
11707 %{
11708 match(Set dummy (ClearArray cnt base));
11709 effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
11710
11711 format %{ "xorl rax, rax\t# ClearArray:\n\t"
11712 "rep stosq\t# Store rax to *rdi++ while rcx--" %}
11713 ins_encode(opc_reg_reg(0x33, RAX, RAX), // xorl %eax, %eax
11714 Opcode(0xF3), Opcode(0x48), Opcode(0xAB)); // rep REX_W stos
11715 ins_pipe(pipe_slow);
11716 %}
11717
11718 instruct string_compare(rdi_RegP str1, rcx_RegI cnt1, rsi_RegP str2, rbx_RegI cnt2,
11719 rax_RegI result, regD tmp1, regD tmp2, rFlagsReg cr)
11720 %{
11721 match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
11722 effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL cr);
11723
11724 format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result // KILL $tmp1, $tmp2" %}
11725 ins_encode %{
11726 __ string_compare($str1$$Register, $str2$$Register,
11727 $cnt1$$Register, $cnt2$$Register, $result$$Register,
11728 $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11729 %}
11730 ins_pipe( pipe_slow );
11731 %}
11732
11733 instruct string_indexof(rdi_RegP str1, rdx_RegI cnt1, rsi_RegP str2, rax_RegI cnt2,
11734 rbx_RegI result, regD tmp1, rcx_RegI tmp2, rFlagsReg cr)
11735 %{
11736 predicate(UseSSE42Intrinsics);
11737 match(Set result (StrIndexOf (Binary str1 cnt1) (Binary str2 cnt2)));
11738 effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL tmp2, KILL cr);
11739
11740 format %{ "String IndexOf $str1,$cnt1,$str2,$cnt2 -> $result // KILL $tmp1, $tmp2" %}
11741 ins_encode %{
11742 __ string_indexof($str1$$Register, $str2$$Register,
11743 $cnt1$$Register, $cnt2$$Register, $result$$Register,
11744 $tmp1$$XMMRegister, $tmp2$$Register);
11745 %}
11746 ins_pipe( pipe_slow );
11747 %}
11748
11749 // fast string equals
11750 instruct string_equals(rdi_RegP str1, rsi_RegP str2, rcx_RegI cnt, rax_RegI result,
11751 regD tmp1, regD tmp2, rbx_RegI tmp3, rFlagsReg cr)
11752 %{
11753 match(Set result (StrEquals (Binary str1 str2) cnt));
11754 effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp3, KILL cr);
11755
11756 format %{ "String Equals $str1,$str2,$cnt -> $result // KILL $tmp1, $tmp2, $tmp3" %}
11757 ins_encode %{
11758 __ char_arrays_equals(false, $str1$$Register, $str2$$Register,
11759 $cnt$$Register, $result$$Register, $tmp3$$Register,
11760 $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11761 %}
11762 ins_pipe( pipe_slow );
11763 %}
11764
11765 // fast array equals
11766 instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI result,
11767 regD tmp1, regD tmp2, rcx_RegI tmp3, rbx_RegI tmp4, rFlagsReg cr)
11768 %{
11769 match(Set result (AryEq ary1 ary2));
11770 effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
11771 //ins_cost(300);
11772
11773 format %{ "Array Equals $ary1,$ary2 -> $result // KILL $tmp1, $tmp2, $tmp3, $tmp4" %}
11774 ins_encode %{
11775 __ char_arrays_equals(true, $ary1$$Register, $ary2$$Register,
11776 $tmp3$$Register, $result$$Register, $tmp4$$Register,
11777 $tmp1$$XMMRegister, $tmp2$$XMMRegister);
11778 %}
11779 ins_pipe( pipe_slow );
11780 %}
11781
11782 //----------Control Flow Instructions------------------------------------------
11783 // Signed compare Instructions
11784
11785 // XXX more variants!!
11786 instruct compI_rReg(rFlagsReg cr, rRegI op1, rRegI op2)
11787 %{
11788 match(Set cr (CmpI op1 op2));
11789 effect(DEF cr, USE op1, USE op2);
11790
11791 format %{ "cmpl $op1, $op2" %}
11792 opcode(0x3B); /* Opcode 3B /r */
11793 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11794 ins_pipe(ialu_cr_reg_reg);
11795 %}
11796
11797 instruct compI_rReg_imm(rFlagsReg cr, rRegI op1, immI op2)
11798 %{
11799 match(Set cr (CmpI op1 op2));
11800
11801 format %{ "cmpl $op1, $op2" %}
11802 opcode(0x81, 0x07); /* Opcode 81 /7 */
11803 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11804 ins_pipe(ialu_cr_reg_imm);
11805 %}
11806
11807 instruct compI_rReg_mem(rFlagsReg cr, rRegI op1, memory op2)
11808 %{
11809 match(Set cr (CmpI op1 (LoadI op2)));
11810
11811 ins_cost(500); // XXX
11812 format %{ "cmpl $op1, $op2" %}
11813 opcode(0x3B); /* Opcode 3B /r */
11814 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11815 ins_pipe(ialu_cr_reg_mem);
11816 %}
11817
11818 instruct testI_reg(rFlagsReg cr, rRegI src, immI0 zero)
11819 %{
11820 match(Set cr (CmpI src zero));
11821
11822 format %{ "testl $src, $src" %}
11823 opcode(0x85);
11824 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11825 ins_pipe(ialu_cr_reg_imm);
11826 %}
11827
11828 instruct testI_reg_imm(rFlagsReg cr, rRegI src, immI con, immI0 zero)
11829 %{
11830 match(Set cr (CmpI (AndI src con) zero));
11831
11832 format %{ "testl $src, $con" %}
11833 opcode(0xF7, 0x00);
11834 ins_encode(REX_reg(src), OpcP, reg_opc(src), Con32(con));
11835 ins_pipe(ialu_cr_reg_imm);
11836 %}
11837
11838 instruct testI_reg_mem(rFlagsReg cr, rRegI src, memory mem, immI0 zero)
11839 %{
11840 match(Set cr (CmpI (AndI src (LoadI mem)) zero));
11841
11842 format %{ "testl $src, $mem" %}
11843 opcode(0x85);
11844 ins_encode(REX_reg_mem(src, mem), OpcP, reg_mem(src, mem));
11845 ins_pipe(ialu_cr_reg_mem);
11846 %}
11847
11848 // Unsigned compare Instructions; really, same as signed except they
11849 // produce an rFlagsRegU instead of rFlagsReg.
11850 instruct compU_rReg(rFlagsRegU cr, rRegI op1, rRegI op2)
11851 %{
11852 match(Set cr (CmpU op1 op2));
11853
11854 format %{ "cmpl $op1, $op2\t# unsigned" %}
11855 opcode(0x3B); /* Opcode 3B /r */
11856 ins_encode(REX_reg_reg(op1, op2), OpcP, reg_reg(op1, op2));
11857 ins_pipe(ialu_cr_reg_reg);
11858 %}
11859
11860 instruct compU_rReg_imm(rFlagsRegU cr, rRegI op1, immI op2)
11861 %{
11862 match(Set cr (CmpU op1 op2));
11863
11864 format %{ "cmpl $op1, $op2\t# unsigned" %}
11865 opcode(0x81,0x07); /* Opcode 81 /7 */
11866 ins_encode(OpcSErm(op1, op2), Con8or32(op2));
11867 ins_pipe(ialu_cr_reg_imm);
11868 %}
11869
11870 instruct compU_rReg_mem(rFlagsRegU cr, rRegI op1, memory op2)
11871 %{
11872 match(Set cr (CmpU op1 (LoadI op2)));
11873
11874 ins_cost(500); // XXX
11875 format %{ "cmpl $op1, $op2\t# unsigned" %}
11876 opcode(0x3B); /* Opcode 3B /r */
11877 ins_encode(REX_reg_mem(op1, op2), OpcP, reg_mem(op1, op2));
11878 ins_pipe(ialu_cr_reg_mem);
11879 %}
11880
11881 // // // Cisc-spilled version of cmpU_rReg
11882 // //instruct compU_mem_rReg(rFlagsRegU cr, memory op1, rRegI op2)
11883 // //%{
11884 // // match(Set cr (CmpU (LoadI op1) op2));
11885 // //
11886 // // format %{ "CMPu $op1,$op2" %}
11887 // // ins_cost(500);
11888 // // opcode(0x39); /* Opcode 39 /r */
11889 // // ins_encode( OpcP, reg_mem( op1, op2) );
11890 // //%}
11891
11892 instruct testU_reg(rFlagsRegU cr, rRegI src, immI0 zero)
11893 %{
11894 match(Set cr (CmpU src zero));
11895
11896 format %{ "testl $src, $src\t# unsigned" %}
11897 opcode(0x85);
11898 ins_encode(REX_reg_reg(src, src), OpcP, reg_reg(src, src));
11899 ins_pipe(ialu_cr_reg_imm);
11900 %}
11901
11902 instruct compP_rReg(rFlagsRegU cr, rRegP op1, rRegP op2)
11903 %{
11904 match(Set cr (CmpP op1 op2));
11905
11906 format %{ "cmpq $op1, $op2\t# ptr" %}
11907 opcode(0x3B); /* Opcode 3B /r */
11908 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
11909 ins_pipe(ialu_cr_reg_reg);
11910 %}
11911
11912 instruct compP_rReg_mem(rFlagsRegU cr, rRegP op1, memory op2)
11913 %{
11914 match(Set cr (CmpP op1 (LoadP op2)));
11915
11916 ins_cost(500); // XXX
11917 format %{ "cmpq $op1, $op2\t# ptr" %}
11918 opcode(0x3B); /* Opcode 3B /r */
11919 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11920 ins_pipe(ialu_cr_reg_mem);
11921 %}
11922
11923 // // // Cisc-spilled version of cmpP_rReg
11924 // //instruct compP_mem_rReg(rFlagsRegU cr, memory op1, rRegP op2)
11925 // //%{
11926 // // match(Set cr (CmpP (LoadP op1) op2));
11927 // //
11928 // // format %{ "CMPu $op1,$op2" %}
11929 // // ins_cost(500);
11930 // // opcode(0x39); /* Opcode 39 /r */
11931 // // ins_encode( OpcP, reg_mem( op1, op2) );
11932 // //%}
11933
11934 // XXX this is generalized by compP_rReg_mem???
11935 // Compare raw pointer (used in out-of-heap check).
11936 // Only works because non-oop pointers must be raw pointers
11937 // and raw pointers have no anti-dependencies.
11938 instruct compP_mem_rReg(rFlagsRegU cr, rRegP op1, memory op2)
11939 %{
11940 predicate(!n->in(2)->in(2)->bottom_type()->isa_oop_ptr());
11941 match(Set cr (CmpP op1 (LoadP op2)));
11942
11943 format %{ "cmpq $op1, $op2\t# raw ptr" %}
11944 opcode(0x3B); /* Opcode 3B /r */
11945 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
11946 ins_pipe(ialu_cr_reg_mem);
11947 %}
11948
11949 // This will generate a signed flags result. This should be OK since
11950 // any compare to a zero should be eq/neq.
11951 instruct testP_reg(rFlagsReg cr, rRegP src, immP0 zero)
11952 %{
11953 match(Set cr (CmpP src zero));
11954
11955 format %{ "testq $src, $src\t# ptr" %}
11956 opcode(0x85);
11957 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
11958 ins_pipe(ialu_cr_reg_imm);
11959 %}
11960
11961 // This will generate a signed flags result. This should be OK since
11962 // any compare to a zero should be eq/neq.
11963 instruct testP_mem(rFlagsReg cr, memory op, immP0 zero)
11964 %{
11965 predicate(!UseCompressedOops || (Universe::narrow_oop_base() != NULL));
11966 match(Set cr (CmpP (LoadP op) zero));
11967
11968 ins_cost(500); // XXX
11969 format %{ "testq $op, 0xffffffffffffffff\t# ptr" %}
11970 opcode(0xF7); /* Opcode F7 /0 */
11971 ins_encode(REX_mem_wide(op),
11972 OpcP, RM_opc_mem(0x00, op), Con_d32(0xFFFFFFFF));
11973 ins_pipe(ialu_cr_reg_imm);
11974 %}
11975
11976 instruct testP_mem_reg0(rFlagsReg cr, memory mem, immP0 zero)
11977 %{
11978 predicate(UseCompressedOops && (Universe::narrow_oop_base() == NULL));
11979 match(Set cr (CmpP (LoadP mem) zero));
11980
11981 format %{ "cmpq R12, $mem\t# ptr (R12_heapbase==0)" %}
11982 ins_encode %{
11983 __ cmpq(r12, $mem$$Address);
11984 %}
11985 ins_pipe(ialu_cr_reg_mem);
11986 %}
11987
11988 instruct compN_rReg(rFlagsRegU cr, rRegN op1, rRegN op2)
11989 %{
11990 match(Set cr (CmpN op1 op2));
11991
11992 format %{ "cmpl $op1, $op2\t# compressed ptr" %}
11993 ins_encode %{ __ cmpl($op1$$Register, $op2$$Register); %}
11994 ins_pipe(ialu_cr_reg_reg);
11995 %}
11996
11997 instruct compN_rReg_mem(rFlagsRegU cr, rRegN src, memory mem)
11998 %{
11999 match(Set cr (CmpN src (LoadN mem)));
12000
12001 format %{ "cmpl $src, $mem\t# compressed ptr" %}
12002 ins_encode %{
12003 __ cmpl($src$$Register, $mem$$Address);
12004 %}
12005 ins_pipe(ialu_cr_reg_mem);
12006 %}
12007
12008 instruct compN_rReg_imm(rFlagsRegU cr, rRegN op1, immN op2) %{
12009 match(Set cr (CmpN op1 op2));
12010
12011 format %{ "cmpl $op1, $op2\t# compressed ptr" %}
12012 ins_encode %{
12013 __ cmp_narrow_oop($op1$$Register, (jobject)$op2$$constant);
12014 %}
12015 ins_pipe(ialu_cr_reg_imm);
12016 %}
12017
12018 instruct compN_mem_imm(rFlagsRegU cr, memory mem, immN src)
12019 %{
12020 match(Set cr (CmpN src (LoadN mem)));
12021
12022 format %{ "cmpl $mem, $src\t# compressed ptr" %}
12023 ins_encode %{
12024 __ cmp_narrow_oop($mem$$Address, (jobject)$src$$constant);
12025 %}
12026 ins_pipe(ialu_cr_reg_mem);
12027 %}
12028
12029 instruct testN_reg(rFlagsReg cr, rRegN src, immN0 zero) %{
12030 match(Set cr (CmpN src zero));
12031
12032 format %{ "testl $src, $src\t# compressed ptr" %}
12033 ins_encode %{ __ testl($src$$Register, $src$$Register); %}
12034 ins_pipe(ialu_cr_reg_imm);
12035 %}
12036
12037 instruct testN_mem(rFlagsReg cr, memory mem, immN0 zero)
12038 %{
12039 predicate(Universe::narrow_oop_base() != NULL);
12040 match(Set cr (CmpN (LoadN mem) zero));
12041
12042 ins_cost(500); // XXX
12043 format %{ "testl $mem, 0xffffffff\t# compressed ptr" %}
12044 ins_encode %{
12045 __ cmpl($mem$$Address, (int)0xFFFFFFFF);
12046 %}
12047 ins_pipe(ialu_cr_reg_mem);
12048 %}
12049
12050 instruct testN_mem_reg0(rFlagsReg cr, memory mem, immN0 zero)
12051 %{
12052 predicate(Universe::narrow_oop_base() == NULL);
12053 match(Set cr (CmpN (LoadN mem) zero));
12054
12055 format %{ "cmpl R12, $mem\t# compressed ptr (R12_heapbase==0)" %}
12056 ins_encode %{
12057 __ cmpl(r12, $mem$$Address);
12058 %}
12059 ins_pipe(ialu_cr_reg_mem);
12060 %}
12061
12062 // Yanked all unsigned pointer compare operations.
12063 // Pointer compares are done with CmpP which is already unsigned.
12064
12065 instruct compL_rReg(rFlagsReg cr, rRegL op1, rRegL op2)
12066 %{
12067 match(Set cr (CmpL op1 op2));
12068
12069 format %{ "cmpq $op1, $op2" %}
12070 opcode(0x3B); /* Opcode 3B /r */
12071 ins_encode(REX_reg_reg_wide(op1, op2), OpcP, reg_reg(op1, op2));
12072 ins_pipe(ialu_cr_reg_reg);
12073 %}
12074
12075 instruct compL_rReg_imm(rFlagsReg cr, rRegL op1, immL32 op2)
12076 %{
12077 match(Set cr (CmpL op1 op2));
12078
12079 format %{ "cmpq $op1, $op2" %}
12080 opcode(0x81, 0x07); /* Opcode 81 /7 */
12081 ins_encode(OpcSErm_wide(op1, op2), Con8or32(op2));
12082 ins_pipe(ialu_cr_reg_imm);
12083 %}
12084
12085 instruct compL_rReg_mem(rFlagsReg cr, rRegL op1, memory op2)
12086 %{
12087 match(Set cr (CmpL op1 (LoadL op2)));
12088
12089 format %{ "cmpq $op1, $op2" %}
12090 opcode(0x3B); /* Opcode 3B /r */
12091 ins_encode(REX_reg_mem_wide(op1, op2), OpcP, reg_mem(op1, op2));
12092 ins_pipe(ialu_cr_reg_mem);
12093 %}
12094
12095 instruct testL_reg(rFlagsReg cr, rRegL src, immL0 zero)
12096 %{
12097 match(Set cr (CmpL src zero));
12098
12099 format %{ "testq $src, $src" %}
12100 opcode(0x85);
12101 ins_encode(REX_reg_reg_wide(src, src), OpcP, reg_reg(src, src));
12102 ins_pipe(ialu_cr_reg_imm);
12103 %}
12104
12105 instruct testL_reg_imm(rFlagsReg cr, rRegL src, immL32 con, immL0 zero)
12106 %{
12107 match(Set cr (CmpL (AndL src con) zero));
12108
12109 format %{ "testq $src, $con\t# long" %}
12110 opcode(0xF7, 0x00);
12111 ins_encode(REX_reg_wide(src), OpcP, reg_opc(src), Con32(con));
12112 ins_pipe(ialu_cr_reg_imm);
12113 %}
12114
12115 instruct testL_reg_mem(rFlagsReg cr, rRegL src, memory mem, immL0 zero)
12116 %{
12117 match(Set cr (CmpL (AndL src (LoadL mem)) zero));
12118
12119 format %{ "testq $src, $mem" %}
12120 opcode(0x85);
12121 ins_encode(REX_reg_mem_wide(src, mem), OpcP, reg_mem(src, mem));
12122 ins_pipe(ialu_cr_reg_mem);
12123 %}
12124
12125 // Manifest a CmpL result in an integer register. Very painful.
12126 // This is the test to avoid.
12127 instruct cmpL3_reg_reg(rRegI dst, rRegL src1, rRegL src2, rFlagsReg flags)
12128 %{
12129 match(Set dst (CmpL3 src1 src2));
12130 effect(KILL flags);
12131
12132 ins_cost(275); // XXX
12133 format %{ "cmpq $src1, $src2\t# CmpL3\n\t"
12134 "movl $dst, -1\n\t"
12135 "jl,s done\n\t"
12136 "setne $dst\n\t"
12137 "movzbl $dst, $dst\n\t"
12138 "done:" %}
12139 ins_encode(cmpl3_flag(src1, src2, dst));
12140 ins_pipe(pipe_slow);
12141 %}
12142
12143 //----------Max and Min--------------------------------------------------------
12144 // Min Instructions
12145
12146 instruct cmovI_reg_g(rRegI dst, rRegI src, rFlagsReg cr)
12147 %{
12148 effect(USE_DEF dst, USE src, USE cr);
12149
12150 format %{ "cmovlgt $dst, $src\t# min" %}
12151 opcode(0x0F, 0x4F);
12152 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
12153 ins_pipe(pipe_cmov_reg);
12154 %}
12155
12156
12157 instruct minI_rReg(rRegI dst, rRegI src)
12158 %{
12159 match(Set dst (MinI dst src));
12160
12161 ins_cost(200);
12162 expand %{
12163 rFlagsReg cr;
12164 compI_rReg(cr, dst, src);
12165 cmovI_reg_g(dst, src, cr);
12166 %}
12167 %}
12168
12169 instruct cmovI_reg_l(rRegI dst, rRegI src, rFlagsReg cr)
12170 %{
12171 effect(USE_DEF dst, USE src, USE cr);
12172
12173 format %{ "cmovllt $dst, $src\t# max" %}
12174 opcode(0x0F, 0x4C);
12175 ins_encode(REX_reg_reg(dst, src), OpcP, OpcS, reg_reg(dst, src));
12176 ins_pipe(pipe_cmov_reg);
12177 %}
12178
12179
12180 instruct maxI_rReg(rRegI dst, rRegI src)
12181 %{
12182 match(Set dst (MaxI dst src));
12183
12184 ins_cost(200);
12185 expand %{
12186 rFlagsReg cr;
12187 compI_rReg(cr, dst, src);
12188 cmovI_reg_l(dst, src, cr);
12189 %}
12190 %}
12191
12192 // ============================================================================
12193 // Branch Instructions
12194
12195 // Jump Direct - Label defines a relative address from JMP+1
12196 instruct jmpDir(label labl)
12197 %{
12198 match(Goto);
12199 effect(USE labl);
12200
12201 ins_cost(300);
12202 format %{ "jmp $labl" %}
12203 size(5);
12204 opcode(0xE9);
12205 ins_encode(OpcP, Lbl(labl));
12206 ins_pipe(pipe_jmp);
12207 ins_pc_relative(1);
12208 %}
12209
12210 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12211 instruct jmpCon(cmpOp cop, rFlagsReg cr, label labl)
12212 %{
12213 match(If cop cr);
12214 effect(USE labl);
12215
12216 ins_cost(300);
12217 format %{ "j$cop $labl" %}
12218 size(6);
12219 opcode(0x0F, 0x80);
12220 ins_encode(Jcc(cop, labl));
12221 ins_pipe(pipe_jcc);
12222 ins_pc_relative(1);
12223 %}
12224
12225 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12226 instruct jmpLoopEnd(cmpOp cop, rFlagsReg cr, label labl)
12227 %{
12228 match(CountedLoopEnd cop cr);
12229 effect(USE labl);
12230
12231 ins_cost(300);
12232 format %{ "j$cop $labl\t# loop end" %}
12233 size(6);
12234 opcode(0x0F, 0x80);
12235 ins_encode(Jcc(cop, labl));
12236 ins_pipe(pipe_jcc);
12237 ins_pc_relative(1);
12238 %}
12239
12240 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12241 instruct jmpLoopEndU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12242 match(CountedLoopEnd cop cmp);
12243 effect(USE labl);
12244
12245 ins_cost(300);
12246 format %{ "j$cop,u $labl\t# loop end" %}
12247 size(6);
12248 opcode(0x0F, 0x80);
12249 ins_encode(Jcc(cop, labl));
12250 ins_pipe(pipe_jcc);
12251 ins_pc_relative(1);
12252 %}
12253
12254 instruct jmpLoopEndUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12255 match(CountedLoopEnd cop cmp);
12256 effect(USE labl);
12257
12258 ins_cost(200);
12259 format %{ "j$cop,u $labl\t# loop end" %}
12260 size(6);
12261 opcode(0x0F, 0x80);
12262 ins_encode(Jcc(cop, labl));
12263 ins_pipe(pipe_jcc);
12264 ins_pc_relative(1);
12265 %}
12266
12267 // Jump Direct Conditional - using unsigned comparison
12268 instruct jmpConU(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12269 match(If cop cmp);
12270 effect(USE labl);
12271
12272 ins_cost(300);
12273 format %{ "j$cop,u $labl" %}
12274 size(6);
12275 opcode(0x0F, 0x80);
12276 ins_encode(Jcc(cop, labl));
12277 ins_pipe(pipe_jcc);
12278 ins_pc_relative(1);
12279 %}
12280
12281 instruct jmpConUCF(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12282 match(If cop cmp);
12283 effect(USE labl);
12284
12285 ins_cost(200);
12286 format %{ "j$cop,u $labl" %}
12287 size(6);
12288 opcode(0x0F, 0x80);
12289 ins_encode(Jcc(cop, labl));
12290 ins_pipe(pipe_jcc);
12291 ins_pc_relative(1);
12292 %}
12293
12294 instruct jmpConUCF2(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
12295 match(If cop cmp);
12296 effect(USE labl);
12297
12298 ins_cost(200);
12299 format %{ $$template
12300 if ($cop$$cmpcode == Assembler::notEqual) {
12301 $$emit$$"jp,u $labl\n\t"
12302 $$emit$$"j$cop,u $labl"
12303 } else {
12304 $$emit$$"jp,u done\n\t"
12305 $$emit$$"j$cop,u $labl\n\t"
12306 $$emit$$"done:"
12307 }
12308 %}
12309 size(12);
12310 opcode(0x0F, 0x80);
12311 ins_encode %{
12312 Label* l = $labl$$label;
12313 $$$emit8$primary;
12314 emit_cc(cbuf, $secondary, Assembler::parity);
12315 int parity_disp = -1;
12316 if ($cop$$cmpcode == Assembler::notEqual) {
12317 // the two jumps 6 bytes apart so the jump distances are too
12318 parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
12319 } else if ($cop$$cmpcode == Assembler::equal) {
12320 parity_disp = 6;
12321 } else {
12322 ShouldNotReachHere();
12323 }
12324 emit_d32(cbuf, parity_disp);
12325 $$$emit8$primary;
12326 emit_cc(cbuf, $secondary, $cop$$cmpcode);
12327 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 4)) : 0;
12328 emit_d32(cbuf, disp);
12329 %}
12330 ins_pipe(pipe_jcc);
12331 ins_pc_relative(1);
12332 %}
12333
12334 // ============================================================================
12335 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary
12336 // superklass array for an instance of the superklass. Set a hidden
12337 // internal cache on a hit (cache is checked with exposed code in
12338 // gen_subtype_check()). Return NZ for a miss or zero for a hit. The
12339 // encoding ALSO sets flags.
12340
12341 instruct partialSubtypeCheck(rdi_RegP result,
12342 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
12343 rFlagsReg cr)
12344 %{
12345 match(Set result (PartialSubtypeCheck sub super));
12346 effect(KILL rcx, KILL cr);
12347
12348 ins_cost(1100); // slightly larger than the next version
12349 format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12350 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12351 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12352 "repne scasq\t# Scan *rdi++ for a match with rax while rcx--\n\t"
12353 "jne,s miss\t\t# Missed: rdi not-zero\n\t"
12354 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12355 "xorq $result, $result\t\t Hit: rdi zero\n\t"
12356 "miss:\t" %}
12357
12358 opcode(0x1); // Force a XOR of RDI
12359 ins_encode(enc_PartialSubtypeCheck());
12360 ins_pipe(pipe_slow);
12361 %}
12362
12363 instruct partialSubtypeCheck_vs_Zero(rFlagsReg cr,
12364 rsi_RegP sub, rax_RegP super, rcx_RegI rcx,
12365 immP0 zero,
12366 rdi_RegP result)
12367 %{
12368 match(Set cr (CmpP (PartialSubtypeCheck sub super) zero));
12369 effect(KILL rcx, KILL result);
12370
12371 ins_cost(1000);
12372 format %{ "movq rdi, [$sub + (sizeof(oopDesc) + Klass::secondary_supers_offset_in_bytes())]\n\t"
12373 "movl rcx, [rdi + arrayOopDesc::length_offset_in_bytes()]\t# length to scan\n\t"
12374 "addq rdi, arrayOopDex::base_offset_in_bytes(T_OBJECT)\t# Skip to start of data; set NZ in case count is zero\n\t"
12375 "repne scasq\t# Scan *rdi++ for a match with rax while cx-- != 0\n\t"
12376 "jne,s miss\t\t# Missed: flags nz\n\t"
12377 "movq [$sub + (sizeof(oopDesc) + Klass::secondary_super_cache_offset_in_bytes())], $super\t# Hit: update cache\n\t"
12378 "miss:\t" %}
12379
12380 opcode(0x0); // No need to XOR RDI
12381 ins_encode(enc_PartialSubtypeCheck());
12382 ins_pipe(pipe_slow);
12383 %}
12384
12385 // ============================================================================
12386 // Branch Instructions -- short offset versions
12387 //
12388 // These instructions are used to replace jumps of a long offset (the default
12389 // match) with jumps of a shorter offset. These instructions are all tagged
12390 // with the ins_short_branch attribute, which causes the ADLC to suppress the
12391 // match rules in general matching. Instead, the ADLC generates a conversion
12392 // method in the MachNode which can be used to do in-place replacement of the
12393 // long variant with the shorter variant. The compiler will determine if a
12394 // branch can be taken by the is_short_branch_offset() predicate in the machine
12395 // specific code section of the file.
12396
12397 // Jump Direct - Label defines a relative address from JMP+1
12398 instruct jmpDir_short(label labl) %{
12399 match(Goto);
12400 effect(USE labl);
12401
12402 ins_cost(300);
12403 format %{ "jmp,s $labl" %}
12404 size(2);
12405 opcode(0xEB);
12406 ins_encode(OpcP, LblShort(labl));
12407 ins_pipe(pipe_jmp);
12408 ins_pc_relative(1);
12409 ins_short_branch(1);
12410 %}
12411
12412 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12413 instruct jmpCon_short(cmpOp cop, rFlagsReg cr, label labl) %{
12414 match(If cop cr);
12415 effect(USE labl);
12416
12417 ins_cost(300);
12418 format %{ "j$cop,s $labl" %}
12419 size(2);
12420 opcode(0x70);
12421 ins_encode(JccShort(cop, labl));
12422 ins_pipe(pipe_jcc);
12423 ins_pc_relative(1);
12424 ins_short_branch(1);
12425 %}
12426
12427 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12428 instruct jmpLoopEnd_short(cmpOp cop, rFlagsReg cr, label labl) %{
12429 match(CountedLoopEnd cop cr);
12430 effect(USE labl);
12431
12432 ins_cost(300);
12433 format %{ "j$cop,s $labl\t# loop end" %}
12434 size(2);
12435 opcode(0x70);
12436 ins_encode(JccShort(cop, labl));
12437 ins_pipe(pipe_jcc);
12438 ins_pc_relative(1);
12439 ins_short_branch(1);
12440 %}
12441
12442 // Jump Direct Conditional - Label defines a relative address from Jcc+1
12443 instruct jmpLoopEndU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12444 match(CountedLoopEnd cop cmp);
12445 effect(USE labl);
12446
12447 ins_cost(300);
12448 format %{ "j$cop,us $labl\t# loop end" %}
12449 size(2);
12450 opcode(0x70);
12451 ins_encode(JccShort(cop, labl));
12452 ins_pipe(pipe_jcc);
12453 ins_pc_relative(1);
12454 ins_short_branch(1);
12455 %}
12456
12457 instruct jmpLoopEndUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12458 match(CountedLoopEnd cop cmp);
12459 effect(USE labl);
12460
12461 ins_cost(300);
12462 format %{ "j$cop,us $labl\t# loop end" %}
12463 size(2);
12464 opcode(0x70);
12465 ins_encode(JccShort(cop, labl));
12466 ins_pipe(pipe_jcc);
12467 ins_pc_relative(1);
12468 ins_short_branch(1);
12469 %}
12470
12471 // Jump Direct Conditional - using unsigned comparison
12472 instruct jmpConU_short(cmpOpU cop, rFlagsRegU cmp, label labl) %{
12473 match(If cop cmp);
12474 effect(USE labl);
12475
12476 ins_cost(300);
12477 format %{ "j$cop,us $labl" %}
12478 size(2);
12479 opcode(0x70);
12480 ins_encode(JccShort(cop, labl));
12481 ins_pipe(pipe_jcc);
12482 ins_pc_relative(1);
12483 ins_short_branch(1);
12484 %}
12485
12486 instruct jmpConUCF_short(cmpOpUCF cop, rFlagsRegUCF cmp, label labl) %{
12487 match(If cop cmp);
12488 effect(USE labl);
12489
12490 ins_cost(300);
12491 format %{ "j$cop,us $labl" %}
12492 size(2);
12493 opcode(0x70);
12494 ins_encode(JccShort(cop, labl));
12495 ins_pipe(pipe_jcc);
12496 ins_pc_relative(1);
12497 ins_short_branch(1);
12498 %}
12499
12500 instruct jmpConUCF2_short(cmpOpUCF2 cop, rFlagsRegUCF cmp, label labl) %{
12501 match(If cop cmp);
12502 effect(USE labl);
12503
12504 ins_cost(300);
12505 format %{ $$template
12506 if ($cop$$cmpcode == Assembler::notEqual) {
12507 $$emit$$"jp,u,s $labl\n\t"
12508 $$emit$$"j$cop,u,s $labl"
12509 } else {
12510 $$emit$$"jp,u,s done\n\t"
12511 $$emit$$"j$cop,u,s $labl\n\t"
12512 $$emit$$"done:"
12513 }
12514 %}
12515 size(4);
12516 opcode(0x70);
12517 ins_encode %{
12518 Label* l = $labl$$label;
12519 emit_cc(cbuf, $primary, Assembler::parity);
12520 int parity_disp = -1;
12521 if ($cop$$cmpcode == Assembler::notEqual) {
12522 parity_disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
12523 } else if ($cop$$cmpcode == Assembler::equal) {
12524 parity_disp = 2;
12525 } else {
12526 ShouldNotReachHere();
12527 }
12528 emit_d8(cbuf, parity_disp);
12529 emit_cc(cbuf, $primary, $cop$$cmpcode);
12530 int disp = l ? (l->loc_pos() - (cbuf.code_size() + 1)) : 0;
12531 emit_d8(cbuf, disp);
12532 assert(-128 <= disp && disp <= 127, "Displacement too large for short jmp");
12533 assert(-128 <= parity_disp && parity_disp <= 127, "Displacement too large for short jmp");
12534 %}
12535 ins_pipe(pipe_jcc);
12536 ins_pc_relative(1);
12537 ins_short_branch(1);
12538 %}
12539
12540 // ============================================================================
12541 // inlined locking and unlocking
12542
12543 instruct cmpFastLock(rFlagsReg cr,
12544 rRegP object, rRegP box, rax_RegI tmp, rRegP scr)
12545 %{
12546 match(Set cr (FastLock object box));
12547 effect(TEMP tmp, TEMP scr);
12548
12549 ins_cost(300);
12550 format %{ "fastlock $object,$box,$tmp,$scr" %}
12551 ins_encode(Fast_Lock(object, box, tmp, scr));
12552 ins_pipe(pipe_slow);
12553 ins_pc_relative(1);
12554 %}
12555
12556 instruct cmpFastUnlock(rFlagsReg cr,
12557 rRegP object, rax_RegP box, rRegP tmp)
12558 %{
12559 match(Set cr (FastUnlock object box));
12560 effect(TEMP tmp);
12561
12562 ins_cost(300);
12563 format %{ "fastunlock $object, $box, $tmp" %}
12564 ins_encode(Fast_Unlock(object, box, tmp));
12565 ins_pipe(pipe_slow);
12566 ins_pc_relative(1);
12567 %}
12568
12569
12570 // ============================================================================
12571 // Safepoint Instructions
12572 instruct safePoint_poll(rFlagsReg cr)
12573 %{
12574 match(SafePoint);
12575 effect(KILL cr);
12576
12577 format %{ "testl rax, [rip + #offset_to_poll_page]\t"
12578 "# Safepoint: poll for GC" %}
12579 size(6); // Opcode + ModRM + Disp32 == 6 bytes
12580 ins_cost(125);
12581 ins_encode(enc_safepoint_poll);
12582 ins_pipe(ialu_reg_mem);
12583 %}
12584
12585 // ============================================================================
12586 // Procedure Call/Return Instructions
12587 // Call Java Static Instruction
12588 // Note: If this code changes, the corresponding ret_addr_offset() and
12589 // compute_padding() functions will have to be adjusted.
12590 instruct CallStaticJavaDirect(method meth) %{
12591 match(CallStaticJava);
12592 predicate(!((CallStaticJavaNode*) n)->is_method_handle_invoke());
12593 effect(USE meth);
12594
12595 ins_cost(300);
12596 format %{ "call,static " %}
12597 opcode(0xE8); /* E8 cd */
12598 ins_encode(Java_Static_Call(meth), call_epilog);
12599 ins_pipe(pipe_slow);
12600 ins_pc_relative(1);
12601 ins_alignment(4);
12602 %}
12603
12604 // Call Java Static Instruction (method handle version)
12605 // Note: If this code changes, the corresponding ret_addr_offset() and
12606 // compute_padding() functions will have to be adjusted.
12607 instruct CallStaticJavaHandle(method meth, rbp_RegP rbp) %{
12608 match(CallStaticJava);
12609 predicate(((CallStaticJavaNode*) n)->is_method_handle_invoke());
12610 effect(USE meth);
12611 // RBP is saved by all callees (for interpreter stack correction).
12612 // We use it here for a similar purpose, in {preserve,restore}_SP.
12613
12614 ins_cost(300);
12615 format %{ "call,static/MethodHandle " %}
12616 opcode(0xE8); /* E8 cd */
12617 ins_encode(preserve_SP,
12618 Java_Static_Call(meth),
12619 restore_SP,
12620 call_epilog);
12621 ins_pipe(pipe_slow);
12622 ins_pc_relative(1);
12623 ins_alignment(4);
12624 %}
12625
12626 // Call Java Dynamic Instruction
12627 // Note: If this code changes, the corresponding ret_addr_offset() and
12628 // compute_padding() functions will have to be adjusted.
12629 instruct CallDynamicJavaDirect(method meth)
12630 %{
12631 match(CallDynamicJava);
12632 effect(USE meth);
12633
12634 ins_cost(300);
12635 format %{ "movq rax, #Universe::non_oop_word()\n\t"
12636 "call,dynamic " %}
12637 opcode(0xE8); /* E8 cd */
12638 ins_encode(Java_Dynamic_Call(meth), call_epilog);
12639 ins_pipe(pipe_slow);
12640 ins_pc_relative(1);
12641 ins_alignment(4);
12642 %}
12643
12644 // Call Runtime Instruction
12645 instruct CallRuntimeDirect(method meth)
12646 %{
12647 match(CallRuntime);
12648 effect(USE meth);
12649
12650 ins_cost(300);
12651 format %{ "call,runtime " %}
12652 opcode(0xE8); /* E8 cd */
12653 ins_encode(Java_To_Runtime(meth));
12654 ins_pipe(pipe_slow);
12655 ins_pc_relative(1);
12656 %}
12657
12658 // Call runtime without safepoint
12659 instruct CallLeafDirect(method meth)
12660 %{
12661 match(CallLeaf);
12662 effect(USE meth);
12663
12664 ins_cost(300);
12665 format %{ "call_leaf,runtime " %}
12666 opcode(0xE8); /* E8 cd */
12667 ins_encode(Java_To_Runtime(meth));
12668 ins_pipe(pipe_slow);
12669 ins_pc_relative(1);
12670 %}
12671
12672 // Call runtime without safepoint
12673 instruct CallLeafNoFPDirect(method meth)
12674 %{
12675 match(CallLeafNoFP);
12676 effect(USE meth);
12677
12678 ins_cost(300);
12679 format %{ "call_leaf_nofp,runtime " %}
12680 opcode(0xE8); /* E8 cd */
12681 ins_encode(Java_To_Runtime(meth));
12682 ins_pipe(pipe_slow);
12683 ins_pc_relative(1);
12684 %}
12685
12686 // Return Instruction
12687 // Remove the return address & jump to it.
12688 // Notice: We always emit a nop after a ret to make sure there is room
12689 // for safepoint patching
12690 instruct Ret()
12691 %{
12692 match(Return);
12693
12694 format %{ "ret" %}
12695 opcode(0xC3);
12696 ins_encode(OpcP);
12697 ins_pipe(pipe_jmp);
12698 %}
12699
12700 // Tail Call; Jump from runtime stub to Java code.
12701 // Also known as an 'interprocedural jump'.
12702 // Target of jump will eventually return to caller.
12703 // TailJump below removes the return address.
12704 instruct TailCalljmpInd(no_rbp_RegP jump_target, rbx_RegP method_oop)
12705 %{
12706 match(TailCall jump_target method_oop);
12707
12708 ins_cost(300);
12709 format %{ "jmp $jump_target\t# rbx holds method oop" %}
12710 opcode(0xFF, 0x4); /* Opcode FF /4 */
12711 ins_encode(REX_reg(jump_target), OpcP, reg_opc(jump_target));
12712 ins_pipe(pipe_jmp);
12713 %}
12714
12715 // Tail Jump; remove the return address; jump to target.
12716 // TailCall above leaves the return address around.
12717 instruct tailjmpInd(no_rbp_RegP jump_target, rax_RegP ex_oop)
12718 %{
12719 match(TailJump jump_target ex_oop);
12720
12721 ins_cost(300);
12722 format %{ "popq rdx\t# pop return address\n\t"
12723 "jmp $jump_target" %}
12724 opcode(0xFF, 0x4); /* Opcode FF /4 */
12725 ins_encode(Opcode(0x5a), // popq rdx
12726 REX_reg(jump_target), OpcP, reg_opc(jump_target));
12727 ins_pipe(pipe_jmp);
12728 %}
12729
12730 // Create exception oop: created by stack-crawling runtime code.
12731 // Created exception is now available to this handler, and is setup
12732 // just prior to jumping to this handler. No code emitted.
12733 instruct CreateException(rax_RegP ex_oop)
12734 %{
12735 match(Set ex_oop (CreateEx));
12736
12737 size(0);
12738 // use the following format syntax
12739 format %{ "# exception oop is in rax; no code emitted" %}
12740 ins_encode();
12741 ins_pipe(empty);
12742 %}
12743
12744 // Rethrow exception:
12745 // The exception oop will come in the first argument position.
12746 // Then JUMP (not call) to the rethrow stub code.
12747 instruct RethrowException()
12748 %{
12749 match(Rethrow);
12750
12751 // use the following format syntax
12752 format %{ "jmp rethrow_stub" %}
12753 ins_encode(enc_rethrow);
12754 ins_pipe(pipe_jmp);
12755 %}
12756
12757
12758 //----------PEEPHOLE RULES-----------------------------------------------------
12759 // These must follow all instruction definitions as they use the names
12760 // defined in the instructions definitions.
12761 //
12762 // peepmatch ( root_instr_name [preceding_instruction]* );
12763 //
12764 // peepconstraint %{
12765 // (instruction_number.operand_name relational_op instruction_number.operand_name
12766 // [, ...] );
12767 // // instruction numbers are zero-based using left to right order in peepmatch
12768 //
12769 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
12770 // // provide an instruction_number.operand_name for each operand that appears
12771 // // in the replacement instruction's match rule
12772 //
12773 // ---------VM FLAGS---------------------------------------------------------
12774 //
12775 // All peephole optimizations can be turned off using -XX:-OptoPeephole
12776 //
12777 // Each peephole rule is given an identifying number starting with zero and
12778 // increasing by one in the order seen by the parser. An individual peephole
12779 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
12780 // on the command-line.
12781 //
12782 // ---------CURRENT LIMITATIONS----------------------------------------------
12783 //
12784 // Only match adjacent instructions in same basic block
12785 // Only equality constraints
12786 // Only constraints between operands, not (0.dest_reg == RAX_enc)
12787 // Only one replacement instruction
12788 //
12789 // ---------EXAMPLE----------------------------------------------------------
12790 //
12791 // // pertinent parts of existing instructions in architecture description
12792 // instruct movI(rRegI dst, rRegI src)
12793 // %{
12794 // match(Set dst (CopyI src));
12795 // %}
12796 //
12797 // instruct incI_rReg(rRegI dst, immI1 src, rFlagsReg cr)
12798 // %{
12799 // match(Set dst (AddI dst src));
12800 // effect(KILL cr);
12801 // %}
12802 //
12803 // // Change (inc mov) to lea
12804 // peephole %{
12805 // // increment preceeded by register-register move
12806 // peepmatch ( incI_rReg movI );
12807 // // require that the destination register of the increment
12808 // // match the destination register of the move
12809 // peepconstraint ( 0.dst == 1.dst );
12810 // // construct a replacement instruction that sets
12811 // // the destination to ( move's source register + one )
12812 // peepreplace ( leaI_rReg_immI( 0.dst 1.src 0.src ) );
12813 // %}
12814 //
12815
12816 // Implementation no longer uses movX instructions since
12817 // machine-independent system no longer uses CopyX nodes.
12818 //
12819 // peephole
12820 // %{
12821 // peepmatch (incI_rReg movI);
12822 // peepconstraint (0.dst == 1.dst);
12823 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12824 // %}
12825
12826 // peephole
12827 // %{
12828 // peepmatch (decI_rReg movI);
12829 // peepconstraint (0.dst == 1.dst);
12830 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12831 // %}
12832
12833 // peephole
12834 // %{
12835 // peepmatch (addI_rReg_imm movI);
12836 // peepconstraint (0.dst == 1.dst);
12837 // peepreplace (leaI_rReg_immI(0.dst 1.src 0.src));
12838 // %}
12839
12840 // peephole
12841 // %{
12842 // peepmatch (incL_rReg movL);
12843 // peepconstraint (0.dst == 1.dst);
12844 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12845 // %}
12846
12847 // peephole
12848 // %{
12849 // peepmatch (decL_rReg movL);
12850 // peepconstraint (0.dst == 1.dst);
12851 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12852 // %}
12853
12854 // peephole
12855 // %{
12856 // peepmatch (addL_rReg_imm movL);
12857 // peepconstraint (0.dst == 1.dst);
12858 // peepreplace (leaL_rReg_immL(0.dst 1.src 0.src));
12859 // %}
12860
12861 // peephole
12862 // %{
12863 // peepmatch (addP_rReg_imm movP);
12864 // peepconstraint (0.dst == 1.dst);
12865 // peepreplace (leaP_rReg_imm(0.dst 1.src 0.src));
12866 // %}
12867
12868 // // Change load of spilled value to only a spill
12869 // instruct storeI(memory mem, rRegI src)
12870 // %{
12871 // match(Set mem (StoreI mem src));
12872 // %}
12873 //
12874 // instruct loadI(rRegI dst, memory mem)
12875 // %{
12876 // match(Set dst (LoadI mem));
12877 // %}
12878 //
12879
12880 peephole
12881 %{
12882 peepmatch (loadI storeI);
12883 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12884 peepreplace (storeI(1.mem 1.mem 1.src));
12885 %}
12886
12887 peephole
12888 %{
12889 peepmatch (loadL storeL);
12890 peepconstraint (1.src == 0.dst, 1.mem == 0.mem);
12891 peepreplace (storeL(1.mem 1.mem 1.src));
12892 %}
12893
12894 //----------SMARTSPILL RULES---------------------------------------------------
12895 // These must follow all instruction definitions as they use the names
12896 // defined in the instructions definitions.